diff --git a/cmd/bridge/main.go b/cmd/bridge/main.go new file mode 100644 index 0000000..5832043 --- /dev/null +++ b/cmd/bridge/main.go @@ -0,0 +1,65 @@ +package main + +import ( + "time" + "fmt" + "github.com/AFASystems/presence/internal/pkg/config" + "github.com/yosssi/gmq/mqtt" + "github.com/yosssi/gmq/mqtt/client" + "github.com/AFASystems/presence/internal/pkg/bridge/mqtthandler" + "github.com/segmentio/kafka-go" +) + +func main() { + cfg := config.Load() + + cli := client.New(&client.Options{ + ErrorHandler: func(err error) { + fmt.Println("Error in initiating MQTT client: ", err) + }, + }) + defer cli.Terminate() + + err := cli.Connect(&client.ConnectOptions{ + Network: "tcp", + Address: cfg.MQTTHost, + ClientID: []byte(cfg.MQTTClientID), + UserName: []byte(cfg.MQTTUser), + Password: []byte(cfg.MQTTPass), + }) + + if err != nil { + fmt.Println("Error comes from here") + panic(err) + } + + writer := kafkaWriter("127.0.0.1:9092", "rawbeacons") + defer writer.Close() + + err = cli.Subscribe(&client.SubscribeOptions{ + SubReqs: []*client.SubReq{ + &client.SubReq{ + TopicFilter: []byte("publish_out/#"), + QoS: mqtt.QoS0, + Handler: func(topicName, message[]byte) { + mqtthandler.MqttHandler(writer, topicName, message) + }, + }, + }, + }) + if err != nil { + panic(err) + } + + select {} +} + +func kafkaWriter(kafkaURL, topic string) *kafka.Writer { + return &kafka.Writer{ + Addr: kafka.TCP(kafkaURL), + Topic: topic, + Balancer: &kafka.LeastBytes{}, + BatchSize: 100, + BatchTimeout: 10 * time.Millisecond, + } +} \ No newline at end of file diff --git a/cmd/decoder/main.go b/cmd/decoder/main.go new file mode 100644 index 0000000..a3f6a93 --- /dev/null +++ b/cmd/decoder/main.go @@ -0,0 +1,162 @@ +package main + +import ( + "context" + "encoding/json" + "fmt" + "strings" + "time" + + "github.com/AFASystems/presence/internal/pkg/model" + "github.com/AFASystems/presence/internal/pkg/mqttclient" + "github.com/redis/go-redis/v9" + "github.com/segmentio/kafka-go" +) + +func main() { + // Load global context to init beacons and latest list + appCtx := model.AppContext{ + Beacons: model.BeaconsList{ + Beacons: make(map[string]model.Beacon), + }, + LatestList: model.LatestBeaconsList{ + LatestList: make(map[string]model.Beacon), + }, + } + + // Kafka writer idk why yet + writer := kafkaWriter("127.0.0.1:9092", "beacons") + + // Kafka reader for Raw MQTT beacons + rawReader := kafkaReader("127.0.0.1:9092", "rawbeacons", "someID") + defer rawReader.Close() + + // Kafka reader for API server updates + apiReader := kafkaReader("127.0.0.1:9092", "apibeacons", "someID") + defer apiReader.Close() + + // Kafka reader for latest list updates + latestReader := kafkaReader("127.0.0.1:9092", "latestbeacons", "someID") + defer latestReader.Close() + + defer writer.Close() + + ctx := context.Background() + + // Init Redis Client + client := redis.NewClient(&redis.Options{ + Addr: "127.0.0.1:6379", + Password: "", + }) + + // Initialize list values from Redis + beaconsList, err := client.Get(ctx, "beaconsList").Result() + if err == redis.Nil { + fmt.Println("no beacons list, starting empty") + } else if err != nil { + panic(err) + } else { + json.Unmarshal([]byte(beaconsList), &appCtx.Beacons.Beacons) + } + + // Initialize list values from Redis + latestList, err := client.Get(ctx, "latestList").Result() + if err == redis.Nil { + fmt.Println("no latest list, starting empty") + } else if err != nil { + panic(err) + } else { + json.Unmarshal([]byte(latestList), &appCtx.LatestList.LatestList) + } + + // declare channel for collecting Kafka messages + chRaw := make(chan model.Incoming_json, 2000) + chApi := make(chan model.Incoming_json, 2000) + chLatest := make(chan model.Incoming_json, 2000) + + go consume(rawReader, chRaw) + go consume(apiReader, chApi) + go consume(latestReader, chLatest) + + for { + select { + case msg := <-chRaw: + processIncoming(msg, &appCtx) + case msg := <-chApi: + fmt.Println("api msg: ", msg) + case msg := <-chLatest: + fmt.Println("latest msg: ", msg) + } + } +} + +func kafkaWriter(kafkaURL, topic string) *kafka.Writer { + return &kafka.Writer{ + Addr: kafka.TCP(kafkaURL), + Topic: topic, + Balancer: &kafka.LeastBytes{}, + BatchSize: 100, + BatchTimeout: 10 * time.Millisecond, + } +} + +func kafkaReader(kafkaURL, topic, groupID string) *kafka.Reader { + brokers := strings.Split(kafkaURL, ",") + return kafka.NewReader(kafka.ReaderConfig{ + Brokers: brokers, + GroupID: groupID, + Topic: topic, + MinBytes: 1, + MaxBytes: 10e6, + }) +} + +func consume(r *kafka.Reader, ch chan<- model.Incoming_json) { + for { + msg, err := r.ReadMessage(context.Background()) + if err != nil { + fmt.Println("error reading message:", err) + continue + } + + var incoming model.Incoming_json + + if err := json.Unmarshal(msg.Value, &incoming); err != nil { + fmt.Println("error in decoding string: ", err) + continue + } + + ch <- incoming + } +} + +func processIncoming(incoming model.Incoming_json, ctx *model.AppContext) { + defer func() { + if err := recover(); err != nil { + fmt.Println("work failed:", err) + } + }() + + incoming = mqttclient.IncomingBeaconFilter(incoming) + id := mqttclient.GetBeaconID(incoming) + now := time.Now().Unix() + + beacons := &ctx.Beacons + + beacons.Lock.Lock() + defer beacons.Lock.Unlock() + + latestList := &ctx.LatestList + + latestList.Lock.Lock() + defer latestList.Lock.Unlock() + + beacon, exists := beacons.Beacons[id] + if !exists { + fmt.Println("beacon does not yet exist") + fmt.Println("time now: ", now) + return + } + + fmt.Println("Beacon does exist: ", beacon) +} diff --git a/cmd/presenSe/presense.go b/cmd/presenSe/presense.go index f3f0e7c..d660152 100644 --- a/cmd/presenSe/presense.go +++ b/cmd/presenSe/presense.go @@ -13,7 +13,7 @@ import ( "github.com/AFASystems/presence/internal/pkg/config" "github.com/AFASystems/presence/internal/pkg/httpserver" "github.com/AFASystems/presence/internal/pkg/model" - "github.com/AFASystems/presence/internal/pkg/mqtt_client" + "github.com/AFASystems/presence/internal/pkg/mqttclient" "github.com/AFASystems/presence/internal/pkg/persistence" "github.com/boltdb/bolt" "github.com/gorilla/websocket" @@ -26,6 +26,8 @@ func main() { signal.Notify(sigc, os.Interrupt) cfg := config.Load() + fmt.Println("hello world") + db, err := bolt.Open("presence.db", 0644, nil) if err != nil { log.Fatal(err) @@ -79,7 +81,7 @@ func main() { } persistence.LoadState(model.Db, ctx) - incomingChan := mqtt_client.IncomingMQTTProcessor(1*time.Second, cli, model.Db, ctx) + incomingChan := mqttclient.IncomingMQTTProcessor(1*time.Second, cli, model.Db, ctx) err = cli.Subscribe(&client.SubscribeOptions{ SubReqs: []*client.SubReq{ @@ -96,7 +98,7 @@ func main() { var readings []model.RawReading err := json.Unmarshal(message, &readings) if err != nil { - log.Printf("Errore parsing JSON: %v", err) + log.Printf("Error parsing JSON: %v", err) return } diff --git a/go.mod b/go.mod index 7b4a49e..33f9aed 100644 --- a/go.mod +++ b/go.mod @@ -9,10 +9,16 @@ require ( github.com/gorilla/handlers v1.5.2 github.com/gorilla/mux v1.8.1 github.com/gorilla/websocket v1.5.3 + github.com/redis/go-redis/v9 v9.16.0 + github.com/segmentio/kafka-go v0.4.49 github.com/yosssi/gmq v0.0.1 ) require ( + github.com/cespare/xxhash/v2 v2.3.0 // indirect + github.com/dgryski/go-rendezvous v0.0.0-20200823014737-9f7001d12a5f // indirect github.com/felixge/httpsnoop v1.0.3 // indirect + github.com/klauspost/compress v1.15.9 // indirect + github.com/pierrec/lz4/v4 v4.1.15 // indirect golang.org/x/sys v0.37.0 // indirect ) diff --git a/go.sum b/go.sum index b844c7b..b51eb9b 100644 --- a/go.sum +++ b/go.sum @@ -1,5 +1,15 @@ github.com/boltdb/bolt v1.3.1 h1:JQmyP4ZBrce+ZQu0dY660FMfatumYDLun9hBCUVIkF4= github.com/boltdb/bolt v1.3.1/go.mod h1:clJnj/oiGkjum5o1McbSZDSLxVThjynRyGBgiAx27Ps= +github.com/bsm/ginkgo/v2 v2.12.0 h1:Ny8MWAHyOepLGlLKYmXG4IEkioBysk6GpaRTLC8zwWs= +github.com/bsm/ginkgo/v2 v2.12.0/go.mod h1:SwYbGRRDovPVboqFv0tPTcG1sN61LM1Z4ARdbAV9g4c= +github.com/bsm/gomega v1.27.10 h1:yeMWxP2pV2fG3FgAODIY8EiRE3dy0aeFYt4l7wh6yKA= +github.com/bsm/gomega v1.27.10/go.mod h1:JyEr/xRbxbtgWNi8tIEVPUYZ5Dzef52k01W3YH0H+O0= +github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UFvs= +github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs= +github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c= +github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38= +github.com/dgryski/go-rendezvous v0.0.0-20200823014737-9f7001d12a5f h1:lO4WD4F/rVNCu3HqELle0jiPLLBs70cWOduZpkS1E78= +github.com/dgryski/go-rendezvous v0.0.0-20200823014737-9f7001d12a5f/go.mod h1:cuUVRXasLTGF7a8hSLbxyZXjz+1KgoB3wDUb6vlszIc= github.com/felixge/httpsnoop v1.0.3 h1:s/nj+GCswXYzN5v2DpNMuMQYe+0DDwt5WVCU6CWBdXk= github.com/felixge/httpsnoop v1.0.3/go.mod h1:m8KPJKqk1gH5J9DgRY2ASl2lWCfGKXixSwevea8zH2U= github.com/gorilla/handlers v1.5.2 h1:cLTUSsNkgcwhgRqvCNmdbRWG0A3N4F+M2nWKdScwyEE= @@ -8,7 +18,31 @@ github.com/gorilla/mux v1.8.1 h1:TuBL49tXwgrFYWhqrNgrUNEY92u81SPhu7sTdzQEiWY= github.com/gorilla/mux v1.8.1/go.mod h1:AKf9I4AEqPTmMytcMc0KkNouC66V3BtZ4qD5fmWSiMQ= github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg= github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE= +github.com/klauspost/compress v1.15.9 h1:wKRjX6JRtDdrE9qwa4b/Cip7ACOshUI4smpCQanqjSY= +github.com/klauspost/compress v1.15.9/go.mod h1:PhcZ0MbTNciWF3rruxRgKxI5NkcHHrHUDtV4Yw2GlzU= +github.com/pierrec/lz4/v4 v4.1.15 h1:MO0/ucJhngq7299dKLwIMtgTfbkoSPF6AoMYDd8Q4q0= +github.com/pierrec/lz4/v4 v4.1.15/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4= +github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM= +github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4= +github.com/redis/go-redis/v9 v9.16.0 h1:OotgqgLSRCmzfqChbQyG1PHC3tLNR89DG4jdOERSEP4= +github.com/redis/go-redis/v9 v9.16.0/go.mod h1:u410H11HMLoB+TP67dz8rL9s6QW2j76l0//kSOd3370= +github.com/segmentio/kafka-go v0.4.49 h1:GJiNX1d/g+kG6ljyJEoi9++PUMdXGAxb7JGPiDCuNmk= +github.com/segmentio/kafka-go v0.4.49/go.mod h1:Y1gn60kzLEEaW28YshXyk2+VCUKbJ3Qr6DrnT3i4+9E= +github.com/stretchr/testify v1.8.0 h1:pSgiaMZlXftHpm5L7V1+rVB+AZJydKsMxsQBIJw4PKk= +github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU= +github.com/xdg-go/pbkdf2 v1.0.0 h1:Su7DPu48wXMwC3bs7MCNG+z4FhcyEuz5dlvchbq0B0c= +github.com/xdg-go/pbkdf2 v1.0.0/go.mod h1:jrpuAogTd400dnrH08LKmI/xc1MbPOebTwRqcT5RDeI= +github.com/xdg-go/scram v1.1.2 h1:FHX5I5B4i4hKRVRBCFRxq1iQRej7WO3hhBuJf+UUySY= +github.com/xdg-go/scram v1.1.2/go.mod h1:RT/sEzTbU5y00aCK8UOx6R7YryM0iF1N2MOmC3kKLN4= +github.com/xdg-go/stringprep v1.0.4 h1:XLI/Ng3O1Atzq0oBs3TWm+5ZVgkq2aqdlvP9JtoZ6c8= +github.com/xdg-go/stringprep v1.0.4/go.mod h1:mPGuuIYwz7CmR2bT9j4GbQqutWS1zV24gijq1dTyGkM= github.com/yosssi/gmq v0.0.1 h1:GhlDVaAQoi3Mvjul/qJXXGfL4JBeE0GQwbWp3eIsja8= github.com/yosssi/gmq v0.0.1/go.mod h1:mReykazh0U1JabvuWh1PEbzzJftqOQWsjr0Lwg5jL1Y= +golang.org/x/net v0.38.0 h1:vRMAPTMaeGqVhG5QyLJHqNDwecKTomGeqbnfZyKlBI8= +golang.org/x/net v0.38.0/go.mod h1:ivrbrMbzFq5J41QOQh0siUuly180yBYtLp+CKbEaFx8= golang.org/x/sys v0.37.0 h1:fdNQudmxPjkdUTPnLn5mdQv7Zwvbvpaxqs831goi9kQ= golang.org/x/sys v0.37.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks= +golang.org/x/text v0.23.0 h1:D71I7dUrlY+VX0gQShAThNGHFxZ13dGLBHQLVl1mJlY= +golang.org/x/text v0.23.0/go.mod h1:/BLNzu4aZCJ1+kcD0DNRotWKage4q2rGVAg4o22unh4= +gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA= +gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM= diff --git a/internal/pkg/bridge/mqtthandler/mqtthandler.go b/internal/pkg/bridge/mqtthandler/mqtthandler.go new file mode 100644 index 0000000..8d9dfa7 --- /dev/null +++ b/internal/pkg/bridge/mqtthandler/mqtthandler.go @@ -0,0 +1,105 @@ +package mqtthandler + +import ( + "fmt" + "encoding/json" + "strings" + "log" + "strconv" + "os" + "context" + "time" + + "github.com/AFASystems/presence/internal/pkg/model" + "github.com/segmentio/kafka-go" +) + +func MqttHandler(writer *kafka.Writer, topicName []byte, message []byte) { + hostname := strings.Split(string(topicName), "/")[1] + msgStr := string(message) + + if strings.HasPrefix(msgStr, "[") { + var readings []model.RawReading + err := json.Unmarshal(message, &readings) + if err != nil { + log.Printf("Error parsing JSON: %v", err) + return + } + + for _, reading := range readings { + if reading.Type == "Gateway" { + continue + } + incoming := model.Incoming_json{ + Hostname: hostname, + MAC: reading.MAC, + RSSI: int64(reading.RSSI), + Data: reading.RawData, + HB_ButtonCounter: parseButtonState(reading.RawData), + } + + encodedMsg, err := json.Marshal(incoming) + if err != nil { + fmt.Println("Error in marshaling: ", err) + } + + msg := kafka.Message{ + Value: encodedMsg, + } + err = writer.WriteMessages(context.Background(), msg) + if err != nil { + fmt.Println("Error in writing to Kafka: ", err) + } + + fmt.Println("message sent: ", time.Now()) + } + } else { + s := strings.Split(string(message), ",") + if len(s) < 6 { + log.Printf("Messaggio CSV non valido: %s", msgStr) + return + } + + rawdata := s[4] + buttonCounter := parseButtonState(rawdata) + if buttonCounter > 0 { + incoming := model.Incoming_json{} + i, _ := strconv.ParseInt(s[3], 10, 64) + incoming.Hostname = hostname + incoming.Beacon_type = "hb_button" + incoming.MAC = s[1] + incoming.RSSI = i + incoming.Data = rawdata + incoming.HB_ButtonCounter = buttonCounter + + read_line := strings.TrimRight(string(s[5]), "\r\n") + it, err33 := strconv.Atoi(read_line) + if err33 != nil { + fmt.Println(it) + fmt.Println(err33) + os.Exit(2) + } + } + } +} + +func parseButtonState(raw string) int64 { + raw = strings.ToUpper(raw) + + if strings.HasPrefix(raw, "0201060303E1FF12") && len(raw) >= 38 { + buttonField := raw[34:38] + if buttonValue, err := strconv.ParseInt(buttonField, 16, 64); err == nil { + return buttonValue + } + } + + if strings.HasPrefix(raw, "02010612FF590") && len(raw) >= 24 { + counterField := raw[22:24] + buttonState, err := strconv.ParseInt(counterField, 16, 64) + if err == nil { + return buttonState + } + } + + return 0 +} \ No newline at end of file diff --git a/internal/pkg/config/config.go b/internal/pkg/config/config.go index e5daa87..b289104 100644 --- a/internal/pkg/config/config.go +++ b/internal/pkg/config/config.go @@ -22,11 +22,11 @@ func getEnv(key, def string) string { func Load() *Config { return &Config{ - HTTPAddr: getEnv("HTTP_HOST_PATH", "0.0.0.0:1902"), - WSAddr: getEnv("HTTPWS_HOST_PATH", "0.0.0.0:1922"), - MQTTHost: getEnv("MQTT_HOST", "localhost:1883"), - MQTTUser: getEnv("MQTT_USERNAME", "chesnek"), - MQTTPass: getEnv("MQTT_PASSWORD", "chesnek"), + HTTPAddr: getEnv("HTTP_HOST_PATH", "0.0.0.0:8080"), + WSAddr: getEnv("HTTPWS_HOST_PATH", "0.0.0.0:8088"), + MQTTHost: getEnv("MQTT_HOST", "127.0.0.1:11883"), + MQTTUser: getEnv("MQTT_USERNAME", "user"), + MQTTPass: getEnv("MQTT_PASSWORD", "sandbox2024"), MQTTClientID: getEnv("MQTT_CLIENT_ID", "presence-detector"), DBPath: getEnv("DB_PATH", "/data/conf/presence/presence.db"), } diff --git a/internal/pkg/httpserver/server.go b/internal/pkg/httpserver/server.go index f6f9f12..39b5df1 100644 --- a/internal/pkg/httpserver/server.go +++ b/internal/pkg/httpserver/server.go @@ -162,12 +162,12 @@ func BeaconsDeleteHandler(beacons *model.BeaconsList, buttonsList map[string]mod func latestBeaconsListHandler(latestList *model.LatestBeaconsList) http.HandlerFunc { return func(w http.ResponseWriter, r *http.Request) { - latestList.LatestListLock.RLock() + latestList.Lock.RLock() var la = make([]model.Beacon, 0) for _, b := range latestList.LatestList { la = append(la, b) } - latestList.LatestListLock.RUnlock() + latestList.Lock.RUnlock() js, err := json.Marshal(la) if err != nil { http.Error(w, err.Error(), http.StatusInternalServerError) @@ -327,7 +327,7 @@ func serveLatestBeaconsWs(latestList *model.LatestBeaconsList) http.HandlerFunc return } - go latestBeaconWriter(ws, latestList.LatestList, &latestList.LatestListLock) + go latestBeaconWriter(ws, latestList.LatestList, &latestList.Lock) reader(ws) } } diff --git a/internal/pkg/model/types.go b/internal/pkg/model/types.go index 9e6fd6a..604fc95 100644 --- a/internal/pkg/model/types.go +++ b/internal/pkg/model/types.go @@ -173,8 +173,8 @@ type RawReading struct { } type LatestBeaconsList struct { - LatestList map[string]Beacon - LatestListLock sync.RWMutex + LatestList map[string]Beacon + Lock sync.RWMutex } type HTTPResultsList struct { diff --git a/internal/pkg/mqtt_client/beacon.go b/internal/pkg/mqttclient/beacon.go similarity index 94% rename from internal/pkg/mqtt_client/beacon.go rename to internal/pkg/mqttclient/beacon.go index 23424ef..e43084a 100644 --- a/internal/pkg/mqtt_client/beacon.go +++ b/internal/pkg/mqttclient/beacon.go @@ -1,4 +1,4 @@ -package mqtt_client +package mqttclient import ( "bytes" @@ -14,17 +14,17 @@ import ( "github.com/yosssi/gmq/mqtt/client" ) -func getBeaconID(incoming model.Incoming_json) string { +func GetBeaconID(incoming model.Incoming_json) string { unique_id := fmt.Sprintf("%s", incoming.MAC) return unique_id } func updateLatestList(incoming model.Incoming_json, now int64, latestList *model.LatestBeaconsList) { - latestList.LatestListLock.Lock() - defer latestList.LatestListLock.Unlock() + latestList.Lock.Lock() + defer latestList.Lock.Unlock() b := model.Beacon{ - Beacon_id: getBeaconID(incoming), + Beacon_id: GetBeaconID(incoming), Beacon_type: incoming.Beacon_type, Last_seen: now, Incoming_JSON: incoming, diff --git a/internal/pkg/mqtt_client/fillter.go b/internal/pkg/mqttclient/fillter.go similarity index 90% rename from internal/pkg/mqtt_client/fillter.go rename to internal/pkg/mqttclient/fillter.go index 455b8be..f6d3039 100644 --- a/internal/pkg/mqtt_client/fillter.go +++ b/internal/pkg/mqttclient/fillter.go @@ -1,4 +1,4 @@ -package mqtt_client +package mqttclient import ( "fmt" @@ -8,7 +8,7 @@ import ( "github.com/AFASystems/presence/internal/pkg/model" ) -func incomingBeaconFilter(incoming model.Incoming_json) model.Incoming_json { +func IncomingBeaconFilter(incoming model.Incoming_json) model.Incoming_json { out_json := incoming if incoming.Beacon_type == "hb_button" { raw_data := incoming.Data diff --git a/internal/pkg/mqtt_client/location.go b/internal/pkg/mqttclient/location.go similarity index 99% rename from internal/pkg/mqtt_client/location.go rename to internal/pkg/mqttclient/location.go index d5e854c..535c927 100644 --- a/internal/pkg/mqtt_client/location.go +++ b/internal/pkg/mqttclient/location.go @@ -1,4 +1,4 @@ -package mqtt_client +package mqttclient import ( "encoding/json" diff --git a/internal/pkg/mqtt_client/processor.go b/internal/pkg/mqttclient/processor.go similarity index 84% rename from internal/pkg/mqtt_client/processor.go rename to internal/pkg/mqttclient/processor.go index 83ee83b..41d311f 100644 --- a/internal/pkg/mqtt_client/processor.go +++ b/internal/pkg/mqttclient/processor.go @@ -1,4 +1,4 @@ -package mqtt_client +package mqttclient import ( "fmt" @@ -27,20 +27,20 @@ func runProcessor(ticker *time.Ticker, cl *client.Client, ch <-chan model.Incomi case <-ticker.C: getLikelyLocations(&ctx.Settings, ctx, cl) case incoming := <-ch: - processIncoming(incoming, cl, ctx) + ProcessIncoming(incoming, cl, ctx) } } } -func processIncoming(incoming model.Incoming_json, cl *client.Client, ctx *model.AppContext) { +func ProcessIncoming(incoming model.Incoming_json, cl *client.Client, ctx *model.AppContext) { defer func() { if err := recover(); err != nil { log.Println("work failed:", err) } }() - incoming = incomingBeaconFilter(incoming) - id := getBeaconID(incoming) + incoming = IncomingBeaconFilter(incoming) + id := GetBeaconID(incoming) now := time.Now().Unix() beacons := &ctx.Beacons @@ -53,7 +53,6 @@ func processIncoming(incoming model.Incoming_json, cl *client.Client, ctx *model beacon, ok := beacons.Beacons[id] if !ok { - fmt.Println("updating latest list") updateLatestList(incoming, now, latestList) return } diff --git a/presence.db b/presence.db new file mode 100644 index 0000000..e73b14e Binary files /dev/null and b/presence.db differ diff --git a/presense.container b/presense.container new file mode 100644 index 0000000..7deb716 --- /dev/null +++ b/presense.container @@ -0,0 +1,28 @@ +[Unit] +Description=Presense +PartOf=podman.service +Wants=network-online.target podman-conf-login.service +After=podman.service network-online.target podman-conf-login.service +StartLimitIntervalSec=0 + +[Container] +Image=presense-go:latest +ContainerName=presense +PodmanArgs=-a stdout -a stderr +Network=sandbox.network +PublishPort=127.0.0.1:1902:8080 +Environment=HTTP_HOST_PATH=0.0.0.0:8080 +Environment=HTTPWS_HOST_PATH=0.0.0.0:8088 +Environment=MQTT_HOST=emqx:1883 +Environment=MQTT_USERNAME=sandbox +Environment=MQTT_PASSWORD=sandbox2025 +Environment=MQTT_CLIENT_ID=presence-detector +Environment=DB_PATH=.presence.db + +[Service] +Restart=always +TimeoutStartSec=infinity +RestartSec=5 + +[Install] +WantedBy=multi-user.target podman.service \ No newline at end of file diff --git a/test/mqtt_test/mqtt_test.go b/test/mqtt_test/mqtt_test.go index 851a111..4603e09 100644 --- a/test/mqtt_test/mqtt_test.go +++ b/test/mqtt_test/mqtt_test.go @@ -6,7 +6,7 @@ import ( "time" "github.com/AFASystems/presence/internal/pkg/model" - "github.com/AFASystems/presence/internal/pkg/mqtt_client" + "github.com/AFASystems/presence/internal/pkg/mqttclient" "github.com/AFASystems/presence/internal/pkg/persistence" "github.com/boltdb/bolt" ) @@ -31,7 +31,7 @@ func TestIncomingMQTTProcessor(t *testing.T) { persistence.LoadState(model.Db, ctx) - ch := mqtt_client.IncomingMQTTProcessor(20*time.Millisecond, nil, model.Db, ctx) + ch := mqttclient.IncomingMQTTProcessor(20*time.Millisecond, nil, model.Db, ctx) msg := model.Incoming_json{MAC: "15:02:31", Hostname: "testHost", RSSI: -55} ch <- msg diff --git a/vendor/github.com/cespare/xxhash/v2/LICENSE.txt b/vendor/github.com/cespare/xxhash/v2/LICENSE.txt new file mode 100644 index 0000000..24b5306 --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/LICENSE.txt @@ -0,0 +1,22 @@ +Copyright (c) 2016 Caleb Spare + +MIT License + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +"Software"), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice shall be +included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND +NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE +LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION +OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/vendor/github.com/cespare/xxhash/v2/README.md b/vendor/github.com/cespare/xxhash/v2/README.md new file mode 100644 index 0000000..33c8830 --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/README.md @@ -0,0 +1,74 @@ +# xxhash + +[![Go Reference](https://pkg.go.dev/badge/github.com/cespare/xxhash/v2.svg)](https://pkg.go.dev/github.com/cespare/xxhash/v2) +[![Test](https://github.com/cespare/xxhash/actions/workflows/test.yml/badge.svg)](https://github.com/cespare/xxhash/actions/workflows/test.yml) + +xxhash is a Go implementation of the 64-bit [xxHash] algorithm, XXH64. This is a +high-quality hashing algorithm that is much faster than anything in the Go +standard library. + +This package provides a straightforward API: + +``` +func Sum64(b []byte) uint64 +func Sum64String(s string) uint64 +type Digest struct{ ... } + func New() *Digest +``` + +The `Digest` type implements hash.Hash64. Its key methods are: + +``` +func (*Digest) Write([]byte) (int, error) +func (*Digest) WriteString(string) (int, error) +func (*Digest) Sum64() uint64 +``` + +The package is written with optimized pure Go and also contains even faster +assembly implementations for amd64 and arm64. If desired, the `purego` build tag +opts into using the Go code even on those architectures. + +[xxHash]: http://cyan4973.github.io/xxHash/ + +## Compatibility + +This package is in a module and the latest code is in version 2 of the module. +You need a version of Go with at least "minimal module compatibility" to use +github.com/cespare/xxhash/v2: + +* 1.9.7+ for Go 1.9 +* 1.10.3+ for Go 1.10 +* Go 1.11 or later + +I recommend using the latest release of Go. + +## Benchmarks + +Here are some quick benchmarks comparing the pure-Go and assembly +implementations of Sum64. + +| input size | purego | asm | +| ---------- | --------- | --------- | +| 4 B | 1.3 GB/s | 1.2 GB/s | +| 16 B | 2.9 GB/s | 3.5 GB/s | +| 100 B | 6.9 GB/s | 8.1 GB/s | +| 4 KB | 11.7 GB/s | 16.7 GB/s | +| 10 MB | 12.0 GB/s | 17.3 GB/s | + +These numbers were generated on Ubuntu 20.04 with an Intel Xeon Platinum 8252C +CPU using the following commands under Go 1.19.2: + +``` +benchstat <(go test -tags purego -benchtime 500ms -count 15 -bench 'Sum64$') +benchstat <(go test -benchtime 500ms -count 15 -bench 'Sum64$') +``` + +## Projects using this package + +- [InfluxDB](https://github.com/influxdata/influxdb) +- [Prometheus](https://github.com/prometheus/prometheus) +- [VictoriaMetrics](https://github.com/VictoriaMetrics/VictoriaMetrics) +- [FreeCache](https://github.com/coocood/freecache) +- [FastCache](https://github.com/VictoriaMetrics/fastcache) +- [Ristretto](https://github.com/dgraph-io/ristretto) +- [Badger](https://github.com/dgraph-io/badger) diff --git a/vendor/github.com/cespare/xxhash/v2/testall.sh b/vendor/github.com/cespare/xxhash/v2/testall.sh new file mode 100644 index 0000000..94b9c44 --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/testall.sh @@ -0,0 +1,10 @@ +#!/bin/bash +set -eu -o pipefail + +# Small convenience script for running the tests with various combinations of +# arch/tags. This assumes we're running on amd64 and have qemu available. + +go test ./... +go test -tags purego ./... +GOARCH=arm64 go test +GOARCH=arm64 go test -tags purego diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash.go b/vendor/github.com/cespare/xxhash/v2/xxhash.go new file mode 100644 index 0000000..78bddf1 --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash.go @@ -0,0 +1,243 @@ +// Package xxhash implements the 64-bit variant of xxHash (XXH64) as described +// at http://cyan4973.github.io/xxHash/. +package xxhash + +import ( + "encoding/binary" + "errors" + "math/bits" +) + +const ( + prime1 uint64 = 11400714785074694791 + prime2 uint64 = 14029467366897019727 + prime3 uint64 = 1609587929392839161 + prime4 uint64 = 9650029242287828579 + prime5 uint64 = 2870177450012600261 +) + +// Store the primes in an array as well. +// +// The consts are used when possible in Go code to avoid MOVs but we need a +// contiguous array for the assembly code. +var primes = [...]uint64{prime1, prime2, prime3, prime4, prime5} + +// Digest implements hash.Hash64. +// +// Note that a zero-valued Digest is not ready to receive writes. +// Call Reset or create a Digest using New before calling other methods. +type Digest struct { + v1 uint64 + v2 uint64 + v3 uint64 + v4 uint64 + total uint64 + mem [32]byte + n int // how much of mem is used +} + +// New creates a new Digest with a zero seed. +func New() *Digest { + return NewWithSeed(0) +} + +// NewWithSeed creates a new Digest with the given seed. +func NewWithSeed(seed uint64) *Digest { + var d Digest + d.ResetWithSeed(seed) + return &d +} + +// Reset clears the Digest's state so that it can be reused. +// It uses a seed value of zero. +func (d *Digest) Reset() { + d.ResetWithSeed(0) +} + +// ResetWithSeed clears the Digest's state so that it can be reused. +// It uses the given seed to initialize the state. +func (d *Digest) ResetWithSeed(seed uint64) { + d.v1 = seed + prime1 + prime2 + d.v2 = seed + prime2 + d.v3 = seed + d.v4 = seed - prime1 + d.total = 0 + d.n = 0 +} + +// Size always returns 8 bytes. +func (d *Digest) Size() int { return 8 } + +// BlockSize always returns 32 bytes. +func (d *Digest) BlockSize() int { return 32 } + +// Write adds more data to d. It always returns len(b), nil. +func (d *Digest) Write(b []byte) (n int, err error) { + n = len(b) + d.total += uint64(n) + + memleft := d.mem[d.n&(len(d.mem)-1):] + + if d.n+n < 32 { + // This new data doesn't even fill the current block. + copy(memleft, b) + d.n += n + return + } + + if d.n > 0 { + // Finish off the partial block. + c := copy(memleft, b) + d.v1 = round(d.v1, u64(d.mem[0:8])) + d.v2 = round(d.v2, u64(d.mem[8:16])) + d.v3 = round(d.v3, u64(d.mem[16:24])) + d.v4 = round(d.v4, u64(d.mem[24:32])) + b = b[c:] + d.n = 0 + } + + if len(b) >= 32 { + // One or more full blocks left. + nw := writeBlocks(d, b) + b = b[nw:] + } + + // Store any remaining partial block. + copy(d.mem[:], b) + d.n = len(b) + + return +} + +// Sum appends the current hash to b and returns the resulting slice. +func (d *Digest) Sum(b []byte) []byte { + s := d.Sum64() + return append( + b, + byte(s>>56), + byte(s>>48), + byte(s>>40), + byte(s>>32), + byte(s>>24), + byte(s>>16), + byte(s>>8), + byte(s), + ) +} + +// Sum64 returns the current hash. +func (d *Digest) Sum64() uint64 { + var h uint64 + + if d.total >= 32 { + v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4 + h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + h = mergeRound(h, v1) + h = mergeRound(h, v2) + h = mergeRound(h, v3) + h = mergeRound(h, v4) + } else { + h = d.v3 + prime5 + } + + h += d.total + + b := d.mem[:d.n&(len(d.mem)-1)] + for ; len(b) >= 8; b = b[8:] { + k1 := round(0, u64(b[:8])) + h ^= k1 + h = rol27(h)*prime1 + prime4 + } + if len(b) >= 4 { + h ^= uint64(u32(b[:4])) * prime1 + h = rol23(h)*prime2 + prime3 + b = b[4:] + } + for ; len(b) > 0; b = b[1:] { + h ^= uint64(b[0]) * prime5 + h = rol11(h) * prime1 + } + + h ^= h >> 33 + h *= prime2 + h ^= h >> 29 + h *= prime3 + h ^= h >> 32 + + return h +} + +const ( + magic = "xxh\x06" + marshaledSize = len(magic) + 8*5 + 32 +) + +// MarshalBinary implements the encoding.BinaryMarshaler interface. +func (d *Digest) MarshalBinary() ([]byte, error) { + b := make([]byte, 0, marshaledSize) + b = append(b, magic...) + b = appendUint64(b, d.v1) + b = appendUint64(b, d.v2) + b = appendUint64(b, d.v3) + b = appendUint64(b, d.v4) + b = appendUint64(b, d.total) + b = append(b, d.mem[:d.n]...) + b = b[:len(b)+len(d.mem)-d.n] + return b, nil +} + +// UnmarshalBinary implements the encoding.BinaryUnmarshaler interface. +func (d *Digest) UnmarshalBinary(b []byte) error { + if len(b) < len(magic) || string(b[:len(magic)]) != magic { + return errors.New("xxhash: invalid hash state identifier") + } + if len(b) != marshaledSize { + return errors.New("xxhash: invalid hash state size") + } + b = b[len(magic):] + b, d.v1 = consumeUint64(b) + b, d.v2 = consumeUint64(b) + b, d.v3 = consumeUint64(b) + b, d.v4 = consumeUint64(b) + b, d.total = consumeUint64(b) + copy(d.mem[:], b) + d.n = int(d.total % uint64(len(d.mem))) + return nil +} + +func appendUint64(b []byte, x uint64) []byte { + var a [8]byte + binary.LittleEndian.PutUint64(a[:], x) + return append(b, a[:]...) +} + +func consumeUint64(b []byte) ([]byte, uint64) { + x := u64(b) + return b[8:], x +} + +func u64(b []byte) uint64 { return binary.LittleEndian.Uint64(b) } +func u32(b []byte) uint32 { return binary.LittleEndian.Uint32(b) } + +func round(acc, input uint64) uint64 { + acc += input * prime2 + acc = rol31(acc) + acc *= prime1 + return acc +} + +func mergeRound(acc, val uint64) uint64 { + val = round(0, val) + acc ^= val + acc = acc*prime1 + prime4 + return acc +} + +func rol1(x uint64) uint64 { return bits.RotateLeft64(x, 1) } +func rol7(x uint64) uint64 { return bits.RotateLeft64(x, 7) } +func rol11(x uint64) uint64 { return bits.RotateLeft64(x, 11) } +func rol12(x uint64) uint64 { return bits.RotateLeft64(x, 12) } +func rol18(x uint64) uint64 { return bits.RotateLeft64(x, 18) } +func rol23(x uint64) uint64 { return bits.RotateLeft64(x, 23) } +func rol27(x uint64) uint64 { return bits.RotateLeft64(x, 27) } +func rol31(x uint64) uint64 { return bits.RotateLeft64(x, 31) } diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash_amd64.s b/vendor/github.com/cespare/xxhash/v2/xxhash_amd64.s new file mode 100644 index 0000000..3e8b132 --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash_amd64.s @@ -0,0 +1,209 @@ +//go:build !appengine && gc && !purego +// +build !appengine +// +build gc +// +build !purego + +#include "textflag.h" + +// Registers: +#define h AX +#define d AX +#define p SI // pointer to advance through b +#define n DX +#define end BX // loop end +#define v1 R8 +#define v2 R9 +#define v3 R10 +#define v4 R11 +#define x R12 +#define prime1 R13 +#define prime2 R14 +#define prime4 DI + +#define round(acc, x) \ + IMULQ prime2, x \ + ADDQ x, acc \ + ROLQ $31, acc \ + IMULQ prime1, acc + +// round0 performs the operation x = round(0, x). +#define round0(x) \ + IMULQ prime2, x \ + ROLQ $31, x \ + IMULQ prime1, x + +// mergeRound applies a merge round on the two registers acc and x. +// It assumes that prime1, prime2, and prime4 have been loaded. +#define mergeRound(acc, x) \ + round0(x) \ + XORQ x, acc \ + IMULQ prime1, acc \ + ADDQ prime4, acc + +// blockLoop processes as many 32-byte blocks as possible, +// updating v1, v2, v3, and v4. It assumes that there is at least one block +// to process. +#define blockLoop() \ +loop: \ + MOVQ +0(p), x \ + round(v1, x) \ + MOVQ +8(p), x \ + round(v2, x) \ + MOVQ +16(p), x \ + round(v3, x) \ + MOVQ +24(p), x \ + round(v4, x) \ + ADDQ $32, p \ + CMPQ p, end \ + JLE loop + +// func Sum64(b []byte) uint64 +TEXT ·Sum64(SB), NOSPLIT|NOFRAME, $0-32 + // Load fixed primes. + MOVQ ·primes+0(SB), prime1 + MOVQ ·primes+8(SB), prime2 + MOVQ ·primes+24(SB), prime4 + + // Load slice. + MOVQ b_base+0(FP), p + MOVQ b_len+8(FP), n + LEAQ (p)(n*1), end + + // The first loop limit will be len(b)-32. + SUBQ $32, end + + // Check whether we have at least one block. + CMPQ n, $32 + JLT noBlocks + + // Set up initial state (v1, v2, v3, v4). + MOVQ prime1, v1 + ADDQ prime2, v1 + MOVQ prime2, v2 + XORQ v3, v3 + XORQ v4, v4 + SUBQ prime1, v4 + + blockLoop() + + MOVQ v1, h + ROLQ $1, h + MOVQ v2, x + ROLQ $7, x + ADDQ x, h + MOVQ v3, x + ROLQ $12, x + ADDQ x, h + MOVQ v4, x + ROLQ $18, x + ADDQ x, h + + mergeRound(h, v1) + mergeRound(h, v2) + mergeRound(h, v3) + mergeRound(h, v4) + + JMP afterBlocks + +noBlocks: + MOVQ ·primes+32(SB), h + +afterBlocks: + ADDQ n, h + + ADDQ $24, end + CMPQ p, end + JG try4 + +loop8: + MOVQ (p), x + ADDQ $8, p + round0(x) + XORQ x, h + ROLQ $27, h + IMULQ prime1, h + ADDQ prime4, h + + CMPQ p, end + JLE loop8 + +try4: + ADDQ $4, end + CMPQ p, end + JG try1 + + MOVL (p), x + ADDQ $4, p + IMULQ prime1, x + XORQ x, h + + ROLQ $23, h + IMULQ prime2, h + ADDQ ·primes+16(SB), h + +try1: + ADDQ $4, end + CMPQ p, end + JGE finalize + +loop1: + MOVBQZX (p), x + ADDQ $1, p + IMULQ ·primes+32(SB), x + XORQ x, h + ROLQ $11, h + IMULQ prime1, h + + CMPQ p, end + JL loop1 + +finalize: + MOVQ h, x + SHRQ $33, x + XORQ x, h + IMULQ prime2, h + MOVQ h, x + SHRQ $29, x + XORQ x, h + IMULQ ·primes+16(SB), h + MOVQ h, x + SHRQ $32, x + XORQ x, h + + MOVQ h, ret+24(FP) + RET + +// func writeBlocks(d *Digest, b []byte) int +TEXT ·writeBlocks(SB), NOSPLIT|NOFRAME, $0-40 + // Load fixed primes needed for round. + MOVQ ·primes+0(SB), prime1 + MOVQ ·primes+8(SB), prime2 + + // Load slice. + MOVQ b_base+8(FP), p + MOVQ b_len+16(FP), n + LEAQ (p)(n*1), end + SUBQ $32, end + + // Load vN from d. + MOVQ s+0(FP), d + MOVQ 0(d), v1 + MOVQ 8(d), v2 + MOVQ 16(d), v3 + MOVQ 24(d), v4 + + // We don't need to check the loop condition here; this function is + // always called with at least one block of data to process. + blockLoop() + + // Copy vN back to d. + MOVQ v1, 0(d) + MOVQ v2, 8(d) + MOVQ v3, 16(d) + MOVQ v4, 24(d) + + // The number of bytes written is p minus the old base pointer. + SUBQ b_base+8(FP), p + MOVQ p, ret+32(FP) + + RET diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash_arm64.s b/vendor/github.com/cespare/xxhash/v2/xxhash_arm64.s new file mode 100644 index 0000000..7e3145a --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash_arm64.s @@ -0,0 +1,183 @@ +//go:build !appengine && gc && !purego +// +build !appengine +// +build gc +// +build !purego + +#include "textflag.h" + +// Registers: +#define digest R1 +#define h R2 // return value +#define p R3 // input pointer +#define n R4 // input length +#define nblocks R5 // n / 32 +#define prime1 R7 +#define prime2 R8 +#define prime3 R9 +#define prime4 R10 +#define prime5 R11 +#define v1 R12 +#define v2 R13 +#define v3 R14 +#define v4 R15 +#define x1 R20 +#define x2 R21 +#define x3 R22 +#define x4 R23 + +#define round(acc, x) \ + MADD prime2, acc, x, acc \ + ROR $64-31, acc \ + MUL prime1, acc + +// round0 performs the operation x = round(0, x). +#define round0(x) \ + MUL prime2, x \ + ROR $64-31, x \ + MUL prime1, x + +#define mergeRound(acc, x) \ + round0(x) \ + EOR x, acc \ + MADD acc, prime4, prime1, acc + +// blockLoop processes as many 32-byte blocks as possible, +// updating v1, v2, v3, and v4. It assumes that n >= 32. +#define blockLoop() \ + LSR $5, n, nblocks \ + PCALIGN $16 \ + loop: \ + LDP.P 16(p), (x1, x2) \ + LDP.P 16(p), (x3, x4) \ + round(v1, x1) \ + round(v2, x2) \ + round(v3, x3) \ + round(v4, x4) \ + SUB $1, nblocks \ + CBNZ nblocks, loop + +// func Sum64(b []byte) uint64 +TEXT ·Sum64(SB), NOSPLIT|NOFRAME, $0-32 + LDP b_base+0(FP), (p, n) + + LDP ·primes+0(SB), (prime1, prime2) + LDP ·primes+16(SB), (prime3, prime4) + MOVD ·primes+32(SB), prime5 + + CMP $32, n + CSEL LT, prime5, ZR, h // if n < 32 { h = prime5 } else { h = 0 } + BLT afterLoop + + ADD prime1, prime2, v1 + MOVD prime2, v2 + MOVD $0, v3 + NEG prime1, v4 + + blockLoop() + + ROR $64-1, v1, x1 + ROR $64-7, v2, x2 + ADD x1, x2 + ROR $64-12, v3, x3 + ROR $64-18, v4, x4 + ADD x3, x4 + ADD x2, x4, h + + mergeRound(h, v1) + mergeRound(h, v2) + mergeRound(h, v3) + mergeRound(h, v4) + +afterLoop: + ADD n, h + + TBZ $4, n, try8 + LDP.P 16(p), (x1, x2) + + round0(x1) + + // NOTE: here and below, sequencing the EOR after the ROR (using a + // rotated register) is worth a small but measurable speedup for small + // inputs. + ROR $64-27, h + EOR x1 @> 64-27, h, h + MADD h, prime4, prime1, h + + round0(x2) + ROR $64-27, h + EOR x2 @> 64-27, h, h + MADD h, prime4, prime1, h + +try8: + TBZ $3, n, try4 + MOVD.P 8(p), x1 + + round0(x1) + ROR $64-27, h + EOR x1 @> 64-27, h, h + MADD h, prime4, prime1, h + +try4: + TBZ $2, n, try2 + MOVWU.P 4(p), x2 + + MUL prime1, x2 + ROR $64-23, h + EOR x2 @> 64-23, h, h + MADD h, prime3, prime2, h + +try2: + TBZ $1, n, try1 + MOVHU.P 2(p), x3 + AND $255, x3, x1 + LSR $8, x3, x2 + + MUL prime5, x1 + ROR $64-11, h + EOR x1 @> 64-11, h, h + MUL prime1, h + + MUL prime5, x2 + ROR $64-11, h + EOR x2 @> 64-11, h, h + MUL prime1, h + +try1: + TBZ $0, n, finalize + MOVBU (p), x4 + + MUL prime5, x4 + ROR $64-11, h + EOR x4 @> 64-11, h, h + MUL prime1, h + +finalize: + EOR h >> 33, h + MUL prime2, h + EOR h >> 29, h + MUL prime3, h + EOR h >> 32, h + + MOVD h, ret+24(FP) + RET + +// func writeBlocks(d *Digest, b []byte) int +TEXT ·writeBlocks(SB), NOSPLIT|NOFRAME, $0-40 + LDP ·primes+0(SB), (prime1, prime2) + + // Load state. Assume v[1-4] are stored contiguously. + MOVD d+0(FP), digest + LDP 0(digest), (v1, v2) + LDP 16(digest), (v3, v4) + + LDP b_base+8(FP), (p, n) + + blockLoop() + + // Store updated state. + STP (v1, v2), 0(digest) + STP (v3, v4), 16(digest) + + BIC $31, n + MOVD n, ret+32(FP) + RET diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash_asm.go b/vendor/github.com/cespare/xxhash/v2/xxhash_asm.go new file mode 100644 index 0000000..78f95f2 --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash_asm.go @@ -0,0 +1,15 @@ +//go:build (amd64 || arm64) && !appengine && gc && !purego +// +build amd64 arm64 +// +build !appengine +// +build gc +// +build !purego + +package xxhash + +// Sum64 computes the 64-bit xxHash digest of b with a zero seed. +// +//go:noescape +func Sum64(b []byte) uint64 + +//go:noescape +func writeBlocks(d *Digest, b []byte) int diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash_other.go b/vendor/github.com/cespare/xxhash/v2/xxhash_other.go new file mode 100644 index 0000000..118e49e --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash_other.go @@ -0,0 +1,76 @@ +//go:build (!amd64 && !arm64) || appengine || !gc || purego +// +build !amd64,!arm64 appengine !gc purego + +package xxhash + +// Sum64 computes the 64-bit xxHash digest of b with a zero seed. +func Sum64(b []byte) uint64 { + // A simpler version would be + // d := New() + // d.Write(b) + // return d.Sum64() + // but this is faster, particularly for small inputs. + + n := len(b) + var h uint64 + + if n >= 32 { + v1 := primes[0] + prime2 + v2 := prime2 + v3 := uint64(0) + v4 := -primes[0] + for len(b) >= 32 { + v1 = round(v1, u64(b[0:8:len(b)])) + v2 = round(v2, u64(b[8:16:len(b)])) + v3 = round(v3, u64(b[16:24:len(b)])) + v4 = round(v4, u64(b[24:32:len(b)])) + b = b[32:len(b):len(b)] + } + h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + h = mergeRound(h, v1) + h = mergeRound(h, v2) + h = mergeRound(h, v3) + h = mergeRound(h, v4) + } else { + h = prime5 + } + + h += uint64(n) + + for ; len(b) >= 8; b = b[8:] { + k1 := round(0, u64(b[:8])) + h ^= k1 + h = rol27(h)*prime1 + prime4 + } + if len(b) >= 4 { + h ^= uint64(u32(b[:4])) * prime1 + h = rol23(h)*prime2 + prime3 + b = b[4:] + } + for ; len(b) > 0; b = b[1:] { + h ^= uint64(b[0]) * prime5 + h = rol11(h) * prime1 + } + + h ^= h >> 33 + h *= prime2 + h ^= h >> 29 + h *= prime3 + h ^= h >> 32 + + return h +} + +func writeBlocks(d *Digest, b []byte) int { + v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4 + n := len(b) + for len(b) >= 32 { + v1 = round(v1, u64(b[0:8:len(b)])) + v2 = round(v2, u64(b[8:16:len(b)])) + v3 = round(v3, u64(b[16:24:len(b)])) + v4 = round(v4, u64(b[24:32:len(b)])) + b = b[32:len(b):len(b)] + } + d.v1, d.v2, d.v3, d.v4 = v1, v2, v3, v4 + return n - len(b) +} diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash_safe.go b/vendor/github.com/cespare/xxhash/v2/xxhash_safe.go new file mode 100644 index 0000000..05f5e7d --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash_safe.go @@ -0,0 +1,16 @@ +//go:build appengine +// +build appengine + +// This file contains the safe implementations of otherwise unsafe-using code. + +package xxhash + +// Sum64String computes the 64-bit xxHash digest of s with a zero seed. +func Sum64String(s string) uint64 { + return Sum64([]byte(s)) +} + +// WriteString adds more data to d. It always returns len(s), nil. +func (d *Digest) WriteString(s string) (n int, err error) { + return d.Write([]byte(s)) +} diff --git a/vendor/github.com/cespare/xxhash/v2/xxhash_unsafe.go b/vendor/github.com/cespare/xxhash/v2/xxhash_unsafe.go new file mode 100644 index 0000000..cf9d42a --- /dev/null +++ b/vendor/github.com/cespare/xxhash/v2/xxhash_unsafe.go @@ -0,0 +1,58 @@ +//go:build !appengine +// +build !appengine + +// This file encapsulates usage of unsafe. +// xxhash_safe.go contains the safe implementations. + +package xxhash + +import ( + "unsafe" +) + +// In the future it's possible that compiler optimizations will make these +// XxxString functions unnecessary by realizing that calls such as +// Sum64([]byte(s)) don't need to copy s. See https://go.dev/issue/2205. +// If that happens, even if we keep these functions they can be replaced with +// the trivial safe code. + +// NOTE: The usual way of doing an unsafe string-to-[]byte conversion is: +// +// var b []byte +// bh := (*reflect.SliceHeader)(unsafe.Pointer(&b)) +// bh.Data = (*reflect.StringHeader)(unsafe.Pointer(&s)).Data +// bh.Len = len(s) +// bh.Cap = len(s) +// +// Unfortunately, as of Go 1.15.3 the inliner's cost model assigns a high enough +// weight to this sequence of expressions that any function that uses it will +// not be inlined. Instead, the functions below use a different unsafe +// conversion designed to minimize the inliner weight and allow both to be +// inlined. There is also a test (TestInlining) which verifies that these are +// inlined. +// +// See https://github.com/golang/go/issues/42739 for discussion. + +// Sum64String computes the 64-bit xxHash digest of s with a zero seed. +// It may be faster than Sum64([]byte(s)) by avoiding a copy. +func Sum64String(s string) uint64 { + b := *(*[]byte)(unsafe.Pointer(&sliceHeader{s, len(s)})) + return Sum64(b) +} + +// WriteString adds more data to d. It always returns len(s), nil. +// It may be faster than Write([]byte(s)) by avoiding a copy. +func (d *Digest) WriteString(s string) (n int, err error) { + d.Write(*(*[]byte)(unsafe.Pointer(&sliceHeader{s, len(s)}))) + // d.Write always returns len(s), nil. + // Ignoring the return output and returning these fixed values buys a + // savings of 6 in the inliner's cost model. + return len(s), nil +} + +// sliceHeader is similar to reflect.SliceHeader, but it assumes that the layout +// of the first two words is the same as the layout of a string. +type sliceHeader struct { + s string + cap int +} diff --git a/vendor/github.com/dgryski/go-rendezvous/LICENSE b/vendor/github.com/dgryski/go-rendezvous/LICENSE new file mode 100644 index 0000000..22080f7 --- /dev/null +++ b/vendor/github.com/dgryski/go-rendezvous/LICENSE @@ -0,0 +1,21 @@ +The MIT License (MIT) + +Copyright (c) 2017-2020 Damian Gryski + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in +all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +THE SOFTWARE. diff --git a/vendor/github.com/dgryski/go-rendezvous/rdv.go b/vendor/github.com/dgryski/go-rendezvous/rdv.go new file mode 100644 index 0000000..7a6f820 --- /dev/null +++ b/vendor/github.com/dgryski/go-rendezvous/rdv.go @@ -0,0 +1,79 @@ +package rendezvous + +type Rendezvous struct { + nodes map[string]int + nstr []string + nhash []uint64 + hash Hasher +} + +type Hasher func(s string) uint64 + +func New(nodes []string, hash Hasher) *Rendezvous { + r := &Rendezvous{ + nodes: make(map[string]int, len(nodes)), + nstr: make([]string, len(nodes)), + nhash: make([]uint64, len(nodes)), + hash: hash, + } + + for i, n := range nodes { + r.nodes[n] = i + r.nstr[i] = n + r.nhash[i] = hash(n) + } + + return r +} + +func (r *Rendezvous) Lookup(k string) string { + // short-circuit if we're empty + if len(r.nodes) == 0 { + return "" + } + + khash := r.hash(k) + + var midx int + var mhash = xorshiftMult64(khash ^ r.nhash[0]) + + for i, nhash := range r.nhash[1:] { + if h := xorshiftMult64(khash ^ nhash); h > mhash { + midx = i + 1 + mhash = h + } + } + + return r.nstr[midx] +} + +func (r *Rendezvous) Add(node string) { + r.nodes[node] = len(r.nstr) + r.nstr = append(r.nstr, node) + r.nhash = append(r.nhash, r.hash(node)) +} + +func (r *Rendezvous) Remove(node string) { + // find index of node to remove + nidx := r.nodes[node] + + // remove from the slices + l := len(r.nstr) + r.nstr[nidx] = r.nstr[l] + r.nstr = r.nstr[:l] + + r.nhash[nidx] = r.nhash[l] + r.nhash = r.nhash[:l] + + // update the map + delete(r.nodes, node) + moved := r.nstr[nidx] + r.nodes[moved] = nidx +} + +func xorshiftMult64(x uint64) uint64 { + x ^= x >> 12 // a + x ^= x << 25 // b + x ^= x >> 27 // c + return x * 2685821657736338717 +} diff --git a/vendor/github.com/klauspost/compress/.gitattributes b/vendor/github.com/klauspost/compress/.gitattributes new file mode 100644 index 0000000..4024335 --- /dev/null +++ b/vendor/github.com/klauspost/compress/.gitattributes @@ -0,0 +1,2 @@ +* -text +*.bin -text -diff diff --git a/vendor/github.com/klauspost/compress/.gitignore b/vendor/github.com/klauspost/compress/.gitignore new file mode 100644 index 0000000..d31b378 --- /dev/null +++ b/vendor/github.com/klauspost/compress/.gitignore @@ -0,0 +1,32 @@ +# Compiled Object files, Static and Dynamic libs (Shared Objects) +*.o +*.a +*.so + +# Folders +_obj +_test + +# Architecture specific extensions/prefixes +*.[568vq] +[568vq].out + +*.cgo1.go +*.cgo2.c +_cgo_defun.c +_cgo_gotypes.go +_cgo_export.* + +_testmain.go + +*.exe +*.test +*.prof +/s2/cmd/_s2sx/sfx-exe + +# Linux perf files +perf.data +perf.data.old + +# gdb history +.gdb_history diff --git a/vendor/github.com/klauspost/compress/.goreleaser.yml b/vendor/github.com/klauspost/compress/.goreleaser.yml new file mode 100644 index 0000000..0af08e6 --- /dev/null +++ b/vendor/github.com/klauspost/compress/.goreleaser.yml @@ -0,0 +1,141 @@ +# This is an example goreleaser.yaml file with some sane defaults. +# Make sure to check the documentation at http://goreleaser.com +before: + hooks: + - ./gen.sh + - go install mvdan.cc/garble@latest + +builds: + - + id: "s2c" + binary: s2c + main: ./s2/cmd/s2c/main.go + flags: + - -trimpath + env: + - CGO_ENABLED=0 + goos: + - aix + - linux + - freebsd + - netbsd + - windows + - darwin + goarch: + - 386 + - amd64 + - arm + - arm64 + - ppc64 + - ppc64le + - mips64 + - mips64le + goarm: + - 7 + gobinary: garble + - + id: "s2d" + binary: s2d + main: ./s2/cmd/s2d/main.go + flags: + - -trimpath + env: + - CGO_ENABLED=0 + goos: + - aix + - linux + - freebsd + - netbsd + - windows + - darwin + goarch: + - 386 + - amd64 + - arm + - arm64 + - ppc64 + - ppc64le + - mips64 + - mips64le + goarm: + - 7 + gobinary: garble + - + id: "s2sx" + binary: s2sx + main: ./s2/cmd/_s2sx/main.go + flags: + - -modfile=s2sx.mod + - -trimpath + env: + - CGO_ENABLED=0 + goos: + - aix + - linux + - freebsd + - netbsd + - windows + - darwin + goarch: + - 386 + - amd64 + - arm + - arm64 + - ppc64 + - ppc64le + - mips64 + - mips64le + goarm: + - 7 + gobinary: garble + +archives: + - + id: s2-binaries + name_template: "s2-{{ .Os }}_{{ .Arch }}_{{ .Version }}" + replacements: + aix: AIX + darwin: OSX + linux: Linux + windows: Windows + 386: i386 + amd64: x86_64 + freebsd: FreeBSD + netbsd: NetBSD + format_overrides: + - goos: windows + format: zip + files: + - unpack/* + - s2/LICENSE + - s2/README.md +checksum: + name_template: 'checksums.txt' +snapshot: + name_template: "{{ .Tag }}-next" +changelog: + sort: asc + filters: + exclude: + - '^doc:' + - '^docs:' + - '^test:' + - '^tests:' + - '^Update\sREADME.md' + +nfpms: + - + file_name_template: "s2_package_{{ .Version }}_{{ .Os }}_{{ .Arch }}" + vendor: Klaus Post + homepage: https://github.com/klauspost/compress + maintainer: Klaus Post + description: S2 Compression Tool + license: BSD 3-Clause + formats: + - deb + - rpm + replacements: + darwin: Darwin + linux: Linux + freebsd: FreeBSD + amd64: x86_64 diff --git a/vendor/github.com/klauspost/compress/LICENSE b/vendor/github.com/klauspost/compress/LICENSE new file mode 100644 index 0000000..87d5574 --- /dev/null +++ b/vendor/github.com/klauspost/compress/LICENSE @@ -0,0 +1,304 @@ +Copyright (c) 2012 The Go Authors. All rights reserved. +Copyright (c) 2019 Klaus Post. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +------------------ + +Files: gzhttp/* + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright 2016-2017 The New York Times Company + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. + +------------------ + +Files: s2/cmd/internal/readahead/* + +The MIT License (MIT) + +Copyright (c) 2015 Klaus Post + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. + +--------------------- +Files: snappy/* +Files: internal/snapref/* + +Copyright (c) 2011 The Snappy-Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +----------------- + +Files: s2/cmd/internal/filepathx/* + +Copyright 2016 The filepathx Authors + +Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/vendor/github.com/klauspost/compress/README.md b/vendor/github.com/klauspost/compress/README.md new file mode 100644 index 0000000..ad5c63a --- /dev/null +++ b/vendor/github.com/klauspost/compress/README.md @@ -0,0 +1,560 @@ +# compress + +This package provides various compression algorithms. + +* [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and decompression in pure Go. +* [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) is a high performance replacement for Snappy. +* Optimized [deflate](https://godoc.org/github.com/klauspost/compress/flate) packages which can be used as a dropin replacement for [gzip](https://godoc.org/github.com/klauspost/compress/gzip), [zip](https://godoc.org/github.com/klauspost/compress/zip) and [zlib](https://godoc.org/github.com/klauspost/compress/zlib). +* [snappy](https://github.com/klauspost/compress/tree/master/snappy) is a drop-in replacement for `github.com/golang/snappy` offering better compression and concurrent streams. +* [huff0](https://github.com/klauspost/compress/tree/master/huff0) and [FSE](https://github.com/klauspost/compress/tree/master/fse) implementations for raw entropy encoding. +* [gzhttp](https://github.com/klauspost/compress/tree/master/gzhttp) Provides client and server wrappers for handling gzipped requests efficiently. +* [pgzip](https://github.com/klauspost/pgzip) is a separate package that provides a very fast parallel gzip implementation. +* [fuzz package](https://github.com/klauspost/compress-fuzz) for fuzz testing all compressors/decompressors here. + +[![Go Reference](https://pkg.go.dev/badge/klauspost/compress.svg)](https://pkg.go.dev/github.com/klauspost/compress?tab=subdirectories) +[![Go](https://github.com/klauspost/compress/actions/workflows/go.yml/badge.svg)](https://github.com/klauspost/compress/actions/workflows/go.yml) +[![Sourcegraph Badge](https://sourcegraph.com/github.com/klauspost/compress/-/badge.svg)](https://sourcegraph.com/github.com/klauspost/compress?badge) + +# changelog + +* July 13, 2022 (v1.15.8) + + * gzip: fix stack exhaustion bug in Reader.Read https://github.com/klauspost/compress/pull/641 + * s2: Add Index header trim/restore https://github.com/klauspost/compress/pull/638 + * zstd: Optimize seqdeq amd64 asm by @greatroar in https://github.com/klauspost/compress/pull/636 + * zstd: Improve decoder memcopy https://github.com/klauspost/compress/pull/637 + * huff0: Pass a single bitReader pointer to asm by @greatroar in https://github.com/klauspost/compress/pull/634 + * zstd: Branchless getBits for amd64 w/o BMI2 by @greatroar in https://github.com/klauspost/compress/pull/640 + * gzhttp: Remove header before writing https://github.com/klauspost/compress/pull/639 + +* June 29, 2022 (v1.15.7) + + * s2: Fix absolute forward seeks https://github.com/klauspost/compress/pull/633 + * zip: Merge upstream https://github.com/klauspost/compress/pull/631 + * zip: Re-add zip64 fix https://github.com/klauspost/compress/pull/624 + * zstd: translate fseDecoder.buildDtable into asm by @WojciechMula in https://github.com/klauspost/compress/pull/598 + * flate: Faster histograms https://github.com/klauspost/compress/pull/620 + * deflate: Use compound hcode https://github.com/klauspost/compress/pull/622 + +* June 3, 2022 (v1.15.6) + * s2: Improve coding for long, close matches https://github.com/klauspost/compress/pull/613 + * s2c: Add Snappy/S2 stream recompression https://github.com/klauspost/compress/pull/611 + * zstd: Always use configured block size https://github.com/klauspost/compress/pull/605 + * zstd: Fix incorrect hash table placement for dict encoding in default https://github.com/klauspost/compress/pull/606 + * zstd: Apply default config to ZipDecompressor without options https://github.com/klauspost/compress/pull/608 + * gzhttp: Exclude more common archive formats https://github.com/klauspost/compress/pull/612 + * s2: Add ReaderIgnoreCRC https://github.com/klauspost/compress/pull/609 + * s2: Remove sanity load on index creation https://github.com/klauspost/compress/pull/607 + * snappy: Use dedicated function for scoring https://github.com/klauspost/compress/pull/614 + * s2c+s2d: Use official snappy framed extension https://github.com/klauspost/compress/pull/610 + +* May 25, 2022 (v1.15.5) + * s2: Add concurrent stream decompression https://github.com/klauspost/compress/pull/602 + * s2: Fix final emit oob read crash on amd64 https://github.com/klauspost/compress/pull/601 + * huff0: asm implementation of Decompress1X by @WojciechMula https://github.com/klauspost/compress/pull/596 + * zstd: Use 1 less goroutine for stream decoding https://github.com/klauspost/compress/pull/588 + * zstd: Copy literal in 16 byte blocks when possible https://github.com/klauspost/compress/pull/592 + * zstd: Speed up when WithDecoderLowmem(false) https://github.com/klauspost/compress/pull/599 + * zstd: faster next state update in BMI2 version of decode by @WojciechMula in https://github.com/klauspost/compress/pull/593 + * huff0: Do not check max size when reading table. https://github.com/klauspost/compress/pull/586 + * flate: Inplace hashing for level 7-9 by @klauspost in https://github.com/klauspost/compress/pull/590 + + +* May 11, 2022 (v1.15.4) + * huff0: decompress directly into output by @WojciechMula in [#577](https://github.com/klauspost/compress/pull/577) + * inflate: Keep dict on stack [#581](https://github.com/klauspost/compress/pull/581) + * zstd: Faster decoding memcopy in asm [#583](https://github.com/klauspost/compress/pull/583) + * zstd: Fix ignored crc [#580](https://github.com/klauspost/compress/pull/580) + +* May 5, 2022 (v1.15.3) + * zstd: Allow to ignore checksum checking by @WojciechMula [#572](https://github.com/klauspost/compress/pull/572) + * s2: Fix incorrect seek for io.SeekEnd in [#575](https://github.com/klauspost/compress/pull/575) + +* Apr 26, 2022 (v1.15.2) + * zstd: Add x86-64 assembly for decompression on streams and blocks. Contributed by [@WojciechMula](https://github.com/WojciechMula). Typically 2x faster. [#528](https://github.com/klauspost/compress/pull/528) [#531](https://github.com/klauspost/compress/pull/531) [#545](https://github.com/klauspost/compress/pull/545) [#537](https://github.com/klauspost/compress/pull/537) + * zstd: Add options to ZipDecompressor and fixes [#539](https://github.com/klauspost/compress/pull/539) + * s2: Use sorted search for index [#555](https://github.com/klauspost/compress/pull/555) + * Minimum version is Go 1.16, added CI test on 1.18. + +* Mar 11, 2022 (v1.15.1) + * huff0: Add x86 assembly of Decode4X by @WojciechMula in [#512](https://github.com/klauspost/compress/pull/512) + * zstd: Reuse zip decoders in [#514](https://github.com/klauspost/compress/pull/514) + * zstd: Detect extra block data and report as corrupted in [#520](https://github.com/klauspost/compress/pull/520) + * zstd: Handle zero sized frame content size stricter in [#521](https://github.com/klauspost/compress/pull/521) + * zstd: Add stricter block size checks in [#523](https://github.com/klauspost/compress/pull/523) + +* Mar 3, 2022 (v1.15.0) + * zstd: Refactor decoder by @klauspost in [#498](https://github.com/klauspost/compress/pull/498) + * zstd: Add stream encoding without goroutines by @klauspost in [#505](https://github.com/klauspost/compress/pull/505) + * huff0: Prevent single blocks exceeding 16 bits by @klauspost in[#507](https://github.com/klauspost/compress/pull/507) + * flate: Inline literal emission by @klauspost in [#509](https://github.com/klauspost/compress/pull/509) + * gzhttp: Add zstd to transport by @klauspost in [#400](https://github.com/klauspost/compress/pull/400) + * gzhttp: Make content-type optional by @klauspost in [#510](https://github.com/klauspost/compress/pull/510) + +
+ See Details +Both compression and decompression now supports "synchronous" stream operations. This means that whenever "concurrency" is set to 1, they will operate without spawning goroutines. + +Stream decompression is now faster on asynchronous, since the goroutine allocation much more effectively splits the workload. On typical streams this will typically use 2 cores fully for decompression. When a stream has finished decoding no goroutines will be left over, so decoders can now safely be pooled and still be garbage collected. + +While the release has been extensively tested, it is recommended to testing when upgrading. +
+ +* Feb 22, 2022 (v1.14.4) + * flate: Fix rare huffman only (-2) corruption. [#503](https://github.com/klauspost/compress/pull/503) + * zip: Update deprecated CreateHeaderRaw to correctly call CreateRaw by @saracen in [#502](https://github.com/klauspost/compress/pull/502) + * zip: don't read data descriptor early by @saracen in [#501](https://github.com/klauspost/compress/pull/501) #501 + * huff0: Use static decompression buffer up to 30% faster by @klauspost in [#499](https://github.com/klauspost/compress/pull/499) [#500](https://github.com/klauspost/compress/pull/500) + +* Feb 17, 2022 (v1.14.3) + * flate: Improve fastest levels compression speed ~10% more throughput. [#482](https://github.com/klauspost/compress/pull/482) [#489](https://github.com/klauspost/compress/pull/489) [#490](https://github.com/klauspost/compress/pull/490) [#491](https://github.com/klauspost/compress/pull/491) [#494](https://github.com/klauspost/compress/pull/494) [#478](https://github.com/klauspost/compress/pull/478) + * flate: Faster decompression speed, ~5-10%. [#483](https://github.com/klauspost/compress/pull/483) + * s2: Faster compression with Go v1.18 and amd64 microarch level 3+. [#484](https://github.com/klauspost/compress/pull/484) [#486](https://github.com/klauspost/compress/pull/486) + +* Jan 25, 2022 (v1.14.2) + * zstd: improve header decoder by @dsnet [#476](https://github.com/klauspost/compress/pull/476) + * zstd: Add bigger default blocks [#469](https://github.com/klauspost/compress/pull/469) + * zstd: Remove unused decompression buffer [#470](https://github.com/klauspost/compress/pull/470) + * zstd: Fix logically dead code by @ningmingxiao [#472](https://github.com/klauspost/compress/pull/472) + * flate: Improve level 7-9 [#471](https://github.com/klauspost/compress/pull/471) [#473](https://github.com/klauspost/compress/pull/473) + * zstd: Add noasm tag for xxhash [#475](https://github.com/klauspost/compress/pull/475) + +* Jan 11, 2022 (v1.14.1) + * s2: Add stream index in [#462](https://github.com/klauspost/compress/pull/462) + * flate: Speed and efficiency improvements in [#439](https://github.com/klauspost/compress/pull/439) [#461](https://github.com/klauspost/compress/pull/461) [#455](https://github.com/klauspost/compress/pull/455) [#452](https://github.com/klauspost/compress/pull/452) [#458](https://github.com/klauspost/compress/pull/458) + * zstd: Performance improvement in [#420]( https://github.com/klauspost/compress/pull/420) [#456](https://github.com/klauspost/compress/pull/456) [#437](https://github.com/klauspost/compress/pull/437) [#467](https://github.com/klauspost/compress/pull/467) [#468](https://github.com/klauspost/compress/pull/468) + * zstd: add arm64 xxhash assembly in [#464](https://github.com/klauspost/compress/pull/464) + * Add garbled for binaries for s2 in [#445](https://github.com/klauspost/compress/pull/445) + +
+ See changes to v1.13.x + +* Aug 30, 2021 (v1.13.5) + * gz/zlib/flate: Alias stdlib errors [#425](https://github.com/klauspost/compress/pull/425) + * s2: Add block support to commandline tools [#413](https://github.com/klauspost/compress/pull/413) + * zstd: pooledZipWriter should return Writers to the same pool [#426](https://github.com/klauspost/compress/pull/426) + * Removed golang/snappy as external dependency for tests [#421](https://github.com/klauspost/compress/pull/421) + +* Aug 12, 2021 (v1.13.4) + * Add [snappy replacement package](https://github.com/klauspost/compress/tree/master/snappy). + * zstd: Fix incorrect encoding in "best" mode [#415](https://github.com/klauspost/compress/pull/415) + +* Aug 3, 2021 (v1.13.3) + * zstd: Improve Best compression [#404](https://github.com/klauspost/compress/pull/404) + * zstd: Fix WriteTo error forwarding [#411](https://github.com/klauspost/compress/pull/411) + * gzhttp: Return http.HandlerFunc instead of http.Handler. Unlikely breaking change. [#406](https://github.com/klauspost/compress/pull/406) + * s2sx: Fix max size error [#399](https://github.com/klauspost/compress/pull/399) + * zstd: Add optional stream content size on reset [#401](https://github.com/klauspost/compress/pull/401) + * zstd: use SpeedBestCompression for level >= 10 [#410](https://github.com/klauspost/compress/pull/410) + +* Jun 14, 2021 (v1.13.1) + * s2: Add full Snappy output support [#396](https://github.com/klauspost/compress/pull/396) + * zstd: Add configurable [Decoder window](https://pkg.go.dev/github.com/klauspost/compress/zstd#WithDecoderMaxWindow) size [#394](https://github.com/klauspost/compress/pull/394) + * gzhttp: Add header to skip compression [#389](https://github.com/klauspost/compress/pull/389) + * s2: Improve speed with bigger output margin [#395](https://github.com/klauspost/compress/pull/395) + +* Jun 3, 2021 (v1.13.0) + * Added [gzhttp](https://github.com/klauspost/compress/tree/master/gzhttp#gzip-handler) which allows wrapping HTTP servers and clients with GZIP compressors. + * zstd: Detect short invalid signatures [#382](https://github.com/klauspost/compress/pull/382) + * zstd: Spawn decoder goroutine only if needed. [#380](https://github.com/klauspost/compress/pull/380) +
+ + +
+ See changes to v1.12.x + +* May 25, 2021 (v1.12.3) + * deflate: Better/faster Huffman encoding [#374](https://github.com/klauspost/compress/pull/374) + * deflate: Allocate less for history. [#375](https://github.com/klauspost/compress/pull/375) + * zstd: Forward read errors [#373](https://github.com/klauspost/compress/pull/373) + +* Apr 27, 2021 (v1.12.2) + * zstd: Improve better/best compression [#360](https://github.com/klauspost/compress/pull/360) [#364](https://github.com/klauspost/compress/pull/364) [#365](https://github.com/klauspost/compress/pull/365) + * zstd: Add helpers to compress/decompress zstd inside zip files [#363](https://github.com/klauspost/compress/pull/363) + * deflate: Improve level 5+6 compression [#367](https://github.com/klauspost/compress/pull/367) + * s2: Improve better/best compression [#358](https://github.com/klauspost/compress/pull/358) [#359](https://github.com/klauspost/compress/pull/358) + * s2: Load after checking src limit on amd64. [#362](https://github.com/klauspost/compress/pull/362) + * s2sx: Limit max executable size [#368](https://github.com/klauspost/compress/pull/368) + +* Apr 14, 2021 (v1.12.1) + * snappy package removed. Upstream added as dependency. + * s2: Better compression in "best" mode [#353](https://github.com/klauspost/compress/pull/353) + * s2sx: Add stdin input and detect pre-compressed from signature [#352](https://github.com/klauspost/compress/pull/352) + * s2c/s2d: Add http as possible input [#348](https://github.com/klauspost/compress/pull/348) + * s2c/s2d/s2sx: Always truncate when writing files [#352](https://github.com/klauspost/compress/pull/352) + * zstd: Reduce memory usage further when using [WithLowerEncoderMem](https://pkg.go.dev/github.com/klauspost/compress/zstd#WithLowerEncoderMem) [#346](https://github.com/klauspost/compress/pull/346) + * s2: Fix potential problem with amd64 assembly and profilers [#349](https://github.com/klauspost/compress/pull/349) +
+ +
+ See changes to v1.11.x + +* Mar 26, 2021 (v1.11.13) + * zstd: Big speedup on small dictionary encodes [#344](https://github.com/klauspost/compress/pull/344) [#345](https://github.com/klauspost/compress/pull/345) + * zstd: Add [WithLowerEncoderMem](https://pkg.go.dev/github.com/klauspost/compress/zstd#WithLowerEncoderMem) encoder option [#336](https://github.com/klauspost/compress/pull/336) + * deflate: Improve entropy compression [#338](https://github.com/klauspost/compress/pull/338) + * s2: Clean up and minor performance improvement in best [#341](https://github.com/klauspost/compress/pull/341) + +* Mar 5, 2021 (v1.11.12) + * s2: Add `s2sx` binary that creates [self extracting archives](https://github.com/klauspost/compress/tree/master/s2#s2sx-self-extracting-archives). + * s2: Speed up decompression on non-assembly platforms [#328](https://github.com/klauspost/compress/pull/328) + +* Mar 1, 2021 (v1.11.9) + * s2: Add ARM64 decompression assembly. Around 2x output speed. [#324](https://github.com/klauspost/compress/pull/324) + * s2: Improve "better" speed and efficiency. [#325](https://github.com/klauspost/compress/pull/325) + * s2: Fix binaries. + +* Feb 25, 2021 (v1.11.8) + * s2: Fixed occational out-of-bounds write on amd64. Upgrade recommended. + * s2: Add AMD64 assembly for better mode. 25-50% faster. [#315](https://github.com/klauspost/compress/pull/315) + * s2: Less upfront decoder allocation. [#322](https://github.com/klauspost/compress/pull/322) + * zstd: Faster "compression" of incompressible data. [#314](https://github.com/klauspost/compress/pull/314) + * zip: Fix zip64 headers. [#313](https://github.com/klauspost/compress/pull/313) + +* Jan 14, 2021 (v1.11.7) + * Use Bytes() interface to get bytes across packages. [#309](https://github.com/klauspost/compress/pull/309) + * s2: Add 'best' compression option. [#310](https://github.com/klauspost/compress/pull/310) + * s2: Add ReaderMaxBlockSize, changes `s2.NewReader` signature to include varargs. [#311](https://github.com/klauspost/compress/pull/311) + * s2: Fix crash on small better buffers. [#308](https://github.com/klauspost/compress/pull/308) + * s2: Clean up decoder. [#312](https://github.com/klauspost/compress/pull/312) + +* Jan 7, 2021 (v1.11.6) + * zstd: Make decoder allocations smaller [#306](https://github.com/klauspost/compress/pull/306) + * zstd: Free Decoder resources when Reset is called with a nil io.Reader [#305](https://github.com/klauspost/compress/pull/305) + +* Dec 20, 2020 (v1.11.4) + * zstd: Add Best compression mode [#304](https://github.com/klauspost/compress/pull/304) + * Add header decoder [#299](https://github.com/klauspost/compress/pull/299) + * s2: Add uncompressed stream option [#297](https://github.com/klauspost/compress/pull/297) + * Simplify/speed up small blocks with known max size. [#300](https://github.com/klauspost/compress/pull/300) + * zstd: Always reset literal dict encoder [#303](https://github.com/klauspost/compress/pull/303) + +* Nov 15, 2020 (v1.11.3) + * inflate: 10-15% faster decompression [#293](https://github.com/klauspost/compress/pull/293) + * zstd: Tweak DecodeAll default allocation [#295](https://github.com/klauspost/compress/pull/295) + +* Oct 11, 2020 (v1.11.2) + * s2: Fix out of bounds read in "better" block compression [#291](https://github.com/klauspost/compress/pull/291) + +* Oct 1, 2020 (v1.11.1) + * zstd: Set allLitEntropy true in default configuration [#286](https://github.com/klauspost/compress/pull/286) + +* Sept 8, 2020 (v1.11.0) + * zstd: Add experimental compression [dictionaries](https://github.com/klauspost/compress/tree/master/zstd#dictionaries) [#281](https://github.com/klauspost/compress/pull/281) + * zstd: Fix mixed Write and ReadFrom calls [#282](https://github.com/klauspost/compress/pull/282) + * inflate/gz: Limit variable shifts, ~5% faster decompression [#274](https://github.com/klauspost/compress/pull/274) +
+ +
+ See changes to v1.10.x + +* July 8, 2020 (v1.10.11) + * zstd: Fix extra block when compressing with ReadFrom. [#278](https://github.com/klauspost/compress/pull/278) + * huff0: Also populate compression table when reading decoding table. [#275](https://github.com/klauspost/compress/pull/275) + +* June 23, 2020 (v1.10.10) + * zstd: Skip entropy compression in fastest mode when no matches. [#270](https://github.com/klauspost/compress/pull/270) + +* June 16, 2020 (v1.10.9): + * zstd: API change for specifying dictionaries. See [#268](https://github.com/klauspost/compress/pull/268) + * zip: update CreateHeaderRaw to handle zip64 fields. [#266](https://github.com/klauspost/compress/pull/266) + * Fuzzit tests removed. The service has been purchased and is no longer available. + +* June 5, 2020 (v1.10.8): + * 1.15x faster zstd block decompression. [#265](https://github.com/klauspost/compress/pull/265) + +* June 1, 2020 (v1.10.7): + * Added zstd decompression [dictionary support](https://github.com/klauspost/compress/tree/master/zstd#dictionaries) + * Increase zstd decompression speed up to 1.19x. [#259](https://github.com/klauspost/compress/pull/259) + * Remove internal reset call in zstd compression and reduce allocations. [#263](https://github.com/klauspost/compress/pull/263) + +* May 21, 2020: (v1.10.6) + * zstd: Reduce allocations while decoding. [#258](https://github.com/klauspost/compress/pull/258), [#252](https://github.com/klauspost/compress/pull/252) + * zstd: Stricter decompression checks. + +* April 12, 2020: (v1.10.5) + * s2-commands: Flush output when receiving SIGINT. [#239](https://github.com/klauspost/compress/pull/239) + +* Apr 8, 2020: (v1.10.4) + * zstd: Minor/special case optimizations. [#251](https://github.com/klauspost/compress/pull/251), [#250](https://github.com/klauspost/compress/pull/250), [#249](https://github.com/klauspost/compress/pull/249), [#247](https://github.com/klauspost/compress/pull/247) +* Mar 11, 2020: (v1.10.3) + * s2: Use S2 encoder in pure Go mode for Snappy output as well. [#245](https://github.com/klauspost/compress/pull/245) + * s2: Fix pure Go block encoder. [#244](https://github.com/klauspost/compress/pull/244) + * zstd: Added "better compression" mode. [#240](https://github.com/klauspost/compress/pull/240) + * zstd: Improve speed of fastest compression mode by 5-10% [#241](https://github.com/klauspost/compress/pull/241) + * zstd: Skip creating encoders when not needed. [#238](https://github.com/klauspost/compress/pull/238) + +* Feb 27, 2020: (v1.10.2) + * Close to 50% speedup in inflate (gzip/zip decompression). [#236](https://github.com/klauspost/compress/pull/236) [#234](https://github.com/klauspost/compress/pull/234) [#232](https://github.com/klauspost/compress/pull/232) + * Reduce deflate level 1-6 memory usage up to 59%. [#227](https://github.com/klauspost/compress/pull/227) + +* Feb 18, 2020: (v1.10.1) + * Fix zstd crash when resetting multiple times without sending data. [#226](https://github.com/klauspost/compress/pull/226) + * deflate: Fix dictionary use on level 1-6. [#224](https://github.com/klauspost/compress/pull/224) + * Remove deflate writer reference when closing. [#224](https://github.com/klauspost/compress/pull/224) + +* Feb 4, 2020: (v1.10.0) + * Add optional dictionary to [stateless deflate](https://pkg.go.dev/github.com/klauspost/compress/flate?tab=doc#StatelessDeflate). Breaking change, send `nil` for previous behaviour. [#216](https://github.com/klauspost/compress/pull/216) + * Fix buffer overflow on repeated small block deflate. [#218](https://github.com/klauspost/compress/pull/218) + * Allow copying content from an existing ZIP file without decompressing+compressing. [#214](https://github.com/klauspost/compress/pull/214) + * Added [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) AMD64 assembler and various optimizations. Stream speed >10GB/s. [#186](https://github.com/klauspost/compress/pull/186) + +
+ +
+ See changes prior to v1.10.0 + +* Jan 20,2020 (v1.9.8) Optimize gzip/deflate with better size estimates and faster table generation. [#207](https://github.com/klauspost/compress/pull/207) by [luyu6056](https://github.com/luyu6056), [#206](https://github.com/klauspost/compress/pull/206). +* Jan 11, 2020: S2 Encode/Decode will use provided buffer if capacity is big enough. [#204](https://github.com/klauspost/compress/pull/204) +* Jan 5, 2020: (v1.9.7) Fix another zstd regression in v1.9.5 - v1.9.6 removed. +* Jan 4, 2020: (v1.9.6) Regression in v1.9.5 fixed causing corrupt zstd encodes in rare cases. +* Jan 4, 2020: Faster IO in [s2c + s2d commandline tools](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) compression/decompression. [#192](https://github.com/klauspost/compress/pull/192) +* Dec 29, 2019: Removed v1.9.5 since fuzz tests showed a compatibility problem with the reference zstandard decoder. +* Dec 29, 2019: (v1.9.5) zstd: 10-20% faster block compression. [#199](https://github.com/klauspost/compress/pull/199) +* Dec 29, 2019: [zip](https://godoc.org/github.com/klauspost/compress/zip) package updated with latest Go features +* Dec 29, 2019: zstd: Single segment flag condintions tweaked. [#197](https://github.com/klauspost/compress/pull/197) +* Dec 18, 2019: s2: Faster compression when ReadFrom is used. [#198](https://github.com/klauspost/compress/pull/198) +* Dec 10, 2019: s2: Fix repeat length output when just above at 16MB limit. +* Dec 10, 2019: zstd: Add function to get decoder as io.ReadCloser. [#191](https://github.com/klauspost/compress/pull/191) +* Dec 3, 2019: (v1.9.4) S2: limit max repeat length. [#188](https://github.com/klauspost/compress/pull/188) +* Dec 3, 2019: Add [WithNoEntropyCompression](https://godoc.org/github.com/klauspost/compress/zstd#WithNoEntropyCompression) to zstd [#187](https://github.com/klauspost/compress/pull/187) +* Dec 3, 2019: Reduce memory use for tests. Check for leaked goroutines. +* Nov 28, 2019 (v1.9.3) Less allocations in stateless deflate. +* Nov 28, 2019: 5-20% Faster huff0 decode. Impacts zstd as well. [#184](https://github.com/klauspost/compress/pull/184) +* Nov 12, 2019 (v1.9.2) Added [Stateless Compression](#stateless-compression) for gzip/deflate. +* Nov 12, 2019: Fixed zstd decompression of large single blocks. [#180](https://github.com/klauspost/compress/pull/180) +* Nov 11, 2019: Set default [s2c](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) block size to 4MB. +* Nov 11, 2019: Reduce inflate memory use by 1KB. +* Nov 10, 2019: Less allocations in deflate bit writer. +* Nov 10, 2019: Fix inconsistent error returned by zstd decoder. +* Oct 28, 2019 (v1.9.1) ztsd: Fix crash when compressing blocks. [#174](https://github.com/klauspost/compress/pull/174) +* Oct 24, 2019 (v1.9.0) zstd: Fix rare data corruption [#173](https://github.com/klauspost/compress/pull/173) +* Oct 24, 2019 zstd: Fix huff0 out of buffer write [#171](https://github.com/klauspost/compress/pull/171) and always return errors [#172](https://github.com/klauspost/compress/pull/172) +* Oct 10, 2019: Big deflate rewrite, 30-40% faster with better compression [#105](https://github.com/klauspost/compress/pull/105) + +
+ +
+ See changes prior to v1.9.0 + +* Oct 10, 2019: (v1.8.6) zstd: Allow partial reads to get flushed data. [#169](https://github.com/klauspost/compress/pull/169) +* Oct 3, 2019: Fix inconsistent results on broken zstd streams. +* Sep 25, 2019: Added `-rm` (remove source files) and `-q` (no output except errors) to `s2c` and `s2d` [commands](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) +* Sep 16, 2019: (v1.8.4) Add `s2c` and `s2d` [commandline tools](https://github.com/klauspost/compress/tree/master/s2#commandline-tools). +* Sep 10, 2019: (v1.8.3) Fix s2 decoder [Skip](https://godoc.org/github.com/klauspost/compress/s2#Reader.Skip). +* Sep 7, 2019: zstd: Added [WithWindowSize](https://godoc.org/github.com/klauspost/compress/zstd#WithWindowSize), contributed by [ianwilkes](https://github.com/ianwilkes). +* Sep 5, 2019: (v1.8.2) Add [WithZeroFrames](https://godoc.org/github.com/klauspost/compress/zstd#WithZeroFrames) which adds full zero payload block encoding option. +* Sep 5, 2019: Lazy initialization of zstandard predefined en/decoder tables. +* Aug 26, 2019: (v1.8.1) S2: 1-2% compression increase in "better" compression mode. +* Aug 26, 2019: zstd: Check maximum size of Huffman 1X compressed literals while decoding. +* Aug 24, 2019: (v1.8.0) Added [S2 compression](https://github.com/klauspost/compress/tree/master/s2#s2-compression), a high performance replacement for Snappy. +* Aug 21, 2019: (v1.7.6) Fixed minor issues found by fuzzer. One could lead to zstd not decompressing. +* Aug 18, 2019: Add [fuzzit](https://fuzzit.dev/) continuous fuzzing. +* Aug 14, 2019: zstd: Skip incompressible data 2x faster. [#147](https://github.com/klauspost/compress/pull/147) +* Aug 4, 2019 (v1.7.5): Better literal compression. [#146](https://github.com/klauspost/compress/pull/146) +* Aug 4, 2019: Faster zstd compression. [#143](https://github.com/klauspost/compress/pull/143) [#144](https://github.com/klauspost/compress/pull/144) +* Aug 4, 2019: Faster zstd decompression. [#145](https://github.com/klauspost/compress/pull/145) [#143](https://github.com/klauspost/compress/pull/143) [#142](https://github.com/klauspost/compress/pull/142) +* July 15, 2019 (v1.7.4): Fix double EOF block in rare cases on zstd encoder. +* July 15, 2019 (v1.7.3): Minor speedup/compression increase in default zstd encoder. +* July 14, 2019: zstd decoder: Fix decompression error on multiple uses with mixed content. +* July 7, 2019 (v1.7.2): Snappy update, zstd decoder potential race fix. +* June 17, 2019: zstd decompression bugfix. +* June 17, 2019: fix 32 bit builds. +* June 17, 2019: Easier use in modules (less dependencies). +* June 9, 2019: New stronger "default" [zstd](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression mode. Matches zstd default compression ratio. +* June 5, 2019: 20-40% throughput in [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and better compression. +* June 5, 2019: deflate/gzip compression: Reduce memory usage of lower compression levels. +* June 2, 2019: Added [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression! +* May 25, 2019: deflate/gzip: 10% faster bit writer, mostly visible in lower levels. +* Apr 22, 2019: [zstd](https://github.com/klauspost/compress/tree/master/zstd#zstd) decompression added. +* Aug 1, 2018: Added [huff0 README](https://github.com/klauspost/compress/tree/master/huff0#huff0-entropy-compression). +* Jul 8, 2018: Added [Performance Update 2018](#performance-update-2018) below. +* Jun 23, 2018: Merged [Go 1.11 inflate optimizations](https://go-review.googlesource.com/c/go/+/102235). Go 1.9 is now required. Backwards compatible version tagged with [v1.3.0](https://github.com/klauspost/compress/releases/tag/v1.3.0). +* Apr 2, 2018: Added [huff0](https://godoc.org/github.com/klauspost/compress/huff0) en/decoder. Experimental for now, API may change. +* Mar 4, 2018: Added [FSE Entropy](https://godoc.org/github.com/klauspost/compress/fse) en/decoder. Experimental for now, API may change. +* Nov 3, 2017: Add compression [Estimate](https://godoc.org/github.com/klauspost/compress#Estimate) function. +* May 28, 2017: Reduce allocations when resetting decoder. +* Apr 02, 2017: Change back to official crc32, since changes were merged in Go 1.7. +* Jan 14, 2017: Reduce stack pressure due to array copies. See [Issue #18625](https://github.com/golang/go/issues/18625). +* Oct 25, 2016: Level 2-4 have been rewritten and now offers significantly better performance than before. +* Oct 20, 2016: Port zlib changes from Go 1.7 to fix zlib writer issue. Please update. +* Oct 16, 2016: Go 1.7 changes merged. Apples to apples this package is a few percent faster, but has a significantly better balance between speed and compression per level. +* Mar 24, 2016: Always attempt Huffman encoding on level 4-7. This improves base 64 encoded data compression. +* Mar 24, 2016: Small speedup for level 1-3. +* Feb 19, 2016: Faster bit writer, level -2 is 15% faster, level 1 is 4% faster. +* Feb 19, 2016: Handle small payloads faster in level 1-3. +* Feb 19, 2016: Added faster level 2 + 3 compression modes. +* Feb 19, 2016: [Rebalanced compression levels](https://blog.klauspost.com/rebalancing-deflate-compression-levels/), so there is a more even progresssion in terms of compression. New default level is 5. +* Feb 14, 2016: Snappy: Merge upstream changes. +* Feb 14, 2016: Snappy: Fix aggressive skipping. +* Feb 14, 2016: Snappy: Update benchmark. +* Feb 13, 2016: Deflate: Fixed assembler problem that could lead to sub-optimal compression. +* Feb 12, 2016: Snappy: Added AMD64 SSE 4.2 optimizations to matching, which makes easy to compress material run faster. Typical speedup is around 25%. +* Feb 9, 2016: Added Snappy package fork. This version is 5-7% faster, much more on hard to compress content. +* Jan 30, 2016: Optimize level 1 to 3 by not considering static dictionary or storing uncompressed. ~4-5% speedup. +* Jan 16, 2016: Optimization on deflate level 1,2,3 compression. +* Jan 8 2016: Merge [CL 18317](https://go-review.googlesource.com/#/c/18317): fix reading, writing of zip64 archives. +* Dec 8 2015: Make level 1 and -2 deterministic even if write size differs. +* Dec 8 2015: Split encoding functions, so hashing and matching can potentially be inlined. 1-3% faster on AMD64. 5% faster on other platforms. +* Dec 8 2015: Fixed rare [one byte out-of bounds read](https://github.com/klauspost/compress/issues/20). Please update! +* Nov 23 2015: Optimization on token writer. ~2-4% faster. Contributed by [@dsnet](https://github.com/dsnet). +* Nov 20 2015: Small optimization to bit writer on 64 bit systems. +* Nov 17 2015: Fixed out-of-bound errors if the underlying Writer returned an error. See [#15](https://github.com/klauspost/compress/issues/15). +* Nov 12 2015: Added [io.WriterTo](https://golang.org/pkg/io/#WriterTo) support to gzip/inflate. +* Nov 11 2015: Merged [CL 16669](https://go-review.googlesource.com/#/c/16669/4): archive/zip: enable overriding (de)compressors per file +* Oct 15 2015: Added skipping on uncompressible data. Random data speed up >5x. + +
+ +# deflate usage + +The packages are drop-in replacements for standard libraries. Simply replace the import path to use them: + +| old import | new import | Documentation +|--------------------|-----------------------------------------|--------------------| +| `compress/gzip` | `github.com/klauspost/compress/gzip` | [gzip](https://pkg.go.dev/github.com/klauspost/compress/gzip?tab=doc) +| `compress/zlib` | `github.com/klauspost/compress/zlib` | [zlib](https://pkg.go.dev/github.com/klauspost/compress/zlib?tab=doc) +| `archive/zip` | `github.com/klauspost/compress/zip` | [zip](https://pkg.go.dev/github.com/klauspost/compress/zip?tab=doc) +| `compress/flate` | `github.com/klauspost/compress/flate` | [flate](https://pkg.go.dev/github.com/klauspost/compress/flate?tab=doc) + +* Optimized [deflate](https://godoc.org/github.com/klauspost/compress/flate) packages which can be used as a dropin replacement for [gzip](https://godoc.org/github.com/klauspost/compress/gzip), [zip](https://godoc.org/github.com/klauspost/compress/zip) and [zlib](https://godoc.org/github.com/klauspost/compress/zlib). + +You may also be interested in [pgzip](https://github.com/klauspost/pgzip), which is a drop in replacement for gzip, which support multithreaded compression on big files and the optimized [crc32](https://github.com/klauspost/crc32) package used by these packages. + +The packages contains the same as the standard library, so you can use the godoc for that: [gzip](http://golang.org/pkg/compress/gzip/), [zip](http://golang.org/pkg/archive/zip/), [zlib](http://golang.org/pkg/compress/zlib/), [flate](http://golang.org/pkg/compress/flate/). + +Currently there is only minor speedup on decompression (mostly CRC32 calculation). + +Memory usage is typically 1MB for a Writer. stdlib is in the same range. +If you expect to have a lot of concurrently allocated Writers consider using +the stateless compress described below. + +For compression performance, see: [this spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing). + +# Stateless compression + +This package offers stateless compression as a special option for gzip/deflate. +It will do compression but without maintaining any state between Write calls. + +This means there will be no memory kept between Write calls, but compression and speed will be suboptimal. + +This is only relevant in cases where you expect to run many thousands of compressors concurrently, +but with very little activity. This is *not* intended for regular web servers serving individual requests. + +Because of this, the size of actual Write calls will affect output size. + +In gzip, specify level `-3` / `gzip.StatelessCompression` to enable. + +For direct deflate use, NewStatelessWriter and StatelessDeflate are available. See [documentation](https://godoc.org/github.com/klauspost/compress/flate#NewStatelessWriter) + +A `bufio.Writer` can of course be used to control write sizes. For example, to use a 4KB buffer: + +``` + // replace 'ioutil.Discard' with your output. + gzw, err := gzip.NewWriterLevel(ioutil.Discard, gzip.StatelessCompression) + if err != nil { + return err + } + defer gzw.Close() + + w := bufio.NewWriterSize(gzw, 4096) + defer w.Flush() + + // Write to 'w' +``` + +This will only use up to 4KB in memory when the writer is idle. + +Compression is almost always worse than the fastest compression level +and each write will allocate (a little) memory. + +# Performance Update 2018 + +It has been a while since we have been looking at the speed of this package compared to the standard library, so I thought I would re-do my tests and give some overall recommendations based on the current state. All benchmarks have been performed with Go 1.10 on my Desktop Intel(R) Core(TM) i7-2600 CPU @3.40GHz. Since I last ran the tests, I have gotten more RAM, which means tests with big files are no longer limited by my SSD. + +The raw results are in my [updated spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing). Due to cgo changes and upstream updates i could not get the cgo version of gzip to compile. Instead I included the [zstd](https://github.com/datadog/zstd) cgo implementation. If I get cgo gzip to work again, I might replace the results in the sheet. + +The columns to take note of are: *MB/s* - the throughput. *Reduction* - the data size reduction in percent of the original. *Rel Speed* relative speed compared to the standard library at the same level. *Smaller* - how many percent smaller is the compressed output compared to stdlib. Negative means the output was bigger. *Loss* means the loss (or gain) in compression as a percentage difference of the input. + +The `gzstd` (standard library gzip) and `gzkp` (this package gzip) only uses one CPU core. [`pgzip`](https://github.com/klauspost/pgzip), [`bgzf`](https://github.com/biogo/hts/tree/master/bgzf) uses all 4 cores. [`zstd`](https://github.com/DataDog/zstd) uses one core, and is a beast (but not Go, yet). + + +## Overall differences. + +There appears to be a roughly 5-10% speed advantage over the standard library when comparing at similar compression levels. + +The biggest difference you will see is the result of [re-balancing](https://blog.klauspost.com/rebalancing-deflate-compression-levels/) the compression levels. I wanted by library to give a smoother transition between the compression levels than the standard library. + +This package attempts to provide a more smooth transition, where "1" is taking a lot of shortcuts, "5" is the reasonable trade-off and "9" is the "give me the best compression", and the values in between gives something reasonable in between. The standard library has big differences in levels 1-4, but levels 5-9 having no significant gains - often spending a lot more time than can be justified by the achieved compression. + +There are links to all the test data in the [spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing) in the top left field on each tab. + +## Web Content + +This test set aims to emulate typical use in a web server. The test-set is 4GB data in 53k files, and is a mixture of (mostly) HTML, JS, CSS. + +Since level 1 and 9 are close to being the same code, they are quite close. But looking at the levels in-between the differences are quite big. + +Looking at level 6, this package is 88% faster, but will output about 6% more data. For a web server, this means you can serve 88% more data, but have to pay for 6% more bandwidth. You can draw your own conclusions on what would be the most expensive for your case. + +## Object files + +This test is for typical data files stored on a server. In this case it is a collection of Go precompiled objects. They are very compressible. + +The picture is similar to the web content, but with small differences since this is very compressible. Levels 2-3 offer good speed, but is sacrificing quite a bit of compression. + +The standard library seems suboptimal on level 3 and 4 - offering both worse compression and speed than level 6 & 7 of this package respectively. + +## Highly Compressible File + +This is a JSON file with very high redundancy. The reduction starts at 95% on level 1, so in real life terms we are dealing with something like a highly redundant stream of data, etc. + +It is definitely visible that we are dealing with specialized content here, so the results are very scattered. This package does not do very well at levels 1-4, but picks up significantly at level 5 and levels 7 and 8 offering great speed for the achieved compression. + +So if you know you content is extremely compressible you might want to go slightly higher than the defaults. The standard library has a huge gap between levels 3 and 4 in terms of speed (2.75x slowdown), so it offers little "middle ground". + +## Medium-High Compressible + +This is a pretty common test corpus: [enwik9](http://mattmahoney.net/dc/textdata.html). It contains the first 10^9 bytes of the English Wikipedia dump on Mar. 3, 2006. This is a very good test of typical text based compression and more data heavy streams. + +We see a similar picture here as in "Web Content". On equal levels some compression is sacrificed for more speed. Level 5 seems to be the best trade-off between speed and size, beating stdlib level 3 in both. + +## Medium Compressible + +I will combine two test sets, one [10GB file set](http://mattmahoney.net/dc/10gb.html) and a VM disk image (~8GB). Both contain different data types and represent a typical backup scenario. + +The most notable thing is how quickly the standard library drops to very low compression speeds around level 5-6 without any big gains in compression. Since this type of data is fairly common, this does not seem like good behavior. + + +## Un-compressible Content + +This is mainly a test of how good the algorithms are at detecting un-compressible input. The standard library only offers this feature with very conservative settings at level 1. Obviously there is no reason for the algorithms to try to compress input that cannot be compressed. The only downside is that it might skip some compressible data on false detections. + + +## Huffman only compression + +This compression library adds a special compression level, named `HuffmanOnly`, which allows near linear time compression. This is done by completely disabling matching of previous data, and only reduce the number of bits to represent each character. + +This means that often used characters, like 'e' and ' ' (space) in text use the fewest bits to represent, and rare characters like '¤' takes more bits to represent. For more information see [wikipedia](https://en.wikipedia.org/wiki/Huffman_coding) or this nice [video](https://youtu.be/ZdooBTdW5bM). + +Since this type of compression has much less variance, the compression speed is mostly unaffected by the input data, and is usually more than *180MB/s* for a single core. + +The downside is that the compression ratio is usually considerably worse than even the fastest conventional compression. The compression ratio can never be better than 8:1 (12.5%). + +The linear time compression can be used as a "better than nothing" mode, where you cannot risk the encoder to slow down on some content. For comparison, the size of the "Twain" text is *233460 bytes* (+29% vs. level 1) and encode speed is 144MB/s (4.5x level 1). So in this case you trade a 30% size increase for a 4 times speedup. + +For more information see my blog post on [Fast Linear Time Compression](http://blog.klauspost.com/constant-time-gzipzip-compression/). + +This is implemented on Go 1.7 as "Huffman Only" mode, though not exposed for gzip. + +# Other packages + +Here are other packages of good quality and pure Go (no cgo wrappers or autoconverted code): + +* [github.com/pierrec/lz4](https://github.com/pierrec/lz4) - strong multithreaded LZ4 compression. +* [github.com/cosnicolaou/pbzip2](https://github.com/cosnicolaou/pbzip2) - multithreaded bzip2 decompression. +* [github.com/dsnet/compress](https://github.com/dsnet/compress) - brotli decompression, bzip2 writer. + +# license + +This code is licensed under the same conditions as the original Go code. See LICENSE file. diff --git a/vendor/github.com/klauspost/compress/compressible.go b/vendor/github.com/klauspost/compress/compressible.go new file mode 100644 index 0000000..ea5a692 --- /dev/null +++ b/vendor/github.com/klauspost/compress/compressible.go @@ -0,0 +1,85 @@ +package compress + +import "math" + +// Estimate returns a normalized compressibility estimate of block b. +// Values close to zero are likely uncompressible. +// Values above 0.1 are likely to be compressible. +// Values above 0.5 are very compressible. +// Very small lengths will return 0. +func Estimate(b []byte) float64 { + if len(b) < 16 { + return 0 + } + + // Correctly predicted order 1 + hits := 0 + lastMatch := false + var o1 [256]byte + var hist [256]int + c1 := byte(0) + for _, c := range b { + if c == o1[c1] { + // We only count a hit if there was two correct predictions in a row. + if lastMatch { + hits++ + } + lastMatch = true + } else { + lastMatch = false + } + o1[c1] = c + c1 = c + hist[c]++ + } + + // Use x^0.6 to give better spread + prediction := math.Pow(float64(hits)/float64(len(b)), 0.6) + + // Calculate histogram distribution + variance := float64(0) + avg := float64(len(b)) / 256 + + for _, v := range hist { + Δ := float64(v) - avg + variance += Δ * Δ + } + + stddev := math.Sqrt(float64(variance)) / float64(len(b)) + exp := math.Sqrt(1 / float64(len(b))) + + // Subtract expected stddev + stddev -= exp + if stddev < 0 { + stddev = 0 + } + stddev *= 1 + exp + + // Use x^0.4 to give better spread + entropy := math.Pow(stddev, 0.4) + + // 50/50 weight between prediction and histogram distribution + return math.Pow((prediction+entropy)/2, 0.9) +} + +// ShannonEntropyBits returns the number of bits minimum required to represent +// an entropy encoding of the input bytes. +// https://en.wiktionary.org/wiki/Shannon_entropy +func ShannonEntropyBits(b []byte) int { + if len(b) == 0 { + return 0 + } + var hist [256]int + for _, c := range b { + hist[c]++ + } + shannon := float64(0) + invTotal := 1.0 / float64(len(b)) + for _, v := range hist[:] { + if v > 0 { + n := float64(v) + shannon += math.Ceil(-math.Log2(n*invTotal) * n) + } + } + return int(math.Ceil(shannon)) +} diff --git a/vendor/github.com/klauspost/compress/flate/deflate.go b/vendor/github.com/klauspost/compress/flate/deflate.go new file mode 100644 index 0000000..f843599 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/deflate.go @@ -0,0 +1,903 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright (c) 2015 Klaus Post +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "encoding/binary" + "fmt" + "io" + "math" +) + +const ( + NoCompression = 0 + BestSpeed = 1 + BestCompression = 9 + DefaultCompression = -1 + + // HuffmanOnly disables Lempel-Ziv match searching and only performs Huffman + // entropy encoding. This mode is useful in compressing data that has + // already been compressed with an LZ style algorithm (e.g. Snappy or LZ4) + // that lacks an entropy encoder. Compression gains are achieved when + // certain bytes in the input stream occur more frequently than others. + // + // Note that HuffmanOnly produces a compressed output that is + // RFC 1951 compliant. That is, any valid DEFLATE decompressor will + // continue to be able to decompress this output. + HuffmanOnly = -2 + ConstantCompression = HuffmanOnly // compatibility alias. + + logWindowSize = 15 + windowSize = 1 << logWindowSize + windowMask = windowSize - 1 + logMaxOffsetSize = 15 // Standard DEFLATE + minMatchLength = 4 // The smallest match that the compressor looks for + maxMatchLength = 258 // The longest match for the compressor + minOffsetSize = 1 // The shortest offset that makes any sense + + // The maximum number of tokens we will encode at the time. + // Smaller sizes usually creates less optimal blocks. + // Bigger can make context switching slow. + // We use this for levels 7-9, so we make it big. + maxFlateBlockTokens = 1 << 15 + maxStoreBlockSize = 65535 + hashBits = 17 // After 17 performance degrades + hashSize = 1 << hashBits + hashMask = (1 << hashBits) - 1 + hashShift = (hashBits + minMatchLength - 1) / minMatchLength + maxHashOffset = 1 << 28 + + skipNever = math.MaxInt32 + + debugDeflate = false +) + +type compressionLevel struct { + good, lazy, nice, chain, fastSkipHashing, level int +} + +// Compression levels have been rebalanced from zlib deflate defaults +// to give a bigger spread in speed and compression. +// See https://blog.klauspost.com/rebalancing-deflate-compression-levels/ +var levels = []compressionLevel{ + {}, // 0 + // Level 1-6 uses specialized algorithm - values not used + {0, 0, 0, 0, 0, 1}, + {0, 0, 0, 0, 0, 2}, + {0, 0, 0, 0, 0, 3}, + {0, 0, 0, 0, 0, 4}, + {0, 0, 0, 0, 0, 5}, + {0, 0, 0, 0, 0, 6}, + // Levels 7-9 use increasingly more lazy matching + // and increasingly stringent conditions for "good enough". + {8, 12, 16, 24, skipNever, 7}, + {16, 30, 40, 64, skipNever, 8}, + {32, 258, 258, 1024, skipNever, 9}, +} + +// advancedState contains state for the advanced levels, with bigger hash tables, etc. +type advancedState struct { + // deflate state + length int + offset int + maxInsertIndex int + chainHead int + hashOffset int + + ii uint16 // position of last match, intended to overflow to reset. + + // input window: unprocessed data is window[index:windowEnd] + index int + estBitsPerByte int + hashMatch [maxMatchLength + minMatchLength]uint32 + + // Input hash chains + // hashHead[hashValue] contains the largest inputIndex with the specified hash value + // If hashHead[hashValue] is within the current window, then + // hashPrev[hashHead[hashValue] & windowMask] contains the previous index + // with the same hash value. + hashHead [hashSize]uint32 + hashPrev [windowSize]uint32 +} + +type compressor struct { + compressionLevel + + h *huffmanEncoder + w *huffmanBitWriter + + // compression algorithm + fill func(*compressor, []byte) int // copy data to window + step func(*compressor) // process window + + window []byte + windowEnd int + blockStart int // window index where current tokens start + err error + + // queued output tokens + tokens tokens + fast fastEnc + state *advancedState + + sync bool // requesting flush + byteAvailable bool // if true, still need to process window[index-1]. +} + +func (d *compressor) fillDeflate(b []byte) int { + s := d.state + if s.index >= 2*windowSize-(minMatchLength+maxMatchLength) { + // shift the window by windowSize + copy(d.window[:], d.window[windowSize:2*windowSize]) + s.index -= windowSize + d.windowEnd -= windowSize + if d.blockStart >= windowSize { + d.blockStart -= windowSize + } else { + d.blockStart = math.MaxInt32 + } + s.hashOffset += windowSize + if s.hashOffset > maxHashOffset { + delta := s.hashOffset - 1 + s.hashOffset -= delta + s.chainHead -= delta + // Iterate over slices instead of arrays to avoid copying + // the entire table onto the stack (Issue #18625). + for i, v := range s.hashPrev[:] { + if int(v) > delta { + s.hashPrev[i] = uint32(int(v) - delta) + } else { + s.hashPrev[i] = 0 + } + } + for i, v := range s.hashHead[:] { + if int(v) > delta { + s.hashHead[i] = uint32(int(v) - delta) + } else { + s.hashHead[i] = 0 + } + } + } + } + n := copy(d.window[d.windowEnd:], b) + d.windowEnd += n + return n +} + +func (d *compressor) writeBlock(tok *tokens, index int, eof bool) error { + if index > 0 || eof { + var window []byte + if d.blockStart <= index { + window = d.window[d.blockStart:index] + } + d.blockStart = index + //d.w.writeBlock(tok, eof, window) + d.w.writeBlockDynamic(tok, eof, window, d.sync) + return d.w.err + } + return nil +} + +// writeBlockSkip writes the current block and uses the number of tokens +// to determine if the block should be stored on no matches, or +// only huffman encoded. +func (d *compressor) writeBlockSkip(tok *tokens, index int, eof bool) error { + if index > 0 || eof { + if d.blockStart <= index { + window := d.window[d.blockStart:index] + // If we removed less than a 64th of all literals + // we huffman compress the block. + if int(tok.n) > len(window)-int(tok.n>>6) { + d.w.writeBlockHuff(eof, window, d.sync) + } else { + // Write a dynamic huffman block. + d.w.writeBlockDynamic(tok, eof, window, d.sync) + } + } else { + d.w.writeBlock(tok, eof, nil) + } + d.blockStart = index + return d.w.err + } + return nil +} + +// fillWindow will fill the current window with the supplied +// dictionary and calculate all hashes. +// This is much faster than doing a full encode. +// Should only be used after a start/reset. +func (d *compressor) fillWindow(b []byte) { + // Do not fill window if we are in store-only or huffman mode. + if d.level <= 0 { + return + } + if d.fast != nil { + // encode the last data, but discard the result + if len(b) > maxMatchOffset { + b = b[len(b)-maxMatchOffset:] + } + d.fast.Encode(&d.tokens, b) + d.tokens.Reset() + return + } + s := d.state + // If we are given too much, cut it. + if len(b) > windowSize { + b = b[len(b)-windowSize:] + } + // Add all to window. + n := copy(d.window[d.windowEnd:], b) + + // Calculate 256 hashes at the time (more L1 cache hits) + loops := (n + 256 - minMatchLength) / 256 + for j := 0; j < loops; j++ { + startindex := j * 256 + end := startindex + 256 + minMatchLength - 1 + if end > n { + end = n + } + tocheck := d.window[startindex:end] + dstSize := len(tocheck) - minMatchLength + 1 + + if dstSize <= 0 { + continue + } + + dst := s.hashMatch[:dstSize] + bulkHash4(tocheck, dst) + var newH uint32 + for i, val := range dst { + di := i + startindex + newH = val & hashMask + // Get previous value with the same hash. + // Our chain should point to the previous value. + s.hashPrev[di&windowMask] = s.hashHead[newH] + // Set the head of the hash chain to us. + s.hashHead[newH] = uint32(di + s.hashOffset) + } + } + // Update window information. + d.windowEnd += n + s.index = n +} + +// Try to find a match starting at index whose length is greater than prevSize. +// We only look at chainCount possibilities before giving up. +// pos = s.index, prevHead = s.chainHead-s.hashOffset, prevLength=minMatchLength-1, lookahead +func (d *compressor) findMatch(pos int, prevHead int, lookahead int) (length, offset int, ok bool) { + minMatchLook := maxMatchLength + if lookahead < minMatchLook { + minMatchLook = lookahead + } + + win := d.window[0 : pos+minMatchLook] + + // We quit when we get a match that's at least nice long + nice := len(win) - pos + if d.nice < nice { + nice = d.nice + } + + // If we've got a match that's good enough, only look in 1/4 the chain. + tries := d.chain + length = minMatchLength - 1 + + wEnd := win[pos+length] + wPos := win[pos:] + minIndex := pos - windowSize + if minIndex < 0 { + minIndex = 0 + } + offset = 0 + + cGain := 0 + if d.chain < 100 { + for i := prevHead; tries > 0; tries-- { + if wEnd == win[i+length] { + n := matchLen(win[i:i+minMatchLook], wPos) + if n > length { + length = n + offset = pos - i + ok = true + if n >= nice { + // The match is good enough that we don't try to find a better one. + break + } + wEnd = win[pos+n] + } + } + if i <= minIndex { + // hashPrev[i & windowMask] has already been overwritten, so stop now. + break + } + i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset + if i < minIndex { + break + } + } + return + } + + // Some like it higher (CSV), some like it lower (JSON) + const baseCost = 6 + // Base is 4 bytes at with an additional cost. + // Matches must be better than this. + for i := prevHead; tries > 0; tries-- { + if wEnd == win[i+length] { + n := matchLen(win[i:i+minMatchLook], wPos) + if n > length { + // Calculate gain. Estimate + newGain := d.h.bitLengthRaw(wPos[:n]) - int(offsetExtraBits[offsetCode(uint32(pos-i))]) - baseCost - int(lengthExtraBits[lengthCodes[(n-3)&255]]) + + //fmt.Println(n, "gain:", newGain, "prev:", cGain, "raw:", d.h.bitLengthRaw(wPos[:n])) + if newGain > cGain { + length = n + offset = pos - i + cGain = newGain + ok = true + if n >= nice { + // The match is good enough that we don't try to find a better one. + break + } + wEnd = win[pos+n] + } + } + } + if i <= minIndex { + // hashPrev[i & windowMask] has already been overwritten, so stop now. + break + } + i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset + if i < minIndex { + break + } + } + return +} + +func (d *compressor) writeStoredBlock(buf []byte) error { + if d.w.writeStoredHeader(len(buf), false); d.w.err != nil { + return d.w.err + } + d.w.writeBytes(buf) + return d.w.err +} + +// hash4 returns a hash representation of the first 4 bytes +// of the supplied slice. +// The caller must ensure that len(b) >= 4. +func hash4(b []byte) uint32 { + return hash4u(binary.LittleEndian.Uint32(b), hashBits) +} + +// bulkHash4 will compute hashes using the same +// algorithm as hash4 +func bulkHash4(b []byte, dst []uint32) { + if len(b) < 4 { + return + } + hb := binary.LittleEndian.Uint32(b) + + dst[0] = hash4u(hb, hashBits) + end := len(b) - 4 + 1 + for i := 1; i < end; i++ { + hb = (hb >> 8) | uint32(b[i+3])<<24 + dst[i] = hash4u(hb, hashBits) + } +} + +func (d *compressor) initDeflate() { + d.window = make([]byte, 2*windowSize) + d.byteAvailable = false + d.err = nil + if d.state == nil { + return + } + s := d.state + s.index = 0 + s.hashOffset = 1 + s.length = minMatchLength - 1 + s.offset = 0 + s.chainHead = -1 +} + +// deflateLazy is the same as deflate, but with d.fastSkipHashing == skipNever, +// meaning it always has lazy matching on. +func (d *compressor) deflateLazy() { + s := d.state + // Sanity enables additional runtime tests. + // It's intended to be used during development + // to supplement the currently ad-hoc unit tests. + const sanity = debugDeflate + + if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync { + return + } + if d.windowEnd != s.index && d.chain > 100 { + // Get literal huffman coder. + if d.h == nil { + d.h = newHuffmanEncoder(maxFlateBlockTokens) + } + var tmp [256]uint16 + for _, v := range d.window[s.index:d.windowEnd] { + tmp[v]++ + } + d.h.generate(tmp[:], 15) + } + + s.maxInsertIndex = d.windowEnd - (minMatchLength - 1) + + for { + if sanity && s.index > d.windowEnd { + panic("index > windowEnd") + } + lookahead := d.windowEnd - s.index + if lookahead < minMatchLength+maxMatchLength { + if !d.sync { + return + } + if sanity && s.index > d.windowEnd { + panic("index > windowEnd") + } + if lookahead == 0 { + // Flush current output block if any. + if d.byteAvailable { + // There is still one pending token that needs to be flushed + d.tokens.AddLiteral(d.window[s.index-1]) + d.byteAvailable = false + } + if d.tokens.n > 0 { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + return + } + } + if s.index < s.maxInsertIndex { + // Update the hash + hash := hash4(d.window[s.index:]) + ch := s.hashHead[hash] + s.chainHead = int(ch) + s.hashPrev[s.index&windowMask] = ch + s.hashHead[hash] = uint32(s.index + s.hashOffset) + } + prevLength := s.length + prevOffset := s.offset + s.length = minMatchLength - 1 + s.offset = 0 + minIndex := s.index - windowSize + if minIndex < 0 { + minIndex = 0 + } + + if s.chainHead-s.hashOffset >= minIndex && lookahead > prevLength && prevLength < d.lazy { + if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, lookahead); ok { + s.length = newLength + s.offset = newOffset + } + } + + if prevLength >= minMatchLength && s.length <= prevLength { + // Check for better match at end... + // + // checkOff must be >=2 since we otherwise risk checking s.index + // Offset of 2 seems to yield best results. + const checkOff = 2 + prevIndex := s.index - 1 + if prevIndex+prevLength+checkOff < s.maxInsertIndex { + end := lookahead + if lookahead > maxMatchLength { + end = maxMatchLength + } + end += prevIndex + idx := prevIndex + prevLength - (4 - checkOff) + h := hash4(d.window[idx:]) + ch2 := int(s.hashHead[h]) - s.hashOffset - prevLength + (4 - checkOff) + if ch2 > minIndex { + length := matchLen(d.window[prevIndex:end], d.window[ch2:]) + // It seems like a pure length metric is best. + if length > prevLength { + prevLength = length + prevOffset = prevIndex - ch2 + } + } + } + // There was a match at the previous step, and the current match is + // not better. Output the previous match. + d.tokens.AddMatch(uint32(prevLength-3), uint32(prevOffset-minOffsetSize)) + + // Insert in the hash table all strings up to the end of the match. + // index and index-1 are already inserted. If there is not enough + // lookahead, the last two strings are not inserted into the hash + // table. + newIndex := s.index + prevLength - 1 + // Calculate missing hashes + end := newIndex + if end > s.maxInsertIndex { + end = s.maxInsertIndex + } + end += minMatchLength - 1 + startindex := s.index + 1 + if startindex > s.maxInsertIndex { + startindex = s.maxInsertIndex + } + tocheck := d.window[startindex:end] + dstSize := len(tocheck) - minMatchLength + 1 + if dstSize > 0 { + dst := s.hashMatch[:dstSize] + bulkHash4(tocheck, dst) + var newH uint32 + for i, val := range dst { + di := i + startindex + newH = val & hashMask + // Get previous value with the same hash. + // Our chain should point to the previous value. + s.hashPrev[di&windowMask] = s.hashHead[newH] + // Set the head of the hash chain to us. + s.hashHead[newH] = uint32(di + s.hashOffset) + } + } + + s.index = newIndex + d.byteAvailable = false + s.length = minMatchLength - 1 + if d.tokens.n == maxFlateBlockTokens { + // The block includes the current character + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + s.ii = 0 + } else { + // Reset, if we got a match this run. + if s.length >= minMatchLength { + s.ii = 0 + } + // We have a byte waiting. Emit it. + if d.byteAvailable { + s.ii++ + d.tokens.AddLiteral(d.window[s.index-1]) + if d.tokens.n == maxFlateBlockTokens { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + s.index++ + + // If we have a long run of no matches, skip additional bytes + // Resets when s.ii overflows after 64KB. + if n := int(s.ii) - d.chain; n > 0 { + n = 1 + int(n>>6) + for j := 0; j < n; j++ { + if s.index >= d.windowEnd-1 { + break + } + d.tokens.AddLiteral(d.window[s.index-1]) + if d.tokens.n == maxFlateBlockTokens { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + // Index... + if s.index < s.maxInsertIndex { + h := hash4(d.window[s.index:]) + ch := s.hashHead[h] + s.chainHead = int(ch) + s.hashPrev[s.index&windowMask] = ch + s.hashHead[h] = uint32(s.index + s.hashOffset) + } + s.index++ + } + // Flush last byte + d.tokens.AddLiteral(d.window[s.index-1]) + d.byteAvailable = false + // s.length = minMatchLength - 1 // not needed, since s.ii is reset above, so it should never be > minMatchLength + if d.tokens.n == maxFlateBlockTokens { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + } + } else { + s.index++ + d.byteAvailable = true + } + } + } +} + +func (d *compressor) store() { + if d.windowEnd > 0 && (d.windowEnd == maxStoreBlockSize || d.sync) { + d.err = d.writeStoredBlock(d.window[:d.windowEnd]) + d.windowEnd = 0 + } +} + +// fillWindow will fill the buffer with data for huffman-only compression. +// The number of bytes copied is returned. +func (d *compressor) fillBlock(b []byte) int { + n := copy(d.window[d.windowEnd:], b) + d.windowEnd += n + return n +} + +// storeHuff will compress and store the currently added data, +// if enough has been accumulated or we at the end of the stream. +// Any error that occurred will be in d.err +func (d *compressor) storeHuff() { + if d.windowEnd < len(d.window) && !d.sync || d.windowEnd == 0 { + return + } + d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync) + d.err = d.w.err + d.windowEnd = 0 +} + +// storeFast will compress and store the currently added data, +// if enough has been accumulated or we at the end of the stream. +// Any error that occurred will be in d.err +func (d *compressor) storeFast() { + // We only compress if we have maxStoreBlockSize. + if d.windowEnd < len(d.window) { + if !d.sync { + return + } + // Handle extremely small sizes. + if d.windowEnd < 128 { + if d.windowEnd == 0 { + return + } + if d.windowEnd <= 32 { + d.err = d.writeStoredBlock(d.window[:d.windowEnd]) + } else { + d.w.writeBlockHuff(false, d.window[:d.windowEnd], true) + d.err = d.w.err + } + d.tokens.Reset() + d.windowEnd = 0 + d.fast.Reset() + return + } + } + + d.fast.Encode(&d.tokens, d.window[:d.windowEnd]) + // If we made zero matches, store the block as is. + if d.tokens.n == 0 { + d.err = d.writeStoredBlock(d.window[:d.windowEnd]) + // If we removed less than 1/16th, huffman compress the block. + } else if int(d.tokens.n) > d.windowEnd-(d.windowEnd>>4) { + d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync) + d.err = d.w.err + } else { + d.w.writeBlockDynamic(&d.tokens, false, d.window[:d.windowEnd], d.sync) + d.err = d.w.err + } + d.tokens.Reset() + d.windowEnd = 0 +} + +// write will add input byte to the stream. +// Unless an error occurs all bytes will be consumed. +func (d *compressor) write(b []byte) (n int, err error) { + if d.err != nil { + return 0, d.err + } + n = len(b) + for len(b) > 0 { + if d.windowEnd == len(d.window) || d.sync { + d.step(d) + } + b = b[d.fill(d, b):] + if d.err != nil { + return 0, d.err + } + } + return n, d.err +} + +func (d *compressor) syncFlush() error { + d.sync = true + if d.err != nil { + return d.err + } + d.step(d) + if d.err == nil { + d.w.writeStoredHeader(0, false) + d.w.flush() + d.err = d.w.err + } + d.sync = false + return d.err +} + +func (d *compressor) init(w io.Writer, level int) (err error) { + d.w = newHuffmanBitWriter(w) + + switch { + case level == NoCompression: + d.window = make([]byte, maxStoreBlockSize) + d.fill = (*compressor).fillBlock + d.step = (*compressor).store + case level == ConstantCompression: + d.w.logNewTablePenalty = 10 + d.window = make([]byte, 32<<10) + d.fill = (*compressor).fillBlock + d.step = (*compressor).storeHuff + case level == DefaultCompression: + level = 5 + fallthrough + case level >= 1 && level <= 6: + d.w.logNewTablePenalty = 7 + d.fast = newFastEnc(level) + d.window = make([]byte, maxStoreBlockSize) + d.fill = (*compressor).fillBlock + d.step = (*compressor).storeFast + case 7 <= level && level <= 9: + d.w.logNewTablePenalty = 8 + d.state = &advancedState{} + d.compressionLevel = levels[level] + d.initDeflate() + d.fill = (*compressor).fillDeflate + d.step = (*compressor).deflateLazy + default: + return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level) + } + d.level = level + return nil +} + +// reset the state of the compressor. +func (d *compressor) reset(w io.Writer) { + d.w.reset(w) + d.sync = false + d.err = nil + // We only need to reset a few things for Snappy. + if d.fast != nil { + d.fast.Reset() + d.windowEnd = 0 + d.tokens.Reset() + return + } + switch d.compressionLevel.chain { + case 0: + // level was NoCompression or ConstantCompresssion. + d.windowEnd = 0 + default: + s := d.state + s.chainHead = -1 + for i := range s.hashHead { + s.hashHead[i] = 0 + } + for i := range s.hashPrev { + s.hashPrev[i] = 0 + } + s.hashOffset = 1 + s.index, d.windowEnd = 0, 0 + d.blockStart, d.byteAvailable = 0, false + d.tokens.Reset() + s.length = minMatchLength - 1 + s.offset = 0 + s.ii = 0 + s.maxInsertIndex = 0 + } +} + +func (d *compressor) close() error { + if d.err != nil { + return d.err + } + d.sync = true + d.step(d) + if d.err != nil { + return d.err + } + if d.w.writeStoredHeader(0, true); d.w.err != nil { + return d.w.err + } + d.w.flush() + d.w.reset(nil) + return d.w.err +} + +// NewWriter returns a new Writer compressing data at the given level. +// Following zlib, levels range from 1 (BestSpeed) to 9 (BestCompression); +// higher levels typically run slower but compress more. +// Level 0 (NoCompression) does not attempt any compression; it only adds the +// necessary DEFLATE framing. +// Level -1 (DefaultCompression) uses the default compression level. +// Level -2 (ConstantCompression) will use Huffman compression only, giving +// a very fast compression for all types of input, but sacrificing considerable +// compression efficiency. +// +// If level is in the range [-2, 9] then the error returned will be nil. +// Otherwise the error returned will be non-nil. +func NewWriter(w io.Writer, level int) (*Writer, error) { + var dw Writer + if err := dw.d.init(w, level); err != nil { + return nil, err + } + return &dw, nil +} + +// NewWriterDict is like NewWriter but initializes the new +// Writer with a preset dictionary. The returned Writer behaves +// as if the dictionary had been written to it without producing +// any compressed output. The compressed data written to w +// can only be decompressed by a Reader initialized with the +// same dictionary. +func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) { + zw, err := NewWriter(w, level) + if err != nil { + return nil, err + } + zw.d.fillWindow(dict) + zw.dict = append(zw.dict, dict...) // duplicate dictionary for Reset method. + return zw, err +} + +// A Writer takes data written to it and writes the compressed +// form of that data to an underlying writer (see NewWriter). +type Writer struct { + d compressor + dict []byte +} + +// Write writes data to w, which will eventually write the +// compressed form of data to its underlying writer. +func (w *Writer) Write(data []byte) (n int, err error) { + return w.d.write(data) +} + +// Flush flushes any pending data to the underlying writer. +// It is useful mainly in compressed network protocols, to ensure that +// a remote reader has enough data to reconstruct a packet. +// Flush does not return until the data has been written. +// Calling Flush when there is no pending data still causes the Writer +// to emit a sync marker of at least 4 bytes. +// If the underlying writer returns an error, Flush returns that error. +// +// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH. +func (w *Writer) Flush() error { + // For more about flushing: + // http://www.bolet.org/~pornin/deflate-flush.html + return w.d.syncFlush() +} + +// Close flushes and closes the writer. +func (w *Writer) Close() error { + return w.d.close() +} + +// Reset discards the writer's state and makes it equivalent to +// the result of NewWriter or NewWriterDict called with dst +// and w's level and dictionary. +func (w *Writer) Reset(dst io.Writer) { + if len(w.dict) > 0 { + // w was created with NewWriterDict + w.d.reset(dst) + if dst != nil { + w.d.fillWindow(w.dict) + } + } else { + // w was created with NewWriter + w.d.reset(dst) + } +} + +// ResetDict discards the writer's state and makes it equivalent to +// the result of NewWriter or NewWriterDict called with dst +// and w's level, but sets a specific dictionary. +func (w *Writer) ResetDict(dst io.Writer, dict []byte) { + w.dict = dict + w.d.reset(dst) + w.d.fillWindow(w.dict) +} diff --git a/vendor/github.com/klauspost/compress/flate/dict_decoder.go b/vendor/github.com/klauspost/compress/flate/dict_decoder.go new file mode 100644 index 0000000..71c75a0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/dict_decoder.go @@ -0,0 +1,184 @@ +// Copyright 2016 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +// dictDecoder implements the LZ77 sliding dictionary as used in decompression. +// LZ77 decompresses data through sequences of two forms of commands: +// +// * Literal insertions: Runs of one or more symbols are inserted into the data +// stream as is. This is accomplished through the writeByte method for a +// single symbol, or combinations of writeSlice/writeMark for multiple symbols. +// Any valid stream must start with a literal insertion if no preset dictionary +// is used. +// +// * Backward copies: Runs of one or more symbols are copied from previously +// emitted data. Backward copies come as the tuple (dist, length) where dist +// determines how far back in the stream to copy from and length determines how +// many bytes to copy. Note that it is valid for the length to be greater than +// the distance. Since LZ77 uses forward copies, that situation is used to +// perform a form of run-length encoding on repeated runs of symbols. +// The writeCopy and tryWriteCopy are used to implement this command. +// +// For performance reasons, this implementation performs little to no sanity +// checks about the arguments. As such, the invariants documented for each +// method call must be respected. +type dictDecoder struct { + hist []byte // Sliding window history + + // Invariant: 0 <= rdPos <= wrPos <= len(hist) + wrPos int // Current output position in buffer + rdPos int // Have emitted hist[:rdPos] already + full bool // Has a full window length been written yet? +} + +// init initializes dictDecoder to have a sliding window dictionary of the given +// size. If a preset dict is provided, it will initialize the dictionary with +// the contents of dict. +func (dd *dictDecoder) init(size int, dict []byte) { + *dd = dictDecoder{hist: dd.hist} + + if cap(dd.hist) < size { + dd.hist = make([]byte, size) + } + dd.hist = dd.hist[:size] + + if len(dict) > len(dd.hist) { + dict = dict[len(dict)-len(dd.hist):] + } + dd.wrPos = copy(dd.hist, dict) + if dd.wrPos == len(dd.hist) { + dd.wrPos = 0 + dd.full = true + } + dd.rdPos = dd.wrPos +} + +// histSize reports the total amount of historical data in the dictionary. +func (dd *dictDecoder) histSize() int { + if dd.full { + return len(dd.hist) + } + return dd.wrPos +} + +// availRead reports the number of bytes that can be flushed by readFlush. +func (dd *dictDecoder) availRead() int { + return dd.wrPos - dd.rdPos +} + +// availWrite reports the available amount of output buffer space. +func (dd *dictDecoder) availWrite() int { + return len(dd.hist) - dd.wrPos +} + +// writeSlice returns a slice of the available buffer to write data to. +// +// This invariant will be kept: len(s) <= availWrite() +func (dd *dictDecoder) writeSlice() []byte { + return dd.hist[dd.wrPos:] +} + +// writeMark advances the writer pointer by cnt. +// +// This invariant must be kept: 0 <= cnt <= availWrite() +func (dd *dictDecoder) writeMark(cnt int) { + dd.wrPos += cnt +} + +// writeByte writes a single byte to the dictionary. +// +// This invariant must be kept: 0 < availWrite() +func (dd *dictDecoder) writeByte(c byte) { + dd.hist[dd.wrPos] = c + dd.wrPos++ +} + +// writeCopy copies a string at a given (dist, length) to the output. +// This returns the number of bytes copied and may be less than the requested +// length if the available space in the output buffer is too small. +// +// This invariant must be kept: 0 < dist <= histSize() +func (dd *dictDecoder) writeCopy(dist, length int) int { + dstBase := dd.wrPos + dstPos := dstBase + srcPos := dstPos - dist + endPos := dstPos + length + if endPos > len(dd.hist) { + endPos = len(dd.hist) + } + + // Copy non-overlapping section after destination position. + // + // This section is non-overlapping in that the copy length for this section + // is always less than or equal to the backwards distance. This can occur + // if a distance refers to data that wraps-around in the buffer. + // Thus, a backwards copy is performed here; that is, the exact bytes in + // the source prior to the copy is placed in the destination. + if srcPos < 0 { + srcPos += len(dd.hist) + dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:]) + srcPos = 0 + } + + // Copy possibly overlapping section before destination position. + // + // This section can overlap if the copy length for this section is larger + // than the backwards distance. This is allowed by LZ77 so that repeated + // strings can be succinctly represented using (dist, length) pairs. + // Thus, a forwards copy is performed here; that is, the bytes copied is + // possibly dependent on the resulting bytes in the destination as the copy + // progresses along. This is functionally equivalent to the following: + // + // for i := 0; i < endPos-dstPos; i++ { + // dd.hist[dstPos+i] = dd.hist[srcPos+i] + // } + // dstPos = endPos + // + for dstPos < endPos { + dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos]) + } + + dd.wrPos = dstPos + return dstPos - dstBase +} + +// tryWriteCopy tries to copy a string at a given (distance, length) to the +// output. This specialized version is optimized for short distances. +// +// This method is designed to be inlined for performance reasons. +// +// This invariant must be kept: 0 < dist <= histSize() +func (dd *dictDecoder) tryWriteCopy(dist, length int) int { + dstPos := dd.wrPos + endPos := dstPos + length + if dstPos < dist || endPos > len(dd.hist) { + return 0 + } + dstBase := dstPos + srcPos := dstPos - dist + + // Copy possibly overlapping section before destination position. +loop: + dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos]) + if dstPos < endPos { + goto loop // Avoid for-loop so that this function can be inlined + } + + dd.wrPos = dstPos + return dstPos - dstBase +} + +// readFlush returns a slice of the historical buffer that is ready to be +// emitted to the user. The data returned by readFlush must be fully consumed +// before calling any other dictDecoder methods. +func (dd *dictDecoder) readFlush() []byte { + toRead := dd.hist[dd.rdPos:dd.wrPos] + dd.rdPos = dd.wrPos + if dd.wrPos == len(dd.hist) { + dd.wrPos, dd.rdPos = 0, 0 + dd.full = true + } + return toRead +} diff --git a/vendor/github.com/klauspost/compress/flate/fast_encoder.go b/vendor/github.com/klauspost/compress/flate/fast_encoder.go new file mode 100644 index 0000000..f781aaa --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/fast_encoder.go @@ -0,0 +1,233 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Modified for deflate by Klaus Post (c) 2015. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "encoding/binary" + "fmt" + "math/bits" +) + +type fastEnc interface { + Encode(dst *tokens, src []byte) + Reset() +} + +func newFastEnc(level int) fastEnc { + switch level { + case 1: + return &fastEncL1{fastGen: fastGen{cur: maxStoreBlockSize}} + case 2: + return &fastEncL2{fastGen: fastGen{cur: maxStoreBlockSize}} + case 3: + return &fastEncL3{fastGen: fastGen{cur: maxStoreBlockSize}} + case 4: + return &fastEncL4{fastGen: fastGen{cur: maxStoreBlockSize}} + case 5: + return &fastEncL5{fastGen: fastGen{cur: maxStoreBlockSize}} + case 6: + return &fastEncL6{fastGen: fastGen{cur: maxStoreBlockSize}} + default: + panic("invalid level specified") + } +} + +const ( + tableBits = 15 // Bits used in the table + tableSize = 1 << tableBits // Size of the table + tableShift = 32 - tableBits // Right-shift to get the tableBits most significant bits of a uint32. + baseMatchOffset = 1 // The smallest match offset + baseMatchLength = 3 // The smallest match length per the RFC section 3.2.5 + maxMatchOffset = 1 << 15 // The largest match offset + + bTableBits = 17 // Bits used in the big tables + bTableSize = 1 << bTableBits // Size of the table + allocHistory = maxStoreBlockSize * 5 // Size to preallocate for history. + bufferReset = (1 << 31) - allocHistory - maxStoreBlockSize - 1 // Reset the buffer offset when reaching this. +) + +const ( + prime3bytes = 506832829 + prime4bytes = 2654435761 + prime5bytes = 889523592379 + prime6bytes = 227718039650203 + prime7bytes = 58295818150454627 + prime8bytes = 0xcf1bbcdcb7a56463 +) + +func load32(b []byte, i int) uint32 { + // Help the compiler eliminate bounds checks on the read so it can be done in a single read. + b = b[i:] + b = b[:4] + return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 +} + +func load64(b []byte, i int) uint64 { + return binary.LittleEndian.Uint64(b[i:]) +} + +func load3232(b []byte, i int32) uint32 { + return binary.LittleEndian.Uint32(b[i:]) +} + +func load6432(b []byte, i int32) uint64 { + return binary.LittleEndian.Uint64(b[i:]) +} + +func hash(u uint32) uint32 { + return (u * 0x1e35a7bd) >> tableShift +} + +type tableEntry struct { + offset int32 +} + +// fastGen maintains the table for matches, +// and the previous byte block for level 2. +// This is the generic implementation. +type fastGen struct { + hist []byte + cur int32 +} + +func (e *fastGen) addBlock(src []byte) int32 { + // check if we have space already + if len(e.hist)+len(src) > cap(e.hist) { + if cap(e.hist) == 0 { + e.hist = make([]byte, 0, allocHistory) + } else { + if cap(e.hist) < maxMatchOffset*2 { + panic("unexpected buffer size") + } + // Move down + offset := int32(len(e.hist)) - maxMatchOffset + copy(e.hist[0:maxMatchOffset], e.hist[offset:]) + e.cur += offset + e.hist = e.hist[:maxMatchOffset] + } + } + s := int32(len(e.hist)) + e.hist = append(e.hist, src...) + return s +} + +// hash4 returns the hash of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <32. +func hash4u(u uint32, h uint8) uint32 { + return (u * prime4bytes) >> (32 - h) +} + +type tableEntryPrev struct { + Cur tableEntry + Prev tableEntry +} + +// hash4x64 returns the hash of the lowest 4 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <32. +func hash4x64(u uint64, h uint8) uint32 { + return (uint32(u) * prime4bytes) >> ((32 - h) & reg8SizeMask32) +} + +// hash7 returns the hash of the lowest 7 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash7(u uint64, h uint8) uint32 { + return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & reg8SizeMask64)) +} + +// hash8 returns the hash of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash8(u uint64, h uint8) uint32 { + return uint32((u * prime8bytes) >> ((64 - h) & reg8SizeMask64)) +} + +// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash6(u uint64, h uint8) uint32 { + return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & reg8SizeMask64)) +} + +// matchlen will return the match length between offsets and t in src. +// The maximum length returned is maxMatchLength - 4. +// It is assumed that s > t, that t >=0 and s < len(src). +func (e *fastGen) matchlen(s, t int32, src []byte) int32 { + if debugDecode { + if t >= s { + panic(fmt.Sprint("t >=s:", t, s)) + } + if int(s) >= len(src) { + panic(fmt.Sprint("s >= len(src):", s, len(src))) + } + if t < 0 { + panic(fmt.Sprint("t < 0:", t)) + } + if s-t > maxMatchOffset { + panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")")) + } + } + s1 := int(s) + maxMatchLength - 4 + if s1 > len(src) { + s1 = len(src) + } + + // Extend the match to be as long as possible. + return int32(matchLen(src[s:s1], src[t:])) +} + +// matchlenLong will return the match length between offsets and t in src. +// It is assumed that s > t, that t >=0 and s < len(src). +func (e *fastGen) matchlenLong(s, t int32, src []byte) int32 { + if debugDeflate { + if t >= s { + panic(fmt.Sprint("t >=s:", t, s)) + } + if int(s) >= len(src) { + panic(fmt.Sprint("s >= len(src):", s, len(src))) + } + if t < 0 { + panic(fmt.Sprint("t < 0:", t)) + } + if s-t > maxMatchOffset { + panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")")) + } + } + // Extend the match to be as long as possible. + return int32(matchLen(src[s:], src[t:])) +} + +// Reset the encoding table. +func (e *fastGen) Reset() { + if cap(e.hist) < allocHistory { + e.hist = make([]byte, 0, allocHistory) + } + // We offset current position so everything will be out of reach. + // If we are above the buffer reset it will be cleared anyway since len(hist) == 0. + if e.cur <= bufferReset { + e.cur += maxMatchOffset + int32(len(e.hist)) + } + e.hist = e.hist[:0] +} + +// matchLen returns the maximum length. +// 'a' must be the shortest of the two. +func matchLen(a, b []byte) int { + var checked int + + for len(a) >= 8 { + if diff := binary.LittleEndian.Uint64(a) ^ binary.LittleEndian.Uint64(b); diff != 0 { + return checked + (bits.TrailingZeros64(diff) >> 3) + } + checked += 8 + a = a[8:] + b = b[8:] + } + b = b[:len(a)] + for i := range a { + if a[i] != b[i] { + return i + checked + } + } + return len(a) + checked +} diff --git a/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go new file mode 100644 index 0000000..40ef45c --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go @@ -0,0 +1,1185 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "encoding/binary" + "fmt" + "io" + "math" +) + +const ( + // The largest offset code. + offsetCodeCount = 30 + + // The special code used to mark the end of a block. + endBlockMarker = 256 + + // The first length code. + lengthCodesStart = 257 + + // The number of codegen codes. + codegenCodeCount = 19 + badCode = 255 + + // maxPredefinedTokens is the maximum number of tokens + // where we check if fixed size is smaller. + maxPredefinedTokens = 250 + + // bufferFlushSize indicates the buffer size + // after which bytes are flushed to the writer. + // Should preferably be a multiple of 6, since + // we accumulate 6 bytes between writes to the buffer. + bufferFlushSize = 246 + + // bufferSize is the actual output byte buffer size. + // It must have additional headroom for a flush + // which can contain up to 8 bytes. + bufferSize = bufferFlushSize + 8 +) + +// Minimum length code that emits bits. +const lengthExtraBitsMinCode = 8 + +// The number of extra bits needed by length code X - LENGTH_CODES_START. +var lengthExtraBits = [32]uint8{ + /* 257 */ 0, 0, 0, + /* 260 */ 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, + /* 270 */ 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, + /* 280 */ 4, 5, 5, 5, 5, 0, +} + +// The length indicated by length code X - LENGTH_CODES_START. +var lengthBase = [32]uint8{ + 0, 1, 2, 3, 4, 5, 6, 7, 8, 10, + 12, 14, 16, 20, 24, 28, 32, 40, 48, 56, + 64, 80, 96, 112, 128, 160, 192, 224, 255, +} + +// Minimum offset code that emits bits. +const offsetExtraBitsMinCode = 4 + +// offset code word extra bits. +var offsetExtraBits = [32]int8{ + 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, + 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, + 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, + /* extended window */ + 14, 14, +} + +var offsetCombined = [32]uint32{} + +func init() { + var offsetBase = [32]uint32{ + /* normal deflate */ + 0x000000, 0x000001, 0x000002, 0x000003, 0x000004, + 0x000006, 0x000008, 0x00000c, 0x000010, 0x000018, + 0x000020, 0x000030, 0x000040, 0x000060, 0x000080, + 0x0000c0, 0x000100, 0x000180, 0x000200, 0x000300, + 0x000400, 0x000600, 0x000800, 0x000c00, 0x001000, + 0x001800, 0x002000, 0x003000, 0x004000, 0x006000, + + /* extended window */ + 0x008000, 0x00c000, + } + + for i := range offsetCombined[:] { + // Don't use extended window values... + if offsetExtraBits[i] == 0 || offsetBase[i] > 0x006000 { + continue + } + offsetCombined[i] = uint32(offsetExtraBits[i]) | (offsetBase[i] << 8) + } +} + +// The odd order in which the codegen code sizes are written. +var codegenOrder = []uint32{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15} + +type huffmanBitWriter struct { + // writer is the underlying writer. + // Do not use it directly; use the write method, which ensures + // that Write errors are sticky. + writer io.Writer + + // Data waiting to be written is bytes[0:nbytes] + // and then the low nbits of bits. + bits uint64 + nbits uint8 + nbytes uint8 + lastHuffMan bool + literalEncoding *huffmanEncoder + tmpLitEncoding *huffmanEncoder + offsetEncoding *huffmanEncoder + codegenEncoding *huffmanEncoder + err error + lastHeader int + // Set between 0 (reused block can be up to 2x the size) + logNewTablePenalty uint + bytes [256 + 8]byte + literalFreq [lengthCodesStart + 32]uint16 + offsetFreq [32]uint16 + codegenFreq [codegenCodeCount]uint16 + + // codegen must have an extra space for the final symbol. + codegen [literalCount + offsetCodeCount + 1]uint8 +} + +// Huffman reuse. +// +// The huffmanBitWriter supports reusing huffman tables and thereby combining block sections. +// +// This is controlled by several variables: +// +// If lastHeader is non-zero the Huffman table can be reused. +// This also indicates that a Huffman table has been generated that can output all +// possible symbols. +// It also indicates that an EOB has not yet been emitted, so if a new tabel is generated +// an EOB with the previous table must be written. +// +// If lastHuffMan is set, a table for outputting literals has been generated and offsets are invalid. +// +// An incoming block estimates the output size of a new table using a 'fresh' by calculating the +// optimal size and adding a penalty in 'logNewTablePenalty'. +// A Huffman table is not optimal, which is why we add a penalty, and generating a new table +// is slower both for compression and decompression. + +func newHuffmanBitWriter(w io.Writer) *huffmanBitWriter { + return &huffmanBitWriter{ + writer: w, + literalEncoding: newHuffmanEncoder(literalCount), + tmpLitEncoding: newHuffmanEncoder(literalCount), + codegenEncoding: newHuffmanEncoder(codegenCodeCount), + offsetEncoding: newHuffmanEncoder(offsetCodeCount), + } +} + +func (w *huffmanBitWriter) reset(writer io.Writer) { + w.writer = writer + w.bits, w.nbits, w.nbytes, w.err = 0, 0, 0, nil + w.lastHeader = 0 + w.lastHuffMan = false +} + +func (w *huffmanBitWriter) canReuse(t *tokens) (ok bool) { + a := t.offHist[:offsetCodeCount] + b := w.offsetEncoding.codes + b = b[:len(a)] + for i, v := range a { + if v != 0 && b[i].zero() { + return false + } + } + + a = t.extraHist[:literalCount-256] + b = w.literalEncoding.codes[256:literalCount] + b = b[:len(a)] + for i, v := range a { + if v != 0 && b[i].zero() { + return false + } + } + + a = t.litHist[:256] + b = w.literalEncoding.codes[:len(a)] + for i, v := range a { + if v != 0 && b[i].zero() { + return false + } + } + return true +} + +func (w *huffmanBitWriter) flush() { + if w.err != nil { + w.nbits = 0 + return + } + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + n := w.nbytes + for w.nbits != 0 { + w.bytes[n] = byte(w.bits) + w.bits >>= 8 + if w.nbits > 8 { // Avoid underflow + w.nbits -= 8 + } else { + w.nbits = 0 + } + n++ + } + w.bits = 0 + w.write(w.bytes[:n]) + w.nbytes = 0 +} + +func (w *huffmanBitWriter) write(b []byte) { + if w.err != nil { + return + } + _, w.err = w.writer.Write(b) +} + +func (w *huffmanBitWriter) writeBits(b int32, nb uint8) { + w.bits |= uint64(b) << (w.nbits & 63) + w.nbits += nb + if w.nbits >= 48 { + w.writeOutBits() + } +} + +func (w *huffmanBitWriter) writeBytes(bytes []byte) { + if w.err != nil { + return + } + n := w.nbytes + if w.nbits&7 != 0 { + w.err = InternalError("writeBytes with unfinished bits") + return + } + for w.nbits != 0 { + w.bytes[n] = byte(w.bits) + w.bits >>= 8 + w.nbits -= 8 + n++ + } + if n != 0 { + w.write(w.bytes[:n]) + } + w.nbytes = 0 + w.write(bytes) +} + +// RFC 1951 3.2.7 specifies a special run-length encoding for specifying +// the literal and offset lengths arrays (which are concatenated into a single +// array). This method generates that run-length encoding. +// +// The result is written into the codegen array, and the frequencies +// of each code is written into the codegenFreq array. +// Codes 0-15 are single byte codes. Codes 16-18 are followed by additional +// information. Code badCode is an end marker +// +// numLiterals The number of literals in literalEncoding +// numOffsets The number of offsets in offsetEncoding +// litenc, offenc The literal and offset encoder to use +func (w *huffmanBitWriter) generateCodegen(numLiterals int, numOffsets int, litEnc, offEnc *huffmanEncoder) { + for i := range w.codegenFreq { + w.codegenFreq[i] = 0 + } + // Note that we are using codegen both as a temporary variable for holding + // a copy of the frequencies, and as the place where we put the result. + // This is fine because the output is always shorter than the input used + // so far. + codegen := w.codegen[:] // cache + // Copy the concatenated code sizes to codegen. Put a marker at the end. + cgnl := codegen[:numLiterals] + for i := range cgnl { + cgnl[i] = litEnc.codes[i].len() + } + + cgnl = codegen[numLiterals : numLiterals+numOffsets] + for i := range cgnl { + cgnl[i] = offEnc.codes[i].len() + } + codegen[numLiterals+numOffsets] = badCode + + size := codegen[0] + count := 1 + outIndex := 0 + for inIndex := 1; size != badCode; inIndex++ { + // INVARIANT: We have seen "count" copies of size that have not yet + // had output generated for them. + nextSize := codegen[inIndex] + if nextSize == size { + count++ + continue + } + // We need to generate codegen indicating "count" of size. + if size != 0 { + codegen[outIndex] = size + outIndex++ + w.codegenFreq[size]++ + count-- + for count >= 3 { + n := 6 + if n > count { + n = count + } + codegen[outIndex] = 16 + outIndex++ + codegen[outIndex] = uint8(n - 3) + outIndex++ + w.codegenFreq[16]++ + count -= n + } + } else { + for count >= 11 { + n := 138 + if n > count { + n = count + } + codegen[outIndex] = 18 + outIndex++ + codegen[outIndex] = uint8(n - 11) + outIndex++ + w.codegenFreq[18]++ + count -= n + } + if count >= 3 { + // count >= 3 && count <= 10 + codegen[outIndex] = 17 + outIndex++ + codegen[outIndex] = uint8(count - 3) + outIndex++ + w.codegenFreq[17]++ + count = 0 + } + } + count-- + for ; count >= 0; count-- { + codegen[outIndex] = size + outIndex++ + w.codegenFreq[size]++ + } + // Set up invariant for next time through the loop. + size = nextSize + count = 1 + } + // Marker indicating the end of the codegen. + codegen[outIndex] = badCode +} + +func (w *huffmanBitWriter) codegens() int { + numCodegens := len(w.codegenFreq) + for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 { + numCodegens-- + } + return numCodegens +} + +func (w *huffmanBitWriter) headerSize() (size, numCodegens int) { + numCodegens = len(w.codegenFreq) + for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 { + numCodegens-- + } + return 3 + 5 + 5 + 4 + (3 * numCodegens) + + w.codegenEncoding.bitLength(w.codegenFreq[:]) + + int(w.codegenFreq[16])*2 + + int(w.codegenFreq[17])*3 + + int(w.codegenFreq[18])*7, numCodegens +} + +// dynamicSize returns the size of dynamically encoded data in bits. +func (w *huffmanBitWriter) dynamicReuseSize(litEnc, offEnc *huffmanEncoder) (size int) { + size = litEnc.bitLength(w.literalFreq[:]) + + offEnc.bitLength(w.offsetFreq[:]) + return size +} + +// dynamicSize returns the size of dynamically encoded data in bits. +func (w *huffmanBitWriter) dynamicSize(litEnc, offEnc *huffmanEncoder, extraBits int) (size, numCodegens int) { + header, numCodegens := w.headerSize() + size = header + + litEnc.bitLength(w.literalFreq[:]) + + offEnc.bitLength(w.offsetFreq[:]) + + extraBits + return size, numCodegens +} + +// extraBitSize will return the number of bits that will be written +// as "extra" bits on matches. +func (w *huffmanBitWriter) extraBitSize() int { + total := 0 + for i, n := range w.literalFreq[257:literalCount] { + total += int(n) * int(lengthExtraBits[i&31]) + } + for i, n := range w.offsetFreq[:offsetCodeCount] { + total += int(n) * int(offsetExtraBits[i&31]) + } + return total +} + +// fixedSize returns the size of dynamically encoded data in bits. +func (w *huffmanBitWriter) fixedSize(extraBits int) int { + return 3 + + fixedLiteralEncoding.bitLength(w.literalFreq[:]) + + fixedOffsetEncoding.bitLength(w.offsetFreq[:]) + + extraBits +} + +// storedSize calculates the stored size, including header. +// The function returns the size in bits and whether the block +// fits inside a single block. +func (w *huffmanBitWriter) storedSize(in []byte) (int, bool) { + if in == nil { + return 0, false + } + if len(in) <= maxStoreBlockSize { + return (len(in) + 5) * 8, true + } + return 0, false +} + +func (w *huffmanBitWriter) writeCode(c hcode) { + // The function does not get inlined if we "& 63" the shift. + w.bits |= c.code64() << (w.nbits & 63) + w.nbits += c.len() + if w.nbits >= 48 { + w.writeOutBits() + } +} + +// writeOutBits will write bits to the buffer. +func (w *huffmanBitWriter) writeOutBits() { + bits := w.bits + w.bits >>= 48 + w.nbits -= 48 + n := w.nbytes + + // We over-write, but faster... + binary.LittleEndian.PutUint64(w.bytes[n:], bits) + n += 6 + + if n >= bufferFlushSize { + if w.err != nil { + n = 0 + return + } + w.write(w.bytes[:n]) + n = 0 + } + + w.nbytes = n +} + +// Write the header of a dynamic Huffman block to the output stream. +// +// numLiterals The number of literals specified in codegen +// numOffsets The number of offsets specified in codegen +// numCodegens The number of codegens used in codegen +func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, numCodegens int, isEof bool) { + if w.err != nil { + return + } + var firstBits int32 = 4 + if isEof { + firstBits = 5 + } + w.writeBits(firstBits, 3) + w.writeBits(int32(numLiterals-257), 5) + w.writeBits(int32(numOffsets-1), 5) + w.writeBits(int32(numCodegens-4), 4) + + for i := 0; i < numCodegens; i++ { + value := uint(w.codegenEncoding.codes[codegenOrder[i]].len()) + w.writeBits(int32(value), 3) + } + + i := 0 + for { + var codeWord = uint32(w.codegen[i]) + i++ + if codeWord == badCode { + break + } + w.writeCode(w.codegenEncoding.codes[codeWord]) + + switch codeWord { + case 16: + w.writeBits(int32(w.codegen[i]), 2) + i++ + case 17: + w.writeBits(int32(w.codegen[i]), 3) + i++ + case 18: + w.writeBits(int32(w.codegen[i]), 7) + i++ + } + } +} + +// writeStoredHeader will write a stored header. +// If the stored block is only used for EOF, +// it is replaced with a fixed huffman block. +func (w *huffmanBitWriter) writeStoredHeader(length int, isEof bool) { + if w.err != nil { + return + } + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + + // To write EOF, use a fixed encoding block. 10 bits instead of 5 bytes. + if length == 0 && isEof { + w.writeFixedHeader(isEof) + // EOB: 7 bits, value: 0 + w.writeBits(0, 7) + w.flush() + return + } + + var flag int32 + if isEof { + flag = 1 + } + w.writeBits(flag, 3) + w.flush() + w.writeBits(int32(length), 16) + w.writeBits(int32(^uint16(length)), 16) +} + +func (w *huffmanBitWriter) writeFixedHeader(isEof bool) { + if w.err != nil { + return + } + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + + // Indicate that we are a fixed Huffman block + var value int32 = 2 + if isEof { + value = 3 + } + w.writeBits(value, 3) +} + +// writeBlock will write a block of tokens with the smallest encoding. +// The original input can be supplied, and if the huffman encoded data +// is larger than the original bytes, the data will be written as a +// stored block. +// If the input is nil, the tokens will always be Huffman encoded. +func (w *huffmanBitWriter) writeBlock(tokens *tokens, eof bool, input []byte) { + if w.err != nil { + return + } + + tokens.AddEOB() + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + numLiterals, numOffsets := w.indexTokens(tokens, false) + w.generate() + var extraBits int + storedSize, storable := w.storedSize(input) + if storable { + extraBits = w.extraBitSize() + } + + // Figure out smallest code. + // Fixed Huffman baseline. + var literalEncoding = fixedLiteralEncoding + var offsetEncoding = fixedOffsetEncoding + var size = math.MaxInt32 + if tokens.n < maxPredefinedTokens { + size = w.fixedSize(extraBits) + } + + // Dynamic Huffman? + var numCodegens int + + // Generate codegen and codegenFrequencies, which indicates how to encode + // the literalEncoding and the offsetEncoding. + w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding) + w.codegenEncoding.generate(w.codegenFreq[:], 7) + dynamicSize, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits) + + if dynamicSize < size { + size = dynamicSize + literalEncoding = w.literalEncoding + offsetEncoding = w.offsetEncoding + } + + // Stored bytes? + if storable && storedSize <= size { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + + // Huffman. + if literalEncoding == fixedLiteralEncoding { + w.writeFixedHeader(eof) + } else { + w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) + } + + // Write the tokens. + w.writeTokens(tokens.Slice(), literalEncoding.codes, offsetEncoding.codes) +} + +// writeBlockDynamic encodes a block using a dynamic Huffman table. +// This should be used if the symbols used have a disproportionate +// histogram distribution. +// If input is supplied and the compression savings are below 1/16th of the +// input size the block is stored. +func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []byte, sync bool) { + if w.err != nil { + return + } + + sync = sync || eof + if sync { + tokens.AddEOB() + } + + // We cannot reuse pure huffman table, and must mark as EOF. + if (w.lastHuffMan || eof) && w.lastHeader > 0 { + // We will not try to reuse. + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + w.lastHuffMan = false + } + + // fillReuse enables filling of empty values. + // This will make encodings always reusable without testing. + // However, this does not appear to benefit on most cases. + const fillReuse = false + + // Check if we can reuse... + if !fillReuse && w.lastHeader > 0 && !w.canReuse(tokens) { + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + + numLiterals, numOffsets := w.indexTokens(tokens, !sync) + extraBits := 0 + ssize, storable := w.storedSize(input) + + const usePrefs = true + if storable || w.lastHeader > 0 { + extraBits = w.extraBitSize() + } + + var size int + + // Check if we should reuse. + if w.lastHeader > 0 { + // Estimate size for using a new table. + // Use the previous header size as the best estimate. + newSize := w.lastHeader + tokens.EstimatedBits() + newSize += int(w.literalEncoding.codes[endBlockMarker].len()) + newSize>>w.logNewTablePenalty + + // The estimated size is calculated as an optimal table. + // We add a penalty to make it more realistic and re-use a bit more. + reuseSize := w.dynamicReuseSize(w.literalEncoding, w.offsetEncoding) + extraBits + + // Check if a new table is better. + if newSize < reuseSize { + // Write the EOB we owe. + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + size = newSize + w.lastHeader = 0 + } else { + size = reuseSize + } + + if tokens.n < maxPredefinedTokens { + if preSize := w.fixedSize(extraBits) + 7; usePrefs && preSize < size { + // Check if we get a reasonable size decrease. + if storable && ssize <= size { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + w.writeFixedHeader(eof) + if !sync { + tokens.AddEOB() + } + w.writeTokens(tokens.Slice(), fixedLiteralEncoding.codes, fixedOffsetEncoding.codes) + return + } + } + // Check if we get a reasonable size decrease. + if storable && ssize <= size { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + } + + // We want a new block/table + if w.lastHeader == 0 { + if fillReuse && !sync { + w.fillTokens() + numLiterals, numOffsets = maxNumLit, maxNumDist + } else { + w.literalFreq[endBlockMarker] = 1 + } + + w.generate() + // Generate codegen and codegenFrequencies, which indicates how to encode + // the literalEncoding and the offsetEncoding. + w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding) + w.codegenEncoding.generate(w.codegenFreq[:], 7) + + var numCodegens int + if fillReuse && !sync { + // Reindex for accurate size... + w.indexTokens(tokens, true) + } + size, numCodegens = w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits) + + // Store predefined, if we don't get a reasonable improvement. + if tokens.n < maxPredefinedTokens { + if preSize := w.fixedSize(extraBits); usePrefs && preSize <= size { + // Store bytes, if we don't get an improvement. + if storable && ssize <= preSize { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + w.writeFixedHeader(eof) + if !sync { + tokens.AddEOB() + } + w.writeTokens(tokens.Slice(), fixedLiteralEncoding.codes, fixedOffsetEncoding.codes) + return + } + } + + if storable && ssize <= size { + // Store bytes, if we don't get an improvement. + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + + // Write Huffman table. + w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) + if !sync { + w.lastHeader, _ = w.headerSize() + } + w.lastHuffMan = false + } + + if sync { + w.lastHeader = 0 + } + // Write the tokens. + w.writeTokens(tokens.Slice(), w.literalEncoding.codes, w.offsetEncoding.codes) +} + +func (w *huffmanBitWriter) fillTokens() { + for i, v := range w.literalFreq[:literalCount] { + if v == 0 { + w.literalFreq[i] = 1 + } + } + for i, v := range w.offsetFreq[:offsetCodeCount] { + if v == 0 { + w.offsetFreq[i] = 1 + } + } +} + +// indexTokens indexes a slice of tokens, and updates +// literalFreq and offsetFreq, and generates literalEncoding +// and offsetEncoding. +// The number of literal and offset tokens is returned. +func (w *huffmanBitWriter) indexTokens(t *tokens, filled bool) (numLiterals, numOffsets int) { + copy(w.literalFreq[:], t.litHist[:]) + copy(w.literalFreq[256:], t.extraHist[:]) + copy(w.offsetFreq[:], t.offHist[:offsetCodeCount]) + + if t.n == 0 { + return + } + if filled { + return maxNumLit, maxNumDist + } + // get the number of literals + numLiterals = len(w.literalFreq) + for w.literalFreq[numLiterals-1] == 0 { + numLiterals-- + } + // get the number of offsets + numOffsets = len(w.offsetFreq) + for numOffsets > 0 && w.offsetFreq[numOffsets-1] == 0 { + numOffsets-- + } + if numOffsets == 0 { + // We haven't found a single match. If we want to go with the dynamic encoding, + // we should count at least one offset to be sure that the offset huffman tree could be encoded. + w.offsetFreq[0] = 1 + numOffsets = 1 + } + return +} + +func (w *huffmanBitWriter) generate() { + w.literalEncoding.generate(w.literalFreq[:literalCount], 15) + w.offsetEncoding.generate(w.offsetFreq[:offsetCodeCount], 15) +} + +// writeTokens writes a slice of tokens to the output. +// codes for literal and offset encoding must be supplied. +func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) { + if w.err != nil { + return + } + if len(tokens) == 0 { + return + } + + // Only last token should be endBlockMarker. + var deferEOB bool + if tokens[len(tokens)-1] == endBlockMarker { + tokens = tokens[:len(tokens)-1] + deferEOB = true + } + + // Create slices up to the next power of two to avoid bounds checks. + lits := leCodes[:256] + offs := oeCodes[:32] + lengths := leCodes[lengthCodesStart:] + lengths = lengths[:32] + + // Go 1.16 LOVES having these on stack. + bits, nbits, nbytes := w.bits, w.nbits, w.nbytes + + for _, t := range tokens { + if t < 256 { + //w.writeCode(lits[t.literal()]) + c := lits[t] + bits |= c.code64() << (nbits & 63) + nbits += c.len() + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } + continue + } + + // Write the length + length := t.length() + lengthCode := lengthCode(length) & 31 + if false { + w.writeCode(lengths[lengthCode]) + } else { + // inlined + c := lengths[lengthCode] + bits |= c.code64() << (nbits & 63) + nbits += c.len() + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } + } + + if lengthCode >= lengthExtraBitsMinCode { + extraLengthBits := lengthExtraBits[lengthCode] + //w.writeBits(extraLength, extraLengthBits) + extraLength := int32(length - lengthBase[lengthCode]) + bits |= uint64(extraLength) << (nbits & 63) + nbits += extraLengthBits + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } + } + // Write the offset + offset := t.offset() + offsetCode := (offset >> 16) & 31 + if false { + w.writeCode(offs[offsetCode]) + } else { + // inlined + c := offs[offsetCode] + bits |= c.code64() << (nbits & 63) + nbits += c.len() + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } + } + + if offsetCode >= offsetExtraBitsMinCode { + offsetComb := offsetCombined[offsetCode] + //w.writeBits(extraOffset, extraOffsetBits) + bits |= uint64((offset-(offsetComb>>8))&matchOffsetOnlyMask) << (nbits & 63) + nbits += uint8(offsetComb) + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } + } + } + // Restore... + w.bits, w.nbits, w.nbytes = bits, nbits, nbytes + + if deferEOB { + w.writeCode(leCodes[endBlockMarker]) + } +} + +// huffOffset is a static offset encoder used for huffman only encoding. +// It can be reused since we will not be encoding offset values. +var huffOffset *huffmanEncoder + +func init() { + w := newHuffmanBitWriter(nil) + w.offsetFreq[0] = 1 + huffOffset = newHuffmanEncoder(offsetCodeCount) + huffOffset.generate(w.offsetFreq[:offsetCodeCount], 15) +} + +// writeBlockHuff encodes a block of bytes as either +// Huffman encoded literals or uncompressed bytes if the +// results only gains very little from compression. +func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) { + if w.err != nil { + return + } + + // Clear histogram + for i := range w.literalFreq[:] { + w.literalFreq[i] = 0 + } + if !w.lastHuffMan { + for i := range w.offsetFreq[:] { + w.offsetFreq[i] = 0 + } + } + + const numLiterals = endBlockMarker + 1 + const numOffsets = 1 + + // Add everything as literals + // We have to estimate the header size. + // Assume header is around 70 bytes: + // https://stackoverflow.com/a/25454430 + const guessHeaderSizeBits = 70 * 8 + histogram(input, w.literalFreq[:numLiterals]) + ssize, storable := w.storedSize(input) + if storable && len(input) > 1024 { + // Quick check for incompressible content. + abs := float64(0) + avg := float64(len(input)) / 256 + max := float64(len(input) * 2) + for _, v := range w.literalFreq[:256] { + diff := float64(v) - avg + abs += diff * diff + if abs > max { + break + } + } + if abs < max { + if debugDeflate { + fmt.Println("stored", abs, "<", max) + } + // No chance we can compress this... + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + } + w.literalFreq[endBlockMarker] = 1 + w.tmpLitEncoding.generate(w.literalFreq[:numLiterals], 15) + estBits := w.tmpLitEncoding.canReuseBits(w.literalFreq[:numLiterals]) + if estBits < math.MaxInt32 { + estBits += w.lastHeader + if w.lastHeader == 0 { + estBits += guessHeaderSizeBits + } + estBits += estBits >> w.logNewTablePenalty + } + + // Store bytes, if we don't get a reasonable improvement. + if storable && ssize <= estBits { + if debugDeflate { + fmt.Println("stored,", ssize, "<=", estBits) + } + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + + if w.lastHeader > 0 { + reuseSize := w.literalEncoding.canReuseBits(w.literalFreq[:256]) + + if estBits < reuseSize { + if debugDeflate { + fmt.Println("NOT reusing, reuse:", reuseSize/8, "> new:", estBits/8, "header est:", w.lastHeader/8, "bytes") + } + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } else if debugDeflate { + fmt.Println("reusing, reuse:", reuseSize/8, "> new:", estBits/8, "- header est:", w.lastHeader/8) + } + } + + count := 0 + if w.lastHeader == 0 { + // Use the temp encoding, so swap. + w.literalEncoding, w.tmpLitEncoding = w.tmpLitEncoding, w.literalEncoding + // Generate codegen and codegenFrequencies, which indicates how to encode + // the literalEncoding and the offsetEncoding. + w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, huffOffset) + w.codegenEncoding.generate(w.codegenFreq[:], 7) + numCodegens := w.codegens() + + // Huffman. + w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) + w.lastHuffMan = true + w.lastHeader, _ = w.headerSize() + if debugDeflate { + count += w.lastHeader + fmt.Println("header:", count/8) + } + } + + encoding := w.literalEncoding.codes[:256] + // Go 1.16 LOVES having these on stack. At least 1.5x the speed. + bits, nbits, nbytes := w.bits, w.nbits, w.nbytes + + if debugDeflate { + count -= int(nbytes)*8 + int(nbits) + } + // Unroll, write 3 codes/loop. + // Fastest number of unrolls. + for len(input) > 3 { + // We must have at least 48 bits free. + if nbits >= 8 { + n := nbits >> 3 + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + bits >>= (n * 8) & 63 + nbits -= n * 8 + nbytes += n + } + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + if debugDeflate { + count += int(nbytes) * 8 + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + a, b := encoding[input[0]], encoding[input[1]] + bits |= a.code64() << (nbits & 63) + bits |= b.code64() << ((nbits + a.len()) & 63) + c := encoding[input[2]] + nbits += b.len() + a.len() + bits |= c.code64() << (nbits & 63) + nbits += c.len() + input = input[3:] + } + + // Remaining... + for _, t := range input { + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + if debugDeflate { + count += int(nbytes) * 8 + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } + // Bitwriting inlined, ~30% speedup + c := encoding[t] + bits |= c.code64() << (nbits & 63) + + nbits += c.len() + if debugDeflate { + count += int(c.len()) + } + } + // Restore... + w.bits, w.nbits, w.nbytes = bits, nbits, nbytes + + if debugDeflate { + nb := count + int(nbytes)*8 + int(nbits) + fmt.Println("wrote", nb, "bits,", nb/8, "bytes.") + } + // Flush if needed to have space. + if w.nbits >= 48 { + w.writeOutBits() + } + + if eof || sync { + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + w.lastHuffMan = false + } +} diff --git a/vendor/github.com/klauspost/compress/flate/huffman_code.go b/vendor/github.com/klauspost/compress/flate/huffman_code.go new file mode 100644 index 0000000..5ac144f --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/huffman_code.go @@ -0,0 +1,412 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "math" + "math/bits" +) + +const ( + maxBitsLimit = 16 + // number of valid literals + literalCount = 286 +) + +// hcode is a huffman code with a bit code and bit length. +type hcode uint32 + +func (h hcode) len() uint8 { + return uint8(h) +} + +func (h hcode) code64() uint64 { + return uint64(h >> 8) +} + +func (h hcode) zero() bool { + return h == 0 +} + +type huffmanEncoder struct { + codes []hcode + bitCount [17]int32 + + // Allocate a reusable buffer with the longest possible frequency table. + // Possible lengths are codegenCodeCount, offsetCodeCount and literalCount. + // The largest of these is literalCount, so we allocate for that case. + freqcache [literalCount + 1]literalNode +} + +type literalNode struct { + literal uint16 + freq uint16 +} + +// A levelInfo describes the state of the constructed tree for a given depth. +type levelInfo struct { + // Our level. for better printing + level int32 + + // The frequency of the last node at this level + lastFreq int32 + + // The frequency of the next character to add to this level + nextCharFreq int32 + + // The frequency of the next pair (from level below) to add to this level. + // Only valid if the "needed" value of the next lower level is 0. + nextPairFreq int32 + + // The number of chains remaining to generate for this level before moving + // up to the next level + needed int32 +} + +// set sets the code and length of an hcode. +func (h *hcode) set(code uint16, length uint8) { + *h = hcode(length) | (hcode(code) << 8) +} + +func newhcode(code uint16, length uint8) hcode { + return hcode(length) | (hcode(code) << 8) +} + +func reverseBits(number uint16, bitLength byte) uint16 { + return bits.Reverse16(number << ((16 - bitLength) & 15)) +} + +func maxNode() literalNode { return literalNode{math.MaxUint16, math.MaxUint16} } + +func newHuffmanEncoder(size int) *huffmanEncoder { + // Make capacity to next power of two. + c := uint(bits.Len32(uint32(size - 1))) + return &huffmanEncoder{codes: make([]hcode, size, 1<= 3 +// The cases of 0, 1, and 2 literals are handled by special case code. +// +// list An array of the literals with non-zero frequencies +// and their associated frequencies. The array is in order of increasing +// frequency, and has as its last element a special element with frequency +// MaxInt32 +// maxBits The maximum number of bits that should be used to encode any literal. +// Must be less than 16. +// return An integer array in which array[i] indicates the number of literals +// that should be encoded in i bits. +func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 { + if maxBits >= maxBitsLimit { + panic("flate: maxBits too large") + } + n := int32(len(list)) + list = list[0 : n+1] + list[n] = maxNode() + + // The tree can't have greater depth than n - 1, no matter what. This + // saves a little bit of work in some small cases + if maxBits > n-1 { + maxBits = n - 1 + } + + // Create information about each of the levels. + // A bogus "Level 0" whose sole purpose is so that + // level1.prev.needed==0. This makes level1.nextPairFreq + // be a legitimate value that never gets chosen. + var levels [maxBitsLimit]levelInfo + // leafCounts[i] counts the number of literals at the left + // of ancestors of the rightmost node at level i. + // leafCounts[i][j] is the number of literals at the left + // of the level j ancestor. + var leafCounts [maxBitsLimit][maxBitsLimit]int32 + + // Descending to only have 1 bounds check. + l2f := int32(list[2].freq) + l1f := int32(list[1].freq) + l0f := int32(list[0].freq) + int32(list[1].freq) + + for level := int32(1); level <= maxBits; level++ { + // For every level, the first two items are the first two characters. + // We initialize the levels as if we had already figured this out. + levels[level] = levelInfo{ + level: level, + lastFreq: l1f, + nextCharFreq: l2f, + nextPairFreq: l0f, + } + leafCounts[level][level] = 2 + if level == 1 { + levels[level].nextPairFreq = math.MaxInt32 + } + } + + // We need a total of 2*n - 2 items at top level and have already generated 2. + levels[maxBits].needed = 2*n - 4 + + level := uint32(maxBits) + for level < 16 { + l := &levels[level] + if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 { + // We've run out of both leafs and pairs. + // End all calculations for this level. + // To make sure we never come back to this level or any lower level, + // set nextPairFreq impossibly large. + l.needed = 0 + levels[level+1].nextPairFreq = math.MaxInt32 + level++ + continue + } + + prevFreq := l.lastFreq + if l.nextCharFreq < l.nextPairFreq { + // The next item on this row is a leaf node. + n := leafCounts[level][level] + 1 + l.lastFreq = l.nextCharFreq + // Lower leafCounts are the same of the previous node. + leafCounts[level][level] = n + e := list[n] + if e.literal < math.MaxUint16 { + l.nextCharFreq = int32(e.freq) + } else { + l.nextCharFreq = math.MaxInt32 + } + } else { + // The next item on this row is a pair from the previous row. + // nextPairFreq isn't valid until we generate two + // more values in the level below + l.lastFreq = l.nextPairFreq + // Take leaf counts from the lower level, except counts[level] remains the same. + if true { + save := leafCounts[level][level] + leafCounts[level] = leafCounts[level-1] + leafCounts[level][level] = save + } else { + copy(leafCounts[level][:level], leafCounts[level-1][:level]) + } + levels[l.level-1].needed = 2 + } + + if l.needed--; l.needed == 0 { + // We've done everything we need to do for this level. + // Continue calculating one level up. Fill in nextPairFreq + // of that level with the sum of the two nodes we've just calculated on + // this level. + if l.level == maxBits { + // All done! + break + } + levels[l.level+1].nextPairFreq = prevFreq + l.lastFreq + level++ + } else { + // If we stole from below, move down temporarily to replenish it. + for levels[level-1].needed > 0 { + level-- + } + } + } + + // Somethings is wrong if at the end, the top level is null or hasn't used + // all of the leaves. + if leafCounts[maxBits][maxBits] != n { + panic("leafCounts[maxBits][maxBits] != n") + } + + bitCount := h.bitCount[:maxBits+1] + bits := 1 + counts := &leafCounts[maxBits] + for level := maxBits; level > 0; level-- { + // chain.leafCount gives the number of literals requiring at least "bits" + // bits to encode. + bitCount[bits] = counts[level] - counts[level-1] + bits++ + } + return bitCount +} + +// Look at the leaves and assign them a bit count and an encoding as specified +// in RFC 1951 3.2.2 +func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalNode) { + code := uint16(0) + for n, bits := range bitCount { + code <<= 1 + if n == 0 || bits == 0 { + continue + } + // The literals list[len(list)-bits] .. list[len(list)-bits] + // are encoded using "bits" bits, and get the values + // code, code + 1, .... The code values are + // assigned in literal order (not frequency order). + chunk := list[len(list)-int(bits):] + + sortByLiteral(chunk) + for _, node := range chunk { + h.codes[node.literal] = newhcode(reverseBits(code, uint8(n)), uint8(n)) + code++ + } + list = list[0 : len(list)-int(bits)] + } +} + +// Update this Huffman Code object to be the minimum code for the specified frequency count. +// +// freq An array of frequencies, in which frequency[i] gives the frequency of literal i. +// maxBits The maximum number of bits to use for any literal. +func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) { + list := h.freqcache[:len(freq)+1] + codes := h.codes[:len(freq)] + // Number of non-zero literals + count := 0 + // Set list to be the set of all non-zero literals and their frequencies + for i, f := range freq { + if f != 0 { + list[count] = literalNode{uint16(i), f} + count++ + } else { + codes[i] = 0 + } + } + list[count] = literalNode{} + + list = list[:count] + if count <= 2 { + // Handle the small cases here, because they are awkward for the general case code. With + // two or fewer literals, everything has bit length 1. + for i, node := range list { + // "list" is in order of increasing literal value. + h.codes[node.literal].set(uint16(i), 1) + } + return + } + sortByFreq(list) + + // Get the number of literals for each bit count + bitCount := h.bitCounts(list, maxBits) + // And do the assignment + h.assignEncodingAndSize(bitCount, list) +} + +// atLeastOne clamps the result between 1 and 15. +func atLeastOne(v float32) float32 { + if v < 1 { + return 1 + } + if v > 15 { + return 15 + } + return v +} + +func histogram(b []byte, h []uint16) { + if true && len(b) >= 8<<10 { + // Split for bigger inputs + histogramSplit(b, h) + } else { + h = h[:256] + for _, t := range b { + h[t]++ + } + } +} + +func histogramSplit(b []byte, h []uint16) { + // Tested, and slightly faster than 2-way. + // Writing to separate arrays and combining is also slightly slower. + h = h[:256] + for len(b)&3 != 0 { + h[b[0]]++ + b = b[1:] + } + n := len(b) / 4 + x, y, z, w := b[:n], b[n:], b[n+n:], b[n+n+n:] + y, z, w = y[:len(x)], z[:len(x)], w[:len(x)] + for i, t := range x { + v0 := &h[t] + v1 := &h[y[i]] + v3 := &h[w[i]] + v2 := &h[z[i]] + *v0++ + *v1++ + *v2++ + *v3++ + } +} diff --git a/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go b/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go new file mode 100644 index 0000000..2077802 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go @@ -0,0 +1,178 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +// Sort sorts data. +// It makes one call to data.Len to determine n, and O(n*log(n)) calls to +// data.Less and data.Swap. The sort is not guaranteed to be stable. +func sortByFreq(data []literalNode) { + n := len(data) + quickSortByFreq(data, 0, n, maxDepth(n)) +} + +func quickSortByFreq(data []literalNode, a, b, maxDepth int) { + for b-a > 12 { // Use ShellSort for slices <= 12 elements + if maxDepth == 0 { + heapSort(data, a, b) + return + } + maxDepth-- + mlo, mhi := doPivotByFreq(data, a, b) + // Avoiding recursion on the larger subproblem guarantees + // a stack depth of at most lg(b-a). + if mlo-a < b-mhi { + quickSortByFreq(data, a, mlo, maxDepth) + a = mhi // i.e., quickSortByFreq(data, mhi, b) + } else { + quickSortByFreq(data, mhi, b, maxDepth) + b = mlo // i.e., quickSortByFreq(data, a, mlo) + } + } + if b-a > 1 { + // Do ShellSort pass with gap 6 + // It could be written in this simplified form cause b-a <= 12 + for i := a + 6; i < b; i++ { + if data[i].freq == data[i-6].freq && data[i].literal < data[i-6].literal || data[i].freq < data[i-6].freq { + data[i], data[i-6] = data[i-6], data[i] + } + } + insertionSortByFreq(data, a, b) + } +} + +// siftDownByFreq implements the heap property on data[lo, hi). +// first is an offset into the array where the root of the heap lies. +func siftDownByFreq(data []literalNode, lo, hi, first int) { + root := lo + for { + child := 2*root + 1 + if child >= hi { + break + } + if child+1 < hi && (data[first+child].freq == data[first+child+1].freq && data[first+child].literal < data[first+child+1].literal || data[first+child].freq < data[first+child+1].freq) { + child++ + } + if data[first+root].freq == data[first+child].freq && data[first+root].literal > data[first+child].literal || data[first+root].freq > data[first+child].freq { + return + } + data[first+root], data[first+child] = data[first+child], data[first+root] + root = child + } +} +func doPivotByFreq(data []literalNode, lo, hi int) (midlo, midhi int) { + m := int(uint(lo+hi) >> 1) // Written like this to avoid integer overflow. + if hi-lo > 40 { + // Tukey's ``Ninther,'' median of three medians of three. + s := (hi - lo) / 8 + medianOfThreeSortByFreq(data, lo, lo+s, lo+2*s) + medianOfThreeSortByFreq(data, m, m-s, m+s) + medianOfThreeSortByFreq(data, hi-1, hi-1-s, hi-1-2*s) + } + medianOfThreeSortByFreq(data, lo, m, hi-1) + + // Invariants are: + // data[lo] = pivot (set up by ChoosePivot) + // data[lo < i < a] < pivot + // data[a <= i < b] <= pivot + // data[b <= i < c] unexamined + // data[c <= i < hi-1] > pivot + // data[hi-1] >= pivot + pivot := lo + a, c := lo+1, hi-1 + + for ; a < c && (data[a].freq == data[pivot].freq && data[a].literal < data[pivot].literal || data[a].freq < data[pivot].freq); a++ { + } + b := a + for { + for ; b < c && (data[pivot].freq == data[b].freq && data[pivot].literal > data[b].literal || data[pivot].freq > data[b].freq); b++ { // data[b] <= pivot + } + for ; b < c && (data[pivot].freq == data[c-1].freq && data[pivot].literal < data[c-1].literal || data[pivot].freq < data[c-1].freq); c-- { // data[c-1] > pivot + } + if b >= c { + break + } + // data[b] > pivot; data[c-1] <= pivot + data[b], data[c-1] = data[c-1], data[b] + b++ + c-- + } + // If hi-c<3 then there are duplicates (by property of median of nine). + // Let's be a bit more conservative, and set border to 5. + protect := hi-c < 5 + if !protect && hi-c < (hi-lo)/4 { + // Lets test some points for equality to pivot + dups := 0 + if data[pivot].freq == data[hi-1].freq && data[pivot].literal > data[hi-1].literal || data[pivot].freq > data[hi-1].freq { // data[hi-1] = pivot + data[c], data[hi-1] = data[hi-1], data[c] + c++ + dups++ + } + if data[b-1].freq == data[pivot].freq && data[b-1].literal > data[pivot].literal || data[b-1].freq > data[pivot].freq { // data[b-1] = pivot + b-- + dups++ + } + // m-lo = (hi-lo)/2 > 6 + // b-lo > (hi-lo)*3/4-1 > 8 + // ==> m < b ==> data[m] <= pivot + if data[m].freq == data[pivot].freq && data[m].literal > data[pivot].literal || data[m].freq > data[pivot].freq { // data[m] = pivot + data[m], data[b-1] = data[b-1], data[m] + b-- + dups++ + } + // if at least 2 points are equal to pivot, assume skewed distribution + protect = dups > 1 + } + if protect { + // Protect against a lot of duplicates + // Add invariant: + // data[a <= i < b] unexamined + // data[b <= i < c] = pivot + for { + for ; a < b && (data[b-1].freq == data[pivot].freq && data[b-1].literal > data[pivot].literal || data[b-1].freq > data[pivot].freq); b-- { // data[b] == pivot + } + for ; a < b && (data[a].freq == data[pivot].freq && data[a].literal < data[pivot].literal || data[a].freq < data[pivot].freq); a++ { // data[a] < pivot + } + if a >= b { + break + } + // data[a] == pivot; data[b-1] < pivot + data[a], data[b-1] = data[b-1], data[a] + a++ + b-- + } + } + // Swap pivot into middle + data[pivot], data[b-1] = data[b-1], data[pivot] + return b - 1, c +} + +// Insertion sort +func insertionSortByFreq(data []literalNode, a, b int) { + for i := a + 1; i < b; i++ { + for j := i; j > a && (data[j].freq == data[j-1].freq && data[j].literal < data[j-1].literal || data[j].freq < data[j-1].freq); j-- { + data[j], data[j-1] = data[j-1], data[j] + } + } +} + +// quickSortByFreq, loosely following Bentley and McIlroy, +// ``Engineering a Sort Function,'' SP&E November 1993. + +// medianOfThreeSortByFreq moves the median of the three values data[m0], data[m1], data[m2] into data[m1]. +func medianOfThreeSortByFreq(data []literalNode, m1, m0, m2 int) { + // sort 3 elements + if data[m1].freq == data[m0].freq && data[m1].literal < data[m0].literal || data[m1].freq < data[m0].freq { + data[m1], data[m0] = data[m0], data[m1] + } + // data[m0] <= data[m1] + if data[m2].freq == data[m1].freq && data[m2].literal < data[m1].literal || data[m2].freq < data[m1].freq { + data[m2], data[m1] = data[m1], data[m2] + // data[m0] <= data[m2] && data[m1] < data[m2] + if data[m1].freq == data[m0].freq && data[m1].literal < data[m0].literal || data[m1].freq < data[m0].freq { + data[m1], data[m0] = data[m0], data[m1] + } + } + // now data[m0] <= data[m1] <= data[m2] +} diff --git a/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go b/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go new file mode 100644 index 0000000..93f1aea --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go @@ -0,0 +1,201 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +// Sort sorts data. +// It makes one call to data.Len to determine n, and O(n*log(n)) calls to +// data.Less and data.Swap. The sort is not guaranteed to be stable. +func sortByLiteral(data []literalNode) { + n := len(data) + quickSort(data, 0, n, maxDepth(n)) +} + +func quickSort(data []literalNode, a, b, maxDepth int) { + for b-a > 12 { // Use ShellSort for slices <= 12 elements + if maxDepth == 0 { + heapSort(data, a, b) + return + } + maxDepth-- + mlo, mhi := doPivot(data, a, b) + // Avoiding recursion on the larger subproblem guarantees + // a stack depth of at most lg(b-a). + if mlo-a < b-mhi { + quickSort(data, a, mlo, maxDepth) + a = mhi // i.e., quickSort(data, mhi, b) + } else { + quickSort(data, mhi, b, maxDepth) + b = mlo // i.e., quickSort(data, a, mlo) + } + } + if b-a > 1 { + // Do ShellSort pass with gap 6 + // It could be written in this simplified form cause b-a <= 12 + for i := a + 6; i < b; i++ { + if data[i].literal < data[i-6].literal { + data[i], data[i-6] = data[i-6], data[i] + } + } + insertionSort(data, a, b) + } +} +func heapSort(data []literalNode, a, b int) { + first := a + lo := 0 + hi := b - a + + // Build heap with greatest element at top. + for i := (hi - 1) / 2; i >= 0; i-- { + siftDown(data, i, hi, first) + } + + // Pop elements, largest first, into end of data. + for i := hi - 1; i >= 0; i-- { + data[first], data[first+i] = data[first+i], data[first] + siftDown(data, lo, i, first) + } +} + +// siftDown implements the heap property on data[lo, hi). +// first is an offset into the array where the root of the heap lies. +func siftDown(data []literalNode, lo, hi, first int) { + root := lo + for { + child := 2*root + 1 + if child >= hi { + break + } + if child+1 < hi && data[first+child].literal < data[first+child+1].literal { + child++ + } + if data[first+root].literal > data[first+child].literal { + return + } + data[first+root], data[first+child] = data[first+child], data[first+root] + root = child + } +} +func doPivot(data []literalNode, lo, hi int) (midlo, midhi int) { + m := int(uint(lo+hi) >> 1) // Written like this to avoid integer overflow. + if hi-lo > 40 { + // Tukey's ``Ninther,'' median of three medians of three. + s := (hi - lo) / 8 + medianOfThree(data, lo, lo+s, lo+2*s) + medianOfThree(data, m, m-s, m+s) + medianOfThree(data, hi-1, hi-1-s, hi-1-2*s) + } + medianOfThree(data, lo, m, hi-1) + + // Invariants are: + // data[lo] = pivot (set up by ChoosePivot) + // data[lo < i < a] < pivot + // data[a <= i < b] <= pivot + // data[b <= i < c] unexamined + // data[c <= i < hi-1] > pivot + // data[hi-1] >= pivot + pivot := lo + a, c := lo+1, hi-1 + + for ; a < c && data[a].literal < data[pivot].literal; a++ { + } + b := a + for { + for ; b < c && data[pivot].literal > data[b].literal; b++ { // data[b] <= pivot + } + for ; b < c && data[pivot].literal < data[c-1].literal; c-- { // data[c-1] > pivot + } + if b >= c { + break + } + // data[b] > pivot; data[c-1] <= pivot + data[b], data[c-1] = data[c-1], data[b] + b++ + c-- + } + // If hi-c<3 then there are duplicates (by property of median of nine). + // Let's be a bit more conservative, and set border to 5. + protect := hi-c < 5 + if !protect && hi-c < (hi-lo)/4 { + // Lets test some points for equality to pivot + dups := 0 + if data[pivot].literal > data[hi-1].literal { // data[hi-1] = pivot + data[c], data[hi-1] = data[hi-1], data[c] + c++ + dups++ + } + if data[b-1].literal > data[pivot].literal { // data[b-1] = pivot + b-- + dups++ + } + // m-lo = (hi-lo)/2 > 6 + // b-lo > (hi-lo)*3/4-1 > 8 + // ==> m < b ==> data[m] <= pivot + if data[m].literal > data[pivot].literal { // data[m] = pivot + data[m], data[b-1] = data[b-1], data[m] + b-- + dups++ + } + // if at least 2 points are equal to pivot, assume skewed distribution + protect = dups > 1 + } + if protect { + // Protect against a lot of duplicates + // Add invariant: + // data[a <= i < b] unexamined + // data[b <= i < c] = pivot + for { + for ; a < b && data[b-1].literal > data[pivot].literal; b-- { // data[b] == pivot + } + for ; a < b && data[a].literal < data[pivot].literal; a++ { // data[a] < pivot + } + if a >= b { + break + } + // data[a] == pivot; data[b-1] < pivot + data[a], data[b-1] = data[b-1], data[a] + a++ + b-- + } + } + // Swap pivot into middle + data[pivot], data[b-1] = data[b-1], data[pivot] + return b - 1, c +} + +// Insertion sort +func insertionSort(data []literalNode, a, b int) { + for i := a + 1; i < b; i++ { + for j := i; j > a && data[j].literal < data[j-1].literal; j-- { + data[j], data[j-1] = data[j-1], data[j] + } + } +} + +// maxDepth returns a threshold at which quicksort should switch +// to heapsort. It returns 2*ceil(lg(n+1)). +func maxDepth(n int) int { + var depth int + for i := n; i > 0; i >>= 1 { + depth++ + } + return depth * 2 +} + +// medianOfThree moves the median of the three values data[m0], data[m1], data[m2] into data[m1]. +func medianOfThree(data []literalNode, m1, m0, m2 int) { + // sort 3 elements + if data[m1].literal < data[m0].literal { + data[m1], data[m0] = data[m0], data[m1] + } + // data[m0] <= data[m1] + if data[m2].literal < data[m1].literal { + data[m2], data[m1] = data[m1], data[m2] + // data[m0] <= data[m2] && data[m1] < data[m2] + if data[m1].literal < data[m0].literal { + data[m1], data[m0] = data[m0], data[m1] + } + } + // now data[m0] <= data[m1] <= data[m2] +} diff --git a/vendor/github.com/klauspost/compress/flate/inflate.go b/vendor/github.com/klauspost/compress/flate/inflate.go new file mode 100644 index 0000000..414c0be --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/inflate.go @@ -0,0 +1,793 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package flate implements the DEFLATE compressed data format, described in +// RFC 1951. The gzip and zlib packages implement access to DEFLATE-based file +// formats. +package flate + +import ( + "bufio" + "compress/flate" + "fmt" + "io" + "math/bits" + "sync" +) + +const ( + maxCodeLen = 16 // max length of Huffman code + maxCodeLenMask = 15 // mask for max length of Huffman code + // The next three numbers come from the RFC section 3.2.7, with the + // additional proviso in section 3.2.5 which implies that distance codes + // 30 and 31 should never occur in compressed data. + maxNumLit = 286 + maxNumDist = 30 + numCodes = 19 // number of codes in Huffman meta-code + + debugDecode = false +) + +// Value of length - 3 and extra bits. +type lengthExtra struct { + length, extra uint8 +} + +var decCodeToLen = [32]lengthExtra{{length: 0x0, extra: 0x0}, {length: 0x1, extra: 0x0}, {length: 0x2, extra: 0x0}, {length: 0x3, extra: 0x0}, {length: 0x4, extra: 0x0}, {length: 0x5, extra: 0x0}, {length: 0x6, extra: 0x0}, {length: 0x7, extra: 0x0}, {length: 0x8, extra: 0x1}, {length: 0xa, extra: 0x1}, {length: 0xc, extra: 0x1}, {length: 0xe, extra: 0x1}, {length: 0x10, extra: 0x2}, {length: 0x14, extra: 0x2}, {length: 0x18, extra: 0x2}, {length: 0x1c, extra: 0x2}, {length: 0x20, extra: 0x3}, {length: 0x28, extra: 0x3}, {length: 0x30, extra: 0x3}, {length: 0x38, extra: 0x3}, {length: 0x40, extra: 0x4}, {length: 0x50, extra: 0x4}, {length: 0x60, extra: 0x4}, {length: 0x70, extra: 0x4}, {length: 0x80, extra: 0x5}, {length: 0xa0, extra: 0x5}, {length: 0xc0, extra: 0x5}, {length: 0xe0, extra: 0x5}, {length: 0xff, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}} + +var bitMask32 = [32]uint32{ + 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF, + 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, + 0x1ffff, 0x3ffff, 0x7FFFF, 0xfFFFF, 0x1fFFFF, 0x3fFFFF, 0x7fFFFF, 0xffFFFF, + 0x1ffFFFF, 0x3ffFFFF, 0x7ffFFFF, 0xfffFFFF, 0x1fffFFFF, 0x3fffFFFF, 0x7fffFFFF, +} // up to 32 bits + +// Initialize the fixedHuffmanDecoder only once upon first use. +var fixedOnce sync.Once +var fixedHuffmanDecoder huffmanDecoder + +// A CorruptInputError reports the presence of corrupt input at a given offset. +type CorruptInputError = flate.CorruptInputError + +// An InternalError reports an error in the flate code itself. +type InternalError string + +func (e InternalError) Error() string { return "flate: internal error: " + string(e) } + +// A ReadError reports an error encountered while reading input. +// +// Deprecated: No longer returned. +type ReadError = flate.ReadError + +// A WriteError reports an error encountered while writing output. +// +// Deprecated: No longer returned. +type WriteError = flate.WriteError + +// Resetter resets a ReadCloser returned by NewReader or NewReaderDict to +// to switch to a new underlying Reader. This permits reusing a ReadCloser +// instead of allocating a new one. +type Resetter interface { + // Reset discards any buffered data and resets the Resetter as if it was + // newly initialized with the given reader. + Reset(r io.Reader, dict []byte) error +} + +// The data structure for decoding Huffman tables is based on that of +// zlib. There is a lookup table of a fixed bit width (huffmanChunkBits), +// For codes smaller than the table width, there are multiple entries +// (each combination of trailing bits has the same value). For codes +// larger than the table width, the table contains a link to an overflow +// table. The width of each entry in the link table is the maximum code +// size minus the chunk width. +// +// Note that you can do a lookup in the table even without all bits +// filled. Since the extra bits are zero, and the DEFLATE Huffman codes +// have the property that shorter codes come before longer ones, the +// bit length estimate in the result is a lower bound on the actual +// number of bits. +// +// See the following: +// http://www.gzip.org/algorithm.txt + +// chunk & 15 is number of bits +// chunk >> 4 is value, including table link + +const ( + huffmanChunkBits = 9 + huffmanNumChunks = 1 << huffmanChunkBits + huffmanCountMask = 15 + huffmanValueShift = 4 +) + +type huffmanDecoder struct { + maxRead int // the maximum number of bits we can read and not overread + chunks *[huffmanNumChunks]uint16 // chunks as described above + links [][]uint16 // overflow links + linkMask uint32 // mask the width of the link table +} + +// Initialize Huffman decoding tables from array of code lengths. +// Following this function, h is guaranteed to be initialized into a complete +// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a +// degenerate case where the tree has only a single symbol with length 1. Empty +// trees are permitted. +func (h *huffmanDecoder) init(lengths []int) bool { + // Sanity enables additional runtime tests during Huffman + // table construction. It's intended to be used during + // development to supplement the currently ad-hoc unit tests. + const sanity = false + + if h.chunks == nil { + h.chunks = &[huffmanNumChunks]uint16{} + } + if h.maxRead != 0 { + *h = huffmanDecoder{chunks: h.chunks, links: h.links} + } + + // Count number of codes of each length, + // compute maxRead and max length. + var count [maxCodeLen]int + var min, max int + for _, n := range lengths { + if n == 0 { + continue + } + if min == 0 || n < min { + min = n + } + if n > max { + max = n + } + count[n&maxCodeLenMask]++ + } + + // Empty tree. The decompressor.huffSym function will fail later if the tree + // is used. Technically, an empty tree is only valid for the HDIST tree and + // not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree + // is guaranteed to fail since it will attempt to use the tree to decode the + // codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is + // guaranteed to fail later since the compressed data section must be + // composed of at least one symbol (the end-of-block marker). + if max == 0 { + return true + } + + code := 0 + var nextcode [maxCodeLen]int + for i := min; i <= max; i++ { + code <<= 1 + nextcode[i&maxCodeLenMask] = code + code += count[i&maxCodeLenMask] + } + + // Check that the coding is complete (i.e., that we've + // assigned all 2-to-the-max possible bit sequences). + // Exception: To be compatible with zlib, we also need to + // accept degenerate single-code codings. See also + // TestDegenerateHuffmanCoding. + if code != 1< huffmanChunkBits { + numLinks := 1 << (uint(max) - huffmanChunkBits) + h.linkMask = uint32(numLinks - 1) + + // create link tables + link := nextcode[huffmanChunkBits+1] >> 1 + if cap(h.links) < huffmanNumChunks-link { + h.links = make([][]uint16, huffmanNumChunks-link) + } else { + h.links = h.links[:huffmanNumChunks-link] + } + for j := uint(link); j < huffmanNumChunks; j++ { + reverse := int(bits.Reverse16(uint16(j))) + reverse >>= uint(16 - huffmanChunkBits) + off := j - uint(link) + if sanity && h.chunks[reverse] != 0 { + panic("impossible: overwriting existing chunk") + } + h.chunks[reverse] = uint16(off<>= uint(16 - n) + if n <= huffmanChunkBits { + for off := reverse; off < len(h.chunks); off += 1 << uint(n) { + // We should never need to overwrite + // an existing chunk. Also, 0 is + // never a valid chunk, because the + // lower 4 "count" bits should be + // between 1 and 15. + if sanity && h.chunks[off] != 0 { + panic("impossible: overwriting existing chunk") + } + h.chunks[off] = chunk + } + } else { + j := reverse & (huffmanNumChunks - 1) + if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 { + // Longer codes should have been + // associated with a link table above. + panic("impossible: not an indirect chunk") + } + value := h.chunks[j] >> huffmanValueShift + linktab := h.links[value] + reverse >>= huffmanChunkBits + for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) { + if sanity && linktab[off] != 0 { + panic("impossible: overwriting existing chunk") + } + linktab[off] = chunk + } + } + } + + if sanity { + // Above we've sanity checked that we never overwrote + // an existing entry. Here we additionally check that + // we filled the tables completely. + for i, chunk := range h.chunks { + if chunk == 0 { + // As an exception, in the degenerate + // single-code case, we allow odd + // chunks to be missing. + if code == 1 && i%2 == 1 { + continue + } + panic("impossible: missing chunk") + } + } + for _, linktab := range h.links { + for _, chunk := range linktab { + if chunk == 0 { + panic("impossible: missing chunk") + } + } + } + } + + return true +} + +// The actual read interface needed by NewReader. +// If the passed in io.Reader does not also have ReadByte, +// the NewReader will introduce its own buffering. +type Reader interface { + io.Reader + io.ByteReader +} + +// Decompress state. +type decompressor struct { + // Input source. + r Reader + roffset int64 + + // Huffman decoders for literal/length, distance. + h1, h2 huffmanDecoder + + // Length arrays used to define Huffman codes. + bits *[maxNumLit + maxNumDist]int + codebits *[numCodes]int + + // Output history, buffer. + dict dictDecoder + + // Next step in the decompression, + // and decompression state. + step func(*decompressor) + stepState int + err error + toRead []byte + hl, hd *huffmanDecoder + copyLen int + copyDist int + + // Temporary buffer (avoids repeated allocation). + buf [4]byte + + // Input bits, in top of b. + b uint32 + + nb uint + final bool +} + +func (f *decompressor) nextBlock() { + for f.nb < 1+2 { + if f.err = f.moreBits(); f.err != nil { + return + } + } + f.final = f.b&1 == 1 + f.b >>= 1 + typ := f.b & 3 + f.b >>= 2 + f.nb -= 1 + 2 + switch typ { + case 0: + f.dataBlock() + if debugDecode { + fmt.Println("stored block") + } + case 1: + // compressed, fixed Huffman tables + f.hl = &fixedHuffmanDecoder + f.hd = nil + f.huffmanBlockDecoder()() + if debugDecode { + fmt.Println("predefinied huffman block") + } + case 2: + // compressed, dynamic Huffman tables + if f.err = f.readHuffman(); f.err != nil { + break + } + f.hl = &f.h1 + f.hd = &f.h2 + f.huffmanBlockDecoder()() + if debugDecode { + fmt.Println("dynamic huffman block") + } + default: + // 3 is reserved. + if debugDecode { + fmt.Println("reserved data block encountered") + } + f.err = CorruptInputError(f.roffset) + } +} + +func (f *decompressor) Read(b []byte) (int, error) { + for { + if len(f.toRead) > 0 { + n := copy(b, f.toRead) + f.toRead = f.toRead[n:] + if len(f.toRead) == 0 { + return n, f.err + } + return n, nil + } + if f.err != nil { + return 0, f.err + } + f.step(f) + if f.err != nil && len(f.toRead) == 0 { + f.toRead = f.dict.readFlush() // Flush what's left in case of error + } + } +} + +// Support the io.WriteTo interface for io.Copy and friends. +func (f *decompressor) WriteTo(w io.Writer) (int64, error) { + total := int64(0) + flushed := false + for { + if len(f.toRead) > 0 { + n, err := w.Write(f.toRead) + total += int64(n) + if err != nil { + f.err = err + return total, err + } + if n != len(f.toRead) { + return total, io.ErrShortWrite + } + f.toRead = f.toRead[:0] + } + if f.err != nil && flushed { + if f.err == io.EOF { + return total, nil + } + return total, f.err + } + if f.err == nil { + f.step(f) + } + if len(f.toRead) == 0 && f.err != nil && !flushed { + f.toRead = f.dict.readFlush() // Flush what's left in case of error + flushed = true + } + } +} + +func (f *decompressor) Close() error { + if f.err == io.EOF { + return nil + } + return f.err +} + +// RFC 1951 section 3.2.7. +// Compression with dynamic Huffman codes + +var codeOrder = [...]int{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15} + +func (f *decompressor) readHuffman() error { + // HLIT[5], HDIST[5], HCLEN[4]. + for f.nb < 5+5+4 { + if err := f.moreBits(); err != nil { + return err + } + } + nlit := int(f.b&0x1F) + 257 + if nlit > maxNumLit { + if debugDecode { + fmt.Println("nlit > maxNumLit", nlit) + } + return CorruptInputError(f.roffset) + } + f.b >>= 5 + ndist := int(f.b&0x1F) + 1 + if ndist > maxNumDist { + if debugDecode { + fmt.Println("ndist > maxNumDist", ndist) + } + return CorruptInputError(f.roffset) + } + f.b >>= 5 + nclen := int(f.b&0xF) + 4 + // numCodes is 19, so nclen is always valid. + f.b >>= 4 + f.nb -= 5 + 5 + 4 + + // (HCLEN+4)*3 bits: code lengths in the magic codeOrder order. + for i := 0; i < nclen; i++ { + for f.nb < 3 { + if err := f.moreBits(); err != nil { + return err + } + } + f.codebits[codeOrder[i]] = int(f.b & 0x7) + f.b >>= 3 + f.nb -= 3 + } + for i := nclen; i < len(codeOrder); i++ { + f.codebits[codeOrder[i]] = 0 + } + if !f.h1.init(f.codebits[0:]) { + if debugDecode { + fmt.Println("init codebits failed") + } + return CorruptInputError(f.roffset) + } + + // HLIT + 257 code lengths, HDIST + 1 code lengths, + // using the code length Huffman code. + for i, n := 0, nlit+ndist; i < n; { + x, err := f.huffSym(&f.h1) + if err != nil { + return err + } + if x < 16 { + // Actual length. + f.bits[i] = x + i++ + continue + } + // Repeat previous length or zero. + var rep int + var nb uint + var b int + switch x { + default: + return InternalError("unexpected length code") + case 16: + rep = 3 + nb = 2 + if i == 0 { + if debugDecode { + fmt.Println("i==0") + } + return CorruptInputError(f.roffset) + } + b = f.bits[i-1] + case 17: + rep = 3 + nb = 3 + b = 0 + case 18: + rep = 11 + nb = 7 + b = 0 + } + for f.nb < nb { + if err := f.moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits:", err) + } + return err + } + } + rep += int(f.b & uint32(1<<(nb®SizeMaskUint32)-1)) + f.b >>= nb & regSizeMaskUint32 + f.nb -= nb + if i+rep > n { + if debugDecode { + fmt.Println("i+rep > n", i, rep, n) + } + return CorruptInputError(f.roffset) + } + for j := 0; j < rep; j++ { + f.bits[i] = b + i++ + } + } + + if !f.h1.init(f.bits[0:nlit]) || !f.h2.init(f.bits[nlit:nlit+ndist]) { + if debugDecode { + fmt.Println("init2 failed") + } + return CorruptInputError(f.roffset) + } + + // As an optimization, we can initialize the maxRead bits to read at a time + // for the HLIT tree to the length of the EOB marker since we know that + // every block must terminate with one. This preserves the property that + // we never read any extra bytes after the end of the DEFLATE stream. + if f.h1.maxRead < f.bits[endBlockMarker] { + f.h1.maxRead = f.bits[endBlockMarker] + } + if !f.final { + // If not the final block, the smallest block possible is + // a predefined table, BTYPE=01, with a single EOB marker. + // This will take up 3 + 7 bits. + f.h1.maxRead += 10 + } + + return nil +} + +// Copy a single uncompressed data block from input to output. +func (f *decompressor) dataBlock() { + // Uncompressed. + // Discard current half-byte. + left := (f.nb) & 7 + f.nb -= left + f.b >>= left + + offBytes := f.nb >> 3 + // Unfilled values will be overwritten. + f.buf[0] = uint8(f.b) + f.buf[1] = uint8(f.b >> 8) + f.buf[2] = uint8(f.b >> 16) + f.buf[3] = uint8(f.b >> 24) + + f.roffset += int64(offBytes) + f.nb, f.b = 0, 0 + + // Length then ones-complement of length. + nr, err := io.ReadFull(f.r, f.buf[offBytes:4]) + f.roffset += int64(nr) + if err != nil { + f.err = noEOF(err) + return + } + n := uint16(f.buf[0]) | uint16(f.buf[1])<<8 + nn := uint16(f.buf[2]) | uint16(f.buf[3])<<8 + if nn != ^n { + if debugDecode { + ncomp := ^n + fmt.Println("uint16(nn) != uint16(^n)", nn, ncomp) + } + f.err = CorruptInputError(f.roffset) + return + } + + if n == 0 { + f.toRead = f.dict.readFlush() + f.finishBlock() + return + } + + f.copyLen = int(n) + f.copyData() +} + +// copyData copies f.copyLen bytes from the underlying reader into f.hist. +// It pauses for reads when f.hist is full. +func (f *decompressor) copyData() { + buf := f.dict.writeSlice() + if len(buf) > f.copyLen { + buf = buf[:f.copyLen] + } + + cnt, err := io.ReadFull(f.r, buf) + f.roffset += int64(cnt) + f.copyLen -= cnt + f.dict.writeMark(cnt) + if err != nil { + f.err = noEOF(err) + return + } + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).copyData + return + } + f.finishBlock() +} + +func (f *decompressor) finishBlock() { + if f.final { + if f.dict.availRead() > 0 { + f.toRead = f.dict.readFlush() + } + f.err = io.EOF + } + f.step = (*decompressor).nextBlock +} + +// noEOF returns err, unless err == io.EOF, in which case it returns io.ErrUnexpectedEOF. +func noEOF(e error) error { + if e == io.EOF { + return io.ErrUnexpectedEOF + } + return e +} + +func (f *decompressor) moreBits() error { + c, err := f.r.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << (f.nb & regSizeMaskUint32) + f.nb += 8 + return nil +} + +// Read the next Huffman-encoded symbol from f according to h. +func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(h.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := f.r.ReadByte() + if err != nil { + f.b = b + f.nb = nb + return 0, noEOF(err) + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := h.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = h.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&h.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return 0, f.err + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + return int(chunk >> huffmanValueShift), nil + } + } +} + +func makeReader(r io.Reader) Reader { + if rr, ok := r.(Reader); ok { + return rr + } + return bufio.NewReader(r) +} + +func fixedHuffmanDecoderInit() { + fixedOnce.Do(func() { + // These come from the RFC section 3.2.6. + var bits [288]int + for i := 0; i < 144; i++ { + bits[i] = 8 + } + for i := 144; i < 256; i++ { + bits[i] = 9 + } + for i := 256; i < 280; i++ { + bits[i] = 7 + } + for i := 280; i < 288; i++ { + bits[i] = 8 + } + fixedHuffmanDecoder.init(bits[:]) + }) +} + +func (f *decompressor) Reset(r io.Reader, dict []byte) error { + *f = decompressor{ + r: makeReader(r), + bits: f.bits, + codebits: f.codebits, + h1: f.h1, + h2: f.h2, + dict: f.dict, + step: (*decompressor).nextBlock, + } + f.dict.init(maxMatchOffset, dict) + return nil +} + +// NewReader returns a new ReadCloser that can be used +// to read the uncompressed version of r. +// If r does not also implement io.ByteReader, +// the decompressor may read more data than necessary from r. +// It is the caller's responsibility to call Close on the ReadCloser +// when finished reading. +// +// The ReadCloser returned by NewReader also implements Resetter. +func NewReader(r io.Reader) io.ReadCloser { + fixedHuffmanDecoderInit() + + var f decompressor + f.r = makeReader(r) + f.bits = new([maxNumLit + maxNumDist]int) + f.codebits = new([numCodes]int) + f.step = (*decompressor).nextBlock + f.dict.init(maxMatchOffset, nil) + return &f +} + +// NewReaderDict is like NewReader but initializes the reader +// with a preset dictionary. The returned Reader behaves as if +// the uncompressed data stream started with the given dictionary, +// which has already been read. NewReaderDict is typically used +// to read data compressed by NewWriterDict. +// +// The ReadCloser returned by NewReader also implements Resetter. +func NewReaderDict(r io.Reader, dict []byte) io.ReadCloser { + fixedHuffmanDecoderInit() + + var f decompressor + f.r = makeReader(r) + f.bits = new([maxNumLit + maxNumDist]int) + f.codebits = new([numCodes]int) + f.step = (*decompressor).nextBlock + f.dict.init(maxMatchOffset, dict) + return &f +} diff --git a/vendor/github.com/klauspost/compress/flate/inflate_gen.go b/vendor/github.com/klauspost/compress/flate/inflate_gen.go new file mode 100644 index 0000000..61342b6 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/inflate_gen.go @@ -0,0 +1,1283 @@ +// Code generated by go generate gen_inflate.go. DO NOT EDIT. + +package flate + +import ( + "bufio" + "bytes" + "fmt" + "math/bits" + "strings" +) + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanBytesBuffer() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*bytes.Buffer) + + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + fnb, fb, dict := f.nb, f.b, &f.dict + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanBytesBuffer + f.stepState = stateInit + f.b, f.nb = fb, fnb + return + } + goto readLiteral + case v == 256: + f.b, f.nb = fb, fnb + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return + } + + var dist uint32 + if f.hd == nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra + default: + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanBytesBuffer // We need to continue this work + f.stepState = stateDict + f.b, f.nb = fb, fnb + return + } + goto readLiteral + } + // Not reached +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanBytesReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*bytes.Reader) + + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + fnb, fb, dict := f.nb, f.b, &f.dict + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanBytesReader + f.stepState = stateInit + f.b, f.nb = fb, fnb + return + } + goto readLiteral + case v == 256: + f.b, f.nb = fb, fnb + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return + } + + var dist uint32 + if f.hd == nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra + default: + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanBytesReader // We need to continue this work + f.stepState = stateDict + f.b, f.nb = fb, fnb + return + } + goto readLiteral + } + // Not reached +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanBufioReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*bufio.Reader) + + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + fnb, fb, dict := f.nb, f.b, &f.dict + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanBufioReader + f.stepState = stateInit + f.b, f.nb = fb, fnb + return + } + goto readLiteral + case v == 256: + f.b, f.nb = fb, fnb + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return + } + + var dist uint32 + if f.hd == nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra + default: + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanBufioReader // We need to continue this work + f.stepState = stateDict + f.b, f.nb = fb, fnb + return + } + goto readLiteral + } + // Not reached +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanStringsReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*strings.Reader) + + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + fnb, fb, dict := f.nb, f.b, &f.dict + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanStringsReader + f.stepState = stateInit + f.b, f.nb = fb, fnb + return + } + goto readLiteral + case v == 256: + f.b, f.nb = fb, fnb + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return + } + + var dist uint32 + if f.hd == nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra + default: + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanStringsReader // We need to continue this work + f.stepState = stateDict + f.b, f.nb = fb, fnb + return + } + goto readLiteral + } + // Not reached +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanGenericReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(Reader) + + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + fnb, fb, dict := f.nb, f.b, &f.dict + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanGenericReader + f.stepState = stateInit + f.b, f.nb = fb, fnb + return + } + goto readLiteral + case v == 256: + f.b, f.nb = fb, fnb + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return + } + + var dist uint32 + if f.hd == nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra + default: + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanGenericReader // We need to continue this work + f.stepState = stateDict + f.b, f.nb = fb, fnb + return + } + goto readLiteral + } + // Not reached +} + +func (f *decompressor) huffmanBlockDecoder() func() { + switch f.r.(type) { + case *bytes.Buffer: + return f.huffmanBytesBuffer + case *bytes.Reader: + return f.huffmanBytesReader + case *bufio.Reader: + return f.huffmanBufioReader + case *strings.Reader: + return f.huffmanStringsReader + case Reader: + return f.huffmanGenericReader + default: + return f.huffmanGenericReader + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level1.go b/vendor/github.com/klauspost/compress/flate/level1.go new file mode 100644 index 0000000..0f14f8d --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/level1.go @@ -0,0 +1,240 @@ +package flate + +import ( + "encoding/binary" + "fmt" + "math/bits" +) + +// fastGen maintains the table for matches, +// and the previous byte block for level 2. +// This is the generic implementation. +type fastEncL1 struct { + fastGen + table [tableSize]tableEntry +} + +// EncodeL1 uses a similar algorithm to level 1 +func (e *fastEncL1) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3232(src, s) + + for { + const skipLog = 5 + const doEvery = 2 + + nextS := s + var candidate tableEntry + for { + nextHash := hash(cv) + candidate = e.table[nextHash] + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + + now := load6432(src, nextS) + e.table[nextHash] = tableEntry{offset: s + e.cur} + nextHash = hash(uint32(now)) + + offset := s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} + break + } + + // Do one right away... + cv = uint32(now) + s = nextS + nextS++ + candidate = e.table[nextHash] + now >>= 8 + e.table[nextHash] = tableEntry{offset: s + e.cur} + + offset = s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} + break + } + cv = uint32(now) + s = nextS + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + + // Extend the 4-byte match as long as possible. + t := candidate.offset - e.cur + var l = int32(4) + if false { + l = e.matchlenLong(s+4, t+4, src) + 4 + } else { + // inlined: + a := src[s+4:] + b := src[t+4:] + for len(a) >= 8 { + if diff := binary.LittleEndian.Uint64(a) ^ binary.LittleEndian.Uint64(b); diff != 0 { + l += int32(bits.TrailingZeros64(diff) >> 3) + break + } + l += 8 + a = a[8:] + b = b[8:] + } + if len(a) < 8 { + b = b[:len(a)] + for i := range a { + if a[i] != b[i] { + break + } + l++ + } + } + } + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + + // Save the match found + if false { + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + } else { + // Inlined... + xoffset := uint32(s - t - baseMatchOffset) + xlength := l + oc := offsetCode(xoffset) + xoffset |= oc << 16 + for xlength > 0 { + xl := xlength + if xl > 258 { + if xl > 258+baseMatchLength { + xl = 258 + } else { + xl = 258 - baseMatchLength + } + } + xlength -= xl + xl -= baseMatchLength + dst.extraHist[lengthCodes1[uint8(xl)]]++ + dst.offHist[oc]++ + dst.tokens[dst.n] = token(matchType | uint32(xl)<= s { + s = nextS + 1 + } + if s >= sLimit { + // Index first pair after match end. + if int(s+l+4) < len(src) { + cv := load3232(src, s) + e.table[hash(cv)] = tableEntry{offset: s + e.cur} + } + goto emitRemainder + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 and at s. If + // another emitCopy is not our next move, also calculate nextHash + // at s+1. At least on GOARCH=amd64, these three hash calculations + // are faster as one load64 call (with some shifts) instead of + // three load32 calls. + x := load6432(src, s-2) + o := e.cur + s - 2 + prevHash := hash(uint32(x)) + e.table[prevHash] = tableEntry{offset: o} + x >>= 16 + currHash := hash(uint32(x)) + candidate = e.table[currHash] + e.table[currHash] = tableEntry{offset: o + 2} + + offset := s - (candidate.offset - e.cur) + if offset > maxMatchOffset || uint32(x) != load3232(src, candidate.offset-e.cur) { + cv = uint32(x >> 8) + s++ + break + } + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level2.go b/vendor/github.com/klauspost/compress/flate/level2.go new file mode 100644 index 0000000..8603fbd --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/level2.go @@ -0,0 +1,213 @@ +package flate + +import "fmt" + +// fastGen maintains the table for matches, +// and the previous byte block for level 2. +// This is the generic implementation. +type fastEncL2 struct { + fastGen + table [bTableSize]tableEntry +} + +// EncodeL2 uses a similar algorithm to level 1, but is capable +// of matching across blocks giving better compression at a small slowdown. +func (e *fastEncL2) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3232(src, s) + for { + // When should we start skipping if we haven't found matches in a long while. + const skipLog = 5 + const doEvery = 2 + + nextS := s + var candidate tableEntry + for { + nextHash := hash4u(cv, bTableBits) + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + candidate = e.table[nextHash] + now := load6432(src, nextS) + e.table[nextHash] = tableEntry{offset: s + e.cur} + nextHash = hash4u(uint32(now), bTableBits) + + offset := s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} + break + } + + // Do one right away... + cv = uint32(now) + s = nextS + nextS++ + candidate = e.table[nextHash] + now >>= 8 + e.table[nextHash] = tableEntry{offset: s + e.cur} + + offset = s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + break + } + cv = uint32(now) + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + + // Call emitCopy, and then see if another emitCopy could be our next + // move. Repeat until we find no match for the input immediately after + // what was consumed by the last emitCopy call. + // + // If we exit this loop normally then we need to call emitLiteral next, + // though we don't yet know how big the literal will be. We handle that + // by proceeding to the next iteration of the main loop. We also can + // exit this loop via goto if we get close to exhausting the input. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + + // Extend the 4-byte match as long as possible. + t := candidate.offset - e.cur + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + // Index first pair after match end. + if int(s+l+4) < len(src) { + cv := load3232(src, s) + e.table[hash4u(cv, bTableBits)] = tableEntry{offset: s + e.cur} + } + goto emitRemainder + } + + // Store every second hash in-between, but offset by 1. + for i := s - l + 2; i < s-5; i += 7 { + x := load6432(src, i) + nextHash := hash4u(uint32(x), bTableBits) + e.table[nextHash] = tableEntry{offset: e.cur + i} + // Skip one + x >>= 16 + nextHash = hash4u(uint32(x), bTableBits) + e.table[nextHash] = tableEntry{offset: e.cur + i + 2} + // Skip one + x >>= 16 + nextHash = hash4u(uint32(x), bTableBits) + e.table[nextHash] = tableEntry{offset: e.cur + i + 4} + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 to s. If + // another emitCopy is not our next move, also calculate nextHash + // at s+1. At least on GOARCH=amd64, these three hash calculations + // are faster as one load64 call (with some shifts) instead of + // three load32 calls. + x := load6432(src, s-2) + o := e.cur + s - 2 + prevHash := hash4u(uint32(x), bTableBits) + prevHash2 := hash4u(uint32(x>>8), bTableBits) + e.table[prevHash] = tableEntry{offset: o} + e.table[prevHash2] = tableEntry{offset: o + 1} + currHash := hash4u(uint32(x>>16), bTableBits) + candidate = e.table[currHash] + e.table[currHash] = tableEntry{offset: o + 2} + + offset := s - (candidate.offset - e.cur) + if offset > maxMatchOffset || uint32(x>>16) != load3232(src, candidate.offset-e.cur) { + cv = uint32(x >> 24) + s++ + break + } + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level3.go b/vendor/github.com/klauspost/compress/flate/level3.go new file mode 100644 index 0000000..039639f --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/level3.go @@ -0,0 +1,240 @@ +package flate + +import "fmt" + +// fastEncL3 +type fastEncL3 struct { + fastGen + table [1 << 16]tableEntryPrev +} + +// Encode uses a similar algorithm to level 2, will check up to two candidates. +func (e *fastEncL3) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 8 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + tableBits = 16 + tableSize = 1 << tableBits + ) + + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntryPrev{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i] + if v.Cur.offset <= minOff { + v.Cur.offset = 0 + } else { + v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset + } + if v.Prev.offset <= minOff { + v.Prev.offset = 0 + } else { + v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset + } + e.table[i] = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // Skip if too small. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3232(src, s) + for { + const skipLog = 6 + nextS := s + var candidate tableEntry + for { + nextHash := hash4u(cv, tableBits) + s = nextS + nextS = s + 1 + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + candidates := e.table[nextHash] + now := load3232(src, nextS) + + // Safe offset distance until s + 4... + minOffset := e.cur + s - (maxMatchOffset - 4) + e.table[nextHash] = tableEntryPrev{Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur}} + + // Check both candidates + candidate = candidates.Cur + if candidate.offset < minOffset { + cv = now + // Previous will also be invalid, we have nothing. + continue + } + + if cv == load3232(src, candidate.offset-e.cur) { + if candidates.Prev.offset < minOffset || cv != load3232(src, candidates.Prev.offset-e.cur) { + break + } + // Both match and are valid, pick longest. + offset := s - (candidate.offset - e.cur) + o2 := s - (candidates.Prev.offset - e.cur) + l1, l2 := matchLen(src[s+4:], src[s-offset+4:]), matchLen(src[s+4:], src[s-o2+4:]) + if l2 > l1 { + candidate = candidates.Prev + } + break + } else { + // We only check if value mismatches. + // Offset will always be invalid in other cases. + candidate = candidates.Prev + if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { + break + } + } + cv = now + } + + // Call emitCopy, and then see if another emitCopy could be our next + // move. Repeat until we find no match for the input immediately after + // what was consumed by the last emitCopy call. + // + // If we exit this loop normally then we need to call emitLiteral next, + // though we don't yet know how big the literal will be. We handle that + // by proceeding to the next iteration of the main loop. We also can + // exit this loop via goto if we get close to exhausting the input. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + + // Extend the 4-byte match as long as possible. + // + t := candidate.offset - e.cur + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + t += l + // Index first pair after match end. + if int(t+4) < len(src) && t > 0 { + cv := load3232(src, t) + nextHash := hash4u(cv, tableBits) + e.table[nextHash] = tableEntryPrev{ + Prev: e.table[nextHash].Cur, + Cur: tableEntry{offset: e.cur + t}, + } + } + goto emitRemainder + } + + // Store every 5th hash in-between. + for i := s - l + 2; i < s-5; i += 5 { + nextHash := hash4u(load3232(src, i), tableBits) + e.table[nextHash] = tableEntryPrev{ + Prev: e.table[nextHash].Cur, + Cur: tableEntry{offset: e.cur + i}} + } + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 to s. + x := load6432(src, s-2) + prevHash := hash4u(uint32(x), tableBits) + + e.table[prevHash] = tableEntryPrev{ + Prev: e.table[prevHash].Cur, + Cur: tableEntry{offset: e.cur + s - 2}, + } + x >>= 8 + prevHash = hash4u(uint32(x), tableBits) + + e.table[prevHash] = tableEntryPrev{ + Prev: e.table[prevHash].Cur, + Cur: tableEntry{offset: e.cur + s - 1}, + } + x >>= 8 + currHash := hash4u(uint32(x), tableBits) + candidates := e.table[currHash] + cv = uint32(x) + e.table[currHash] = tableEntryPrev{ + Prev: candidates.Cur, + Cur: tableEntry{offset: s + e.cur}, + } + + // Check both candidates + candidate = candidates.Cur + minOffset := e.cur + s - (maxMatchOffset - 4) + + if candidate.offset > minOffset { + if cv == load3232(src, candidate.offset-e.cur) { + // Found a match... + continue + } + candidate = candidates.Prev + if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { + // Match at prev... + continue + } + } + cv = uint32(x >> 8) + s++ + break + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level4.go b/vendor/github.com/klauspost/compress/flate/level4.go new file mode 100644 index 0000000..1cbffa1 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/level4.go @@ -0,0 +1,220 @@ +package flate + +import "fmt" + +type fastEncL4 struct { + fastGen + table [tableSize]tableEntry + bTable [tableSize]tableEntry +} + +func (e *fastEncL4) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.bTable[:] { + e.bTable[i] = tableEntry{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + for i := range e.bTable[:] { + v := e.bTable[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.bTable[i].offset = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load6432(src, s) + for { + const skipLog = 6 + const doEvery = 1 + + nextS := s + var t int32 + for { + nextHashS := hash4x64(cv, tableBits) + nextHashL := hash7(cv, tableBits) + + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + // Fetch a short+long candidate + sCandidate := e.table[nextHashS] + lCandidate := e.bTable[nextHashL] + next := load6432(src, nextS) + entry := tableEntry{offset: s + e.cur} + e.table[nextHashS] = entry + e.bTable[nextHashL] = entry + + t = lCandidate.offset - e.cur + if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.offset-e.cur) { + // We got a long match. Use that. + break + } + + t = sCandidate.offset - e.cur + if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) { + // Found a 4 match... + lCandidate = e.bTable[hash7(next, tableBits)] + + // If the next long is a candidate, check if we should use that instead... + lOff := nextS - (lCandidate.offset - e.cur) + if lOff < maxMatchOffset && load3232(src, lCandidate.offset-e.cur) == uint32(next) { + l1, l2 := matchLen(src[s+4:], src[t+4:]), matchLen(src[nextS+4:], src[nextS-lOff+4:]) + if l2 > l1 { + s = nextS + t = lCandidate.offset - e.cur + } + } + break + } + cv = next + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + + // Extend the 4-byte match as long as possible. + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + if debugDeflate { + if t >= s { + panic("s-t") + } + if (s - t) > maxMatchOffset { + panic(fmt.Sprintln("mmo", t)) + } + if l < baseMatchLength { + panic("bml") + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + // Index first pair after match end. + if int(s+8) < len(src) { + cv := load6432(src, s) + e.table[hash4x64(cv, tableBits)] = tableEntry{offset: s + e.cur} + e.bTable[hash7(cv, tableBits)] = tableEntry{offset: s + e.cur} + } + goto emitRemainder + } + + // Store every 3rd hash in-between + if true { + i := nextS + if i < s-1 { + cv := load6432(src, i) + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} + e.bTable[hash7(cv, tableBits)] = t + e.bTable[hash7(cv>>8, tableBits)] = t2 + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 + + i += 3 + for ; i < s-1; i += 3 { + cv := load6432(src, i) + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} + e.bTable[hash7(cv, tableBits)] = t + e.bTable[hash7(cv>>8, tableBits)] = t2 + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 + } + } + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. + x := load6432(src, s-1) + o := e.cur + s - 1 + prevHashS := hash4x64(x, tableBits) + prevHashL := hash7(x, tableBits) + e.table[prevHashS] = tableEntry{offset: o} + e.bTable[prevHashL] = tableEntry{offset: o} + cv = x >> 8 + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level5.go b/vendor/github.com/klauspost/compress/flate/level5.go new file mode 100644 index 0000000..4b97576 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/level5.go @@ -0,0 +1,302 @@ +package flate + +import "fmt" + +type fastEncL5 struct { + fastGen + table [tableSize]tableEntry + bTable [tableSize]tableEntryPrev +} + +func (e *fastEncL5) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.bTable[:] { + e.bTable[i] = tableEntryPrev{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + for i := range e.bTable[:] { + v := e.bTable[i] + if v.Cur.offset <= minOff { + v.Cur.offset = 0 + v.Prev.offset = 0 + } else { + v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset + if v.Prev.offset <= minOff { + v.Prev.offset = 0 + } else { + v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset + } + } + e.bTable[i] = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load6432(src, s) + for { + const skipLog = 6 + const doEvery = 1 + + nextS := s + var l int32 + var t int32 + for { + nextHashS := hash4x64(cv, tableBits) + nextHashL := hash7(cv, tableBits) + + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + // Fetch a short+long candidate + sCandidate := e.table[nextHashS] + lCandidate := e.bTable[nextHashL] + next := load6432(src, nextS) + entry := tableEntry{offset: s + e.cur} + e.table[nextHashS] = entry + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = entry, eLong.Cur + + nextHashS = hash4x64(next, tableBits) + nextHashL = hash7(next, tableBits) + + t = lCandidate.Cur.offset - e.cur + if s-t < maxMatchOffset { + if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) { + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + + t2 := lCandidate.Prev.offset - e.cur + if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { + l = e.matchlen(s+4, t+4, src) + 4 + ml1 := e.matchlen(s+4, t2+4, src) + 4 + if ml1 > l { + t = t2 + l = ml1 + break + } + } + break + } + t = lCandidate.Prev.offset - e.cur + if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + break + } + } + + t = sCandidate.offset - e.cur + if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) { + // Found a 4 match... + l = e.matchlen(s+4, t+4, src) + 4 + lCandidate = e.bTable[nextHashL] + // Store the next match + + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + + // If the next long is a candidate, use that... + t2 := lCandidate.Cur.offset - e.cur + if nextS-t2 < maxMatchOffset { + if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) { + ml := e.matchlen(nextS+4, t2+4, src) + 4 + if ml > l { + t = t2 + s = nextS + l = ml + break + } + } + // If the previous long is a candidate, use that... + t2 = lCandidate.Prev.offset - e.cur + if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) { + ml := e.matchlen(nextS+4, t2+4, src) + 4 + if ml > l { + t = t2 + s = nextS + l = ml + break + } + } + } + break + } + cv = next + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + + if l == 0 { + // Extend the 4-byte match as long as possible. + l = e.matchlenLong(s+4, t+4, src) + 4 + } else if l == maxMatchLength { + l += e.matchlenLong(s+l, t+l, src) + } + + // Try to locate a better match by checking the end of best match... + if sAt := s + l; l < 30 && sAt < sLimit { + eLong := e.bTable[hash7(load6432(src, sAt), tableBits)].Cur.offset + // Test current + t2 := eLong - e.cur - l + off := s - t2 + if t2 >= 0 && off < maxMatchOffset && off > 0 { + if l2 := e.matchlenLong(s, t2, src); l2 > l { + t = t2 + l = l2 + } + } + } + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + if debugDeflate { + if t >= s { + panic(fmt.Sprintln("s-t", s, t)) + } + if (s - t) > maxMatchOffset { + panic(fmt.Sprintln("mmo", s-t)) + } + if l < baseMatchLength { + panic("bml") + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + goto emitRemainder + } + + // Store every 3rd hash in-between. + if true { + const hashEvery = 3 + i := s - l + 1 + if i < s-1 { + cv := load6432(src, i) + t := tableEntry{offset: i + e.cur} + e.table[hash4x64(cv, tableBits)] = t + eLong := &e.bTable[hash7(cv, tableBits)] + eLong.Cur, eLong.Prev = t, eLong.Cur + + // Do an long at i+1 + cv >>= 8 + t = tableEntry{offset: t.offset + 1} + eLong = &e.bTable[hash7(cv, tableBits)] + eLong.Cur, eLong.Prev = t, eLong.Cur + + // We only have enough bits for a short entry at i+2 + cv >>= 8 + t = tableEntry{offset: t.offset + 1} + e.table[hash4x64(cv, tableBits)] = t + + // Skip one - otherwise we risk hitting 's' + i += 4 + for ; i < s-1; i += hashEvery { + cv := load6432(src, i) + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} + eLong := &e.bTable[hash7(cv, tableBits)] + eLong.Cur, eLong.Prev = t, eLong.Cur + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 + } + } + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. + x := load6432(src, s-1) + o := e.cur + s - 1 + prevHashS := hash4x64(x, tableBits) + prevHashL := hash7(x, tableBits) + e.table[prevHashS] = tableEntry{offset: o} + eLong := &e.bTable[prevHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: o}, eLong.Cur + cv = x >> 8 + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level6.go b/vendor/github.com/klauspost/compress/flate/level6.go new file mode 100644 index 0000000..62888ed --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/level6.go @@ -0,0 +1,315 @@ +package flate + +import "fmt" + +type fastEncL6 struct { + fastGen + table [tableSize]tableEntry + bTable [tableSize]tableEntryPrev +} + +func (e *fastEncL6) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.bTable[:] { + e.bTable[i] = tableEntryPrev{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + for i := range e.bTable[:] { + v := e.bTable[i] + if v.Cur.offset <= minOff { + v.Cur.offset = 0 + v.Prev.offset = 0 + } else { + v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset + if v.Prev.offset <= minOff { + v.Prev.offset = 0 + } else { + v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset + } + } + e.bTable[i] = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load6432(src, s) + // Repeat MUST be > 1 and within range + repeat := int32(1) + for { + const skipLog = 7 + const doEvery = 1 + + nextS := s + var l int32 + var t int32 + for { + nextHashS := hash4x64(cv, tableBits) + nextHashL := hash7(cv, tableBits) + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + // Fetch a short+long candidate + sCandidate := e.table[nextHashS] + lCandidate := e.bTable[nextHashL] + next := load6432(src, nextS) + entry := tableEntry{offset: s + e.cur} + e.table[nextHashS] = entry + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = entry, eLong.Cur + + // Calculate hashes of 'next' + nextHashS = hash4x64(next, tableBits) + nextHashL = hash7(next, tableBits) + + t = lCandidate.Cur.offset - e.cur + if s-t < maxMatchOffset { + if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) { + // Long candidate matches at least 4 bytes. + + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + + // Check the previous long candidate as well. + t2 := lCandidate.Prev.offset - e.cur + if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { + l = e.matchlen(s+4, t+4, src) + 4 + ml1 := e.matchlen(s+4, t2+4, src) + 4 + if ml1 > l { + t = t2 + l = ml1 + break + } + } + break + } + // Current value did not match, but check if previous long value does. + t = lCandidate.Prev.offset - e.cur + if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + break + } + } + + t = sCandidate.offset - e.cur + if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) { + // Found a 4 match... + l = e.matchlen(s+4, t+4, src) + 4 + + // Look up next long candidate (at nextS) + lCandidate = e.bTable[nextHashL] + + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + + // Check repeat at s + repOff + const repOff = 1 + t2 := s - repeat + repOff + if load3232(src, t2) == uint32(cv>>(8*repOff)) { + ml := e.matchlen(s+4+repOff, t2+4, src) + 4 + if ml > l { + t = t2 + l = ml + s += repOff + // Not worth checking more. + break + } + } + + // If the next long is a candidate, use that... + t2 = lCandidate.Cur.offset - e.cur + if nextS-t2 < maxMatchOffset { + if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) { + ml := e.matchlen(nextS+4, t2+4, src) + 4 + if ml > l { + t = t2 + s = nextS + l = ml + // This is ok, but check previous as well. + } + } + // If the previous long is a candidate, use that... + t2 = lCandidate.Prev.offset - e.cur + if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) { + ml := e.matchlen(nextS+4, t2+4, src) + 4 + if ml > l { + t = t2 + s = nextS + l = ml + break + } + } + } + break + } + cv = next + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + + // Extend the 4-byte match as long as possible. + if l == 0 { + l = e.matchlenLong(s+4, t+4, src) + 4 + } else if l == maxMatchLength { + l += e.matchlenLong(s+l, t+l, src) + } + + // Try to locate a better match by checking the end-of-match... + if sAt := s + l; sAt < sLimit { + eLong := &e.bTable[hash7(load6432(src, sAt), tableBits)] + // Test current + t2 := eLong.Cur.offset - e.cur - l + off := s - t2 + if off < maxMatchOffset { + if off > 0 && t2 >= 0 { + if l2 := e.matchlenLong(s, t2, src); l2 > l { + t = t2 + l = l2 + } + } + // Test next: + t2 = eLong.Prev.offset - e.cur - l + off := s - t2 + if off > 0 && off < maxMatchOffset && t2 >= 0 { + if l2 := e.matchlenLong(s, t2, src); l2 > l { + t = t2 + l = l2 + } + } + } + } + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + if false { + if t >= s { + panic(fmt.Sprintln("s-t", s, t)) + } + if (s - t) > maxMatchOffset { + panic(fmt.Sprintln("mmo", s-t)) + } + if l < baseMatchLength { + panic("bml") + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + repeat = s - t + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + // Index after match end. + for i := nextS + 1; i < int32(len(src))-8; i += 2 { + cv := load6432(src, i) + e.table[hash4x64(cv, tableBits)] = tableEntry{offset: i + e.cur} + eLong := &e.bTable[hash7(cv, tableBits)] + eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur}, eLong.Cur + } + goto emitRemainder + } + + // Store every long hash in-between and every second short. + if true { + for i := nextS + 1; i < s-1; i += 2 { + cv := load6432(src, i) + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} + eLong := &e.bTable[hash7(cv, tableBits)] + eLong2 := &e.bTable[hash7(cv>>8, tableBits)] + e.table[hash4x64(cv, tableBits)] = t + eLong.Cur, eLong.Prev = t, eLong.Cur + eLong2.Cur, eLong2.Prev = t2, eLong2.Cur + } + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. + cv = load6432(src, s) + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/regmask_amd64.go b/vendor/github.com/klauspost/compress/flate/regmask_amd64.go new file mode 100644 index 0000000..6ed2806 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/regmask_amd64.go @@ -0,0 +1,37 @@ +package flate + +const ( + // Masks for shifts with register sizes of the shift value. + // This can be used to work around the x86 design of shifting by mod register size. + // It can be used when a variable shift is always smaller than the register size. + + // reg8SizeMaskX - shift value is 8 bits, shifted is X + reg8SizeMask8 = 7 + reg8SizeMask16 = 15 + reg8SizeMask32 = 31 + reg8SizeMask64 = 63 + + // reg16SizeMaskX - shift value is 16 bits, shifted is X + reg16SizeMask8 = reg8SizeMask8 + reg16SizeMask16 = reg8SizeMask16 + reg16SizeMask32 = reg8SizeMask32 + reg16SizeMask64 = reg8SizeMask64 + + // reg32SizeMaskX - shift value is 32 bits, shifted is X + reg32SizeMask8 = reg8SizeMask8 + reg32SizeMask16 = reg8SizeMask16 + reg32SizeMask32 = reg8SizeMask32 + reg32SizeMask64 = reg8SizeMask64 + + // reg64SizeMaskX - shift value is 64 bits, shifted is X + reg64SizeMask8 = reg8SizeMask8 + reg64SizeMask16 = reg8SizeMask16 + reg64SizeMask32 = reg8SizeMask32 + reg64SizeMask64 = reg8SizeMask64 + + // regSizeMaskUintX - shift value is uint, shifted is X + regSizeMaskUint8 = reg8SizeMask8 + regSizeMaskUint16 = reg8SizeMask16 + regSizeMaskUint32 = reg8SizeMask32 + regSizeMaskUint64 = reg8SizeMask64 +) diff --git a/vendor/github.com/klauspost/compress/flate/regmask_other.go b/vendor/github.com/klauspost/compress/flate/regmask_other.go new file mode 100644 index 0000000..1b7a2cb --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/regmask_other.go @@ -0,0 +1,40 @@ +//go:build !amd64 +// +build !amd64 + +package flate + +const ( + // Masks for shifts with register sizes of the shift value. + // This can be used to work around the x86 design of shifting by mod register size. + // It can be used when a variable shift is always smaller than the register size. + + // reg8SizeMaskX - shift value is 8 bits, shifted is X + reg8SizeMask8 = 0xff + reg8SizeMask16 = 0xff + reg8SizeMask32 = 0xff + reg8SizeMask64 = 0xff + + // reg16SizeMaskX - shift value is 16 bits, shifted is X + reg16SizeMask8 = 0xffff + reg16SizeMask16 = 0xffff + reg16SizeMask32 = 0xffff + reg16SizeMask64 = 0xffff + + // reg32SizeMaskX - shift value is 32 bits, shifted is X + reg32SizeMask8 = 0xffffffff + reg32SizeMask16 = 0xffffffff + reg32SizeMask32 = 0xffffffff + reg32SizeMask64 = 0xffffffff + + // reg64SizeMaskX - shift value is 64 bits, shifted is X + reg64SizeMask8 = 0xffffffffffffffff + reg64SizeMask16 = 0xffffffffffffffff + reg64SizeMask32 = 0xffffffffffffffff + reg64SizeMask64 = 0xffffffffffffffff + + // regSizeMaskUintX - shift value is uint, shifted is X + regSizeMaskUint8 = ^uint(0) + regSizeMaskUint16 = ^uint(0) + regSizeMaskUint32 = ^uint(0) + regSizeMaskUint64 = ^uint(0) +) diff --git a/vendor/github.com/klauspost/compress/flate/stateless.go b/vendor/github.com/klauspost/compress/flate/stateless.go new file mode 100644 index 0000000..93a1d15 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/stateless.go @@ -0,0 +1,305 @@ +package flate + +import ( + "io" + "math" + "sync" +) + +const ( + maxStatelessBlock = math.MaxInt16 + // dictionary will be taken from maxStatelessBlock, so limit it. + maxStatelessDict = 8 << 10 + + slTableBits = 13 + slTableSize = 1 << slTableBits + slTableShift = 32 - slTableBits +) + +type statelessWriter struct { + dst io.Writer + closed bool +} + +func (s *statelessWriter) Close() error { + if s.closed { + return nil + } + s.closed = true + // Emit EOF block + return StatelessDeflate(s.dst, nil, true, nil) +} + +func (s *statelessWriter) Write(p []byte) (n int, err error) { + err = StatelessDeflate(s.dst, p, false, nil) + if err != nil { + return 0, err + } + return len(p), nil +} + +func (s *statelessWriter) Reset(w io.Writer) { + s.dst = w + s.closed = false +} + +// NewStatelessWriter will do compression but without maintaining any state +// between Write calls. +// There will be no memory kept between Write calls, +// but compression and speed will be suboptimal. +// Because of this, the size of actual Write calls will affect output size. +func NewStatelessWriter(dst io.Writer) io.WriteCloser { + return &statelessWriter{dst: dst} +} + +// bitWriterPool contains bit writers that can be reused. +var bitWriterPool = sync.Pool{ + New: func() interface{} { + return newHuffmanBitWriter(nil) + }, +} + +// StatelessDeflate allows compressing directly to a Writer without retaining state. +// When returning everything will be flushed. +// Up to 8KB of an optional dictionary can be given which is presumed to precede the block. +// Longer dictionaries will be truncated and will still produce valid output. +// Sending nil dictionary is perfectly fine. +func StatelessDeflate(out io.Writer, in []byte, eof bool, dict []byte) error { + var dst tokens + bw := bitWriterPool.Get().(*huffmanBitWriter) + bw.reset(out) + defer func() { + // don't keep a reference to our output + bw.reset(nil) + bitWriterPool.Put(bw) + }() + if eof && len(in) == 0 { + // Just write an EOF block. + // Could be faster... + bw.writeStoredHeader(0, true) + bw.flush() + return bw.err + } + + // Truncate dict + if len(dict) > maxStatelessDict { + dict = dict[len(dict)-maxStatelessDict:] + } + + for len(in) > 0 { + todo := in + if len(todo) > maxStatelessBlock-len(dict) { + todo = todo[:maxStatelessBlock-len(dict)] + } + in = in[len(todo):] + uncompressed := todo + if len(dict) > 0 { + // combine dict and source + bufLen := len(todo) + len(dict) + combined := make([]byte, bufLen) + copy(combined, dict) + copy(combined[len(dict):], todo) + todo = combined + } + // Compress + statelessEnc(&dst, todo, int16(len(dict))) + isEof := eof && len(in) == 0 + + if dst.n == 0 { + bw.writeStoredHeader(len(uncompressed), isEof) + if bw.err != nil { + return bw.err + } + bw.writeBytes(uncompressed) + } else if int(dst.n) > len(uncompressed)-len(uncompressed)>>4 { + // If we removed less than 1/16th, huffman compress the block. + bw.writeBlockHuff(isEof, uncompressed, len(in) == 0) + } else { + bw.writeBlockDynamic(&dst, isEof, uncompressed, len(in) == 0) + } + if len(in) > 0 { + // Retain a dict if we have more + dict = todo[len(todo)-maxStatelessDict:] + dst.Reset() + } + if bw.err != nil { + return bw.err + } + } + if !eof { + // Align, only a stored block can do that. + bw.writeStoredHeader(0, false) + } + bw.flush() + return bw.err +} + +func hashSL(u uint32) uint32 { + return (u * 0x1e35a7bd) >> slTableShift +} + +func load3216(b []byte, i int16) uint32 { + // Help the compiler eliminate bounds checks on the read so it can be done in a single read. + b = b[i:] + b = b[:4] + return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 +} + +func load6416(b []byte, i int16) uint64 { + // Help the compiler eliminate bounds checks on the read so it can be done in a single read. + b = b[i:] + b = b[:8] + return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 | + uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56 +} + +func statelessEnc(dst *tokens, src []byte, startAt int16) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + + type tableEntry struct { + offset int16 + } + + var table [slTableSize]tableEntry + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src)-int(startAt) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = 0 + return + } + // Index until startAt + if startAt > 0 { + cv := load3232(src, 0) + for i := int16(0); i < startAt; i++ { + table[hashSL(cv)] = tableEntry{offset: i} + cv = (cv >> 8) | (uint32(src[i+4]) << 24) + } + } + + s := startAt + 1 + nextEmit := startAt + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := int16(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3216(src, s) + + for { + const skipLog = 5 + const doEvery = 2 + + nextS := s + var candidate tableEntry + for { + nextHash := hashSL(cv) + candidate = table[nextHash] + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit || nextS <= 0 { + goto emitRemainder + } + + now := load6416(src, nextS) + table[nextHash] = tableEntry{offset: s} + nextHash = hashSL(uint32(now)) + + if cv == load3216(src, candidate.offset) { + table[nextHash] = tableEntry{offset: nextS} + break + } + + // Do one right away... + cv = uint32(now) + s = nextS + nextS++ + candidate = table[nextHash] + now >>= 8 + table[nextHash] = tableEntry{offset: s} + + if cv == load3216(src, candidate.offset) { + table[nextHash] = tableEntry{offset: nextS} + break + } + cv = uint32(now) + s = nextS + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + + // Extend the 4-byte match as long as possible. + t := candidate.offset + l := int16(matchLen(src[s+4:], src[t+4:]) + 4) + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } + } + + // Save the match found + dst.AddMatchLong(int32(l), uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + if s >= sLimit { + goto emitRemainder + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 and at s. If + // another emitCopy is not our next move, also calculate nextHash + // at s+1. At least on GOARCH=amd64, these three hash calculations + // are faster as one load64 call (with some shifts) instead of + // three load32 calls. + x := load6416(src, s-2) + o := s - 2 + prevHash := hashSL(uint32(x)) + table[prevHash] = tableEntry{offset: o} + x >>= 16 + currHash := hashSL(uint32(x)) + candidate = table[currHash] + table[currHash] = tableEntry{offset: o + 2} + + if uint32(x) != load3216(src, candidate.offset) { + cv = uint32(x >> 8) + s++ + break + } + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/vendor/github.com/klauspost/compress/flate/token.go b/vendor/github.com/klauspost/compress/flate/token.go new file mode 100644 index 0000000..d818790 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/token.go @@ -0,0 +1,379 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "bytes" + "encoding/binary" + "fmt" + "io" + "math" +) + +const ( + // bits 0-16 xoffset = offset - MIN_OFFSET_SIZE, or literal - 16 bits + // bits 16-22 offsetcode - 5 bits + // bits 22-30 xlength = length - MIN_MATCH_LENGTH - 8 bits + // bits 30-32 type 0 = literal 1=EOF 2=Match 3=Unused - 2 bits + lengthShift = 22 + offsetMask = 1<maxnumlit + offHist [32]uint16 // offset codes + litHist [256]uint16 // codes 0->255 + nFilled int + n uint16 // Must be able to contain maxStoreBlockSize + tokens [maxStoreBlockSize + 1]token +} + +func (t *tokens) Reset() { + if t.n == 0 { + return + } + t.n = 0 + t.nFilled = 0 + for i := range t.litHist[:] { + t.litHist[i] = 0 + } + for i := range t.extraHist[:] { + t.extraHist[i] = 0 + } + for i := range t.offHist[:] { + t.offHist[i] = 0 + } +} + +func (t *tokens) Fill() { + if t.n == 0 { + return + } + for i, v := range t.litHist[:] { + if v == 0 { + t.litHist[i] = 1 + t.nFilled++ + } + } + for i, v := range t.extraHist[:literalCount-256] { + if v == 0 { + t.nFilled++ + t.extraHist[i] = 1 + } + } + for i, v := range t.offHist[:offsetCodeCount] { + if v == 0 { + t.offHist[i] = 1 + } + } +} + +func indexTokens(in []token) tokens { + var t tokens + t.indexTokens(in) + return t +} + +func (t *tokens) indexTokens(in []token) { + t.Reset() + for _, tok := range in { + if tok < matchType { + t.AddLiteral(tok.literal()) + continue + } + t.AddMatch(uint32(tok.length()), tok.offset()&matchOffsetOnlyMask) + } +} + +// emitLiteral writes a literal chunk and returns the number of bytes written. +func emitLiteral(dst *tokens, lit []byte) { + for _, v := range lit { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } +} + +func (t *tokens) AddLiteral(lit byte) { + t.tokens[t.n] = token(lit) + t.litHist[lit]++ + t.n++ +} + +// from https://stackoverflow.com/a/28730362 +func mFastLog2(val float32) float32 { + ux := int32(math.Float32bits(val)) + log2 := (float32)(((ux >> 23) & 255) - 128) + ux &= -0x7f800001 + ux += 127 << 23 + uval := math.Float32frombits(uint32(ux)) + log2 += ((-0.34484843)*uval+2.02466578)*uval - 0.67487759 + return log2 +} + +// EstimatedBits will return an minimum size estimated by an *optimal* +// compression of the block. +// The size of the block +func (t *tokens) EstimatedBits() int { + shannon := float32(0) + bits := int(0) + nMatches := 0 + total := int(t.n) + t.nFilled + if total > 0 { + invTotal := 1.0 / float32(total) + for _, v := range t.litHist[:] { + if v > 0 { + n := float32(v) + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n + } + } + // Just add 15 for EOB + shannon += 15 + for i, v := range t.extraHist[1 : literalCount-256] { + if v > 0 { + n := float32(v) + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n + bits += int(lengthExtraBits[i&31]) * int(v) + nMatches += int(v) + } + } + } + if nMatches > 0 { + invTotal := 1.0 / float32(nMatches) + for i, v := range t.offHist[:offsetCodeCount] { + if v > 0 { + n := float32(v) + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n + bits += int(offsetExtraBits[i&31]) * int(v) + } + } + } + return int(shannon) + bits +} + +// AddMatch adds a match to the tokens. +// This function is very sensitive to inlining and right on the border. +func (t *tokens) AddMatch(xlength uint32, xoffset uint32) { + if debugDeflate { + if xlength >= maxMatchLength+baseMatchLength { + panic(fmt.Errorf("invalid length: %v", xlength)) + } + if xoffset >= maxMatchOffset+baseMatchOffset { + panic(fmt.Errorf("invalid offset: %v", xoffset)) + } + } + oCode := offsetCode(xoffset) + xoffset |= oCode << 16 + + t.extraHist[lengthCodes1[uint8(xlength)]]++ + t.offHist[oCode&31]++ + t.tokens[t.n] = token(matchType | xlength<= maxMatchOffset+baseMatchOffset { + panic(fmt.Errorf("invalid offset: %v", xoffset)) + } + } + oc := offsetCode(xoffset) + xoffset |= oc << 16 + for xlength > 0 { + xl := xlength + if xl > 258 { + // We need to have at least baseMatchLength left over for next loop. + if xl > 258+baseMatchLength { + xl = 258 + } else { + xl = 258 - baseMatchLength + } + } + xlength -= xl + xl -= baseMatchLength + t.extraHist[lengthCodes1[uint8(xl)]]++ + t.offHist[oc&31]++ + t.tokens[t.n] = token(matchType | uint32(xl)<> lengthShift) } + +// Convert length to code. +func lengthCode(len uint8) uint8 { return lengthCodes[len] } + +// Returns the offset code corresponding to a specific offset +func offsetCode(off uint32) uint32 { + if false { + if off < uint32(len(offsetCodes)) { + return offsetCodes[off&255] + } else if off>>7 < uint32(len(offsetCodes)) { + return offsetCodes[(off>>7)&255] + 14 + } else { + return offsetCodes[(off>>14)&255] + 28 + } + } + if off < uint32(len(offsetCodes)) { + return offsetCodes[uint8(off)] + } + return offsetCodes14[uint8(off>>7)] +} diff --git a/vendor/github.com/klauspost/compress/fse/README.md b/vendor/github.com/klauspost/compress/fse/README.md new file mode 100644 index 0000000..ea7324d --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/README.md @@ -0,0 +1,79 @@ +# Finite State Entropy + +This package provides Finite State Entropy encoding and decoding. + +Finite State Entropy (also referenced as [tANS](https://en.wikipedia.org/wiki/Asymmetric_numeral_systems#tANS)) +encoding provides a fast near-optimal symbol encoding/decoding +for byte blocks as implemented in [zstandard](https://github.com/facebook/zstd). + +This can be used for compressing input with a lot of similar input values to the smallest number of bytes. +This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, +but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. + +* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/fse) + +## News + + * Feb 2018: First implementation released. Consider this beta software for now. + +# Usage + +This package provides a low level interface that allows to compress single independent blocks. + +Each block is separate, and there is no built in integrity checks. +This means that the caller should keep track of block sizes and also do checksums if needed. + +Compressing a block is done via the [`Compress`](https://godoc.org/github.com/klauspost/compress/fse#Compress) function. +You must provide input and will receive the output and maybe an error. + +These error values can be returned: + +| Error | Description | +|---------------------|-----------------------------------------------------------------------------| +| `` | Everything ok, output is returned | +| `ErrIncompressible` | Returned when input is judged to be too hard to compress | +| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | +| `(error)` | An internal error occurred. | + +As can be seen above there are errors that will be returned even under normal operation so it is important to handle these. + +To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/fse#Scratch) object +that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same +object can be used for both. + +Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this +you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. + +Decompressing is done by calling the [`Decompress`](https://godoc.org/github.com/klauspost/compress/fse#Decompress) function. +You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back +your input was likely corrupted. + +It is important to note that a successful decoding does *not* mean your output matches your original input. +There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. + +For more detailed usage, see examples in the [godoc documentation](https://godoc.org/github.com/klauspost/compress/fse#pkg-examples). + +# Performance + +A lot of factors are affecting speed. Block sizes and compressibility of the material are primary factors. +All compression functions are currently only running on the calling goroutine so only one core will be used per block. + +The compressor is significantly faster if symbols are kept as small as possible. The highest byte value of the input +is used to reduce some of the processing, so if all your input is above byte value 64 for instance, it may be +beneficial to transpose all your input values down by 64. + +With moderate block sizes around 64k speed are typically 200MB/s per core for compression and +around 300MB/s decompression speed. + +The same hardware typically does Huffman (deflate) encoding at 125MB/s and decompression at 100MB/s. + +# Plans + +At one point, more internals will be exposed to facilitate more "expert" usage of the components. + +A streaming interface is also likely to be implemented. Likely compatible with [FSE stream format](https://github.com/Cyan4973/FiniteStateEntropy/blob/dev/programs/fileio.c#L261). + +# Contributing + +Contributions are always welcome. Be aware that adding public functions will require good justification and breaking +changes will likely not be accepted. If in doubt open an issue before writing the PR. \ No newline at end of file diff --git a/vendor/github.com/klauspost/compress/fse/bitreader.go b/vendor/github.com/klauspost/compress/fse/bitreader.go new file mode 100644 index 0000000..f65eb39 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/bitreader.go @@ -0,0 +1,122 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +import ( + "encoding/binary" + "errors" + "io" +) + +// bitReader reads a bitstream in reverse. +// The last set bit indicates the start of the stream and is used +// for aligning the input. +type bitReader struct { + in []byte + off uint // next byte to read is at in[off - 1] + value uint64 + bitsRead uint8 +} + +// init initializes and resets the bit reader. +func (b *bitReader) init(in []byte) error { + if len(in) < 1 { + return errors.New("corrupt stream: too short") + } + b.in = in + b.off = uint(len(in)) + // The highest bit of the last byte indicates where to start + v := in[len(in)-1] + if v == 0 { + return errors.New("corrupt stream, did not find end of stream") + } + b.bitsRead = 64 + b.value = 0 + if len(in) >= 8 { + b.fillFastStart() + } else { + b.fill() + b.fill() + } + b.bitsRead += 8 - uint8(highBits(uint32(v))) + return nil +} + +// getBits will return n bits. n can be 0. +func (b *bitReader) getBits(n uint8) uint16 { + if n == 0 || b.bitsRead >= 64 { + return 0 + } + return b.getBitsFast(n) +} + +// getBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReader) getBitsFast(n uint8) uint16 { + const regMask = 64 - 1 + v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask)) + b.bitsRead += n + return v +} + +// fillFast() will make sure at least 32 bits are available. +// There must be at least 4 bytes available. +func (b *bitReader) fillFast() { + if b.bitsRead < 32 { + return + } + // 2 bounds checks. + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value = (b.value << 32) | uint64(low) + b.bitsRead -= 32 + b.off -= 4 +} + +// fill() will make sure at least 32 bits are available. +func (b *bitReader) fill() { + if b.bitsRead < 32 { + return + } + if b.off > 4 { + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value = (b.value << 32) | uint64(low) + b.bitsRead -= 32 + b.off -= 4 + return + } + for b.off > 0 { + b.value = (b.value << 8) | uint64(b.in[b.off-1]) + b.bitsRead -= 8 + b.off-- + } +} + +// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read. +func (b *bitReader) fillFastStart() { + // Do single re-slice to avoid bounds checks. + b.value = binary.LittleEndian.Uint64(b.in[b.off-8:]) + b.bitsRead = 0 + b.off -= 8 +} + +// finished returns true if all bits have been read from the bit stream. +func (b *bitReader) finished() bool { + return b.bitsRead >= 64 && b.off == 0 +} + +// close the bitstream and returns an error if out-of-buffer reads occurred. +func (b *bitReader) close() error { + // Release reference. + b.in = nil + if b.bitsRead > 64 { + return io.ErrUnexpectedEOF + } + return nil +} diff --git a/vendor/github.com/klauspost/compress/fse/bitwriter.go b/vendor/github.com/klauspost/compress/fse/bitwriter.go new file mode 100644 index 0000000..43e4636 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/bitwriter.go @@ -0,0 +1,168 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +import "fmt" + +// bitWriter will write bits. +// First bit will be LSB of the first byte of output. +type bitWriter struct { + bitContainer uint64 + nBits uint8 + out []byte +} + +// bitMask16 is bitmasks. Has extra to avoid bounds check. +var bitMask16 = [32]uint16{ + 0, 1, 3, 7, 0xF, 0x1F, + 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, + 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF} /* up to 16 bits */ + +// addBits16NC will add up to 16 bits. +// It will not check if there is space for them, +// so the caller must ensure that it has flushed recently. +func (b *bitWriter) addBits16NC(value uint16, bits uint8) { + b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63) + b.nBits += bits +} + +// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated. +// It will not check if there is space for them, so the caller must ensure that it has flushed recently. +func (b *bitWriter) addBits16Clean(value uint16, bits uint8) { + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// addBits16ZeroNC will add up to 16 bits. +// It will not check if there is space for them, +// so the caller must ensure that it has flushed recently. +// This is fastest if bits can be zero. +func (b *bitWriter) addBits16ZeroNC(value uint16, bits uint8) { + if bits == 0 { + return + } + value <<= (16 - bits) & 15 + value >>= (16 - bits) & 15 + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// flush will flush all pending full bytes. +// There will be at least 56 bits available for writing when this has been called. +// Using flush32 is faster, but leaves less space for writing. +func (b *bitWriter) flush() { + v := b.nBits >> 3 + switch v { + case 0: + case 1: + b.out = append(b.out, + byte(b.bitContainer), + ) + case 2: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + ) + case 3: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + ) + case 4: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24), + ) + case 5: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24), + byte(b.bitContainer>>32), + ) + case 6: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24), + byte(b.bitContainer>>32), + byte(b.bitContainer>>40), + ) + case 7: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24), + byte(b.bitContainer>>32), + byte(b.bitContainer>>40), + byte(b.bitContainer>>48), + ) + case 8: + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24), + byte(b.bitContainer>>32), + byte(b.bitContainer>>40), + byte(b.bitContainer>>48), + byte(b.bitContainer>>56), + ) + default: + panic(fmt.Errorf("bits (%d) > 64", b.nBits)) + } + b.bitContainer >>= v << 3 + b.nBits &= 7 +} + +// flush32 will flush out, so there are at least 32 bits available for writing. +func (b *bitWriter) flush32() { + if b.nBits < 32 { + return + } + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24)) + b.nBits -= 32 + b.bitContainer >>= 32 +} + +// flushAlign will flush remaining full bytes and align to next byte boundary. +func (b *bitWriter) flushAlign() { + nbBytes := (b.nBits + 7) >> 3 + for i := uint8(0); i < nbBytes; i++ { + b.out = append(b.out, byte(b.bitContainer>>(i*8))) + } + b.nBits = 0 + b.bitContainer = 0 +} + +// close will write the alignment bit and write the final byte(s) +// to the output. +func (b *bitWriter) close() error { + // End mark + b.addBits16Clean(1, 1) + // flush until next byte. + b.flushAlign() + return nil +} + +// reset and continue writing by appending to out. +func (b *bitWriter) reset(out []byte) { + b.bitContainer = 0 + b.nBits = 0 + b.out = out +} diff --git a/vendor/github.com/klauspost/compress/fse/bytereader.go b/vendor/github.com/klauspost/compress/fse/bytereader.go new file mode 100644 index 0000000..abade2d --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/bytereader.go @@ -0,0 +1,47 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +// byteReader provides a byte reader that reads +// little endian values from a byte stream. +// The input stream is manually advanced. +// The reader performs no bounds checks. +type byteReader struct { + b []byte + off int +} + +// init will initialize the reader and set the input. +func (b *byteReader) init(in []byte) { + b.b = in + b.off = 0 +} + +// advance the stream b n bytes. +func (b *byteReader) advance(n uint) { + b.off += int(n) +} + +// Uint32 returns a little endian uint32 starting at current offset. +func (b byteReader) Uint32() uint32 { + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := uint32(b2[3]) + v2 := uint32(b2[2]) + v1 := uint32(b2[1]) + v0 := uint32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// unread returns the unread portion of the input. +func (b byteReader) unread() []byte { + return b.b[b.off:] +} + +// remain will return the number of bytes remaining. +func (b byteReader) remain() int { + return len(b.b) - b.off +} diff --git a/vendor/github.com/klauspost/compress/fse/compress.go b/vendor/github.com/klauspost/compress/fse/compress.go new file mode 100644 index 0000000..6f34191 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/compress.go @@ -0,0 +1,683 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +import ( + "errors" + "fmt" +) + +// Compress the input bytes. Input must be < 2GB. +// Provide a Scratch buffer to avoid memory allocations. +// Note that the output is also kept in the scratch buffer. +// If input is too hard to compress, ErrIncompressible is returned. +// If input is a single byte value repeated ErrUseRLE is returned. +func Compress(in []byte, s *Scratch) ([]byte, error) { + if len(in) <= 1 { + return nil, ErrIncompressible + } + if len(in) > (2<<30)-1 { + return nil, errors.New("input too big, must be < 2GB") + } + s, err := s.prepare(in) + if err != nil { + return nil, err + } + + // Create histogram, if none was provided. + maxCount := s.maxCount + if maxCount == 0 { + maxCount = s.countSimple(in) + } + // Reset for next run. + s.clearCount = true + s.maxCount = 0 + if maxCount == len(in) { + // One symbol, use RLE + return nil, ErrUseRLE + } + if maxCount == 1 || maxCount < (len(in)>>7) { + // Each symbol present maximum once or too well distributed. + return nil, ErrIncompressible + } + s.optimalTableLog() + err = s.normalizeCount() + if err != nil { + return nil, err + } + err = s.writeCount() + if err != nil { + return nil, err + } + + if false { + err = s.validateNorm() + if err != nil { + return nil, err + } + } + + err = s.buildCTable() + if err != nil { + return nil, err + } + err = s.compress(in) + if err != nil { + return nil, err + } + s.Out = s.bw.out + // Check if we compressed. + if len(s.Out) >= len(in) { + return nil, ErrIncompressible + } + return s.Out, nil +} + +// cState contains the compression state of a stream. +type cState struct { + bw *bitWriter + stateTable []uint16 + state uint16 +} + +// init will initialize the compression state to the first symbol of the stream. +func (c *cState) init(bw *bitWriter, ct *cTable, tableLog uint8, first symbolTransform) { + c.bw = bw + c.stateTable = ct.stateTable + + nbBitsOut := (first.deltaNbBits + (1 << 15)) >> 16 + im := int32((nbBitsOut << 16) - first.deltaNbBits) + lu := (im >> nbBitsOut) + first.deltaFindState + c.state = c.stateTable[lu] +} + +// encode the output symbol provided and write it to the bitstream. +func (c *cState) encode(symbolTT symbolTransform) { + nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16 + dstState := int32(c.state>>(nbBitsOut&15)) + symbolTT.deltaFindState + c.bw.addBits16NC(c.state, uint8(nbBitsOut)) + c.state = c.stateTable[dstState] +} + +// encode the output symbol provided and write it to the bitstream. +func (c *cState) encodeZero(symbolTT symbolTransform) { + nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16 + dstState := int32(c.state>>(nbBitsOut&15)) + symbolTT.deltaFindState + c.bw.addBits16ZeroNC(c.state, uint8(nbBitsOut)) + c.state = c.stateTable[dstState] +} + +// flush will write the tablelog to the output and flush the remaining full bytes. +func (c *cState) flush(tableLog uint8) { + c.bw.flush32() + c.bw.addBits16NC(c.state, tableLog) + c.bw.flush() +} + +// compress is the main compression loop that will encode the input from the last byte to the first. +func (s *Scratch) compress(src []byte) error { + if len(src) <= 2 { + return errors.New("compress: src too small") + } + tt := s.ct.symbolTT[:256] + s.bw.reset(s.Out) + + // Our two states each encodes every second byte. + // Last byte encoded (first byte decoded) will always be encoded by c1. + var c1, c2 cState + + // Encode so remaining size is divisible by 4. + ip := len(src) + if ip&1 == 1 { + c1.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-1]]) + c2.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-2]]) + c1.encodeZero(tt[src[ip-3]]) + ip -= 3 + } else { + c2.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-1]]) + c1.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-2]]) + ip -= 2 + } + if ip&2 != 0 { + c2.encodeZero(tt[src[ip-1]]) + c1.encodeZero(tt[src[ip-2]]) + ip -= 2 + } + + // Main compression loop. + switch { + case !s.zeroBits && s.actualTableLog <= 8: + // We can encode 4 symbols without requiring a flush. + // We do not need to check if any output is 0 bits. + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encode(tt[v0]) + c1.encode(tt[v1]) + c2.encode(tt[v2]) + c1.encode(tt[v3]) + ip -= 4 + } + case !s.zeroBits: + // We do not need to check if any output is 0 bits. + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encode(tt[v0]) + c1.encode(tt[v1]) + s.bw.flush32() + c2.encode(tt[v2]) + c1.encode(tt[v3]) + ip -= 4 + } + case s.actualTableLog <= 8: + // We can encode 4 symbols without requiring a flush + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encodeZero(tt[v0]) + c1.encodeZero(tt[v1]) + c2.encodeZero(tt[v2]) + c1.encodeZero(tt[v3]) + ip -= 4 + } + default: + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encodeZero(tt[v0]) + c1.encodeZero(tt[v1]) + s.bw.flush32() + c2.encodeZero(tt[v2]) + c1.encodeZero(tt[v3]) + ip -= 4 + } + } + + // Flush final state. + // Used to initialize state when decoding. + c2.flush(s.actualTableLog) + c1.flush(s.actualTableLog) + + return s.bw.close() +} + +// writeCount will write the normalized histogram count to header. +// This is read back by readNCount. +func (s *Scratch) writeCount() error { + var ( + tableLog = s.actualTableLog + tableSize = 1 << tableLog + previous0 bool + charnum uint16 + + maxHeaderSize = ((int(s.symbolLen) * int(tableLog)) >> 3) + 3 + + // Write Table Size + bitStream = uint32(tableLog - minTablelog) + bitCount = uint(4) + remaining = int16(tableSize + 1) /* +1 for extra accuracy */ + threshold = int16(tableSize) + nbBits = uint(tableLog + 1) + ) + if cap(s.Out) < maxHeaderSize { + s.Out = make([]byte, 0, s.br.remain()+maxHeaderSize) + } + outP := uint(0) + out := s.Out[:maxHeaderSize] + + // stops at 1 + for remaining > 1 { + if previous0 { + start := charnum + for s.norm[charnum] == 0 { + charnum++ + } + for charnum >= start+24 { + start += 24 + bitStream += uint32(0xFFFF) << bitCount + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + } + for charnum >= start+3 { + start += 3 + bitStream += 3 << bitCount + bitCount += 2 + } + bitStream += uint32(charnum-start) << bitCount + bitCount += 2 + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + count := s.norm[charnum] + charnum++ + max := (2*threshold - 1) - remaining + if count < 0 { + remaining += count + } else { + remaining -= count + } + count++ // +1 for extra accuracy + if count >= threshold { + count += max // [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ + } + bitStream += uint32(count) << bitCount + bitCount += nbBits + if count < max { + bitCount-- + } + + previous0 = count == 1 + if remaining < 1 { + return errors.New("internal error: remaining<1") + } + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += (bitCount + 7) / 8 + + if charnum > s.symbolLen { + return errors.New("internal error: charnum > s.symbolLen") + } + s.Out = out[:outP] + return nil +} + +// symbolTransform contains the state transform for a symbol. +type symbolTransform struct { + deltaFindState int32 + deltaNbBits uint32 +} + +// String prints values as a human readable string. +func (s symbolTransform) String() string { + return fmt.Sprintf("dnbits: %08x, fs:%d", s.deltaNbBits, s.deltaFindState) +} + +// cTable contains tables used for compression. +type cTable struct { + tableSymbol []byte + stateTable []uint16 + symbolTT []symbolTransform +} + +// allocCtable will allocate tables needed for compression. +// If existing tables a re big enough, they are simply re-used. +func (s *Scratch) allocCtable() { + tableSize := 1 << s.actualTableLog + // get tableSymbol that is big enough. + if cap(s.ct.tableSymbol) < tableSize { + s.ct.tableSymbol = make([]byte, tableSize) + } + s.ct.tableSymbol = s.ct.tableSymbol[:tableSize] + + ctSize := tableSize + if cap(s.ct.stateTable) < ctSize { + s.ct.stateTable = make([]uint16, ctSize) + } + s.ct.stateTable = s.ct.stateTable[:ctSize] + + if cap(s.ct.symbolTT) < 256 { + s.ct.symbolTT = make([]symbolTransform, 256) + } + s.ct.symbolTT = s.ct.symbolTT[:256] +} + +// buildCTable will populate the compression table so it is ready to be used. +func (s *Scratch) buildCTable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + var cumul [maxSymbolValue + 2]int16 + + s.allocCtable() + tableSymbol := s.ct.tableSymbol[:tableSize] + // symbol start positions + { + cumul[0] = 0 + for ui, v := range s.norm[:s.symbolLen-1] { + u := byte(ui) // one less than reference + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = u + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + } + // Encode last symbol separately to avoid overflowing u + u := int(s.symbolLen - 1) + v := s.norm[s.symbolLen-1] + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = byte(u) + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + if uint32(cumul[s.symbolLen]) != tableSize { + return fmt.Errorf("internal error: expected cumul[s.symbolLen] (%d) == tableSize (%d)", cumul[s.symbolLen], tableSize) + } + cumul[s.symbolLen] = int16(tableSize) + 1 + } + // Spread symbols + s.zeroBits = false + { + step := tableStep(tableSize) + tableMask := tableSize - 1 + var position uint32 + // if any symbol > largeLimit, we may have 0 bits output. + largeLimit := int16(1 << (s.actualTableLog - 1)) + for ui, v := range s.norm[:s.symbolLen] { + symbol := byte(ui) + if v > largeLimit { + s.zeroBits = true + } + for nbOccurrences := int16(0); nbOccurrences < v; nbOccurrences++ { + tableSymbol[position] = symbol + position = (position + step) & tableMask + for position > highThreshold { + position = (position + step) & tableMask + } /* Low proba area */ + } + } + + // Check if we have gone through all positions + if position != 0 { + return errors.New("position!=0") + } + } + + // Build table + table := s.ct.stateTable + { + tsi := int(tableSize) + for u, v := range tableSymbol { + // TableU16 : sorted by symbol order; gives next state value + table[cumul[v]] = uint16(tsi + u) + cumul[v]++ + } + } + + // Build Symbol Transformation Table + { + total := int16(0) + symbolTT := s.ct.symbolTT[:s.symbolLen] + tableLog := s.actualTableLog + tl := (uint32(tableLog) << 16) - (1 << tableLog) + for i, v := range s.norm[:s.symbolLen] { + switch v { + case 0: + case -1, 1: + symbolTT[i].deltaNbBits = tl + symbolTT[i].deltaFindState = int32(total - 1) + total++ + default: + maxBitsOut := uint32(tableLog) - highBits(uint32(v-1)) + minStatePlus := uint32(v) << maxBitsOut + symbolTT[i].deltaNbBits = (maxBitsOut << 16) - minStatePlus + symbolTT[i].deltaFindState = int32(total - v) + total += v + } + } + if total != int16(tableSize) { + return fmt.Errorf("total mismatch %d (got) != %d (want)", total, tableSize) + } + } + return nil +} + +// countSimple will create a simple histogram in s.count. +// Returns the biggest count. +// Does not update s.clearCount. +func (s *Scratch) countSimple(in []byte) (max int) { + for _, v := range in { + s.count[v]++ + } + m := uint32(0) + for i, v := range s.count[:] { + if v > m { + m = v + } + if v > 0 { + s.symbolLen = uint16(i) + 1 + } + } + return int(m) +} + +// minTableLog provides the minimum logSize to safely represent a distribution. +func (s *Scratch) minTableLog() uint8 { + minBitsSrc := highBits(uint32(s.br.remain()-1)) + 1 + minBitsSymbols := highBits(uint32(s.symbolLen-1)) + 2 + if minBitsSrc < minBitsSymbols { + return uint8(minBitsSrc) + } + return uint8(minBitsSymbols) +} + +// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog +func (s *Scratch) optimalTableLog() { + tableLog := s.TableLog + minBits := s.minTableLog() + maxBitsSrc := uint8(highBits(uint32(s.br.remain()-1))) - 2 + if maxBitsSrc < tableLog { + // Accuracy can be reduced + tableLog = maxBitsSrc + } + if minBits > tableLog { + tableLog = minBits + } + // Need a minimum to safely represent all symbol values + if tableLog < minTablelog { + tableLog = minTablelog + } + if tableLog > maxTableLog { + tableLog = maxTableLog + } + s.actualTableLog = tableLog +} + +var rtbTable = [...]uint32{0, 473195, 504333, 520860, 550000, 700000, 750000, 830000} + +// normalizeCount will normalize the count of the symbols so +// the total is equal to the table size. +func (s *Scratch) normalizeCount() error { + var ( + tableLog = s.actualTableLog + scale = 62 - uint64(tableLog) + step = (1 << 62) / uint64(s.br.remain()) + vStep = uint64(1) << (scale - 20) + stillToDistribute = int16(1 << tableLog) + largest int + largestP int16 + lowThreshold = (uint32)(s.br.remain() >> tableLog) + ) + + for i, cnt := range s.count[:s.symbolLen] { + // already handled + // if (count[s] == s.length) return 0; /* rle special case */ + + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + stillToDistribute-- + } else { + proba := (int16)((uint64(cnt) * step) >> scale) + if proba < 8 { + restToBeat := vStep * uint64(rtbTable[proba]) + v := uint64(cnt)*step - (uint64(proba) << scale) + if v > restToBeat { + proba++ + } + } + if proba > largestP { + largestP = proba + largest = i + } + s.norm[i] = proba + stillToDistribute -= proba + } + } + + if -stillToDistribute >= (s.norm[largest] >> 1) { + // corner case, need another normalization method + return s.normalizeCount2() + } + s.norm[largest] += stillToDistribute + return nil +} + +// Secondary normalization method. +// To be used when primary method fails. +func (s *Scratch) normalizeCount2() error { + const notYetAssigned = -2 + var ( + distributed uint32 + total = uint32(s.br.remain()) + tableLog = s.actualTableLog + lowThreshold = total >> tableLog + lowOne = (total * 3) >> (tableLog + 1) + ) + for i, cnt := range s.count[:s.symbolLen] { + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + distributed++ + total -= cnt + continue + } + if cnt <= lowOne { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + s.norm[i] = notYetAssigned + } + toDistribute := (1 << tableLog) - distributed + + if (total / toDistribute) > lowOne { + // risk of rounding to zero + lowOne = (total * 3) / (toDistribute * 2) + for i, cnt := range s.count[:s.symbolLen] { + if (s.norm[i] == notYetAssigned) && (cnt <= lowOne) { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + } + toDistribute = (1 << tableLog) - distributed + } + if distributed == uint32(s.symbolLen)+1 { + // all values are pretty poor; + // probably incompressible data (should have already been detected); + // find max, then give all remaining points to max + var maxV int + var maxC uint32 + for i, cnt := range s.count[:s.symbolLen] { + if cnt > maxC { + maxV = i + maxC = cnt + } + } + s.norm[maxV] += int16(toDistribute) + return nil + } + + if total == 0 { + // all of the symbols were low enough for the lowOne or lowThreshold + for i := uint32(0); toDistribute > 0; i = (i + 1) % (uint32(s.symbolLen)) { + if s.norm[i] > 0 { + toDistribute-- + s.norm[i]++ + } + } + return nil + } + + var ( + vStepLog = 62 - uint64(tableLog) + mid = uint64((1 << (vStepLog - 1)) - 1) + rStep = (((1 << vStepLog) * uint64(toDistribute)) + mid) / uint64(total) // scale on remaining + tmpTotal = mid + ) + for i, cnt := range s.count[:s.symbolLen] { + if s.norm[i] == notYetAssigned { + var ( + end = tmpTotal + uint64(cnt)*rStep + sStart = uint32(tmpTotal >> vStepLog) + sEnd = uint32(end >> vStepLog) + weight = sEnd - sStart + ) + if weight < 1 { + return errors.New("weight < 1") + } + s.norm[i] = int16(weight) + tmpTotal = end + } + } + return nil +} + +// validateNorm validates the normalized histogram table. +func (s *Scratch) validateNorm() (err error) { + var total int + for _, v := range s.norm[:s.symbolLen] { + if v >= 0 { + total += int(v) + } else { + total -= int(v) + } + } + defer func() { + if err == nil { + return + } + fmt.Printf("selected TableLog: %d, Symbol length: %d\n", s.actualTableLog, s.symbolLen) + for i, v := range s.norm[:s.symbolLen] { + fmt.Printf("%3d: %5d -> %4d \n", i, s.count[i], v) + } + }() + if total != (1 << s.actualTableLog) { + return fmt.Errorf("warning: Total == %d != %d", total, 1< tablelogAbsoluteMax { + return errors.New("tableLog too large") + } + bitStream >>= 4 + bitCount := uint(4) + + s.actualTableLog = uint8(nbBits) + remaining := int32((1 << nbBits) + 1) + threshold := int32(1 << nbBits) + gotTotal := int32(0) + nbBits++ + + for remaining > 1 { + if previous0 { + n0 := charnum + for (bitStream & 0xFFFF) == 0xFFFF { + n0 += 24 + if b.off < iend-5 { + b.advance(2) + bitStream = b.Uint32() >> bitCount + } else { + bitStream >>= 16 + bitCount += 16 + } + } + for (bitStream & 3) == 3 { + n0 += 3 + bitStream >>= 2 + bitCount += 2 + } + n0 += uint16(bitStream & 3) + bitCount += 2 + if n0 > maxSymbolValue { + return errors.New("maxSymbolValue too small") + } + for charnum < n0 { + s.norm[charnum&0xff] = 0 + charnum++ + } + + if b.off <= iend-7 || b.off+int(bitCount>>3) <= iend-4 { + b.advance(bitCount >> 3) + bitCount &= 7 + bitStream = b.Uint32() >> bitCount + } else { + bitStream >>= 2 + } + } + + max := (2*(threshold) - 1) - (remaining) + var count int32 + + if (int32(bitStream) & (threshold - 1)) < max { + count = int32(bitStream) & (threshold - 1) + bitCount += nbBits - 1 + } else { + count = int32(bitStream) & (2*threshold - 1) + if count >= threshold { + count -= max + } + bitCount += nbBits + } + + count-- // extra accuracy + if count < 0 { + // -1 means +1 + remaining += count + gotTotal -= count + } else { + remaining -= count + gotTotal += count + } + s.norm[charnum&0xff] = int16(count) + charnum++ + previous0 = count == 0 + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + if b.off <= iend-7 || b.off+int(bitCount>>3) <= iend-4 { + b.advance(bitCount >> 3) + bitCount &= 7 + } else { + bitCount -= (uint)(8 * (len(b.b) - 4 - b.off)) + b.off = len(b.b) - 4 + } + bitStream = b.Uint32() >> (bitCount & 31) + } + s.symbolLen = charnum + + if s.symbolLen <= 1 { + return fmt.Errorf("symbolLen (%d) too small", s.symbolLen) + } + if s.symbolLen > maxSymbolValue+1 { + return fmt.Errorf("symbolLen (%d) too big", s.symbolLen) + } + if remaining != 1 { + return fmt.Errorf("corruption detected (remaining %d != 1)", remaining) + } + if bitCount > 32 { + return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount) + } + if gotTotal != 1<> 3) + return nil +} + +// decSymbol contains information about a state entry, +// Including the state offset base, the output symbol and +// the number of bits to read for the low part of the destination state. +type decSymbol struct { + newState uint16 + symbol uint8 + nbBits uint8 +} + +// allocDtable will allocate decoding tables if they are not big enough. +func (s *Scratch) allocDtable() { + tableSize := 1 << s.actualTableLog + if cap(s.decTable) < tableSize { + s.decTable = make([]decSymbol, tableSize) + } + s.decTable = s.decTable[:tableSize] + + if cap(s.ct.tableSymbol) < 256 { + s.ct.tableSymbol = make([]byte, 256) + } + s.ct.tableSymbol = s.ct.tableSymbol[:256] + + if cap(s.ct.stateTable) < 256 { + s.ct.stateTable = make([]uint16, 256) + } + s.ct.stateTable = s.ct.stateTable[:256] +} + +// buildDtable will build the decoding table. +func (s *Scratch) buildDtable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + s.allocDtable() + symbolNext := s.ct.stateTable[:256] + + // Init, lay down lowprob symbols + s.zeroBits = false + { + largeLimit := int16(1 << (s.actualTableLog - 1)) + for i, v := range s.norm[:s.symbolLen] { + if v == -1 { + s.decTable[highThreshold].symbol = uint8(i) + highThreshold-- + symbolNext[i] = 1 + } else { + if v >= largeLimit { + s.zeroBits = true + } + symbolNext[i] = uint16(v) + } + } + } + // Spread symbols + { + tableMask := tableSize - 1 + step := tableStep(tableSize) + position := uint32(0) + for ss, v := range s.norm[:s.symbolLen] { + for i := 0; i < int(v); i++ { + s.decTable[position].symbol = uint8(ss) + position = (position + step) & tableMask + for position > highThreshold { + // lowprob area + position = (position + step) & tableMask + } + } + } + if position != 0 { + // position must reach all cells once, otherwise normalizedCounter is incorrect + return errors.New("corrupted input (position != 0)") + } + } + + // Build Decoding table + { + tableSize := uint16(1 << s.actualTableLog) + for u, v := range s.decTable { + symbol := v.symbol + nextState := symbolNext[symbol] + symbolNext[symbol] = nextState + 1 + nBits := s.actualTableLog - byte(highBits(uint32(nextState))) + s.decTable[u].nbBits = nBits + newState := (nextState << nBits) - tableSize + if newState >= tableSize { + return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize) + } + if newState == uint16(u) && nBits == 0 { + // Seems weird that this is possible with nbits > 0. + return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u) + } + s.decTable[u].newState = newState + } + } + return nil +} + +// decompress will decompress the bitstream. +// If the buffer is over-read an error is returned. +func (s *Scratch) decompress() error { + br := &s.bits + br.init(s.br.unread()) + + var s1, s2 decoder + // Initialize and decode first state and symbol. + s1.init(br, s.decTable, s.actualTableLog) + s2.init(br, s.decTable, s.actualTableLog) + + // Use temp table to avoid bound checks/append penalty. + var tmp = s.ct.tableSymbol[:256] + var off uint8 + + // Main part + if !s.zeroBits { + for br.off >= 8 { + br.fillFast() + tmp[off+0] = s1.nextFast() + tmp[off+1] = s2.nextFast() + br.fillFast() + tmp[off+2] = s1.nextFast() + tmp[off+3] = s2.nextFast() + off += 4 + // When off is 0, we have overflowed and should write. + if off == 0 { + s.Out = append(s.Out, tmp...) + if len(s.Out) >= s.DecompressLimit { + return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit) + } + } + } + } else { + for br.off >= 8 { + br.fillFast() + tmp[off+0] = s1.next() + tmp[off+1] = s2.next() + br.fillFast() + tmp[off+2] = s1.next() + tmp[off+3] = s2.next() + off += 4 + if off == 0 { + s.Out = append(s.Out, tmp...) + // When off is 0, we have overflowed and should write. + if len(s.Out) >= s.DecompressLimit { + return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit) + } + } + } + } + s.Out = append(s.Out, tmp[:off]...) + + // Final bits, a bit more expensive check + for { + if s1.finished() { + s.Out = append(s.Out, s1.final(), s2.final()) + break + } + br.fill() + s.Out = append(s.Out, s1.next()) + if s2.finished() { + s.Out = append(s.Out, s2.final(), s1.final()) + break + } + s.Out = append(s.Out, s2.next()) + if len(s.Out) >= s.DecompressLimit { + return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit) + } + } + return br.close() +} + +// decoder keeps track of the current state and updates it from the bitstream. +type decoder struct { + state uint16 + br *bitReader + dt []decSymbol +} + +// init will initialize the decoder and read the first state from the stream. +func (d *decoder) init(in *bitReader, dt []decSymbol, tableLog uint8) { + d.dt = dt + d.br = in + d.state = in.getBits(tableLog) +} + +// next returns the next symbol and sets the next state. +// At least tablelog bits must be available in the bit reader. +func (d *decoder) next() uint8 { + n := &d.dt[d.state] + lowBits := d.br.getBits(n.nbBits) + d.state = n.newState + lowBits + return n.symbol +} + +// finished returns true if all bits have been read from the bitstream +// and the next state would require reading bits from the input. +func (d *decoder) finished() bool { + return d.br.finished() && d.dt[d.state].nbBits > 0 +} + +// final returns the current state symbol without decoding the next. +func (d *decoder) final() uint8 { + return d.dt[d.state].symbol +} + +// nextFast returns the next symbol and sets the next state. +// This can only be used if no symbols are 0 bits. +// At least tablelog bits must be available in the bit reader. +func (d *decoder) nextFast() uint8 { + n := d.dt[d.state] + lowBits := d.br.getBitsFast(n.nbBits) + d.state = n.newState + lowBits + return n.symbol +} diff --git a/vendor/github.com/klauspost/compress/fse/fse.go b/vendor/github.com/klauspost/compress/fse/fse.go new file mode 100644 index 0000000..535cbad --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/fse.go @@ -0,0 +1,144 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +// Package fse provides Finite State Entropy encoding and decoding. +// +// Finite State Entropy encoding provides a fast near-optimal symbol encoding/decoding +// for byte blocks as implemented in zstd. +// +// See https://github.com/klauspost/compress/tree/master/fse for more information. +package fse + +import ( + "errors" + "fmt" + "math/bits" +) + +const ( + /*!MEMORY_USAGE : + * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.) + * Increasing memory usage improves compression ratio + * Reduced memory usage can improve speed, due to cache effect + * Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */ + maxMemoryUsage = 14 + defaultMemoryUsage = 13 + + maxTableLog = maxMemoryUsage - 2 + maxTablesize = 1 << maxTableLog + defaultTablelog = defaultMemoryUsage - 2 + minTablelog = 5 + maxSymbolValue = 255 +) + +var ( + // ErrIncompressible is returned when input is judged to be too hard to compress. + ErrIncompressible = errors.New("input is not compressible") + + // ErrUseRLE is returned from the compressor when the input is a single byte value repeated. + ErrUseRLE = errors.New("input is single value repeated") +) + +// Scratch provides temporary storage for compression and decompression. +type Scratch struct { + // Private + count [maxSymbolValue + 1]uint32 + norm [maxSymbolValue + 1]int16 + br byteReader + bits bitReader + bw bitWriter + ct cTable // Compression tables. + decTable []decSymbol // Decompression table. + maxCount int // count of the most probable symbol + + // Per block parameters. + // These can be used to override compression parameters of the block. + // Do not touch, unless you know what you are doing. + + // Out is output buffer. + // If the scratch is re-used before the caller is done processing the output, + // set this field to nil. + // Otherwise the output buffer will be re-used for next Compression/Decompression step + // and allocation will be avoided. + Out []byte + + // DecompressLimit limits the maximum decoded size acceptable. + // If > 0 decompression will stop when approximately this many bytes + // has been decoded. + // If 0, maximum size will be 2GB. + DecompressLimit int + + symbolLen uint16 // Length of active part of the symbol table. + actualTableLog uint8 // Selected tablelog. + zeroBits bool // no bits has prob > 50%. + clearCount bool // clear count + + // MaxSymbolValue will override the maximum symbol value of the next block. + MaxSymbolValue uint8 + + // TableLog will attempt to override the tablelog for the next block. + TableLog uint8 +} + +// Histogram allows to populate the histogram and skip that step in the compression, +// It otherwise allows to inspect the histogram when compression is done. +// To indicate that you have populated the histogram call HistogramFinished +// with the value of the highest populated symbol, as well as the number of entries +// in the most populated entry. These are accepted at face value. +// The returned slice will always be length 256. +func (s *Scratch) Histogram() []uint32 { + return s.count[:] +} + +// HistogramFinished can be called to indicate that the histogram has been populated. +// maxSymbol is the index of the highest set symbol of the next data segment. +// maxCount is the number of entries in the most populated entry. +// These are accepted at face value. +func (s *Scratch) HistogramFinished(maxSymbol uint8, maxCount int) { + s.maxCount = maxCount + s.symbolLen = uint16(maxSymbol) + 1 + s.clearCount = maxCount != 0 +} + +// prepare will prepare and allocate scratch tables used for both compression and decompression. +func (s *Scratch) prepare(in []byte) (*Scratch, error) { + if s == nil { + s = &Scratch{} + } + if s.MaxSymbolValue == 0 { + s.MaxSymbolValue = 255 + } + if s.TableLog == 0 { + s.TableLog = defaultTablelog + } + if s.TableLog > maxTableLog { + return nil, fmt.Errorf("tableLog (%d) > maxTableLog (%d)", s.TableLog, maxTableLog) + } + if cap(s.Out) == 0 { + s.Out = make([]byte, 0, len(in)) + } + if s.clearCount && s.maxCount == 0 { + for i := range s.count { + s.count[i] = 0 + } + s.clearCount = false + } + s.br.init(in) + if s.DecompressLimit == 0 { + // Max size 2GB. + s.DecompressLimit = (2 << 30) - 1 + } + + return s, nil +} + +// tableStep returns the next table index. +func tableStep(tableSize uint32) uint32 { + return (tableSize >> 1) + (tableSize >> 3) + 3 +} + +func highBits(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} diff --git a/vendor/github.com/klauspost/compress/gen.sh b/vendor/github.com/klauspost/compress/gen.sh new file mode 100644 index 0000000..aff9422 --- /dev/null +++ b/vendor/github.com/klauspost/compress/gen.sh @@ -0,0 +1,4 @@ +#!/bin/sh + +cd s2/cmd/_s2sx/ || exit 1 +go generate . diff --git a/vendor/github.com/klauspost/compress/gzip/gunzip.go b/vendor/github.com/klauspost/compress/gzip/gunzip.go new file mode 100644 index 0000000..66fe5dd --- /dev/null +++ b/vendor/github.com/klauspost/compress/gzip/gunzip.go @@ -0,0 +1,349 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package gzip implements reading and writing of gzip format compressed files, +// as specified in RFC 1952. +package gzip + +import ( + "bufio" + "compress/gzip" + "encoding/binary" + "hash/crc32" + "io" + "time" + + "github.com/klauspost/compress/flate" +) + +const ( + gzipID1 = 0x1f + gzipID2 = 0x8b + gzipDeflate = 8 + flagText = 1 << 0 + flagHdrCrc = 1 << 1 + flagExtra = 1 << 2 + flagName = 1 << 3 + flagComment = 1 << 4 +) + +var ( + // ErrChecksum is returned when reading GZIP data that has an invalid checksum. + ErrChecksum = gzip.ErrChecksum + // ErrHeader is returned when reading GZIP data that has an invalid header. + ErrHeader = gzip.ErrHeader +) + +var le = binary.LittleEndian + +// noEOF converts io.EOF to io.ErrUnexpectedEOF. +func noEOF(err error) error { + if err == io.EOF { + return io.ErrUnexpectedEOF + } + return err +} + +// The gzip file stores a header giving metadata about the compressed file. +// That header is exposed as the fields of the Writer and Reader structs. +// +// Strings must be UTF-8 encoded and may only contain Unicode code points +// U+0001 through U+00FF, due to limitations of the GZIP file format. +type Header struct { + Comment string // comment + Extra []byte // "extra data" + ModTime time.Time // modification time + Name string // file name + OS byte // operating system type +} + +// A Reader is an io.Reader that can be read to retrieve +// uncompressed data from a gzip-format compressed file. +// +// In general, a gzip file can be a concatenation of gzip files, +// each with its own header. Reads from the Reader +// return the concatenation of the uncompressed data of each. +// Only the first header is recorded in the Reader fields. +// +// Gzip files store a length and checksum of the uncompressed data. +// The Reader will return a ErrChecksum when Read +// reaches the end of the uncompressed data if it does not +// have the expected length or checksum. Clients should treat data +// returned by Read as tentative until they receive the io.EOF +// marking the end of the data. +type Reader struct { + Header // valid after NewReader or Reader.Reset + r flate.Reader + br *bufio.Reader + decompressor io.ReadCloser + digest uint32 // CRC-32, IEEE polynomial (section 8) + size uint32 // Uncompressed size (section 2.3.1) + buf [512]byte + err error + multistream bool +} + +// NewReader creates a new Reader reading the given reader. +// If r does not also implement io.ByteReader, +// the decompressor may read more data than necessary from r. +// +// It is the caller's responsibility to call Close on the Reader when done. +// +// The Reader.Header fields will be valid in the Reader returned. +func NewReader(r io.Reader) (*Reader, error) { + z := new(Reader) + if err := z.Reset(r); err != nil { + return nil, err + } + return z, nil +} + +// Reset discards the Reader z's state and makes it equivalent to the +// result of its original state from NewReader, but reading from r instead. +// This permits reusing a Reader rather than allocating a new one. +func (z *Reader) Reset(r io.Reader) error { + *z = Reader{ + decompressor: z.decompressor, + multistream: true, + } + if rr, ok := r.(flate.Reader); ok { + z.r = rr + } else { + // Reuse if we can. + if z.br != nil { + z.br.Reset(r) + } else { + z.br = bufio.NewReader(r) + } + z.r = z.br + } + z.Header, z.err = z.readHeader() + return z.err +} + +// Multistream controls whether the reader supports multistream files. +// +// If enabled (the default), the Reader expects the input to be a sequence +// of individually gzipped data streams, each with its own header and +// trailer, ending at EOF. The effect is that the concatenation of a sequence +// of gzipped files is treated as equivalent to the gzip of the concatenation +// of the sequence. This is standard behavior for gzip readers. +// +// Calling Multistream(false) disables this behavior; disabling the behavior +// can be useful when reading file formats that distinguish individual gzip +// data streams or mix gzip data streams with other data streams. +// In this mode, when the Reader reaches the end of the data stream, +// Read returns io.EOF. If the underlying reader implements io.ByteReader, +// it will be left positioned just after the gzip stream. +// To start the next stream, call z.Reset(r) followed by z.Multistream(false). +// If there is no next stream, z.Reset(r) will return io.EOF. +func (z *Reader) Multistream(ok bool) { + z.multistream = ok +} + +// readString reads a NUL-terminated string from z.r. +// It treats the bytes read as being encoded as ISO 8859-1 (Latin-1) and +// will output a string encoded using UTF-8. +// This method always updates z.digest with the data read. +func (z *Reader) readString() (string, error) { + var err error + needConv := false + for i := 0; ; i++ { + if i >= len(z.buf) { + return "", ErrHeader + } + z.buf[i], err = z.r.ReadByte() + if err != nil { + return "", err + } + if z.buf[i] > 0x7f { + needConv = true + } + if z.buf[i] == 0 { + // Digest covers the NUL terminator. + z.digest = crc32.Update(z.digest, crc32.IEEETable, z.buf[:i+1]) + + // Strings are ISO 8859-1, Latin-1 (RFC 1952, section 2.3.1). + if needConv { + s := make([]rune, 0, i) + for _, v := range z.buf[:i] { + s = append(s, rune(v)) + } + return string(s), nil + } + return string(z.buf[:i]), nil + } + } +} + +// readHeader reads the GZIP header according to section 2.3.1. +// This method does not set z.err. +func (z *Reader) readHeader() (hdr Header, err error) { + if _, err = io.ReadFull(z.r, z.buf[:10]); err != nil { + // RFC 1952, section 2.2, says the following: + // A gzip file consists of a series of "members" (compressed data sets). + // + // Other than this, the specification does not clarify whether a + // "series" is defined as "one or more" or "zero or more". To err on the + // side of caution, Go interprets this to mean "zero or more". + // Thus, it is okay to return io.EOF here. + return hdr, err + } + if z.buf[0] != gzipID1 || z.buf[1] != gzipID2 || z.buf[2] != gzipDeflate { + return hdr, ErrHeader + } + flg := z.buf[3] + hdr.ModTime = time.Unix(int64(le.Uint32(z.buf[4:8])), 0) + // z.buf[8] is XFL and is currently ignored. + hdr.OS = z.buf[9] + z.digest = crc32.ChecksumIEEE(z.buf[:10]) + + if flg&flagExtra != 0 { + if _, err = io.ReadFull(z.r, z.buf[:2]); err != nil { + return hdr, noEOF(err) + } + z.digest = crc32.Update(z.digest, crc32.IEEETable, z.buf[:2]) + data := make([]byte, le.Uint16(z.buf[:2])) + if _, err = io.ReadFull(z.r, data); err != nil { + return hdr, noEOF(err) + } + z.digest = crc32.Update(z.digest, crc32.IEEETable, data) + hdr.Extra = data + } + + var s string + if flg&flagName != 0 { + if s, err = z.readString(); err != nil { + return hdr, err + } + hdr.Name = s + } + + if flg&flagComment != 0 { + if s, err = z.readString(); err != nil { + return hdr, err + } + hdr.Comment = s + } + + if flg&flagHdrCrc != 0 { + if _, err = io.ReadFull(z.r, z.buf[:2]); err != nil { + return hdr, noEOF(err) + } + digest := le.Uint16(z.buf[:2]) + if digest != uint16(z.digest) { + return hdr, ErrHeader + } + } + + z.digest = 0 + if z.decompressor == nil { + z.decompressor = flate.NewReader(z.r) + } else { + z.decompressor.(flate.Resetter).Reset(z.r, nil) + } + return hdr, nil +} + +// Read implements io.Reader, reading uncompressed bytes from its underlying Reader. +func (z *Reader) Read(p []byte) (n int, err error) { + if z.err != nil { + return 0, z.err + } + + for n == 0 { + n, z.err = z.decompressor.Read(p) + z.digest = crc32.Update(z.digest, crc32.IEEETable, p[:n]) + z.size += uint32(n) + if z.err != io.EOF { + // In the normal case we return here. + return n, z.err + } + + // Finished file; check checksum and size. + if _, err := io.ReadFull(z.r, z.buf[:8]); err != nil { + z.err = noEOF(err) + return n, z.err + } + digest := le.Uint32(z.buf[:4]) + size := le.Uint32(z.buf[4:8]) + if digest != z.digest || size != z.size { + z.err = ErrChecksum + return n, z.err + } + z.digest, z.size = 0, 0 + + // File is ok; check if there is another. + if !z.multistream { + return n, io.EOF + } + z.err = nil // Remove io.EOF + + if _, z.err = z.readHeader(); z.err != nil { + return n, z.err + } + } + + return n, nil +} + +// Support the io.WriteTo interface for io.Copy and friends. +func (z *Reader) WriteTo(w io.Writer) (int64, error) { + total := int64(0) + crcWriter := crc32.NewIEEE() + for { + if z.err != nil { + if z.err == io.EOF { + return total, nil + } + return total, z.err + } + + // We write both to output and digest. + mw := io.MultiWriter(w, crcWriter) + n, err := z.decompressor.(io.WriterTo).WriteTo(mw) + total += n + z.size += uint32(n) + if err != nil { + z.err = err + return total, z.err + } + + // Finished file; check checksum + size. + if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil { + if err == io.EOF { + err = io.ErrUnexpectedEOF + } + z.err = err + return total, err + } + z.digest = crcWriter.Sum32() + digest := le.Uint32(z.buf[:4]) + size := le.Uint32(z.buf[4:8]) + if digest != z.digest || size != z.size { + z.err = ErrChecksum + return total, z.err + } + z.digest, z.size = 0, 0 + + // File is ok; check if there is another. + if !z.multistream { + return total, nil + } + crcWriter.Reset() + z.err = nil // Remove io.EOF + + if _, z.err = z.readHeader(); z.err != nil { + if z.err == io.EOF { + return total, nil + } + return total, z.err + } + } +} + +// Close closes the Reader. It does not close the underlying io.Reader. +// In order for the GZIP checksum to be verified, the reader must be +// fully consumed until the io.EOF. +func (z *Reader) Close() error { return z.decompressor.Close() } diff --git a/vendor/github.com/klauspost/compress/gzip/gzip.go b/vendor/github.com/klauspost/compress/gzip/gzip.go new file mode 100644 index 0000000..2620385 --- /dev/null +++ b/vendor/github.com/klauspost/compress/gzip/gzip.go @@ -0,0 +1,269 @@ +// Copyright 2010 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package gzip + +import ( + "errors" + "fmt" + "hash/crc32" + "io" + + "github.com/klauspost/compress/flate" +) + +// These constants are copied from the flate package, so that code that imports +// "compress/gzip" does not also have to import "compress/flate". +const ( + NoCompression = flate.NoCompression + BestSpeed = flate.BestSpeed + BestCompression = flate.BestCompression + DefaultCompression = flate.DefaultCompression + ConstantCompression = flate.ConstantCompression + HuffmanOnly = flate.HuffmanOnly + + // StatelessCompression will do compression but without maintaining any state + // between Write calls. + // There will be no memory kept between Write calls, + // but compression and speed will be suboptimal. + // Because of this, the size of actual Write calls will affect output size. + StatelessCompression = -3 +) + +// A Writer is an io.WriteCloser. +// Writes to a Writer are compressed and written to w. +type Writer struct { + Header // written at first call to Write, Flush, or Close + w io.Writer + level int + err error + compressor *flate.Writer + digest uint32 // CRC-32, IEEE polynomial (section 8) + size uint32 // Uncompressed size (section 2.3.1) + wroteHeader bool + closed bool + buf [10]byte +} + +// NewWriter returns a new Writer. +// Writes to the returned writer are compressed and written to w. +// +// It is the caller's responsibility to call Close on the WriteCloser when done. +// Writes may be buffered and not flushed until Close. +// +// Callers that wish to set the fields in Writer.Header must do so before +// the first call to Write, Flush, or Close. +func NewWriter(w io.Writer) *Writer { + z, _ := NewWriterLevel(w, DefaultCompression) + return z +} + +// NewWriterLevel is like NewWriter but specifies the compression level instead +// of assuming DefaultCompression. +// +// The compression level can be DefaultCompression, NoCompression, or any +// integer value between BestSpeed and BestCompression inclusive. The error +// returned will be nil if the level is valid. +func NewWriterLevel(w io.Writer, level int) (*Writer, error) { + if level < StatelessCompression || level > BestCompression { + return nil, fmt.Errorf("gzip: invalid compression level: %d", level) + } + z := new(Writer) + z.init(w, level) + return z, nil +} + +func (z *Writer) init(w io.Writer, level int) { + compressor := z.compressor + if level != StatelessCompression { + if compressor != nil { + compressor.Reset(w) + } + } + + *z = Writer{ + Header: Header{ + OS: 255, // unknown + }, + w: w, + level: level, + compressor: compressor, + } +} + +// Reset discards the Writer z's state and makes it equivalent to the +// result of its original state from NewWriter or NewWriterLevel, but +// writing to w instead. This permits reusing a Writer rather than +// allocating a new one. +func (z *Writer) Reset(w io.Writer) { + z.init(w, z.level) +} + +// writeBytes writes a length-prefixed byte slice to z.w. +func (z *Writer) writeBytes(b []byte) error { + if len(b) > 0xffff { + return errors.New("gzip.Write: Extra data is too large") + } + le.PutUint16(z.buf[:2], uint16(len(b))) + _, err := z.w.Write(z.buf[:2]) + if err != nil { + return err + } + _, err = z.w.Write(b) + return err +} + +// writeString writes a UTF-8 string s in GZIP's format to z.w. +// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1). +func (z *Writer) writeString(s string) (err error) { + // GZIP stores Latin-1 strings; error if non-Latin-1; convert if non-ASCII. + needconv := false + for _, v := range s { + if v == 0 || v > 0xff { + return errors.New("gzip.Write: non-Latin-1 header string") + } + if v > 0x7f { + needconv = true + } + } + if needconv { + b := make([]byte, 0, len(s)) + for _, v := range s { + b = append(b, byte(v)) + } + _, err = z.w.Write(b) + } else { + _, err = io.WriteString(z.w, s) + } + if err != nil { + return err + } + // GZIP strings are NUL-terminated. + z.buf[0] = 0 + _, err = z.w.Write(z.buf[:1]) + return err +} + +// Write writes a compressed form of p to the underlying io.Writer. The +// compressed bytes are not necessarily flushed until the Writer is closed. +func (z *Writer) Write(p []byte) (int, error) { + if z.err != nil { + return 0, z.err + } + var n int + // Write the GZIP header lazily. + if !z.wroteHeader { + z.wroteHeader = true + z.buf[0] = gzipID1 + z.buf[1] = gzipID2 + z.buf[2] = gzipDeflate + z.buf[3] = 0 + if z.Extra != nil { + z.buf[3] |= 0x04 + } + if z.Name != "" { + z.buf[3] |= 0x08 + } + if z.Comment != "" { + z.buf[3] |= 0x10 + } + le.PutUint32(z.buf[4:8], uint32(z.ModTime.Unix())) + if z.level == BestCompression { + z.buf[8] = 2 + } else if z.level == BestSpeed { + z.buf[8] = 4 + } else { + z.buf[8] = 0 + } + z.buf[9] = z.OS + n, z.err = z.w.Write(z.buf[:10]) + if z.err != nil { + return n, z.err + } + if z.Extra != nil { + z.err = z.writeBytes(z.Extra) + if z.err != nil { + return n, z.err + } + } + if z.Name != "" { + z.err = z.writeString(z.Name) + if z.err != nil { + return n, z.err + } + } + if z.Comment != "" { + z.err = z.writeString(z.Comment) + if z.err != nil { + return n, z.err + } + } + + if z.compressor == nil && z.level != StatelessCompression { + z.compressor, _ = flate.NewWriter(z.w, z.level) + } + } + z.size += uint32(len(p)) + z.digest = crc32.Update(z.digest, crc32.IEEETable, p) + if z.level == StatelessCompression { + return len(p), flate.StatelessDeflate(z.w, p, false, nil) + } + n, z.err = z.compressor.Write(p) + return n, z.err +} + +// Flush flushes any pending compressed data to the underlying writer. +// +// It is useful mainly in compressed network protocols, to ensure that +// a remote reader has enough data to reconstruct a packet. Flush does +// not return until the data has been written. If the underlying +// writer returns an error, Flush returns that error. +// +// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH. +func (z *Writer) Flush() error { + if z.err != nil { + return z.err + } + if z.closed || z.level == StatelessCompression { + return nil + } + if !z.wroteHeader { + z.Write(nil) + if z.err != nil { + return z.err + } + } + z.err = z.compressor.Flush() + return z.err +} + +// Close closes the Writer, flushing any unwritten data to the underlying +// io.Writer, but does not close the underlying io.Writer. +func (z *Writer) Close() error { + if z.err != nil { + return z.err + } + if z.closed { + return nil + } + z.closed = true + if !z.wroteHeader { + z.Write(nil) + if z.err != nil { + return z.err + } + } + if z.level == StatelessCompression { + z.err = flate.StatelessDeflate(z.w, nil, true, nil) + } else { + z.err = z.compressor.Close() + } + if z.err != nil { + return z.err + } + le.PutUint32(z.buf[:4], z.digest) + le.PutUint32(z.buf[4:8], z.size) + _, z.err = z.w.Write(z.buf[:8]) + return z.err +} diff --git a/vendor/github.com/klauspost/compress/huff0/.gitignore b/vendor/github.com/klauspost/compress/huff0/.gitignore new file mode 100644 index 0000000..b3d2629 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/.gitignore @@ -0,0 +1 @@ +/huff0-fuzz.zip diff --git a/vendor/github.com/klauspost/compress/huff0/README.md b/vendor/github.com/klauspost/compress/huff0/README.md new file mode 100644 index 0000000..8b6e5c6 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/README.md @@ -0,0 +1,89 @@ +# Huff0 entropy compression + +This package provides Huff0 encoding and decoding as used in zstd. + +[Huff0](https://github.com/Cyan4973/FiniteStateEntropy#new-generation-entropy-coders), +a Huffman codec designed for modern CPU, featuring OoO (Out of Order) operations on multiple ALU +(Arithmetic Logic Unit), achieving extremely fast compression and decompression speeds. + +This can be used for compressing input with a lot of similar input values to the smallest number of bytes. +This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, +but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. + +* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/huff0) + +## News + +This is used as part of the [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and decompression package. + +This ensures that most functionality is well tested. + +# Usage + +This package provides a low level interface that allows to compress single independent blocks. + +Each block is separate, and there is no built in integrity checks. +This means that the caller should keep track of block sizes and also do checksums if needed. + +Compressing a block is done via the [`Compress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress1X) and +[`Compress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress4X) functions. +You must provide input and will receive the output and maybe an error. + +These error values can be returned: + +| Error | Description | +|---------------------|-----------------------------------------------------------------------------| +| `` | Everything ok, output is returned | +| `ErrIncompressible` | Returned when input is judged to be too hard to compress | +| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | +| `ErrTooBig` | Returned if the input block exceeds the maximum allowed size (128 Kib) | +| `(error)` | An internal error occurred. | + + +As can be seen above some of there are errors that will be returned even under normal operation so it is important to handle these. + +To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object +that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same +object can be used for both. + +Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this +you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. + +The `Scratch` object will retain state that allows to re-use previous tables for encoding and decoding. + +## Tables and re-use + +Huff0 allows for reusing tables from the previous block to save space if that is expected to give better/faster results. + +The Scratch object allows you to set a [`ReusePolicy`](https://godoc.org/github.com/klauspost/compress/huff0#ReusePolicy) +that controls this behaviour. See the documentation for details. This can be altered between each block. + +Do however note that this information is *not* stored in the output block and it is up to the users of the package to +record whether [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable) should be called, +based on the boolean reported back from the CompressXX call. + +If you want to store the table separate from the data, you can access them as `OutData` and `OutTable` on the +[`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object. + +## Decompressing + +The first part of decoding is to initialize the decoding table through [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable). +This will initialize the decoding tables. +You can supply the complete block to `ReadTable` and it will return the data part of the block +which can be given to the decompressor. + +Decompressing is done by calling the [`Decompress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress1X) +or [`Decompress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress4X) function. + +For concurrently decompressing content with a fixed table a stateless [`Decoder`](https://godoc.org/github.com/klauspost/compress/huff0#Decoder) can be requested which will remain correct as long as the scratch is unchanged. The capacity of the provided slice indicates the expected output size. + +You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back +your input was likely corrupted. + +It is important to note that a successful decoding does *not* mean your output matches your original input. +There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. + +# Contributing + +Contributions are always welcome. Be aware that adding public functions will require good justification and breaking +changes will likely not be accepted. If in doubt open an issue before writing the PR. diff --git a/vendor/github.com/klauspost/compress/huff0/bitreader.go b/vendor/github.com/klauspost/compress/huff0/bitreader.go new file mode 100644 index 0000000..504a7be --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/bitreader.go @@ -0,0 +1,233 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package huff0 + +import ( + "encoding/binary" + "errors" + "fmt" + "io" +) + +// bitReader reads a bitstream in reverse. +// The last set bit indicates the start of the stream and is used +// for aligning the input. +type bitReaderBytes struct { + in []byte + off uint // next byte to read is at in[off - 1] + value uint64 + bitsRead uint8 +} + +// init initializes and resets the bit reader. +func (b *bitReaderBytes) init(in []byte) error { + if len(in) < 1 { + return errors.New("corrupt stream: too short") + } + b.in = in + b.off = uint(len(in)) + // The highest bit of the last byte indicates where to start + v := in[len(in)-1] + if v == 0 { + return errors.New("corrupt stream, did not find end of stream") + } + b.bitsRead = 64 + b.value = 0 + if len(in) >= 8 { + b.fillFastStart() + } else { + b.fill() + b.fill() + } + b.advance(8 - uint8(highBit32(uint32(v)))) + return nil +} + +// peekBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReaderBytes) peekByteFast() uint8 { + got := uint8(b.value >> 56) + return got +} + +func (b *bitReaderBytes) advance(n uint8) { + b.bitsRead += n + b.value <<= n & 63 +} + +// fillFast() will make sure at least 32 bits are available. +// There must be at least 4 bytes available. +func (b *bitReaderBytes) fillFast() { + if b.bitsRead < 32 { + return + } + + // 2 bounds checks. + v := b.in[b.off-4 : b.off] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << (b.bitsRead - 32) + b.bitsRead -= 32 + b.off -= 4 +} + +// fillFastStart() assumes the bitReaderBytes is empty and there is at least 8 bytes to read. +func (b *bitReaderBytes) fillFastStart() { + // Do single re-slice to avoid bounds checks. + b.value = binary.LittleEndian.Uint64(b.in[b.off-8:]) + b.bitsRead = 0 + b.off -= 8 +} + +// fill() will make sure at least 32 bits are available. +func (b *bitReaderBytes) fill() { + if b.bitsRead < 32 { + return + } + if b.off > 4 { + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << (b.bitsRead - 32) + b.bitsRead -= 32 + b.off -= 4 + return + } + for b.off > 0 { + b.value |= uint64(b.in[b.off-1]) << (b.bitsRead - 8) + b.bitsRead -= 8 + b.off-- + } +} + +// finished returns true if all bits have been read from the bit stream. +func (b *bitReaderBytes) finished() bool { + return b.off == 0 && b.bitsRead >= 64 +} + +func (b *bitReaderBytes) remaining() uint { + return b.off*8 + uint(64-b.bitsRead) +} + +// close the bitstream and returns an error if out-of-buffer reads occurred. +func (b *bitReaderBytes) close() error { + // Release reference. + b.in = nil + if b.remaining() > 0 { + return fmt.Errorf("corrupt input: %d bits remain on stream", b.remaining()) + } + if b.bitsRead > 64 { + return io.ErrUnexpectedEOF + } + return nil +} + +// bitReaderShifted reads a bitstream in reverse. +// The last set bit indicates the start of the stream and is used +// for aligning the input. +type bitReaderShifted struct { + in []byte + off uint // next byte to read is at in[off - 1] + value uint64 + bitsRead uint8 +} + +// init initializes and resets the bit reader. +func (b *bitReaderShifted) init(in []byte) error { + if len(in) < 1 { + return errors.New("corrupt stream: too short") + } + b.in = in + b.off = uint(len(in)) + // The highest bit of the last byte indicates where to start + v := in[len(in)-1] + if v == 0 { + return errors.New("corrupt stream, did not find end of stream") + } + b.bitsRead = 64 + b.value = 0 + if len(in) >= 8 { + b.fillFastStart() + } else { + b.fill() + b.fill() + } + b.advance(8 - uint8(highBit32(uint32(v)))) + return nil +} + +// peekBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReaderShifted) peekBitsFast(n uint8) uint16 { + return uint16(b.value >> ((64 - n) & 63)) +} + +func (b *bitReaderShifted) advance(n uint8) { + b.bitsRead += n + b.value <<= n & 63 +} + +// fillFast() will make sure at least 32 bits are available. +// There must be at least 4 bytes available. +func (b *bitReaderShifted) fillFast() { + if b.bitsRead < 32 { + return + } + + // 2 bounds checks. + v := b.in[b.off-4 : b.off] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << ((b.bitsRead - 32) & 63) + b.bitsRead -= 32 + b.off -= 4 +} + +// fillFastStart() assumes the bitReaderShifted is empty and there is at least 8 bytes to read. +func (b *bitReaderShifted) fillFastStart() { + // Do single re-slice to avoid bounds checks. + b.value = binary.LittleEndian.Uint64(b.in[b.off-8:]) + b.bitsRead = 0 + b.off -= 8 +} + +// fill() will make sure at least 32 bits are available. +func (b *bitReaderShifted) fill() { + if b.bitsRead < 32 { + return + } + if b.off > 4 { + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << ((b.bitsRead - 32) & 63) + b.bitsRead -= 32 + b.off -= 4 + return + } + for b.off > 0 { + b.value |= uint64(b.in[b.off-1]) << ((b.bitsRead - 8) & 63) + b.bitsRead -= 8 + b.off-- + } +} + +func (b *bitReaderShifted) remaining() uint { + return b.off*8 + uint(64-b.bitsRead) +} + +// close the bitstream and returns an error if out-of-buffer reads occurred. +func (b *bitReaderShifted) close() error { + // Release reference. + b.in = nil + if b.remaining() > 0 { + return fmt.Errorf("corrupt input: %d bits remain on stream", b.remaining()) + } + if b.bitsRead > 64 { + return io.ErrUnexpectedEOF + } + return nil +} diff --git a/vendor/github.com/klauspost/compress/huff0/bitwriter.go b/vendor/github.com/klauspost/compress/huff0/bitwriter.go new file mode 100644 index 0000000..ec71f7a --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/bitwriter.go @@ -0,0 +1,95 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package huff0 + +// bitWriter will write bits. +// First bit will be LSB of the first byte of output. +type bitWriter struct { + bitContainer uint64 + nBits uint8 + out []byte +} + +// bitMask16 is bitmasks. Has extra to avoid bounds check. +var bitMask16 = [32]uint16{ + 0, 1, 3, 7, 0xF, 0x1F, + 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, + 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF} /* up to 16 bits */ + +// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated. +// It will not check if there is space for them, so the caller must ensure that it has flushed recently. +func (b *bitWriter) addBits16Clean(value uint16, bits uint8) { + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// encSymbol will add up to 16 bits. value may not contain more set bits than indicated. +// It will not check if there is space for them, so the caller must ensure that it has flushed recently. +func (b *bitWriter) encSymbol(ct cTable, symbol byte) { + enc := ct[symbol] + b.bitContainer |= uint64(enc.val) << (b.nBits & 63) + if false { + if enc.nBits == 0 { + panic("nbits 0") + } + } + b.nBits += enc.nBits +} + +// encTwoSymbols will add up to 32 bits. value may not contain more set bits than indicated. +// It will not check if there is space for them, so the caller must ensure that it has flushed recently. +func (b *bitWriter) encTwoSymbols(ct cTable, av, bv byte) { + encA := ct[av] + encB := ct[bv] + sh := b.nBits & 63 + combined := uint64(encA.val) | (uint64(encB.val) << (encA.nBits & 63)) + b.bitContainer |= combined << sh + if false { + if encA.nBits == 0 { + panic("nbitsA 0") + } + if encB.nBits == 0 { + panic("nbitsB 0") + } + } + b.nBits += encA.nBits + encB.nBits +} + +// flush32 will flush out, so there are at least 32 bits available for writing. +func (b *bitWriter) flush32() { + if b.nBits < 32 { + return + } + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24)) + b.nBits -= 32 + b.bitContainer >>= 32 +} + +// flushAlign will flush remaining full bytes and align to next byte boundary. +func (b *bitWriter) flushAlign() { + nbBytes := (b.nBits + 7) >> 3 + for i := uint8(0); i < nbBytes; i++ { + b.out = append(b.out, byte(b.bitContainer>>(i*8))) + } + b.nBits = 0 + b.bitContainer = 0 +} + +// close will write the alignment bit and write the final byte(s) +// to the output. +func (b *bitWriter) close() error { + // End mark + b.addBits16Clean(1, 1) + // flush until next byte. + b.flushAlign() + return nil +} diff --git a/vendor/github.com/klauspost/compress/huff0/bytereader.go b/vendor/github.com/klauspost/compress/huff0/bytereader.go new file mode 100644 index 0000000..4dcab8d --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/bytereader.go @@ -0,0 +1,44 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package huff0 + +// byteReader provides a byte reader that reads +// little endian values from a byte stream. +// The input stream is manually advanced. +// The reader performs no bounds checks. +type byteReader struct { + b []byte + off int +} + +// init will initialize the reader and set the input. +func (b *byteReader) init(in []byte) { + b.b = in + b.off = 0 +} + +// Int32 returns a little endian int32 starting at current offset. +func (b byteReader) Int32() int32 { + v3 := int32(b.b[b.off+3]) + v2 := int32(b.b[b.off+2]) + v1 := int32(b.b[b.off+1]) + v0 := int32(b.b[b.off]) + return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0 +} + +// Uint32 returns a little endian uint32 starting at current offset. +func (b byteReader) Uint32() uint32 { + v3 := uint32(b.b[b.off+3]) + v2 := uint32(b.b[b.off+2]) + v1 := uint32(b.b[b.off+1]) + v0 := uint32(b.b[b.off]) + return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0 +} + +// remain will return the number of bytes remaining. +func (b byteReader) remain() int { + return len(b.b) - b.off +} diff --git a/vendor/github.com/klauspost/compress/huff0/compress.go b/vendor/github.com/klauspost/compress/huff0/compress.go new file mode 100644 index 0000000..4d14542 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/compress.go @@ -0,0 +1,730 @@ +package huff0 + +import ( + "fmt" + "math" + "runtime" + "sync" +) + +// Compress1X will compress the input. +// The output can be decoded using Decompress1X. +// Supply a Scratch object. The scratch object contains state about re-use, +// So when sharing across independent encodes, be sure to set the re-use policy. +func Compress1X(in []byte, s *Scratch) (out []byte, reUsed bool, err error) { + s, err = s.prepare(in) + if err != nil { + return nil, false, err + } + return compress(in, s, s.compress1X) +} + +// Compress4X will compress the input. The input is split into 4 independent blocks +// and compressed similar to Compress1X. +// The output can be decoded using Decompress4X. +// Supply a Scratch object. The scratch object contains state about re-use, +// So when sharing across independent encodes, be sure to set the re-use policy. +func Compress4X(in []byte, s *Scratch) (out []byte, reUsed bool, err error) { + s, err = s.prepare(in) + if err != nil { + return nil, false, err + } + if false { + // TODO: compress4Xp only slightly faster. + const parallelThreshold = 8 << 10 + if len(in) < parallelThreshold || runtime.GOMAXPROCS(0) == 1 { + return compress(in, s, s.compress4X) + } + return compress(in, s, s.compress4Xp) + } + return compress(in, s, s.compress4X) +} + +func compress(in []byte, s *Scratch, compressor func(src []byte) ([]byte, error)) (out []byte, reUsed bool, err error) { + // Nuke previous table if we cannot reuse anyway. + if s.Reuse == ReusePolicyNone { + s.prevTable = s.prevTable[:0] + } + + // Create histogram, if none was provided. + maxCount := s.maxCount + var canReuse = false + if maxCount == 0 { + maxCount, canReuse = s.countSimple(in) + } else { + canReuse = s.canUseTable(s.prevTable) + } + + // We want the output size to be less than this: + wantSize := len(in) + if s.WantLogLess > 0 { + wantSize -= wantSize >> s.WantLogLess + } + + // Reset for next run. + s.clearCount = true + s.maxCount = 0 + if maxCount >= len(in) { + if maxCount > len(in) { + return nil, false, fmt.Errorf("maxCount (%d) > length (%d)", maxCount, len(in)) + } + if len(in) == 1 { + return nil, false, ErrIncompressible + } + // One symbol, use RLE + return nil, false, ErrUseRLE + } + if maxCount == 1 || maxCount < (len(in)>>7) { + // Each symbol present maximum once or too well distributed. + return nil, false, ErrIncompressible + } + if s.Reuse == ReusePolicyMust && !canReuse { + // We must reuse, but we can't. + return nil, false, ErrIncompressible + } + if (s.Reuse == ReusePolicyPrefer || s.Reuse == ReusePolicyMust) && canReuse { + keepTable := s.cTable + keepTL := s.actualTableLog + s.cTable = s.prevTable + s.actualTableLog = s.prevTableLog + s.Out, err = compressor(in) + s.cTable = keepTable + s.actualTableLog = keepTL + if err == nil && len(s.Out) < wantSize { + s.OutData = s.Out + return s.Out, true, nil + } + if s.Reuse == ReusePolicyMust { + return nil, false, ErrIncompressible + } + // Do not attempt to re-use later. + s.prevTable = s.prevTable[:0] + } + + // Calculate new table. + err = s.buildCTable() + if err != nil { + return nil, false, err + } + + if false && !s.canUseTable(s.cTable) { + panic("invalid table generated") + } + + if s.Reuse == ReusePolicyAllow && canReuse { + hSize := len(s.Out) + oldSize := s.prevTable.estimateSize(s.count[:s.symbolLen]) + newSize := s.cTable.estimateSize(s.count[:s.symbolLen]) + if oldSize <= hSize+newSize || hSize+12 >= wantSize { + // Retain cTable even if we re-use. + keepTable := s.cTable + keepTL := s.actualTableLog + + s.cTable = s.prevTable + s.actualTableLog = s.prevTableLog + s.Out, err = compressor(in) + + // Restore ctable. + s.cTable = keepTable + s.actualTableLog = keepTL + if err != nil { + return nil, false, err + } + if len(s.Out) >= wantSize { + return nil, false, ErrIncompressible + } + s.OutData = s.Out + return s.Out, true, nil + } + } + + // Use new table + err = s.cTable.write(s) + if err != nil { + s.OutTable = nil + return nil, false, err + } + s.OutTable = s.Out + + // Compress using new table + s.Out, err = compressor(in) + if err != nil { + s.OutTable = nil + return nil, false, err + } + if len(s.Out) >= wantSize { + s.OutTable = nil + return nil, false, ErrIncompressible + } + // Move current table into previous. + s.prevTable, s.prevTableLog, s.cTable = s.cTable, s.actualTableLog, s.prevTable[:0] + s.OutData = s.Out[len(s.OutTable):] + return s.Out, false, nil +} + +// EstimateSizes will estimate the data sizes +func EstimateSizes(in []byte, s *Scratch) (tableSz, dataSz, reuseSz int, err error) { + s, err = s.prepare(in) + if err != nil { + return 0, 0, 0, err + } + + // Create histogram, if none was provided. + tableSz, dataSz, reuseSz = -1, -1, -1 + maxCount := s.maxCount + var canReuse = false + if maxCount == 0 { + maxCount, canReuse = s.countSimple(in) + } else { + canReuse = s.canUseTable(s.prevTable) + } + + // We want the output size to be less than this: + wantSize := len(in) + if s.WantLogLess > 0 { + wantSize -= wantSize >> s.WantLogLess + } + + // Reset for next run. + s.clearCount = true + s.maxCount = 0 + if maxCount >= len(in) { + if maxCount > len(in) { + return 0, 0, 0, fmt.Errorf("maxCount (%d) > length (%d)", maxCount, len(in)) + } + if len(in) == 1 { + return 0, 0, 0, ErrIncompressible + } + // One symbol, use RLE + return 0, 0, 0, ErrUseRLE + } + if maxCount == 1 || maxCount < (len(in)>>7) { + // Each symbol present maximum once or too well distributed. + return 0, 0, 0, ErrIncompressible + } + + // Calculate new table. + err = s.buildCTable() + if err != nil { + return 0, 0, 0, err + } + + if false && !s.canUseTable(s.cTable) { + panic("invalid table generated") + } + + tableSz, err = s.cTable.estTableSize(s) + if err != nil { + return 0, 0, 0, err + } + if canReuse { + reuseSz = s.prevTable.estimateSize(s.count[:s.symbolLen]) + } + dataSz = s.cTable.estimateSize(s.count[:s.symbolLen]) + + // Restore + return tableSz, dataSz, reuseSz, nil +} + +func (s *Scratch) compress1X(src []byte) ([]byte, error) { + return s.compress1xDo(s.Out, src) +} + +func (s *Scratch) compress1xDo(dst, src []byte) ([]byte, error) { + var bw = bitWriter{out: dst} + + // N is length divisible by 4. + n := len(src) + n -= n & 3 + cTable := s.cTable[:256] + + // Encode last bytes. + for i := len(src) & 3; i > 0; i-- { + bw.encSymbol(cTable, src[n+i-1]) + } + n -= 4 + if s.actualTableLog <= 8 { + for ; n >= 0; n -= 4 { + tmp := src[n : n+4] + // tmp should be len 4 + bw.flush32() + bw.encTwoSymbols(cTable, tmp[3], tmp[2]) + bw.encTwoSymbols(cTable, tmp[1], tmp[0]) + } + } else { + for ; n >= 0; n -= 4 { + tmp := src[n : n+4] + // tmp should be len 4 + bw.flush32() + bw.encTwoSymbols(cTable, tmp[3], tmp[2]) + bw.flush32() + bw.encTwoSymbols(cTable, tmp[1], tmp[0]) + } + } + err := bw.close() + return bw.out, err +} + +var sixZeros [6]byte + +func (s *Scratch) compress4X(src []byte) ([]byte, error) { + if len(src) < 12 { + return nil, ErrIncompressible + } + segmentSize := (len(src) + 3) / 4 + + // Add placeholder for output length + offsetIdx := len(s.Out) + s.Out = append(s.Out, sixZeros[:]...) + + for i := 0; i < 4; i++ { + toDo := src + if len(toDo) > segmentSize { + toDo = toDo[:segmentSize] + } + src = src[len(toDo):] + + var err error + idx := len(s.Out) + s.Out, err = s.compress1xDo(s.Out, toDo) + if err != nil { + return nil, err + } + if len(s.Out)-idx > math.MaxUint16 { + // We cannot store the size in the jump table + return nil, ErrIncompressible + } + // Write compressed length as little endian before block. + if i < 3 { + // Last length is not written. + length := len(s.Out) - idx + s.Out[i*2+offsetIdx] = byte(length) + s.Out[i*2+offsetIdx+1] = byte(length >> 8) + } + } + + return s.Out, nil +} + +// compress4Xp will compress 4 streams using separate goroutines. +func (s *Scratch) compress4Xp(src []byte) ([]byte, error) { + if len(src) < 12 { + return nil, ErrIncompressible + } + // Add placeholder for output length + s.Out = s.Out[:6] + + segmentSize := (len(src) + 3) / 4 + var wg sync.WaitGroup + var errs [4]error + wg.Add(4) + for i := 0; i < 4; i++ { + toDo := src + if len(toDo) > segmentSize { + toDo = toDo[:segmentSize] + } + src = src[len(toDo):] + + // Separate goroutine for each block. + go func(i int) { + s.tmpOut[i], errs[i] = s.compress1xDo(s.tmpOut[i][:0], toDo) + wg.Done() + }(i) + } + wg.Wait() + for i := 0; i < 4; i++ { + if errs[i] != nil { + return nil, errs[i] + } + o := s.tmpOut[i] + if len(o) > math.MaxUint16 { + // We cannot store the size in the jump table + return nil, ErrIncompressible + } + // Write compressed length as little endian before block. + if i < 3 { + // Last length is not written. + s.Out[i*2] = byte(len(o)) + s.Out[i*2+1] = byte(len(o) >> 8) + } + + // Write output. + s.Out = append(s.Out, o...) + } + return s.Out, nil +} + +// countSimple will create a simple histogram in s.count. +// Returns the biggest count. +// Does not update s.clearCount. +func (s *Scratch) countSimple(in []byte) (max int, reuse bool) { + reuse = true + for _, v := range in { + s.count[v]++ + } + m := uint32(0) + if len(s.prevTable) > 0 { + for i, v := range s.count[:] { + if v > m { + m = v + } + if v > 0 { + s.symbolLen = uint16(i) + 1 + if i >= len(s.prevTable) { + reuse = false + } else { + if s.prevTable[i].nBits == 0 { + reuse = false + } + } + } + } + return int(m), reuse + } + for i, v := range s.count[:] { + if v > m { + m = v + } + if v > 0 { + s.symbolLen = uint16(i) + 1 + } + } + return int(m), false +} + +func (s *Scratch) canUseTable(c cTable) bool { + if len(c) < int(s.symbolLen) { + return false + } + for i, v := range s.count[:s.symbolLen] { + if v != 0 && c[i].nBits == 0 { + return false + } + } + return true +} + +//lint:ignore U1000 used for debugging +func (s *Scratch) validateTable(c cTable) bool { + if len(c) < int(s.symbolLen) { + return false + } + for i, v := range s.count[:s.symbolLen] { + if v != 0 { + if c[i].nBits == 0 { + return false + } + if c[i].nBits > s.actualTableLog { + return false + } + } + } + return true +} + +// minTableLog provides the minimum logSize to safely represent a distribution. +func (s *Scratch) minTableLog() uint8 { + minBitsSrc := highBit32(uint32(s.br.remain())) + 1 + minBitsSymbols := highBit32(uint32(s.symbolLen-1)) + 2 + if minBitsSrc < minBitsSymbols { + return uint8(minBitsSrc) + } + return uint8(minBitsSymbols) +} + +// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog +func (s *Scratch) optimalTableLog() { + tableLog := s.TableLog + minBits := s.minTableLog() + maxBitsSrc := uint8(highBit32(uint32(s.br.remain()-1))) - 1 + if maxBitsSrc < tableLog { + // Accuracy can be reduced + tableLog = maxBitsSrc + } + if minBits > tableLog { + tableLog = minBits + } + // Need a minimum to safely represent all symbol values + if tableLog < minTablelog { + tableLog = minTablelog + } + if tableLog > tableLogMax { + tableLog = tableLogMax + } + s.actualTableLog = tableLog +} + +type cTableEntry struct { + val uint16 + nBits uint8 + // We have 8 bits extra +} + +const huffNodesMask = huffNodesLen - 1 + +func (s *Scratch) buildCTable() error { + s.optimalTableLog() + s.huffSort() + if cap(s.cTable) < maxSymbolValue+1 { + s.cTable = make([]cTableEntry, s.symbolLen, maxSymbolValue+1) + } else { + s.cTable = s.cTable[:s.symbolLen] + for i := range s.cTable { + s.cTable[i] = cTableEntry{} + } + } + + var startNode = int16(s.symbolLen) + nonNullRank := s.symbolLen - 1 + + nodeNb := startNode + huffNode := s.nodes[1 : huffNodesLen+1] + + // This overlays the slice above, but allows "-1" index lookups. + // Different from reference implementation. + huffNode0 := s.nodes[0 : huffNodesLen+1] + + for huffNode[nonNullRank].count == 0 { + nonNullRank-- + } + + lowS := int16(nonNullRank) + nodeRoot := nodeNb + lowS - 1 + lowN := nodeNb + huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count + huffNode[lowS].parent, huffNode[lowS-1].parent = uint16(nodeNb), uint16(nodeNb) + nodeNb++ + lowS -= 2 + for n := nodeNb; n <= nodeRoot; n++ { + huffNode[n].count = 1 << 30 + } + // fake entry, strong barrier + huffNode0[0].count = 1 << 31 + + // create parents + for nodeNb <= nodeRoot { + var n1, n2 int16 + if huffNode0[lowS+1].count < huffNode0[lowN+1].count { + n1 = lowS + lowS-- + } else { + n1 = lowN + lowN++ + } + if huffNode0[lowS+1].count < huffNode0[lowN+1].count { + n2 = lowS + lowS-- + } else { + n2 = lowN + lowN++ + } + + huffNode[nodeNb].count = huffNode0[n1+1].count + huffNode0[n2+1].count + huffNode0[n1+1].parent, huffNode0[n2+1].parent = uint16(nodeNb), uint16(nodeNb) + nodeNb++ + } + + // distribute weights (unlimited tree height) + huffNode[nodeRoot].nbBits = 0 + for n := nodeRoot - 1; n >= startNode; n-- { + huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1 + } + for n := uint16(0); n <= nonNullRank; n++ { + huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1 + } + s.actualTableLog = s.setMaxHeight(int(nonNullRank)) + maxNbBits := s.actualTableLog + + // fill result into tree (val, nbBits) + if maxNbBits > tableLogMax { + return fmt.Errorf("internal error: maxNbBits (%d) > tableLogMax (%d)", maxNbBits, tableLogMax) + } + var nbPerRank [tableLogMax + 1]uint16 + var valPerRank [16]uint16 + for _, v := range huffNode[:nonNullRank+1] { + nbPerRank[v.nbBits]++ + } + // determine stating value per rank + { + min := uint16(0) + for n := maxNbBits; n > 0; n-- { + // get starting value within each rank + valPerRank[n] = min + min += nbPerRank[n] + min >>= 1 + } + } + + // push nbBits per symbol, symbol order + for _, v := range huffNode[:nonNullRank+1] { + s.cTable[v.symbol].nBits = v.nbBits + } + + // assign value within rank, symbol order + t := s.cTable[:s.symbolLen] + for n, val := range t { + nbits := val.nBits & 15 + v := valPerRank[nbits] + t[n].val = v + valPerRank[nbits] = v + 1 + } + + return nil +} + +// huffSort will sort symbols, decreasing order. +func (s *Scratch) huffSort() { + type rankPos struct { + base uint32 + current uint32 + } + + // Clear nodes + nodes := s.nodes[:huffNodesLen+1] + s.nodes = nodes + nodes = nodes[1 : huffNodesLen+1] + + // Sort into buckets based on length of symbol count. + var rank [32]rankPos + for _, v := range s.count[:s.symbolLen] { + r := highBit32(v+1) & 31 + rank[r].base++ + } + // maxBitLength is log2(BlockSizeMax) + 1 + const maxBitLength = 18 + 1 + for n := maxBitLength; n > 0; n-- { + rank[n-1].base += rank[n].base + } + for n := range rank[:maxBitLength] { + rank[n].current = rank[n].base + } + for n, c := range s.count[:s.symbolLen] { + r := (highBit32(c+1) + 1) & 31 + pos := rank[r].current + rank[r].current++ + prev := nodes[(pos-1)&huffNodesMask] + for pos > rank[r].base && c > prev.count { + nodes[pos&huffNodesMask] = prev + pos-- + prev = nodes[(pos-1)&huffNodesMask] + } + nodes[pos&huffNodesMask] = nodeElt{count: c, symbol: byte(n)} + } +} + +func (s *Scratch) setMaxHeight(lastNonNull int) uint8 { + maxNbBits := s.actualTableLog + huffNode := s.nodes[1 : huffNodesLen+1] + //huffNode = huffNode[: huffNodesLen] + + largestBits := huffNode[lastNonNull].nbBits + + // early exit : no elt > maxNbBits + if largestBits <= maxNbBits { + return largestBits + } + totalCost := int(0) + baseCost := int(1) << (largestBits - maxNbBits) + n := uint32(lastNonNull) + + for huffNode[n].nbBits > maxNbBits { + totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits)) + huffNode[n].nbBits = maxNbBits + n-- + } + // n stops at huffNode[n].nbBits <= maxNbBits + + for huffNode[n].nbBits == maxNbBits { + n-- + } + // n end at index of smallest symbol using < maxNbBits + + // renorm totalCost + totalCost >>= largestBits - maxNbBits /* note : totalCost is necessarily a multiple of baseCost */ + + // repay normalized cost + { + const noSymbol = 0xF0F0F0F0 + var rankLast [tableLogMax + 2]uint32 + + for i := range rankLast[:] { + rankLast[i] = noSymbol + } + + // Get pos of last (smallest) symbol per rank + { + currentNbBits := maxNbBits + for pos := int(n); pos >= 0; pos-- { + if huffNode[pos].nbBits >= currentNbBits { + continue + } + currentNbBits = huffNode[pos].nbBits // < maxNbBits + rankLast[maxNbBits-currentNbBits] = uint32(pos) + } + } + + for totalCost > 0 { + nBitsToDecrease := uint8(highBit32(uint32(totalCost))) + 1 + + for ; nBitsToDecrease > 1; nBitsToDecrease-- { + highPos := rankLast[nBitsToDecrease] + lowPos := rankLast[nBitsToDecrease-1] + if highPos == noSymbol { + continue + } + if lowPos == noSymbol { + break + } + highTotal := huffNode[highPos].count + lowTotal := 2 * huffNode[lowPos].count + if highTotal <= lowTotal { + break + } + } + // only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) + // HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary + // FIXME: try to remove + for (nBitsToDecrease <= tableLogMax) && (rankLast[nBitsToDecrease] == noSymbol) { + nBitsToDecrease++ + } + totalCost -= 1 << (nBitsToDecrease - 1) + if rankLast[nBitsToDecrease-1] == noSymbol { + // this rank is no longer empty + rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease] + } + huffNode[rankLast[nBitsToDecrease]].nbBits++ + if rankLast[nBitsToDecrease] == 0 { + /* special case, reached largest symbol */ + rankLast[nBitsToDecrease] = noSymbol + } else { + rankLast[nBitsToDecrease]-- + if huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease { + rankLast[nBitsToDecrease] = noSymbol /* this rank is now empty */ + } + } + } + + for totalCost < 0 { /* Sometimes, cost correction overshoot */ + if rankLast[1] == noSymbol { /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */ + for huffNode[n].nbBits == maxNbBits { + n-- + } + huffNode[n+1].nbBits-- + rankLast[1] = n + 1 + totalCost++ + continue + } + huffNode[rankLast[1]+1].nbBits-- + rankLast[1]++ + totalCost++ + } + } + return maxNbBits +} + +type nodeElt struct { + count uint32 + parent uint16 + symbol byte + nbBits uint8 +} diff --git a/vendor/github.com/klauspost/compress/huff0/decompress.go b/vendor/github.com/klauspost/compress/huff0/decompress.go new file mode 100644 index 0000000..c0c48bd --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/decompress.go @@ -0,0 +1,1159 @@ +package huff0 + +import ( + "errors" + "fmt" + "io" + "sync" + + "github.com/klauspost/compress/fse" +) + +type dTable struct { + single []dEntrySingle +} + +// single-symbols decoding +type dEntrySingle struct { + entry uint16 +} + +// Uses special code for all tables that are < 8 bits. +const use8BitTables = true + +// ReadTable will read a table from the input. +// The size of the input may be larger than the table definition. +// Any content remaining after the table definition will be returned. +// If no Scratch is provided a new one is allocated. +// The returned Scratch can be used for encoding or decoding input using this table. +func ReadTable(in []byte, s *Scratch) (s2 *Scratch, remain []byte, err error) { + s, err = s.prepare(nil) + if err != nil { + return s, nil, err + } + if len(in) <= 1 { + return s, nil, errors.New("input too small for table") + } + iSize := in[0] + in = in[1:] + if iSize >= 128 { + // Uncompressed + oSize := iSize - 127 + iSize = (oSize + 1) / 2 + if int(iSize) > len(in) { + return s, nil, errors.New("input too small for table") + } + for n := uint8(0); n < oSize; n += 2 { + v := in[n/2] + s.huffWeight[n] = v >> 4 + s.huffWeight[n+1] = v & 15 + } + s.symbolLen = uint16(oSize) + in = in[iSize:] + } else { + if len(in) < int(iSize) { + return s, nil, fmt.Errorf("input too small for table, want %d bytes, have %d", iSize, len(in)) + } + // FSE compressed weights + s.fse.DecompressLimit = 255 + hw := s.huffWeight[:] + s.fse.Out = hw + b, err := fse.Decompress(in[:iSize], s.fse) + s.fse.Out = nil + if err != nil { + return s, nil, err + } + if len(b) > 255 { + return s, nil, errors.New("corrupt input: output table too large") + } + s.symbolLen = uint16(len(b)) + in = in[iSize:] + } + + // collect weight stats + var rankStats [16]uint32 + weightTotal := uint32(0) + for _, v := range s.huffWeight[:s.symbolLen] { + if v > tableLogMax { + return s, nil, errors.New("corrupt input: weight too large") + } + v2 := v & 15 + rankStats[v2]++ + // (1 << (v2-1)) is slower since the compiler cannot prove that v2 isn't 0. + weightTotal += (1 << v2) >> 1 + } + if weightTotal == 0 { + return s, nil, errors.New("corrupt input: weights zero") + } + + // get last non-null symbol weight (implied, total must be 2^n) + { + tableLog := highBit32(weightTotal) + 1 + if tableLog > tableLogMax { + return s, nil, errors.New("corrupt input: tableLog too big") + } + s.actualTableLog = uint8(tableLog) + // determine last weight + { + total := uint32(1) << tableLog + rest := total - weightTotal + verif := uint32(1) << highBit32(rest) + lastWeight := highBit32(rest) + 1 + if verif != rest { + // last value must be a clean power of 2 + return s, nil, errors.New("corrupt input: last value not power of two") + } + s.huffWeight[s.symbolLen] = uint8(lastWeight) + s.symbolLen++ + rankStats[lastWeight]++ + } + } + + if (rankStats[1] < 2) || (rankStats[1]&1 != 0) { + // by construction : at least 2 elts of rank 1, must be even + return s, nil, errors.New("corrupt input: min elt size, even check failed ") + } + + // TODO: Choose between single/double symbol decoding + + // Calculate starting value for each rank + { + var nextRankStart uint32 + for n := uint8(1); n < s.actualTableLog+1; n++ { + current := nextRankStart + nextRankStart += rankStats[n] << (n - 1) + rankStats[n] = current + } + } + + // fill DTable (always full size) + tSize := 1 << tableLogMax + if len(s.dt.single) != tSize { + s.dt.single = make([]dEntrySingle, tSize) + } + cTable := s.prevTable + if cap(cTable) < maxSymbolValue+1 { + cTable = make([]cTableEntry, 0, maxSymbolValue+1) + } + cTable = cTable[:maxSymbolValue+1] + s.prevTable = cTable[:s.symbolLen] + s.prevTableLog = s.actualTableLog + + for n, w := range s.huffWeight[:s.symbolLen] { + if w == 0 { + cTable[n] = cTableEntry{ + val: 0, + nBits: 0, + } + continue + } + length := (uint32(1) << w) >> 1 + d := dEntrySingle{ + entry: uint16(s.actualTableLog+1-w) | (uint16(n) << 8), + } + + rank := &rankStats[w] + cTable[n] = cTableEntry{ + val: uint16(*rank >> (w - 1)), + nBits: uint8(d.entry), + } + + single := s.dt.single[*rank : *rank+length] + for i := range single { + single[i] = d + } + *rank += length + } + + return s, in, nil +} + +// Decompress1X will decompress a 1X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// Before this is called, the table must be initialized with ReadTable unless +// the encoder re-used the table. +// deprecated: Use the stateless Decoder() to get a concurrent version. +func (s *Scratch) Decompress1X(in []byte) (out []byte, err error) { + if cap(s.Out) < s.MaxDecodedSize { + s.Out = make([]byte, s.MaxDecodedSize) + } + s.Out = s.Out[:0:s.MaxDecodedSize] + s.Out, err = s.Decoder().Decompress1X(s.Out, in) + return s.Out, err +} + +// Decompress4X will decompress a 4X encoded stream. +// Before this is called, the table must be initialized with ReadTable unless +// the encoder re-used the table. +// The length of the supplied input must match the end of a block exactly. +// The destination size of the uncompressed data must be known and provided. +// deprecated: Use the stateless Decoder() to get a concurrent version. +func (s *Scratch) Decompress4X(in []byte, dstSize int) (out []byte, err error) { + if dstSize > s.MaxDecodedSize { + return nil, ErrMaxDecodedSizeExceeded + } + if cap(s.Out) < dstSize { + s.Out = make([]byte, s.MaxDecodedSize) + } + s.Out = s.Out[:0:dstSize] + s.Out, err = s.Decoder().Decompress4X(s.Out, in) + return s.Out, err +} + +// Decoder will return a stateless decoder that can be used by multiple +// decompressors concurrently. +// Before this is called, the table must be initialized with ReadTable. +// The Decoder is still linked to the scratch buffer so that cannot be reused. +// However, it is safe to discard the scratch. +func (s *Scratch) Decoder() *Decoder { + return &Decoder{ + dt: s.dt, + actualTableLog: s.actualTableLog, + bufs: &s.decPool, + } +} + +// Decoder provides stateless decoding. +type Decoder struct { + dt dTable + actualTableLog uint8 + bufs *sync.Pool +} + +func (d *Decoder) buffer() *[4][256]byte { + buf, ok := d.bufs.Get().(*[4][256]byte) + if ok { + return buf + } + return &[4][256]byte{} +} + +// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) decompress1X8Bit(dst, src []byte) ([]byte, error) { + if d.actualTableLog == 8 { + return d.decompress1X8BitExactly(dst, src) + } + var br bitReaderBytes + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:0] + + // Avoid bounds check by always having full sized table. + dt := d.dt.single[:256] + + // Use temp table to avoid bound checks/append penalty. + bufs := d.buffer() + buf := &bufs[0] + var off uint8 + + switch d.actualTableLog { + case 8: + const shift = 8 - 8 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + br.close() + d.bufs.Put(bufs) + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 7: + const shift = 8 - 7 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + br.close() + d.bufs.Put(bufs) + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 6: + const shift = 8 - 6 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 5: + const shift = 8 - 5 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 4: + const shift = 8 - 4 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 3: + const shift = 8 - 3 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 2: + const shift = 8 - 2 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + case 1: + const shift = 8 - 1 + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>(56+shift))] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + default: + d.bufs.Put(bufs) + return nil, fmt.Errorf("invalid tablelog: %d", d.actualTableLog) + } + + if len(dst)+int(off) > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:off]...) + + // br < 4, so uint8 is fine + bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead)) + shift := (8 - d.actualTableLog) & 7 + + for bitsLeft > 0 { + if br.bitsRead >= 64-8 { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + if len(dst) >= maxDecodedSize { + br.close() + d.bufs.Put(bufs) + return nil, ErrMaxDecodedSizeExceeded + } + v := dt[br.peekByteFast()>>shift] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= int8(nBits) + dst = append(dst, uint8(v.entry>>8)) + } + d.bufs.Put(bufs) + return dst, br.close() +} + +// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) decompress1X8BitExactly(dst, src []byte) ([]byte, error) { + var br bitReaderBytes + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:0] + + // Avoid bounds check by always having full sized table. + dt := d.dt.single[:256] + + // Use temp table to avoid bound checks/append penalty. + bufs := d.buffer() + buf := &bufs[0] + var off uint8 + + const shift = 56 + + //fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog) + for br.off >= 4 { + br.fillFast() + v := dt[uint8(br.value>>shift)] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>shift)] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>shift)] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[uint8(br.value>>shift)] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + + if len(dst)+int(off) > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:off]...) + + // br < 4, so uint8 is fine + bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead)) + for bitsLeft > 0 { + if br.bitsRead >= 64-8 { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + if len(dst) >= maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + v := dt[br.peekByteFast()] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= int8(nBits) + dst = append(dst, uint8(v.entry>>8)) + } + d.bufs.Put(bufs) + return dst, br.close() +} + +// Decompress4X will decompress a 4X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// The *capacity* of the dst slice must match the destination size of +// the uncompressed data exactly. +func (d *Decoder) decompress4X8bit(dst, src []byte) ([]byte, error) { + if d.actualTableLog == 8 { + return d.decompress4X8bitExactly(dst, src) + } + + var br [4]bitReaderBytes + start := 6 + for i := 0; i < 3; i++ { + length := int(src[i*2]) | (int(src[i*2+1]) << 8) + if start+length >= len(src) { + return nil, errors.New("truncated input (or invalid offset)") + } + err := br[i].init(src[start : start+length]) + if err != nil { + return nil, err + } + start += length + } + err := br[3].init(src[start:]) + if err != nil { + return nil, err + } + + // destination, offset to match first output + dstSize := cap(dst) + dst = dst[:dstSize] + out := dst + dstEvery := (dstSize + 3) / 4 + + shift := (56 + (8 - d.actualTableLog)) & 63 + + const tlSize = 1 << 8 + single := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + buf := d.buffer() + var off uint8 + var decoded int + + // Decode 4 values from each decoder/loop. + const bufoff = 256 + for { + if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 { + break + } + + { + // Interleave 2 decodes. + const stream = 0 + const stream2 = 1 + br1 := &br[stream] + br2 := &br[stream2] + br1.fillFast() + br2.fillFast() + + v := single[uint8(br1.value>>shift)].entry + v2 := single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off] = uint8(v >> 8) + buf[stream2][off] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+1] = uint8(v >> 8) + buf[stream2][off+1] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+2] = uint8(v >> 8) + buf[stream2][off+2] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+3] = uint8(v >> 8) + buf[stream2][off+3] = uint8(v2 >> 8) + } + + { + const stream = 2 + const stream2 = 3 + br1 := &br[stream] + br2 := &br[stream2] + br1.fillFast() + br2.fillFast() + + v := single[uint8(br1.value>>shift)].entry + v2 := single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off] = uint8(v >> 8) + buf[stream2][off] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+1] = uint8(v >> 8) + buf[stream2][off+1] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+2] = uint8(v >> 8) + buf[stream2][off+2] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+3] = uint8(v >> 8) + buf[stream2][off+3] = uint8(v2 >> 8) + } + + off += 4 + + if off == 0 { + if bufoff > dstEvery { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 1") + } + copy(out, buf[0][:]) + copy(out[dstEvery:], buf[1][:]) + copy(out[dstEvery*2:], buf[2][:]) + copy(out[dstEvery*3:], buf[3][:]) + out = out[bufoff:] + decoded += bufoff * 4 + // There must at least be 3 buffers left. + if len(out) < dstEvery*3 { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 2") + } + } + } + if off > 0 { + ioff := int(off) + if len(out) < dstEvery*3+ioff { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 3") + } + copy(out, buf[0][:off]) + copy(out[dstEvery:], buf[1][:off]) + copy(out[dstEvery*2:], buf[2][:off]) + copy(out[dstEvery*3:], buf[3][:off]) + decoded += int(off) * 4 + out = out[off:] + } + + // Decode remaining. + // Decode remaining. + remainBytes := dstEvery - (decoded / 4) + for i := range br { + offset := dstEvery * i + endsAt := offset + remainBytes + if endsAt > len(out) { + endsAt = len(out) + } + br := &br[i] + bitsLeft := br.remaining() + for bitsLeft > 0 { + if br.finished() { + d.bufs.Put(buf) + return nil, io.ErrUnexpectedEOF + } + if br.bitsRead >= 56 { + if br.off >= 4 { + v := br.in[br.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + br.value |= uint64(low) << (br.bitsRead - 32) + br.bitsRead -= 32 + br.off -= 4 + } else { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + } + // end inline... + if offset >= endsAt { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 4") + } + + // Read value and increment offset. + v := single[uint8(br.value>>shift)].entry + nBits := uint8(v) + br.advance(nBits) + bitsLeft -= uint(nBits) + out[offset] = uint8(v >> 8) + offset++ + } + if offset != endsAt { + d.bufs.Put(buf) + return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt) + } + decoded += offset - dstEvery*i + err = br.close() + if err != nil { + d.bufs.Put(buf) + return nil, err + } + } + d.bufs.Put(buf) + if dstSize != decoded { + return nil, errors.New("corruption detected: short output block") + } + return dst, nil +} + +// Decompress4X will decompress a 4X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// The *capacity* of the dst slice must match the destination size of +// the uncompressed data exactly. +func (d *Decoder) decompress4X8bitExactly(dst, src []byte) ([]byte, error) { + var br [4]bitReaderBytes + start := 6 + for i := 0; i < 3; i++ { + length := int(src[i*2]) | (int(src[i*2+1]) << 8) + if start+length >= len(src) { + return nil, errors.New("truncated input (or invalid offset)") + } + err := br[i].init(src[start : start+length]) + if err != nil { + return nil, err + } + start += length + } + err := br[3].init(src[start:]) + if err != nil { + return nil, err + } + + // destination, offset to match first output + dstSize := cap(dst) + dst = dst[:dstSize] + out := dst + dstEvery := (dstSize + 3) / 4 + + const shift = 56 + const tlSize = 1 << 8 + single := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + buf := d.buffer() + var off uint8 + var decoded int + + // Decode 4 values from each decoder/loop. + const bufoff = 256 + for { + if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 { + break + } + + { + // Interleave 2 decodes. + const stream = 0 + const stream2 = 1 + br1 := &br[stream] + br2 := &br[stream2] + br1.fillFast() + br2.fillFast() + + v := single[uint8(br1.value>>shift)].entry + v2 := single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off] = uint8(v >> 8) + buf[stream2][off] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+1] = uint8(v >> 8) + buf[stream2][off+1] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+2] = uint8(v >> 8) + buf[stream2][off+2] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+3] = uint8(v >> 8) + buf[stream2][off+3] = uint8(v2 >> 8) + } + + { + const stream = 2 + const stream2 = 3 + br1 := &br[stream] + br2 := &br[stream2] + br1.fillFast() + br2.fillFast() + + v := single[uint8(br1.value>>shift)].entry + v2 := single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off] = uint8(v >> 8) + buf[stream2][off] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+1] = uint8(v >> 8) + buf[stream2][off+1] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+2] = uint8(v >> 8) + buf[stream2][off+2] = uint8(v2 >> 8) + + v = single[uint8(br1.value>>shift)].entry + v2 = single[uint8(br2.value>>shift)].entry + br1.bitsRead += uint8(v) + br1.value <<= v & 63 + br2.bitsRead += uint8(v2) + br2.value <<= v2 & 63 + buf[stream][off+3] = uint8(v >> 8) + buf[stream2][off+3] = uint8(v2 >> 8) + } + + off += 4 + + if off == 0 { + if bufoff > dstEvery { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 1") + } + copy(out, buf[0][:]) + copy(out[dstEvery:], buf[1][:]) + copy(out[dstEvery*2:], buf[2][:]) + copy(out[dstEvery*3:], buf[3][:]) + out = out[bufoff:] + decoded += bufoff * 4 + // There must at least be 3 buffers left. + if len(out) < dstEvery*3 { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 2") + } + } + } + if off > 0 { + ioff := int(off) + if len(out) < dstEvery*3+ioff { + return nil, errors.New("corruption detected: stream overrun 3") + } + copy(out, buf[0][:off]) + copy(out[dstEvery:], buf[1][:off]) + copy(out[dstEvery*2:], buf[2][:off]) + copy(out[dstEvery*3:], buf[3][:off]) + decoded += int(off) * 4 + out = out[off:] + } + + // Decode remaining. + remainBytes := dstEvery - (decoded / 4) + for i := range br { + offset := dstEvery * i + endsAt := offset + remainBytes + if endsAt > len(out) { + endsAt = len(out) + } + br := &br[i] + bitsLeft := br.remaining() + for bitsLeft > 0 { + if br.finished() { + d.bufs.Put(buf) + return nil, io.ErrUnexpectedEOF + } + if br.bitsRead >= 56 { + if br.off >= 4 { + v := br.in[br.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + br.value |= uint64(low) << (br.bitsRead - 32) + br.bitsRead -= 32 + br.off -= 4 + } else { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + } + // end inline... + if offset >= endsAt { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 4") + } + + // Read value and increment offset. + v := single[br.peekByteFast()].entry + nBits := uint8(v) + br.advance(nBits) + bitsLeft -= uint(nBits) + out[offset] = uint8(v >> 8) + offset++ + } + if offset != endsAt { + d.bufs.Put(buf) + return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt) + } + + decoded += offset - dstEvery*i + err = br.close() + if err != nil { + d.bufs.Put(buf) + return nil, err + } + } + d.bufs.Put(buf) + if dstSize != decoded { + return nil, errors.New("corruption detected: short output block") + } + return dst, nil +} + +// matches will compare a decoding table to a coding table. +// Errors are written to the writer. +// Nothing will be written if table is ok. +func (s *Scratch) matches(ct cTable, w io.Writer) { + if s == nil || len(s.dt.single) == 0 { + return + } + dt := s.dt.single[:1<>8) == byte(sym) { + fmt.Fprintf(w, "symbol %x has decoder, but no encoder\n", sym) + errs++ + break + } + } + if errs == 0 { + broken-- + } + continue + } + // Unused bits in input + ub := tablelog - enc.nBits + top := enc.val << ub + // decoder looks at top bits. + dec := dt[top] + if uint8(dec.entry) != enc.nBits { + fmt.Fprintf(w, "symbol 0x%x bit size mismatch (enc: %d, dec:%d).\n", sym, enc.nBits, uint8(dec.entry)) + errs++ + } + if uint8(dec.entry>>8) != uint8(sym) { + fmt.Fprintf(w, "symbol 0x%x decoder output mismatch (enc: %d, dec:%d).\n", sym, sym, uint8(dec.entry>>8)) + errs++ + } + if errs > 0 { + fmt.Fprintf(w, "%d errros in base, stopping\n", errs) + continue + } + // Ensure that all combinations are covered. + for i := uint16(0); i < (1 << ub); i++ { + vval := top | i + dec := dt[vval] + if uint8(dec.entry) != enc.nBits { + fmt.Fprintf(w, "symbol 0x%x bit size mismatch (enc: %d, dec:%d).\n", vval, enc.nBits, uint8(dec.entry)) + errs++ + } + if uint8(dec.entry>>8) != uint8(sym) { + fmt.Fprintf(w, "symbol 0x%x decoder output mismatch (enc: %d, dec:%d).\n", vval, sym, uint8(dec.entry>>8)) + errs++ + } + if errs > 20 { + fmt.Fprintf(w, "%d errros, stopping\n", errs) + break + } + } + if errs == 0 { + ok++ + broken-- + } + } + if broken > 0 { + fmt.Fprintf(w, "%d broken, %d ok\n", broken, ok) + } +} diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go new file mode 100644 index 0000000..9f3e9f7 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go @@ -0,0 +1,222 @@ +//go:build amd64 && !appengine && !noasm && gc +// +build amd64,!appengine,!noasm,gc + +// This file contains the specialisation of Decoder.Decompress4X +// and Decoder.Decompress1X that use an asm implementation of thir main loops. +package huff0 + +import ( + "errors" + "fmt" + + "github.com/klauspost/compress/internal/cpuinfo" +) + +// decompress4x_main_loop_x86 is an x86 assembler implementation +// of Decompress4X when tablelog > 8. +//go:noescape +func decompress4x_main_loop_amd64(ctx *decompress4xContext) + +// decompress4x_8b_loop_x86 is an x86 assembler implementation +// of Decompress4X when tablelog <= 8 which decodes 4 entries +// per loop. +//go:noescape +func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext) + +// fallback8BitSize is the size where using Go version is faster. +const fallback8BitSize = 800 + +type decompress4xContext struct { + pbr *[4]bitReaderShifted + peekBits uint8 + out *byte + dstEvery int + tbl *dEntrySingle + decoded int + limit *byte +} + +// Decompress4X will decompress a 4X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// The *capacity* of the dst slice must match the destination size of +// the uncompressed data exactly. +func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) { + if len(d.dt.single) == 0 { + return nil, errors.New("no table loaded") + } + if len(src) < 6+(4*1) { + return nil, errors.New("input too small") + } + + use8BitTables := d.actualTableLog <= 8 + if cap(dst) < fallback8BitSize && use8BitTables { + return d.decompress4X8bit(dst, src) + } + + var br [4]bitReaderShifted + // Decode "jump table" + start := 6 + for i := 0; i < 3; i++ { + length := int(src[i*2]) | (int(src[i*2+1]) << 8) + if start+length >= len(src) { + return nil, errors.New("truncated input (or invalid offset)") + } + err := br[i].init(src[start : start+length]) + if err != nil { + return nil, err + } + start += length + } + err := br[3].init(src[start:]) + if err != nil { + return nil, err + } + + // destination, offset to match first output + dstSize := cap(dst) + dst = dst[:dstSize] + out := dst + dstEvery := (dstSize + 3) / 4 + + const tlSize = 1 << tableLogMax + const tlMask = tlSize - 1 + single := d.dt.single[:tlSize] + + var decoded int + + if len(out) > 4*4 && !(br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4) { + ctx := decompress4xContext{ + pbr: &br, + peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast() + out: &out[0], + dstEvery: dstEvery, + tbl: &single[0], + limit: &out[dstEvery-4], // Always stop decoding when first buffer gets here to avoid writing OOB on last. + } + if use8BitTables { + decompress4x_8b_main_loop_amd64(&ctx) + } else { + decompress4x_main_loop_amd64(&ctx) + } + + decoded = ctx.decoded + out = out[decoded/4:] + } + + // Decode remaining. + remainBytes := dstEvery - (decoded / 4) + for i := range br { + offset := dstEvery * i + endsAt := offset + remainBytes + if endsAt > len(out) { + endsAt = len(out) + } + br := &br[i] + bitsLeft := br.remaining() + for bitsLeft > 0 { + br.fill() + if offset >= endsAt { + return nil, errors.New("corruption detected: stream overrun 4") + } + + // Read value and increment offset. + val := br.peekBitsFast(d.actualTableLog) + v := single[val&tlMask].entry + nBits := uint8(v) + br.advance(nBits) + bitsLeft -= uint(nBits) + out[offset] = uint8(v >> 8) + offset++ + } + if offset != endsAt { + return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt) + } + decoded += offset - dstEvery*i + err = br.close() + if err != nil { + return nil, err + } + } + if dstSize != decoded { + return nil, errors.New("corruption detected: short output block") + } + return dst, nil +} + +// decompress4x_main_loop_x86 is an x86 assembler implementation +// of Decompress1X when tablelog > 8. +//go:noescape +func decompress1x_main_loop_amd64(ctx *decompress1xContext) + +// decompress4x_main_loop_x86 is an x86 with BMI2 assembler implementation +// of Decompress1X when tablelog > 8. +//go:noescape +func decompress1x_main_loop_bmi2(ctx *decompress1xContext) + +type decompress1xContext struct { + pbr *bitReaderShifted + peekBits uint8 + out *byte + outCap int + tbl *dEntrySingle + decoded int +} + +// Error reported by asm implementations +const error_max_decoded_size_exeeded = -1 + +// Decompress1X will decompress a 1X encoded stream. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) { + if len(d.dt.single) == 0 { + return nil, errors.New("no table loaded") + } + var br bitReaderShifted + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:maxDecodedSize] + + const tlSize = 1 << tableLogMax + const tlMask = tlSize - 1 + + if maxDecodedSize >= 4 { + ctx := decompress1xContext{ + pbr: &br, + out: &dst[0], + outCap: maxDecodedSize, + peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast() + tbl: &d.dt.single[0], + } + + if cpuinfo.HasBMI2() { + decompress1x_main_loop_bmi2(&ctx) + } else { + decompress1x_main_loop_amd64(&ctx) + } + if ctx.decoded == error_max_decoded_size_exeeded { + return nil, ErrMaxDecodedSizeExceeded + } + + dst = dst[:ctx.decoded] + } + + // br < 8, so uint8 is fine + bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead + for bitsLeft > 0 { + br.fill() + if len(dst) >= maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= nBits + dst = append(dst, uint8(v.entry>>8)) + } + return dst, br.close() +} diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s new file mode 100644 index 0000000..dd1a5ae --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s @@ -0,0 +1,847 @@ +// Code generated by command: go run gen.go -out ../decompress_amd64.s -pkg=huff0. DO NOT EDIT. + +//go:build amd64 && !appengine && !noasm && gc +// +build amd64,!appengine,!noasm,gc + +// func decompress4x_main_loop_amd64(ctx *decompress4xContext) +TEXT ·decompress4x_main_loop_amd64(SB), $0-8 + XORQ DX, DX + + // Preload values + MOVQ ctx+0(FP), AX + MOVBQZX 8(AX), DI + MOVQ 16(AX), SI + MOVQ 48(AX), BX + MOVQ 24(AX), R9 + MOVQ 32(AX), R10 + MOVQ (AX), R11 + + // Main loop +main_loop: + MOVQ SI, R8 + CMPQ R8, BX + SETGE DL + + // br0.fillFast32() + MOVQ 32(R11), R12 + MOVBQZX 40(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill0 + MOVQ 24(R11), AX + SUBQ $0x20, R13 + SUBQ $0x04, AX + MOVQ (R11), R14 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (AX)(R14*1), R14 + MOVQ R13, CX + SHLQ CL, R14 + MOVQ AX, 24(R11) + ORQ R14, R12 + + // exhausted = exhausted || (br0.off < 4) + CMPQ AX, $0x04 + SETLT AL + ORB AL, DL + +skip_fill0: + // val0 := br0.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br0.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br0.peekTopBits(peekBits) + MOVQ DI, CX + MOVQ R12, R14 + SHRQ CL, R14 + + // v1 := table[val1&mask] + MOVW (R10)(R14*2), CX + + // br0.advance(uint8(v1.entry)) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // these two writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + MOVW AX, (R8) + + // update the bitreader structure + MOVQ R12, 32(R11) + MOVB R13, 40(R11) + ADDQ R9, R8 + + // br1.fillFast32() + MOVQ 80(R11), R12 + MOVBQZX 88(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill1 + MOVQ 72(R11), AX + SUBQ $0x20, R13 + SUBQ $0x04, AX + MOVQ 48(R11), R14 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (AX)(R14*1), R14 + MOVQ R13, CX + SHLQ CL, R14 + MOVQ AX, 72(R11) + ORQ R14, R12 + + // exhausted = exhausted || (br1.off < 4) + CMPQ AX, $0x04 + SETLT AL + ORB AL, DL + +skip_fill1: + // val0 := br1.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br1.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br1.peekTopBits(peekBits) + MOVQ DI, CX + MOVQ R12, R14 + SHRQ CL, R14 + + // v1 := table[val1&mask] + MOVW (R10)(R14*2), CX + + // br1.advance(uint8(v1.entry)) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // these two writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + MOVW AX, (R8) + + // update the bitreader structure + MOVQ R12, 80(R11) + MOVB R13, 88(R11) + ADDQ R9, R8 + + // br2.fillFast32() + MOVQ 128(R11), R12 + MOVBQZX 136(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill2 + MOVQ 120(R11), AX + SUBQ $0x20, R13 + SUBQ $0x04, AX + MOVQ 96(R11), R14 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (AX)(R14*1), R14 + MOVQ R13, CX + SHLQ CL, R14 + MOVQ AX, 120(R11) + ORQ R14, R12 + + // exhausted = exhausted || (br2.off < 4) + CMPQ AX, $0x04 + SETLT AL + ORB AL, DL + +skip_fill2: + // val0 := br2.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br2.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br2.peekTopBits(peekBits) + MOVQ DI, CX + MOVQ R12, R14 + SHRQ CL, R14 + + // v1 := table[val1&mask] + MOVW (R10)(R14*2), CX + + // br2.advance(uint8(v1.entry)) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // these two writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + MOVW AX, (R8) + + // update the bitreader structure + MOVQ R12, 128(R11) + MOVB R13, 136(R11) + ADDQ R9, R8 + + // br3.fillFast32() + MOVQ 176(R11), R12 + MOVBQZX 184(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill3 + MOVQ 168(R11), AX + SUBQ $0x20, R13 + SUBQ $0x04, AX + MOVQ 144(R11), R14 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (AX)(R14*1), R14 + MOVQ R13, CX + SHLQ CL, R14 + MOVQ AX, 168(R11) + ORQ R14, R12 + + // exhausted = exhausted || (br3.off < 4) + CMPQ AX, $0x04 + SETLT AL + ORB AL, DL + +skip_fill3: + // val0 := br3.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br3.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br3.peekTopBits(peekBits) + MOVQ DI, CX + MOVQ R12, R14 + SHRQ CL, R14 + + // v1 := table[val1&mask] + MOVW (R10)(R14*2), CX + + // br3.advance(uint8(v1.entry)) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // these two writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + MOVW AX, (R8) + + // update the bitreader structure + MOVQ R12, 176(R11) + MOVB R13, 184(R11) + ADDQ $0x02, SI + TESTB DL, DL + JZ main_loop + MOVQ ctx+0(FP), AX + SUBQ 16(AX), SI + SHLQ $0x02, SI + MOVQ SI, 40(AX) + RET + +// func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext) +TEXT ·decompress4x_8b_main_loop_amd64(SB), $0-8 + XORQ DX, DX + + // Preload values + MOVQ ctx+0(FP), CX + MOVBQZX 8(CX), DI + MOVQ 16(CX), BX + MOVQ 48(CX), SI + MOVQ 24(CX), R9 + MOVQ 32(CX), R10 + MOVQ (CX), R11 + + // Main loop +main_loop: + MOVQ BX, R8 + CMPQ R8, SI + SETGE DL + + // br0.fillFast32() + MOVQ 32(R11), R12 + MOVBQZX 40(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill0 + MOVQ 24(R11), R14 + SUBQ $0x20, R13 + SUBQ $0x04, R14 + MOVQ (R11), R15 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (R14)(R15*1), R15 + MOVQ R13, CX + SHLQ CL, R15 + MOVQ R14, 24(R11) + ORQ R15, R12 + + // exhausted = exhausted || (br0.off < 4) + CMPQ R14, $0x04 + SETLT AL + ORB AL, DL + +skip_fill0: + // val0 := br0.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br0.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br0.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v1 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br0.advance(uint8(v1.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // val2 := br0.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v2 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br0.advance(uint8(v2.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // val3 := br0.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v3 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br0.advance(uint8(v3.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // these four writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + // out[id * dstEvery + 3] = uint8(v2.entry >> 8) + // out[id * dstEvery + 4] = uint8(v3.entry >> 8) + MOVL AX, (R8) + + // update the bitreader structure + MOVQ R12, 32(R11) + MOVB R13, 40(R11) + ADDQ R9, R8 + + // br1.fillFast32() + MOVQ 80(R11), R12 + MOVBQZX 88(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill1 + MOVQ 72(R11), R14 + SUBQ $0x20, R13 + SUBQ $0x04, R14 + MOVQ 48(R11), R15 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (R14)(R15*1), R15 + MOVQ R13, CX + SHLQ CL, R15 + MOVQ R14, 72(R11) + ORQ R15, R12 + + // exhausted = exhausted || (br1.off < 4) + CMPQ R14, $0x04 + SETLT AL + ORB AL, DL + +skip_fill1: + // val0 := br1.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br1.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br1.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v1 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br1.advance(uint8(v1.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // val2 := br1.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v2 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br1.advance(uint8(v2.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // val3 := br1.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v3 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br1.advance(uint8(v3.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // these four writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + // out[id * dstEvery + 3] = uint8(v2.entry >> 8) + // out[id * dstEvery + 4] = uint8(v3.entry >> 8) + MOVL AX, (R8) + + // update the bitreader structure + MOVQ R12, 80(R11) + MOVB R13, 88(R11) + ADDQ R9, R8 + + // br2.fillFast32() + MOVQ 128(R11), R12 + MOVBQZX 136(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill2 + MOVQ 120(R11), R14 + SUBQ $0x20, R13 + SUBQ $0x04, R14 + MOVQ 96(R11), R15 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (R14)(R15*1), R15 + MOVQ R13, CX + SHLQ CL, R15 + MOVQ R14, 120(R11) + ORQ R15, R12 + + // exhausted = exhausted || (br2.off < 4) + CMPQ R14, $0x04 + SETLT AL + ORB AL, DL + +skip_fill2: + // val0 := br2.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br2.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br2.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v1 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br2.advance(uint8(v1.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // val2 := br2.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v2 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br2.advance(uint8(v2.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // val3 := br2.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v3 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br2.advance(uint8(v3.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // these four writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + // out[id * dstEvery + 3] = uint8(v2.entry >> 8) + // out[id * dstEvery + 4] = uint8(v3.entry >> 8) + MOVL AX, (R8) + + // update the bitreader structure + MOVQ R12, 128(R11) + MOVB R13, 136(R11) + ADDQ R9, R8 + + // br3.fillFast32() + MOVQ 176(R11), R12 + MOVBQZX 184(R11), R13 + CMPQ R13, $0x20 + JBE skip_fill3 + MOVQ 168(R11), R14 + SUBQ $0x20, R13 + SUBQ $0x04, R14 + MOVQ 144(R11), R15 + + // b.value |= uint64(low) << (b.bitsRead & 63) + MOVL (R14)(R15*1), R15 + MOVQ R13, CX + SHLQ CL, R15 + MOVQ R14, 168(R11) + ORQ R15, R12 + + // exhausted = exhausted || (br3.off < 4) + CMPQ R14, $0x04 + SETLT AL + ORB AL, DL + +skip_fill3: + // val0 := br3.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v0 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br3.advance(uint8(v0.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + + // val1 := br3.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v1 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br3.advance(uint8(v1.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // val2 := br3.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v2 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br3.advance(uint8(v2.entry) + MOVB CH, AH + SHLQ CL, R12 + ADDB CL, R13 + + // val3 := br3.peekTopBits(peekBits) + MOVQ R12, R14 + MOVQ DI, CX + SHRQ CL, R14 + + // v3 := table[val0&mask] + MOVW (R10)(R14*2), CX + + // br3.advance(uint8(v3.entry) + MOVB CH, AL + SHLQ CL, R12 + ADDB CL, R13 + BSWAPL AX + + // these four writes get coalesced + // out[id * dstEvery + 0] = uint8(v0.entry >> 8) + // out[id * dstEvery + 1] = uint8(v1.entry >> 8) + // out[id * dstEvery + 3] = uint8(v2.entry >> 8) + // out[id * dstEvery + 4] = uint8(v3.entry >> 8) + MOVL AX, (R8) + + // update the bitreader structure + MOVQ R12, 176(R11) + MOVB R13, 184(R11) + ADDQ $0x04, BX + TESTB DL, DL + JZ main_loop + MOVQ ctx+0(FP), AX + SUBQ 16(AX), BX + SHLQ $0x02, BX + MOVQ BX, 40(AX) + RET + +// func decompress1x_main_loop_amd64(ctx *decompress1xContext) +TEXT ·decompress1x_main_loop_amd64(SB), $0-8 + MOVQ ctx+0(FP), CX + MOVQ 16(CX), DX + MOVQ 24(CX), BX + CMPQ BX, $0x04 + JB error_max_decoded_size_exeeded + LEAQ (DX)(BX*1), BX + MOVQ (CX), SI + MOVQ (SI), R8 + MOVQ 24(SI), R9 + MOVQ 32(SI), R10 + MOVBQZX 40(SI), R11 + MOVQ 32(CX), SI + MOVBQZX 8(CX), DI + JMP loop_condition + +main_loop: + // Check if we have room for 4 bytes in the output buffer + LEAQ 4(DX), CX + CMPQ CX, BX + JGE error_max_decoded_size_exeeded + + // Decode 4 values + CMPQ R11, $0x20 + JL bitReader_fillFast_1_end + SUBQ $0x20, R11 + SUBQ $0x04, R9 + MOVL (R8)(R9*1), R12 + MOVQ R11, CX + SHLQ CL, R12 + ORQ R12, R10 + +bitReader_fillFast_1_end: + MOVQ DI, CX + MOVQ R10, R12 + SHRQ CL, R12 + MOVW (SI)(R12*2), CX + MOVB CH, AL + MOVBQZX CL, CX + ADDQ CX, R11 + SHLQ CL, R10 + MOVQ DI, CX + MOVQ R10, R12 + SHRQ CL, R12 + MOVW (SI)(R12*2), CX + MOVB CH, AH + MOVBQZX CL, CX + ADDQ CX, R11 + SHLQ CL, R10 + BSWAPL AX + CMPQ R11, $0x20 + JL bitReader_fillFast_2_end + SUBQ $0x20, R11 + SUBQ $0x04, R9 + MOVL (R8)(R9*1), R12 + MOVQ R11, CX + SHLQ CL, R12 + ORQ R12, R10 + +bitReader_fillFast_2_end: + MOVQ DI, CX + MOVQ R10, R12 + SHRQ CL, R12 + MOVW (SI)(R12*2), CX + MOVB CH, AH + MOVBQZX CL, CX + ADDQ CX, R11 + SHLQ CL, R10 + MOVQ DI, CX + MOVQ R10, R12 + SHRQ CL, R12 + MOVW (SI)(R12*2), CX + MOVB CH, AL + MOVBQZX CL, CX + ADDQ CX, R11 + SHLQ CL, R10 + BSWAPL AX + + // Store the decoded values + MOVL AX, (DX) + ADDQ $0x04, DX + +loop_condition: + CMPQ R9, $0x08 + JGE main_loop + + // Update ctx structure + MOVQ ctx+0(FP), AX + SUBQ 16(AX), DX + MOVQ DX, 40(AX) + MOVQ (AX), AX + MOVQ R9, 24(AX) + MOVQ R10, 32(AX) + MOVB R11, 40(AX) + RET + + // Report error +error_max_decoded_size_exeeded: + MOVQ ctx+0(FP), AX + MOVQ $-1, CX + MOVQ CX, 40(AX) + RET + +// func decompress1x_main_loop_bmi2(ctx *decompress1xContext) +// Requires: BMI2 +TEXT ·decompress1x_main_loop_bmi2(SB), $0-8 + MOVQ ctx+0(FP), CX + MOVQ 16(CX), DX + MOVQ 24(CX), BX + CMPQ BX, $0x04 + JB error_max_decoded_size_exeeded + LEAQ (DX)(BX*1), BX + MOVQ (CX), SI + MOVQ (SI), R8 + MOVQ 24(SI), R9 + MOVQ 32(SI), R10 + MOVBQZX 40(SI), R11 + MOVQ 32(CX), SI + MOVBQZX 8(CX), DI + JMP loop_condition + +main_loop: + // Check if we have room for 4 bytes in the output buffer + LEAQ 4(DX), CX + CMPQ CX, BX + JGE error_max_decoded_size_exeeded + + // Decode 4 values + CMPQ R11, $0x20 + JL bitReader_fillFast_1_end + SUBQ $0x20, R11 + SUBQ $0x04, R9 + MOVL (R8)(R9*1), CX + SHLXQ R11, CX, CX + ORQ CX, R10 + +bitReader_fillFast_1_end: + SHRXQ DI, R10, CX + MOVW (SI)(CX*2), CX + MOVB CH, AL + MOVBQZX CL, CX + ADDQ CX, R11 + SHLXQ CX, R10, R10 + SHRXQ DI, R10, CX + MOVW (SI)(CX*2), CX + MOVB CH, AH + MOVBQZX CL, CX + ADDQ CX, R11 + SHLXQ CX, R10, R10 + BSWAPL AX + CMPQ R11, $0x20 + JL bitReader_fillFast_2_end + SUBQ $0x20, R11 + SUBQ $0x04, R9 + MOVL (R8)(R9*1), CX + SHLXQ R11, CX, CX + ORQ CX, R10 + +bitReader_fillFast_2_end: + SHRXQ DI, R10, CX + MOVW (SI)(CX*2), CX + MOVB CH, AH + MOVBQZX CL, CX + ADDQ CX, R11 + SHLXQ CX, R10, R10 + SHRXQ DI, R10, CX + MOVW (SI)(CX*2), CX + MOVB CH, AL + MOVBQZX CL, CX + ADDQ CX, R11 + SHLXQ CX, R10, R10 + BSWAPL AX + + // Store the decoded values + MOVL AX, (DX) + ADDQ $0x04, DX + +loop_condition: + CMPQ R9, $0x08 + JGE main_loop + + // Update ctx structure + MOVQ ctx+0(FP), AX + SUBQ 16(AX), DX + MOVQ DX, 40(AX) + MOVQ (AX), AX + MOVQ R9, 24(AX) + MOVQ R10, 32(AX) + MOVB R11, 40(AX) + RET + + // Report error +error_max_decoded_size_exeeded: + MOVQ ctx+0(FP), AX + MOVQ $-1, CX + MOVQ CX, 40(AX) + RET diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_generic.go b/vendor/github.com/klauspost/compress/huff0/decompress_generic.go new file mode 100644 index 0000000..4f6f37c --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/decompress_generic.go @@ -0,0 +1,295 @@ +//go:build !amd64 || appengine || !gc || noasm +// +build !amd64 appengine !gc noasm + +// This file contains a generic implementation of Decoder.Decompress4X. +package huff0 + +import ( + "errors" + "fmt" +) + +// Decompress4X will decompress a 4X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// The *capacity* of the dst slice must match the destination size of +// the uncompressed data exactly. +func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) { + if len(d.dt.single) == 0 { + return nil, errors.New("no table loaded") + } + if len(src) < 6+(4*1) { + return nil, errors.New("input too small") + } + if use8BitTables && d.actualTableLog <= 8 { + return d.decompress4X8bit(dst, src) + } + + var br [4]bitReaderShifted + // Decode "jump table" + start := 6 + for i := 0; i < 3; i++ { + length := int(src[i*2]) | (int(src[i*2+1]) << 8) + if start+length >= len(src) { + return nil, errors.New("truncated input (or invalid offset)") + } + err := br[i].init(src[start : start+length]) + if err != nil { + return nil, err + } + start += length + } + err := br[3].init(src[start:]) + if err != nil { + return nil, err + } + + // destination, offset to match first output + dstSize := cap(dst) + dst = dst[:dstSize] + out := dst + dstEvery := (dstSize + 3) / 4 + + const tlSize = 1 << tableLogMax + const tlMask = tlSize - 1 + single := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + buf := d.buffer() + var off uint8 + var decoded int + + // Decode 2 values from each decoder/loop. + const bufoff = 256 + for { + if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 { + break + } + + { + const stream = 0 + const stream2 = 1 + br[stream].fillFast() + br[stream2].fillFast() + + val := br[stream].peekBitsFast(d.actualTableLog) + val2 := br[stream2].peekBitsFast(d.actualTableLog) + v := single[val&tlMask] + v2 := single[val2&tlMask] + br[stream].advance(uint8(v.entry)) + br[stream2].advance(uint8(v2.entry)) + buf[stream][off] = uint8(v.entry >> 8) + buf[stream2][off] = uint8(v2.entry >> 8) + + val = br[stream].peekBitsFast(d.actualTableLog) + val2 = br[stream2].peekBitsFast(d.actualTableLog) + v = single[val&tlMask] + v2 = single[val2&tlMask] + br[stream].advance(uint8(v.entry)) + br[stream2].advance(uint8(v2.entry)) + buf[stream][off+1] = uint8(v.entry >> 8) + buf[stream2][off+1] = uint8(v2.entry >> 8) + } + + { + const stream = 2 + const stream2 = 3 + br[stream].fillFast() + br[stream2].fillFast() + + val := br[stream].peekBitsFast(d.actualTableLog) + val2 := br[stream2].peekBitsFast(d.actualTableLog) + v := single[val&tlMask] + v2 := single[val2&tlMask] + br[stream].advance(uint8(v.entry)) + br[stream2].advance(uint8(v2.entry)) + buf[stream][off] = uint8(v.entry >> 8) + buf[stream2][off] = uint8(v2.entry >> 8) + + val = br[stream].peekBitsFast(d.actualTableLog) + val2 = br[stream2].peekBitsFast(d.actualTableLog) + v = single[val&tlMask] + v2 = single[val2&tlMask] + br[stream].advance(uint8(v.entry)) + br[stream2].advance(uint8(v2.entry)) + buf[stream][off+1] = uint8(v.entry >> 8) + buf[stream2][off+1] = uint8(v2.entry >> 8) + } + + off += 2 + + if off == 0 { + if bufoff > dstEvery { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 1") + } + copy(out, buf[0][:]) + copy(out[dstEvery:], buf[1][:]) + copy(out[dstEvery*2:], buf[2][:]) + copy(out[dstEvery*3:], buf[3][:]) + out = out[bufoff:] + decoded += bufoff * 4 + // There must at least be 3 buffers left. + if len(out) < dstEvery*3 { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 2") + } + } + } + if off > 0 { + ioff := int(off) + if len(out) < dstEvery*3+ioff { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 3") + } + copy(out, buf[0][:off]) + copy(out[dstEvery:], buf[1][:off]) + copy(out[dstEvery*2:], buf[2][:off]) + copy(out[dstEvery*3:], buf[3][:off]) + decoded += int(off) * 4 + out = out[off:] + } + + // Decode remaining. + remainBytes := dstEvery - (decoded / 4) + for i := range br { + offset := dstEvery * i + endsAt := offset + remainBytes + if endsAt > len(out) { + endsAt = len(out) + } + br := &br[i] + bitsLeft := br.remaining() + for bitsLeft > 0 { + br.fill() + if offset >= endsAt { + d.bufs.Put(buf) + return nil, errors.New("corruption detected: stream overrun 4") + } + + // Read value and increment offset. + val := br.peekBitsFast(d.actualTableLog) + v := single[val&tlMask].entry + nBits := uint8(v) + br.advance(nBits) + bitsLeft -= uint(nBits) + out[offset] = uint8(v >> 8) + offset++ + } + if offset != endsAt { + d.bufs.Put(buf) + return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt) + } + decoded += offset - dstEvery*i + err = br.close() + if err != nil { + return nil, err + } + } + d.bufs.Put(buf) + if dstSize != decoded { + return nil, errors.New("corruption detected: short output block") + } + return dst, nil +} + +// Decompress1X will decompress a 1X encoded stream. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) { + if len(d.dt.single) == 0 { + return nil, errors.New("no table loaded") + } + if use8BitTables && d.actualTableLog <= 8 { + return d.decompress1X8Bit(dst, src) + } + var br bitReaderShifted + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:0] + + // Avoid bounds check by always having full sized table. + const tlSize = 1 << tableLogMax + const tlMask = tlSize - 1 + dt := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + bufs := d.buffer() + buf := &bufs[0] + var off uint8 + + for br.off >= 8 { + br.fillFast() + v := dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + // Refill + br.fillFast() + + v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + br.close() + d.bufs.Put(bufs) + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + + if len(dst)+int(off) > maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:off]...) + + // br < 8, so uint8 is fine + bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead + for bitsLeft > 0 { + br.fill() + if false && br.bitsRead >= 32 { + if br.off >= 4 { + v := br.in[br.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + br.value = (br.value << 32) | uint64(low) + br.bitsRead -= 32 + br.off -= 4 + } else { + for br.off > 0 { + br.value = (br.value << 8) | uint64(br.in[br.off-1]) + br.bitsRead -= 8 + br.off-- + } + } + } + if len(dst) >= maxDecodedSize { + d.bufs.Put(bufs) + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= nBits + dst = append(dst, uint8(v.entry>>8)) + } + d.bufs.Put(bufs) + return dst, br.close() +} diff --git a/vendor/github.com/klauspost/compress/huff0/huff0.go b/vendor/github.com/klauspost/compress/huff0/huff0.go new file mode 100644 index 0000000..e8ad17a --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/huff0.go @@ -0,0 +1,337 @@ +// Package huff0 provides fast huffman encoding as used in zstd. +// +// See README.md at https://github.com/klauspost/compress/tree/master/huff0 for details. +package huff0 + +import ( + "errors" + "fmt" + "math" + "math/bits" + "sync" + + "github.com/klauspost/compress/fse" +) + +const ( + maxSymbolValue = 255 + + // zstandard limits tablelog to 11, see: + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#huffman-tree-description + tableLogMax = 11 + tableLogDefault = 11 + minTablelog = 5 + huffNodesLen = 512 + + // BlockSizeMax is maximum input size for a single block uncompressed. + BlockSizeMax = 1<<18 - 1 +) + +var ( + // ErrIncompressible is returned when input is judged to be too hard to compress. + ErrIncompressible = errors.New("input is not compressible") + + // ErrUseRLE is returned from the compressor when the input is a single byte value repeated. + ErrUseRLE = errors.New("input is single value repeated") + + // ErrTooBig is return if input is too large for a single block. + ErrTooBig = errors.New("input too big") + + // ErrMaxDecodedSizeExceeded is return if input is too large for a single block. + ErrMaxDecodedSizeExceeded = errors.New("maximum output size exceeded") +) + +type ReusePolicy uint8 + +const ( + // ReusePolicyAllow will allow reuse if it produces smaller output. + ReusePolicyAllow ReusePolicy = iota + + // ReusePolicyPrefer will re-use aggressively if possible. + // This will not check if a new table will produce smaller output, + // except if the current table is impossible to use or + // compressed output is bigger than input. + ReusePolicyPrefer + + // ReusePolicyNone will disable re-use of tables. + // This is slightly faster than ReusePolicyAllow but may produce larger output. + ReusePolicyNone + + // ReusePolicyMust must allow reuse and produce smaller output. + ReusePolicyMust +) + +type Scratch struct { + count [maxSymbolValue + 1]uint32 + + // Per block parameters. + // These can be used to override compression parameters of the block. + // Do not touch, unless you know what you are doing. + + // Out is output buffer. + // If the scratch is re-used before the caller is done processing the output, + // set this field to nil. + // Otherwise the output buffer will be re-used for next Compression/Decompression step + // and allocation will be avoided. + Out []byte + + // OutTable will contain the table data only, if a new table has been generated. + // Slice of the returned data. + OutTable []byte + + // OutData will contain the compressed data. + // Slice of the returned data. + OutData []byte + + // MaxDecodedSize will set the maximum allowed output size. + // This value will automatically be set to BlockSizeMax if not set. + // Decoders will return ErrMaxDecodedSizeExceeded is this limit is exceeded. + MaxDecodedSize int + + br byteReader + + // MaxSymbolValue will override the maximum symbol value of the next block. + MaxSymbolValue uint8 + + // TableLog will attempt to override the tablelog for the next block. + // Must be <= 11 and >= 5. + TableLog uint8 + + // Reuse will specify the reuse policy + Reuse ReusePolicy + + // WantLogLess allows to specify a log 2 reduction that should at least be achieved, + // otherwise the block will be returned as incompressible. + // The reduction should then at least be (input size >> WantLogLess) + // If WantLogLess == 0 any improvement will do. + WantLogLess uint8 + + symbolLen uint16 // Length of active part of the symbol table. + maxCount int // count of the most probable symbol + clearCount bool // clear count + actualTableLog uint8 // Selected tablelog. + prevTableLog uint8 // Tablelog for previous table + prevTable cTable // Table used for previous compression. + cTable cTable // compression table + dt dTable // decompression table + nodes []nodeElt + tmpOut [4][]byte + fse *fse.Scratch + decPool sync.Pool // *[4][256]byte buffers. + huffWeight [maxSymbolValue + 1]byte +} + +// TransferCTable will transfer the previously used compression table. +func (s *Scratch) TransferCTable(src *Scratch) { + if cap(s.prevTable) < len(src.prevTable) { + s.prevTable = make(cTable, 0, maxSymbolValue+1) + } + s.prevTable = s.prevTable[:len(src.prevTable)] + copy(s.prevTable, src.prevTable) + s.prevTableLog = src.prevTableLog +} + +func (s *Scratch) prepare(in []byte) (*Scratch, error) { + if len(in) > BlockSizeMax { + return nil, ErrTooBig + } + if s == nil { + s = &Scratch{} + } + if s.MaxSymbolValue == 0 { + s.MaxSymbolValue = maxSymbolValue + } + if s.TableLog == 0 { + s.TableLog = tableLogDefault + } + if s.TableLog > tableLogMax || s.TableLog < minTablelog { + return nil, fmt.Errorf(" invalid tableLog %d (%d -> %d)", s.TableLog, minTablelog, tableLogMax) + } + if s.MaxDecodedSize <= 0 || s.MaxDecodedSize > BlockSizeMax { + s.MaxDecodedSize = BlockSizeMax + } + if s.clearCount && s.maxCount == 0 { + for i := range s.count { + s.count[i] = 0 + } + s.clearCount = false + } + if cap(s.Out) == 0 { + s.Out = make([]byte, 0, len(in)) + } + s.Out = s.Out[:0] + + s.OutTable = nil + s.OutData = nil + if cap(s.nodes) < huffNodesLen+1 { + s.nodes = make([]nodeElt, 0, huffNodesLen+1) + } + s.nodes = s.nodes[:0] + if s.fse == nil { + s.fse = &fse.Scratch{} + } + s.br.init(in) + + return s, nil +} + +type cTable []cTableEntry + +func (c cTable) write(s *Scratch) error { + var ( + // precomputed conversion table + bitsToWeight [tableLogMax + 1]byte + huffLog = s.actualTableLog + // last weight is not saved. + maxSymbolValue = uint8(s.symbolLen - 1) + huffWeight = s.huffWeight[:256] + ) + const ( + maxFSETableLog = 6 + ) + // convert to weight + bitsToWeight[0] = 0 + for n := uint8(1); n < huffLog+1; n++ { + bitsToWeight[n] = huffLog + 1 - n + } + + // Acquire histogram for FSE. + hist := s.fse.Histogram() + hist = hist[:256] + for i := range hist[:16] { + hist[i] = 0 + } + for n := uint8(0); n < maxSymbolValue; n++ { + v := bitsToWeight[c[n].nBits] & 15 + huffWeight[n] = v + hist[v]++ + } + + // FSE compress if feasible. + if maxSymbolValue >= 2 { + huffMaxCnt := uint32(0) + huffMax := uint8(0) + for i, v := range hist[:16] { + if v == 0 { + continue + } + huffMax = byte(i) + if v > huffMaxCnt { + huffMaxCnt = v + } + } + s.fse.HistogramFinished(huffMax, int(huffMaxCnt)) + s.fse.TableLog = maxFSETableLog + b, err := fse.Compress(huffWeight[:maxSymbolValue], s.fse) + if err == nil && len(b) < int(s.symbolLen>>1) { + s.Out = append(s.Out, uint8(len(b))) + s.Out = append(s.Out, b...) + return nil + } + // Unable to compress (RLE/uncompressible) + } + // write raw values as 4-bits (max : 15) + if maxSymbolValue > (256 - 128) { + // should not happen : likely means source cannot be compressed + return ErrIncompressible + } + op := s.Out + // special case, pack weights 4 bits/weight. + op = append(op, 128|(maxSymbolValue-1)) + // be sure it doesn't cause msan issue in final combination + huffWeight[maxSymbolValue] = 0 + for n := uint16(0); n < uint16(maxSymbolValue); n += 2 { + op = append(op, (huffWeight[n]<<4)|huffWeight[n+1]) + } + s.Out = op + return nil +} + +func (c cTable) estTableSize(s *Scratch) (sz int, err error) { + var ( + // precomputed conversion table + bitsToWeight [tableLogMax + 1]byte + huffLog = s.actualTableLog + // last weight is not saved. + maxSymbolValue = uint8(s.symbolLen - 1) + huffWeight = s.huffWeight[:256] + ) + const ( + maxFSETableLog = 6 + ) + // convert to weight + bitsToWeight[0] = 0 + for n := uint8(1); n < huffLog+1; n++ { + bitsToWeight[n] = huffLog + 1 - n + } + + // Acquire histogram for FSE. + hist := s.fse.Histogram() + hist = hist[:256] + for i := range hist[:16] { + hist[i] = 0 + } + for n := uint8(0); n < maxSymbolValue; n++ { + v := bitsToWeight[c[n].nBits] & 15 + huffWeight[n] = v + hist[v]++ + } + + // FSE compress if feasible. + if maxSymbolValue >= 2 { + huffMaxCnt := uint32(0) + huffMax := uint8(0) + for i, v := range hist[:16] { + if v == 0 { + continue + } + huffMax = byte(i) + if v > huffMaxCnt { + huffMaxCnt = v + } + } + s.fse.HistogramFinished(huffMax, int(huffMaxCnt)) + s.fse.TableLog = maxFSETableLog + b, err := fse.Compress(huffWeight[:maxSymbolValue], s.fse) + if err == nil && len(b) < int(s.symbolLen>>1) { + sz += 1 + len(b) + return sz, nil + } + // Unable to compress (RLE/uncompressible) + } + // write raw values as 4-bits (max : 15) + if maxSymbolValue > (256 - 128) { + // should not happen : likely means source cannot be compressed + return 0, ErrIncompressible + } + // special case, pack weights 4 bits/weight. + sz += 1 + int(maxSymbolValue/2) + return sz, nil +} + +// estimateSize returns the estimated size in bytes of the input represented in the +// histogram supplied. +func (c cTable) estimateSize(hist []uint32) int { + nbBits := uint32(7) + for i, v := range c[:len(hist)] { + nbBits += uint32(v.nBits) * hist[i] + } + return int(nbBits >> 3) +} + +// minSize returns the minimum possible size considering the shannon limit. +func (s *Scratch) minSize(total int) int { + nbBits := float64(7) + fTotal := float64(total) + for _, v := range s.count[:s.symbolLen] { + n := float64(v) + if n > 0 { + nbBits += math.Log2(fTotal/n) * n + } + } + return int(nbBits) >> 3 +} + +func highBit32(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} diff --git a/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go new file mode 100644 index 0000000..3954c51 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go @@ -0,0 +1,34 @@ +// Package cpuinfo gives runtime info about the current CPU. +// +// This is a very limited module meant for use internally +// in this project. For more versatile solution check +// https://github.com/klauspost/cpuid. +package cpuinfo + +// HasBMI1 checks whether an x86 CPU supports the BMI1 extension. +func HasBMI1() bool { + return hasBMI1 +} + +// HasBMI2 checks whether an x86 CPU supports the BMI2 extension. +func HasBMI2() bool { + return hasBMI2 +} + +// DisableBMI2 will disable BMI2, for testing purposes. +// Call returned function to restore previous state. +func DisableBMI2() func() { + old := hasBMI2 + hasBMI2 = false + return func() { + hasBMI2 = old + } +} + +// HasBMI checks whether an x86 CPU supports both BMI1 and BMI2 extensions. +func HasBMI() bool { + return HasBMI1() && HasBMI2() +} + +var hasBMI1 bool +var hasBMI2 bool diff --git a/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go new file mode 100644 index 0000000..e802579 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go @@ -0,0 +1,11 @@ +//go:build amd64 && !appengine && !noasm && gc +// +build amd64,!appengine,!noasm,gc + +package cpuinfo + +// go:noescape +func x86extensions() (bmi1, bmi2 bool) + +func init() { + hasBMI1, hasBMI2 = x86extensions() +} diff --git a/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s new file mode 100644 index 0000000..4465fbe --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s @@ -0,0 +1,36 @@ +// +build !appengine +// +build gc +// +build !noasm + +#include "textflag.h" +#include "funcdata.h" +#include "go_asm.h" + +TEXT ·x86extensions(SB), NOSPLIT, $0 + // 1. determine max EAX value + XORQ AX, AX + CPUID + + CMPQ AX, $7 + JB unsupported + + // 2. EAX = 7, ECX = 0 --- see Table 3-8 "Information Returned by CPUID Instruction" + MOVQ $7, AX + MOVQ $0, CX + CPUID + + BTQ $3, BX // bit 3 = BMI1 + SETCS AL + + BTQ $8, BX // bit 8 = BMI2 + SETCS AH + + MOVB AL, bmi1+0(FP) + MOVB AH, bmi2+1(FP) + RET + +unsupported: + XORQ AX, AX + MOVB AL, bmi1+0(FP) + MOVB AL, bmi2+1(FP) + RET diff --git a/vendor/github.com/klauspost/compress/internal/snapref/LICENSE b/vendor/github.com/klauspost/compress/internal/snapref/LICENSE new file mode 100644 index 0000000..6050c10 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/snapref/LICENSE @@ -0,0 +1,27 @@ +Copyright (c) 2011 The Snappy-Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/vendor/github.com/klauspost/compress/internal/snapref/decode.go b/vendor/github.com/klauspost/compress/internal/snapref/decode.go new file mode 100644 index 0000000..40796a4 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/snapref/decode.go @@ -0,0 +1,264 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package snapref + +import ( + "encoding/binary" + "errors" + "io" +) + +var ( + // ErrCorrupt reports that the input is invalid. + ErrCorrupt = errors.New("snappy: corrupt input") + // ErrTooLarge reports that the uncompressed length is too large. + ErrTooLarge = errors.New("snappy: decoded block is too large") + // ErrUnsupported reports that the input isn't supported. + ErrUnsupported = errors.New("snappy: unsupported input") + + errUnsupportedLiteralLength = errors.New("snappy: unsupported literal length") +) + +// DecodedLen returns the length of the decoded block. +func DecodedLen(src []byte) (int, error) { + v, _, err := decodedLen(src) + return v, err +} + +// decodedLen returns the length of the decoded block and the number of bytes +// that the length header occupied. +func decodedLen(src []byte) (blockLen, headerLen int, err error) { + v, n := binary.Uvarint(src) + if n <= 0 || v > 0xffffffff { + return 0, 0, ErrCorrupt + } + + const wordSize = 32 << (^uint(0) >> 32 & 1) + if wordSize == 32 && v > 0x7fffffff { + return 0, 0, ErrTooLarge + } + return int(v), n, nil +} + +const ( + decodeErrCodeCorrupt = 1 + decodeErrCodeUnsupportedLiteralLength = 2 +) + +// Decode returns the decoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire decoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// Decode handles the Snappy block format, not the Snappy stream format. +func Decode(dst, src []byte) ([]byte, error) { + dLen, s, err := decodedLen(src) + if err != nil { + return nil, err + } + if dLen <= len(dst) { + dst = dst[:dLen] + } else { + dst = make([]byte, dLen) + } + switch decode(dst, src[s:]) { + case 0: + return dst, nil + case decodeErrCodeUnsupportedLiteralLength: + return nil, errUnsupportedLiteralLength + } + return nil, ErrCorrupt +} + +// NewReader returns a new Reader that decompresses from r, using the framing +// format described at +// https://github.com/google/snappy/blob/master/framing_format.txt +func NewReader(r io.Reader) *Reader { + return &Reader{ + r: r, + decoded: make([]byte, maxBlockSize), + buf: make([]byte, maxEncodedLenOfMaxBlockSize+checksumSize), + } +} + +// Reader is an io.Reader that can read Snappy-compressed bytes. +// +// Reader handles the Snappy stream format, not the Snappy block format. +type Reader struct { + r io.Reader + err error + decoded []byte + buf []byte + // decoded[i:j] contains decoded bytes that have not yet been passed on. + i, j int + readHeader bool +} + +// Reset discards any buffered data, resets all state, and switches the Snappy +// reader to read from r. This permits reusing a Reader rather than allocating +// a new one. +func (r *Reader) Reset(reader io.Reader) { + r.r = reader + r.err = nil + r.i = 0 + r.j = 0 + r.readHeader = false +} + +func (r *Reader) readFull(p []byte, allowEOF bool) (ok bool) { + if _, r.err = io.ReadFull(r.r, p); r.err != nil { + if r.err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) { + r.err = ErrCorrupt + } + return false + } + return true +} + +func (r *Reader) fill() error { + for r.i >= r.j { + if !r.readFull(r.buf[:4], true) { + return r.err + } + chunkType := r.buf[0] + if !r.readHeader { + if chunkType != chunkTypeStreamIdentifier { + r.err = ErrCorrupt + return r.err + } + r.readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + if chunkLen > len(r.buf) { + r.err = ErrUnsupported + return r.err + } + + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + // Section 4.2. Compressed data (chunk type 0x00). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[checksumSize:] + + n, err := DecodedLen(buf) + if err != nil { + r.err = err + return r.err + } + if n > len(r.decoded) { + r.err = ErrCorrupt + return r.err + } + if _, err := Decode(r.decoded, buf); err != nil { + r.err = err + return r.err + } + if crc(r.decoded[:n]) != checksum { + r.err = ErrCorrupt + return r.err + } + r.i, r.j = 0, n + continue + + case chunkTypeUncompressedData: + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return r.err + } + buf := r.buf[:checksumSize] + if !r.readFull(buf, false) { + return r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + // Read directly into r.decoded instead of via r.buf. + n := chunkLen - checksumSize + if n > len(r.decoded) { + r.err = ErrCorrupt + return r.err + } + if !r.readFull(r.decoded[:n], false) { + return r.err + } + if crc(r.decoded[:n]) != checksum { + r.err = ErrCorrupt + return r.err + } + r.i, r.j = 0, n + continue + + case chunkTypeStreamIdentifier: + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(magicBody) { + r.err = ErrCorrupt + return r.err + } + if !r.readFull(r.buf[:len(magicBody)], false) { + return r.err + } + for i := 0; i < len(magicBody); i++ { + if r.buf[i] != magicBody[i] { + r.err = ErrCorrupt + return r.err + } + } + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + r.err = ErrUnsupported + return r.err + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + if !r.readFull(r.buf[:chunkLen], false) { + return r.err + } + } + + return nil +} + +// Read satisfies the io.Reader interface. +func (r *Reader) Read(p []byte) (int, error) { + if r.err != nil { + return 0, r.err + } + + if err := r.fill(); err != nil { + return 0, err + } + + n := copy(p, r.decoded[r.i:r.j]) + r.i += n + return n, nil +} + +// ReadByte satisfies the io.ByteReader interface. +func (r *Reader) ReadByte() (byte, error) { + if r.err != nil { + return 0, r.err + } + + if err := r.fill(); err != nil { + return 0, err + } + + c := r.decoded[r.i] + r.i++ + return c, nil +} diff --git a/vendor/github.com/klauspost/compress/internal/snapref/decode_other.go b/vendor/github.com/klauspost/compress/internal/snapref/decode_other.go new file mode 100644 index 0000000..77395a6 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/snapref/decode_other.go @@ -0,0 +1,113 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package snapref + +// decode writes the decoding of src to dst. It assumes that the varint-encoded +// length of the decompressed bytes has already been read, and that len(dst) +// equals that length. +// +// It returns 0 on success or a decodeErrCodeXxx error code on failure. +func decode(dst, src []byte) int { + var d, s, offset, length int + for s < len(src) { + switch src[s] & 0x03 { + case tagLiteral: + x := uint32(src[s] >> 2) + switch { + case x < 60: + s++ + case x == 60: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-1]) + case x == 61: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-2]) | uint32(src[s-1])<<8 + case x == 62: + s += 4 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + case x == 63: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + length = int(x) + 1 + if length <= 0 { + return decodeErrCodeUnsupportedLiteralLength + } + if length > len(dst)-d || length > len(src)-s { + return decodeErrCodeCorrupt + } + copy(dst[d:], src[s:s+length]) + d += length + s += length + continue + + case tagCopy1: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = 4 + int(src[s-2])>>2&0x7 + offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + + case tagCopy2: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = 1 + int(src[s-3])>>2 + offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + + case tagCopy4: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = 1 + int(src[s-5])>>2 + offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + } + + if offset <= 0 || d < offset || length > len(dst)-d { + return decodeErrCodeCorrupt + } + // Copy from an earlier sub-slice of dst to a later sub-slice. + // If no overlap, use the built-in copy: + if offset >= length { + copy(dst[d:d+length], dst[d-offset:]) + d += length + continue + } + + // Unlike the built-in copy function, this byte-by-byte copy always runs + // forwards, even if the slices overlap. Conceptually, this is: + // + // d += forwardCopy(dst[d:d+length], dst[d-offset:]) + // + // We align the slices into a and b and show the compiler they are the same size. + // This allows the loop to run without bounds checks. + a := dst[d : d+length] + b := dst[d-offset:] + b = b[:len(a)] + for i := range a { + a[i] = b[i] + } + d += length + } + if d != len(dst) { + return decodeErrCodeCorrupt + } + return 0 +} diff --git a/vendor/github.com/klauspost/compress/internal/snapref/encode.go b/vendor/github.com/klauspost/compress/internal/snapref/encode.go new file mode 100644 index 0000000..13c6040 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/snapref/encode.go @@ -0,0 +1,289 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package snapref + +import ( + "encoding/binary" + "errors" + "io" +) + +// Encode returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// Encode handles the Snappy block format, not the Snappy stream format. +func Encode(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if len(dst) < n { + dst = make([]byte, n) + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + for len(src) > 0 { + p := src + src = nil + if len(p) > maxBlockSize { + p, src = p[:maxBlockSize], p[maxBlockSize:] + } + if len(p) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], p) + } else { + d += encodeBlock(dst[d:], p) + } + } + return dst[:d] +} + +// inputMargin is the minimum number of extra input bytes to keep, inside +// encodeBlock's inner loop. On some architectures, this margin lets us +// implement a fast path for emitLiteral, where the copy of short (<= 16 byte) +// literals can be implemented as a single load to and store from a 16-byte +// register. That literal's actual length can be as short as 1 byte, so this +// can copy up to 15 bytes too much, but that's OK as subsequent iterations of +// the encoding loop will fix up the copy overrun, and this inputMargin ensures +// that we don't overrun the dst and src buffers. +const inputMargin = 16 - 1 + +// minNonLiteralBlockSize is the minimum size of the input to encodeBlock that +// could be encoded with a copy tag. This is the minimum with respect to the +// algorithm used by encodeBlock, not a minimum enforced by the file format. +// +// The encoded output must start with at least a 1 byte literal, as there are +// no previous bytes to copy. A minimal (1 byte) copy after that, generated +// from an emitCopy call in encodeBlock's main loop, would require at least +// another inputMargin bytes, for the reason above: we want any emitLiteral +// calls inside encodeBlock's main loop to use the fast path if possible, which +// requires being able to overrun by inputMargin bytes. Thus, +// minNonLiteralBlockSize equals 1 + 1 + inputMargin. +// +// The C++ code doesn't use this exact threshold, but it could, as discussed at +// https://groups.google.com/d/topic/snappy-compression/oGbhsdIJSJ8/discussion +// The difference between Go (2+inputMargin) and C++ (inputMargin) is purely an +// optimization. It should not affect the encoded form. This is tested by +// TestSameEncodingAsCppShortCopies. +const minNonLiteralBlockSize = 1 + 1 + inputMargin + +// MaxEncodedLen returns the maximum length of a snappy block, given its +// uncompressed length. +// +// It will return a negative value if srcLen is too large to encode. +func MaxEncodedLen(srcLen int) int { + n := uint64(srcLen) + if n > 0xffffffff { + return -1 + } + // Compressed data can be defined as: + // compressed := item* literal* + // item := literal* copy + // + // The trailing literal sequence has a space blowup of at most 62/60 + // since a literal of length 60 needs one tag byte + one extra byte + // for length information. + // + // Item blowup is trickier to measure. Suppose the "copy" op copies + // 4 bytes of data. Because of a special check in the encoding code, + // we produce a 4-byte copy only if the offset is < 65536. Therefore + // the copy op takes 3 bytes to encode, and this type of item leads + // to at most the 62/60 blowup for representing literals. + // + // Suppose the "copy" op copies 5 bytes of data. If the offset is big + // enough, it will take 5 bytes to encode the copy op. Therefore the + // worst case here is a one-byte literal followed by a five-byte copy. + // That is, 6 bytes of input turn into 7 bytes of "compressed" data. + // + // This last factor dominates the blowup, so the final estimate is: + n = 32 + n + n/6 + if n > 0xffffffff { + return -1 + } + return int(n) +} + +var errClosed = errors.New("snappy: Writer is closed") + +// NewWriter returns a new Writer that compresses to w. +// +// The Writer returned does not buffer writes. There is no need to Flush or +// Close such a Writer. +// +// Deprecated: the Writer returned is not suitable for many small writes, only +// for few large writes. Use NewBufferedWriter instead, which is efficient +// regardless of the frequency and shape of the writes, and remember to Close +// that Writer when done. +func NewWriter(w io.Writer) *Writer { + return &Writer{ + w: w, + obuf: make([]byte, obufLen), + } +} + +// NewBufferedWriter returns a new Writer that compresses to w, using the +// framing format described at +// https://github.com/google/snappy/blob/master/framing_format.txt +// +// The Writer returned buffers writes. Users must call Close to guarantee all +// data has been forwarded to the underlying io.Writer. They may also call +// Flush zero or more times before calling Close. +func NewBufferedWriter(w io.Writer) *Writer { + return &Writer{ + w: w, + ibuf: make([]byte, 0, maxBlockSize), + obuf: make([]byte, obufLen), + } +} + +// Writer is an io.Writer that can write Snappy-compressed bytes. +// +// Writer handles the Snappy stream format, not the Snappy block format. +type Writer struct { + w io.Writer + err error + + // ibuf is a buffer for the incoming (uncompressed) bytes. + // + // Its use is optional. For backwards compatibility, Writers created by the + // NewWriter function have ibuf == nil, do not buffer incoming bytes, and + // therefore do not need to be Flush'ed or Close'd. + ibuf []byte + + // obuf is a buffer for the outgoing (compressed) bytes. + obuf []byte + + // wroteStreamHeader is whether we have written the stream header. + wroteStreamHeader bool +} + +// Reset discards the writer's state and switches the Snappy writer to write to +// w. This permits reusing a Writer rather than allocating a new one. +func (w *Writer) Reset(writer io.Writer) { + w.w = writer + w.err = nil + if w.ibuf != nil { + w.ibuf = w.ibuf[:0] + } + w.wroteStreamHeader = false +} + +// Write satisfies the io.Writer interface. +func (w *Writer) Write(p []byte) (nRet int, errRet error) { + if w.ibuf == nil { + // Do not buffer incoming bytes. This does not perform or compress well + // if the caller of Writer.Write writes many small slices. This + // behavior is therefore deprecated, but still supported for backwards + // compatibility with code that doesn't explicitly Flush or Close. + return w.write(p) + } + + // The remainder of this method is based on bufio.Writer.Write from the + // standard library. + + for len(p) > (cap(w.ibuf)-len(w.ibuf)) && w.err == nil { + var n int + if len(w.ibuf) == 0 { + // Large write, empty buffer. + // Write directly from p to avoid copy. + n, _ = w.write(p) + } else { + n = copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p) + w.ibuf = w.ibuf[:len(w.ibuf)+n] + w.Flush() + } + nRet += n + p = p[n:] + } + if w.err != nil { + return nRet, w.err + } + n := copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p) + w.ibuf = w.ibuf[:len(w.ibuf)+n] + nRet += n + return nRet, nil +} + +func (w *Writer) write(p []byte) (nRet int, errRet error) { + if w.err != nil { + return 0, w.err + } + for len(p) > 0 { + obufStart := len(magicChunk) + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + copy(w.obuf, magicChunk) + obufStart = 0 + } + + var uncompressed []byte + if len(p) > maxBlockSize { + uncompressed, p = p[:maxBlockSize], p[maxBlockSize:] + } else { + uncompressed, p = p, nil + } + checksum := crc(uncompressed) + + // Compress the buffer, discarding the result if the improvement + // isn't at least 12.5%. + compressed := Encode(w.obuf[obufHeaderLen:], uncompressed) + chunkType := uint8(chunkTypeCompressedData) + chunkLen := 4 + len(compressed) + obufEnd := obufHeaderLen + len(compressed) + if len(compressed) >= len(uncompressed)-len(uncompressed)/8 { + chunkType = chunkTypeUncompressedData + chunkLen = 4 + len(uncompressed) + obufEnd = obufHeaderLen + } + + // Fill in the per-chunk header that comes before the body. + w.obuf[len(magicChunk)+0] = chunkType + w.obuf[len(magicChunk)+1] = uint8(chunkLen >> 0) + w.obuf[len(magicChunk)+2] = uint8(chunkLen >> 8) + w.obuf[len(magicChunk)+3] = uint8(chunkLen >> 16) + w.obuf[len(magicChunk)+4] = uint8(checksum >> 0) + w.obuf[len(magicChunk)+5] = uint8(checksum >> 8) + w.obuf[len(magicChunk)+6] = uint8(checksum >> 16) + w.obuf[len(magicChunk)+7] = uint8(checksum >> 24) + + if _, err := w.w.Write(w.obuf[obufStart:obufEnd]); err != nil { + w.err = err + return nRet, err + } + if chunkType == chunkTypeUncompressedData { + if _, err := w.w.Write(uncompressed); err != nil { + w.err = err + return nRet, err + } + } + nRet += len(uncompressed) + } + return nRet, nil +} + +// Flush flushes the Writer to its underlying io.Writer. +func (w *Writer) Flush() error { + if w.err != nil { + return w.err + } + if len(w.ibuf) == 0 { + return nil + } + w.write(w.ibuf) + w.ibuf = w.ibuf[:0] + return w.err +} + +// Close calls Flush and then closes the Writer. +func (w *Writer) Close() error { + w.Flush() + ret := w.err + if w.err == nil { + w.err = errClosed + } + return ret +} diff --git a/vendor/github.com/klauspost/compress/internal/snapref/encode_other.go b/vendor/github.com/klauspost/compress/internal/snapref/encode_other.go new file mode 100644 index 0000000..511bba6 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/snapref/encode_other.go @@ -0,0 +1,236 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package snapref + +func load32(b []byte, i int) uint32 { + b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line. + return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 +} + +func load64(b []byte, i int) uint64 { + b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line. + return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 | + uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56 +} + +// emitLiteral writes a literal chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 1 <= len(lit) && len(lit) <= 65536 +func emitLiteral(dst, lit []byte) int { + i, n := 0, uint(len(lit)-1) + switch { + case n < 60: + dst[0] = uint8(n)<<2 | tagLiteral + i = 1 + case n < 1<<8: + dst[0] = 60<<2 | tagLiteral + dst[1] = uint8(n) + i = 2 + default: + dst[0] = 61<<2 | tagLiteral + dst[1] = uint8(n) + dst[2] = uint8(n >> 8) + i = 3 + } + return i + copy(dst[i:], lit) +} + +// emitCopy writes a copy chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 1 <= offset && offset <= 65535 +// 4 <= length && length <= 65535 +func emitCopy(dst []byte, offset, length int) int { + i := 0 + // The maximum length for a single tagCopy1 or tagCopy2 op is 64 bytes. The + // threshold for this loop is a little higher (at 68 = 64 + 4), and the + // length emitted down below is is a little lower (at 60 = 64 - 4), because + // it's shorter to encode a length 67 copy as a length 60 tagCopy2 followed + // by a length 7 tagCopy1 (which encodes as 3+2 bytes) than to encode it as + // a length 64 tagCopy2 followed by a length 3 tagCopy2 (which encodes as + // 3+3 bytes). The magic 4 in the 64±4 is because the minimum length for a + // tagCopy1 op is 4 bytes, which is why a length 3 copy has to be an + // encodes-as-3-bytes tagCopy2 instead of an encodes-as-2-bytes tagCopy1. + for length >= 68 { + // Emit a length 64 copy, encoded as 3 bytes. + dst[i+0] = 63<<2 | tagCopy2 + dst[i+1] = uint8(offset) + dst[i+2] = uint8(offset >> 8) + i += 3 + length -= 64 + } + if length > 64 { + // Emit a length 60 copy, encoded as 3 bytes. + dst[i+0] = 59<<2 | tagCopy2 + dst[i+1] = uint8(offset) + dst[i+2] = uint8(offset >> 8) + i += 3 + length -= 60 + } + if length >= 12 || offset >= 2048 { + // Emit the remaining copy, encoded as 3 bytes. + dst[i+0] = uint8(length-1)<<2 | tagCopy2 + dst[i+1] = uint8(offset) + dst[i+2] = uint8(offset >> 8) + return i + 3 + } + // Emit the remaining copy, encoded as 2 bytes. + dst[i+0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1 + dst[i+1] = uint8(offset) + return i + 2 +} + +// extendMatch returns the largest k such that k <= len(src) and that +// src[i:i+k-j] and src[j:k] have the same contents. +// +// It assumes that: +// 0 <= i && i < j && j <= len(src) +func extendMatch(src []byte, i, j int) int { + for ; j < len(src) && src[i] == src[j]; i, j = i+1, j+1 { + } + return j +} + +func hash(u, shift uint32) uint32 { + return (u * 0x1e35a7bd) >> shift +} + +// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlock(dst, src []byte) (d int) { + // Initialize the hash table. Its size ranges from 1<<8 to 1<<14 inclusive. + // The table element type is uint16, as s < sLimit and sLimit < len(src) + // and len(src) <= maxBlockSize and maxBlockSize == 65536. + const ( + maxTableSize = 1 << 14 + // tableMask is redundant, but helps the compiler eliminate bounds + // checks. + tableMask = maxTableSize - 1 + ) + shift := uint32(32 - 8) + for tableSize := 1 << 8; tableSize < maxTableSize && tableSize < len(src); tableSize *= 2 { + shift-- + } + // In Go, all array elements are zero-initialized, so there is no advantage + // to a smaller tableSize per se. However, it matches the C++ algorithm, + // and in the asm versions of this code, we can get away with zeroing only + // the first tableSize elements. + var table [maxTableSize]uint16 + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + nextHash := hash(load32(src, s), shift) + + for { + // Copied from the C++ snappy implementation: + // + // Heuristic match skipping: If 32 bytes are scanned with no matches + // found, start looking only at every other byte. If 32 more bytes are + // scanned (or skipped), look at every third byte, etc.. When a match + // is found, immediately go back to looking at every byte. This is a + // small loss (~5% performance, ~0.1% density) for compressible data + // due to more bookkeeping, but for non-compressible data (such as + // JPEG) it's a huge win since the compressor quickly "realizes" the + // data is incompressible and doesn't bother looking for matches + // everywhere. + // + // The "skip" variable keeps track of how many bytes there are since + // the last match; dividing it by 32 (ie. right-shifting by five) gives + // the number of bytes to move ahead for each iteration. + skip := 32 + + nextS := s + candidate := 0 + for { + s = nextS + bytesBetweenHashLookups := skip >> 5 + nextS = s + bytesBetweenHashLookups + skip += bytesBetweenHashLookups + if nextS > sLimit { + goto emitRemainder + } + candidate = int(table[nextHash&tableMask]) + table[nextHash&tableMask] = uint16(s) + nextHash = hash(load32(src, nextS), shift) + if load32(src, s) == load32(src, candidate) { + break + } + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + d += emitLiteral(dst[d:], src[nextEmit:s]) + + // Call emitCopy, and then see if another emitCopy could be our next + // move. Repeat until we find no match for the input immediately after + // what was consumed by the last emitCopy call. + // + // If we exit this loop normally then we need to call emitLiteral next, + // though we don't yet know how big the literal will be. We handle that + // by proceeding to the next iteration of the main loop. We also can + // exit this loop via goto if we get close to exhausting the input. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + base := s + + // Extend the 4-byte match as long as possible. + // + // This is an inlined version of: + // s = extendMatch(src, candidate+4, s+4) + s += 4 + for i := candidate + 4; s < len(src) && src[i] == src[s]; i, s = i+1, s+1 { + } + + d += emitCopy(dst[d:], base-candidate, s-base) + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. If + // another emitCopy is not our next move, also calculate nextHash + // at s+1. At least on GOARCH=amd64, these three hash calculations + // are faster as one load64 call (with some shifts) instead of + // three load32 calls. + x := load64(src, s-1) + prevHash := hash(uint32(x>>0), shift) + table[prevHash&tableMask] = uint16(s - 1) + currHash := hash(uint32(x>>8), shift) + candidate = int(table[currHash&tableMask]) + table[currHash&tableMask] = uint16(s) + if uint32(x>>8) != load32(src, candidate) { + nextHash = hash(uint32(x>>16), shift) + s++ + break + } + } + } + +emitRemainder: + if nextEmit < len(src) { + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} diff --git a/vendor/github.com/klauspost/compress/internal/snapref/snappy.go b/vendor/github.com/klauspost/compress/internal/snapref/snappy.go new file mode 100644 index 0000000..34d01f4 --- /dev/null +++ b/vendor/github.com/klauspost/compress/internal/snapref/snappy.go @@ -0,0 +1,98 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package snapref implements the Snappy compression format. It aims for very +// high speeds and reasonable compression. +// +// There are actually two Snappy formats: block and stream. They are related, +// but different: trying to decompress block-compressed data as a Snappy stream +// will fail, and vice versa. The block format is the Decode and Encode +// functions and the stream format is the Reader and Writer types. +// +// The block format, the more common case, is used when the complete size (the +// number of bytes) of the original data is known upfront, at the time +// compression starts. The stream format, also known as the framing format, is +// for when that isn't always true. +// +// The canonical, C++ implementation is at https://github.com/google/snappy and +// it only implements the block format. +package snapref + +import ( + "hash/crc32" +) + +/* +Each encoded block begins with the varint-encoded length of the decoded data, +followed by a sequence of chunks. Chunks begin and end on byte boundaries. The +first byte of each chunk is broken into its 2 least and 6 most significant bits +called l and m: l ranges in [0, 4) and m ranges in [0, 64). l is the chunk tag. +Zero means a literal tag. All other values mean a copy tag. + +For literal tags: + - If m < 60, the next 1 + m bytes are literal bytes. + - Otherwise, let n be the little-endian unsigned integer denoted by the next + m - 59 bytes. The next 1 + n bytes after that are literal bytes. + +For copy tags, length bytes are copied from offset bytes ago, in the style of +Lempel-Ziv compression algorithms. In particular: + - For l == 1, the offset ranges in [0, 1<<11) and the length in [4, 12). + The length is 4 + the low 3 bits of m. The high 3 bits of m form bits 8-10 + of the offset. The next byte is bits 0-7 of the offset. + - For l == 2, the offset ranges in [0, 1<<16) and the length in [1, 65). + The length is 1 + m. The offset is the little-endian unsigned integer + denoted by the next 2 bytes. + - For l == 3, this tag is a legacy format that is no longer issued by most + encoders. Nonetheless, the offset ranges in [0, 1<<32) and the length in + [1, 65). The length is 1 + m. The offset is the little-endian unsigned + integer denoted by the next 4 bytes. +*/ +const ( + tagLiteral = 0x00 + tagCopy1 = 0x01 + tagCopy2 = 0x02 + tagCopy4 = 0x03 +) + +const ( + checksumSize = 4 + chunkHeaderSize = 4 + magicChunk = "\xff\x06\x00\x00" + magicBody + magicBody = "sNaPpY" + + // maxBlockSize is the maximum size of the input to encodeBlock. It is not + // part of the wire format per se, but some parts of the encoder assume + // that an offset fits into a uint16. + // + // Also, for the framing format (Writer type instead of Encode function), + // https://github.com/google/snappy/blob/master/framing_format.txt says + // that "the uncompressed data in a chunk must be no longer than 65536 + // bytes". + maxBlockSize = 65536 + + // maxEncodedLenOfMaxBlockSize equals MaxEncodedLen(maxBlockSize), but is + // hard coded to be a const instead of a variable, so that obufLen can also + // be a const. Their equivalence is confirmed by + // TestMaxEncodedLenOfMaxBlockSize. + maxEncodedLenOfMaxBlockSize = 76490 + + obufHeaderLen = len(magicChunk) + checksumSize + chunkHeaderSize + obufLen = obufHeaderLen + maxEncodedLenOfMaxBlockSize +) + +const ( + chunkTypeCompressedData = 0x00 + chunkTypeUncompressedData = 0x01 + chunkTypePadding = 0xfe + chunkTypeStreamIdentifier = 0xff +) + +var crcTable = crc32.MakeTable(crc32.Castagnoli) + +// crc implements the checksum specified in section 3 of +// https://github.com/google/snappy/blob/master/framing_format.txt +func crc(b []byte) uint32 { + c := crc32.Update(0, crcTable, b) + return uint32(c>>15|c<<17) + 0xa282ead8 +} diff --git a/vendor/github.com/klauspost/compress/s2/.gitignore b/vendor/github.com/klauspost/compress/s2/.gitignore new file mode 100644 index 0000000..3a89c6e --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/.gitignore @@ -0,0 +1,15 @@ +testdata/bench + +# These explicitly listed benchmark data files are for an obsolete version of +# snappy_test.go. +testdata/alice29.txt +testdata/asyoulik.txt +testdata/fireworks.jpeg +testdata/geo.protodata +testdata/html +testdata/html_x_4 +testdata/kppkn.gtb +testdata/lcet10.txt +testdata/paper-100k.pdf +testdata/plrabn12.txt +testdata/urls.10K diff --git a/vendor/github.com/klauspost/compress/s2/LICENSE b/vendor/github.com/klauspost/compress/s2/LICENSE new file mode 100644 index 0000000..1d2d645 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/LICENSE @@ -0,0 +1,28 @@ +Copyright (c) 2011 The Snappy-Go Authors. All rights reserved. +Copyright (c) 2019 Klaus Post. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/vendor/github.com/klauspost/compress/s2/README.md b/vendor/github.com/klauspost/compress/s2/README.md new file mode 100644 index 0000000..73c0c46 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/README.md @@ -0,0 +1,965 @@ +# S2 Compression + +S2 is an extension of [Snappy](https://github.com/google/snappy). + +S2 is aimed for high throughput, which is why it features concurrent compression for bigger payloads. + +Decoding is compatible with Snappy compressed content, but content compressed with S2 cannot be decompressed by Snappy. +This means that S2 can seamlessly replace Snappy without converting compressed content. + +S2 can produce Snappy compatible output, faster and better than Snappy. +If you want full benefit of the changes you should use s2 without Snappy compatibility. + +S2 is designed to have high throughput on content that cannot be compressed. +This is important, so you don't have to worry about spending CPU cycles on already compressed data. + +## Benefits over Snappy + +* Better compression +* Adjustable compression (3 levels) +* Concurrent stream compression +* Faster decompression, even for Snappy compatible content +* Concurrent Snappy/S2 stream decompression +* Ability to quickly skip forward in compressed stream +* Random seeking with indexes +* Compatible with reading Snappy compressed content +* Smaller block size overhead on incompressible blocks +* Block concatenation +* Uncompressed stream mode +* Automatic stream size padding +* Snappy compatible block compression + +## Drawbacks over Snappy + +* Not optimized for 32 bit systems +* Streams use slightly more memory due to larger blocks and concurrency (configurable) + +# Usage + +Installation: `go get -u github.com/klauspost/compress/s2` + +Full package documentation: + +[![godoc][1]][2] + +[1]: https://godoc.org/github.com/klauspost/compress?status.svg +[2]: https://godoc.org/github.com/klauspost/compress/s2 + +## Compression + +```Go +func EncodeStream(src io.Reader, dst io.Writer) error { + enc := s2.NewWriter(dst) + _, err := io.Copy(enc, src) + if err != nil { + enc.Close() + return err + } + // Blocks until compression is done. + return enc.Close() +} +``` + +You should always call `enc.Close()`, otherwise you will leak resources and your encode will be incomplete. + +For the best throughput, you should attempt to reuse the `Writer` using the `Reset()` method. + +The Writer in S2 is always buffered, therefore `NewBufferedWriter` in Snappy can be replaced with `NewWriter` in S2. +It is possible to flush any buffered data using the `Flush()` method. +This will block until all data sent to the encoder has been written to the output. + +S2 also supports the `io.ReaderFrom` interface, which will consume all input from a reader. + +As a final method to compress data, if you have a single block of data you would like to have encoded as a stream, +a slightly more efficient method is to use the `EncodeBuffer` method. +This will take ownership of the buffer until the stream is closed. + +```Go +func EncodeStream(src []byte, dst io.Writer) error { + enc := s2.NewWriter(dst) + // The encoder owns the buffer until Flush or Close is called. + err := enc.EncodeBuffer(buf) + if err != nil { + enc.Close() + return err + } + // Blocks until compression is done. + return enc.Close() +} +``` + +Each call to `EncodeBuffer` will result in discrete blocks being created without buffering, +so it should only be used a single time per stream. +If you need to write several blocks, you should use the regular io.Writer interface. + + +## Decompression + +```Go +func DecodeStream(src io.Reader, dst io.Writer) error { + dec := s2.NewReader(src) + _, err := io.Copy(dst, dec) + return err +} +``` + +Similar to the Writer, a Reader can be reused using the `Reset` method. + +For the best possible throughput, there is a `EncodeBuffer(buf []byte)` function available. +However, it requires that the provided buffer isn't used after it is handed over to S2 and until the stream is flushed or closed. + +For smaller data blocks, there is also a non-streaming interface: `Encode()`, `EncodeBetter()` and `Decode()`. +Do however note that these functions (similar to Snappy) does not provide validation of data, +so data corruption may be undetected. Stream encoding provides CRC checks of data. + +It is possible to efficiently skip forward in a compressed stream using the `Skip()` method. +For big skips the decompressor is able to skip blocks without decompressing them. + +## Single Blocks + +Similar to Snappy S2 offers single block compression. +Blocks do not offer the same flexibility and safety as streams, +but may be preferable for very small payloads, less than 100K. + +Using a simple `dst := s2.Encode(nil, src)` will compress `src` and return the compressed result. +It is possible to provide a destination buffer. +If the buffer has a capacity of `s2.MaxEncodedLen(len(src))` it will be used. +If not a new will be allocated. + +Alternatively `EncodeBetter`/`EncodeBest` can also be used for better, but slightly slower compression. + +Similarly to decompress a block you can use `dst, err := s2.Decode(nil, src)`. +Again an optional destination buffer can be supplied. +The `s2.DecodedLen(src)` can be used to get the minimum capacity needed. +If that is not satisfied a new buffer will be allocated. + +Block function always operate on a single goroutine since it should only be used for small payloads. + +# Commandline tools + +Some very simply commandline tools are provided; `s2c` for compression and `s2d` for decompression. + +Binaries can be downloaded on the [Releases Page](https://github.com/klauspost/compress/releases). + +Installing then requires Go to be installed. To install them, use: + +`go install github.com/klauspost/compress/s2/cmd/s2c@latest && go install github.com/klauspost/compress/s2/cmd/s2d@latest` + +To build binaries to the current folder use: + +`go build github.com/klauspost/compress/s2/cmd/s2c && go build github.com/klauspost/compress/s2/cmd/s2d` + + +## s2c + +``` +Usage: s2c [options] file1 file2 + +Compresses all files supplied as input separately. +Output files are written as 'filename.ext.s2' or 'filename.ext.snappy'. +By default output files will be overwritten. +Use - as the only file name to read from stdin and write to stdout. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +File names beginning with 'http://' and 'https://' will be downloaded and compressed. +Only http response code 200 is accepted. + +Options: + -bench int + Run benchmark n times. No output will be written + -blocksize string + Max block size. Examples: 64K, 256K, 1M, 4M. Must be power of two and <= 4MB (default "4M") + -c Write all output to stdout. Multiple input files will be concatenated + -cpu int + Compress using this amount of threads (default 32) + -faster + Compress faster, but with a minor compression loss + -help + Display help + -index + Add seek index (default true) + -o string + Write output to another file. Single input file only + -pad string + Pad size to a multiple of this value, Examples: 500, 64K, 256K, 1M, 4M, etc (default "1") + -q Don't write any output to terminal, except errors + -rm + Delete source file(s) after successful compression + -safe + Do not overwrite output files + -slower + Compress more, but a lot slower + -snappy + Generate Snappy compatible output stream + -verify + Verify written files + +``` + +## s2d + +``` +Usage: s2d [options] file1 file2 + +Decompresses all files supplied as input. Input files must end with '.s2' or '.snappy'. +Output file names have the extension removed. By default output files will be overwritten. +Use - as the only file name to read from stdin and write to stdout. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +File names beginning with 'http://' and 'https://' will be downloaded and decompressed. +Extensions on downloaded files are ignored. Only http response code 200 is accepted. + +Options: + -bench int + Run benchmark n times. No output will be written + -c Write all output to stdout. Multiple input files will be concatenated + -help + Display help + -o string + Write output to another file. Single input file only + -offset string + Start at offset. Examples: 92, 64K, 256K, 1M, 4M. Requires Index + -q Don't write any output to terminal, except errors + -rm + Delete source file(s) after successful decompression + -safe + Do not overwrite output files + -tail string + Return last of compressed file. Examples: 92, 64K, 256K, 1M, 4M. Requires Index + -verify + Verify files, but do not write output +``` + +## s2sx: self-extracting archives + +s2sx allows creating self-extracting archives with no dependencies. + +By default, executables are created for the same platforms as the host os, +but this can be overridden with `-os` and `-arch` parameters. + +Extracted files have 0666 permissions, except when untar option used. + +``` +Usage: s2sx [options] file1 file2 + +Compresses all files supplied as input separately. +If files have '.s2' extension they are assumed to be compressed already. +Output files are written as 'filename.s2sx' and with '.exe' for windows targets. +If output is big, an additional file with ".more" is written. This must be included as well. +By default output files will be overwritten. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +Options: + -arch string + Destination architecture (default "amd64") + -c Write all output to stdout. Multiple input files will be concatenated + -cpu int + Compress using this amount of threads (default 32) + -help + Display help + -max string + Maximum executable size. Rest will be written to another file. (default "1G") + -os string + Destination operating system (default "windows") + -q Don't write any output to terminal, except errors + -rm + Delete source file(s) after successful compression + -safe + Do not overwrite output files + -untar + Untar on destination +``` + +Available platforms are: + + * darwin-amd64 + * darwin-arm64 + * linux-amd64 + * linux-arm + * linux-arm64 + * linux-mips64 + * linux-ppc64le + * windows-386 + * windows-amd64 + +By default, there is a size limit of 1GB for the output executable. + +When this is exceeded the remaining file content is written to a file called +output+`.more`. This file must be included for a successful extraction and +placed alongside the executable for a successful extraction. + +This file *must* have the same name as the executable, so if the executable is renamed, +so must the `.more` file. + +This functionality is disabled with stdin/stdout. + +### Self-extracting TAR files + +If you wrap a TAR file you can specify `-untar` to make it untar on the destination host. + +Files are extracted to the current folder with the path specified in the tar file. + +Note that tar files are not validated before they are wrapped. + +For security reasons files that move below the root folder are not allowed. + +# Performance + +This section will focus on comparisons to Snappy. +This package is solely aimed at replacing Snappy as a high speed compression package. +If you are mainly looking for better compression [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) +gives better compression, but typically at speeds slightly below "better" mode in this package. + +Compression is increased compared to Snappy, mostly around 5-20% and the throughput is typically 25-40% increased (single threaded) compared to the Snappy Go implementation. + +Streams are concurrently compressed. The stream will be distributed among all available CPU cores for the best possible throughput. + +A "better" compression mode is also available. This allows to trade a bit of speed for a minor compression gain. +The content compressed in this mode is fully compatible with the standard decoder. + +Snappy vs S2 **compression** speed on 16 core (32 thread) computer, using all threads and a single thread (1 CPU): + +| File | S2 speed | S2 Throughput | S2 % smaller | S2 "better" | "better" throughput | "better" % smaller | +|-----------------------------------------------------------------------------------------------------|----------|---------------|--------------|-------------|---------------------|--------------------| +| [rawstudio-mint14.tar](https://files.klauspost.com/compress/rawstudio-mint14.7z) | 12.70x | 10556 MB/s | 7.35% | 4.15x | 3455 MB/s | 12.79% | +| (1 CPU) | 1.14x | 948 MB/s | - | 0.42x | 349 MB/s | - | +| [github-june-2days-2019.json](https://files.klauspost.com/compress/github-june-2days-2019.json.zst) | 17.13x | 14484 MB/s | 31.60% | 10.09x | 8533 MB/s | 37.71% | +| (1 CPU) | 1.33x | 1127 MB/s | - | 0.70x | 589 MB/s | - | +| [github-ranks-backup.bin](https://files.klauspost.com/compress/github-ranks-backup.bin.zst) | 15.14x | 12000 MB/s | -5.79% | 6.59x | 5223 MB/s | 5.80% | +| (1 CPU) | 1.11x | 877 MB/s | - | 0.47x | 370 MB/s | - | +| [consensus.db.10gb](https://files.klauspost.com/compress/consensus.db.10gb.zst) | 14.62x | 12116 MB/s | 15.90% | 5.35x | 4430 MB/s | 16.08% | +| (1 CPU) | 1.38x | 1146 MB/s | - | 0.38x | 312 MB/s | - | +| [adresser.json](https://files.klauspost.com/compress/adresser.json.zst) | 8.83x | 17579 MB/s | 43.86% | 6.54x | 13011 MB/s | 47.23% | +| (1 CPU) | 1.14x | 2259 MB/s | - | 0.74x | 1475 MB/s | - | +| [gob-stream](https://files.klauspost.com/compress/gob-stream.7z) | 16.72x | 14019 MB/s | 24.02% | 10.11x | 8477 MB/s | 30.48% | +| (1 CPU) | 1.24x | 1043 MB/s | - | 0.70x | 586 MB/s | - | +| [10gb.tar](http://mattmahoney.net/dc/10gb.html) | 13.33x | 9254 MB/s | 1.84% | 6.75x | 4686 MB/s | 6.72% | +| (1 CPU) | 0.97x | 672 MB/s | - | 0.53x | 366 MB/s | - | +| sharnd.out.2gb | 2.11x | 12639 MB/s | 0.01% | 1.98x | 11833 MB/s | 0.01% | +| (1 CPU) | 0.93x | 5594 MB/s | - | 1.34x | 8030 MB/s | - | +| [enwik9](http://mattmahoney.net/dc/textdata.html) | 19.34x | 8220 MB/s | 3.98% | 7.87x | 3345 MB/s | 15.82% | +| (1 CPU) | 1.06x | 452 MB/s | - | 0.50x | 213 MB/s | - | +| [silesia.tar](http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip) | 10.48x | 6124 MB/s | 5.67% | 3.76x | 2197 MB/s | 12.60% | +| (1 CPU) | 0.97x | 568 MB/s | - | 0.46x | 271 MB/s | - | +| [enwik10](https://encode.su/threads/3315-enwik10-benchmark-results) | 21.07x | 9020 MB/s | 6.36% | 6.91x | 2959 MB/s | 16.95% | +| (1 CPU) | 1.07x | 460 MB/s | - | 0.51x | 220 MB/s | - | + +### Legend + +* `S2 speed`: Speed of S2 compared to Snappy, using 16 cores and 1 core. +* `S2 throughput`: Throughput of S2 in MB/s. +* `S2 % smaller`: How many percent of the Snappy output size is S2 better. +* `S2 "better"`: Speed when enabling "better" compression mode in S2 compared to Snappy. +* `"better" throughput`: Speed when enabling "better" compression mode in S2 compared to Snappy. +* `"better" % smaller`: How many percent of the Snappy output size is S2 better when using "better" compression. + +There is a good speedup across the board when using a single thread and a significant speedup when using multiple threads. + +Machine generated data gets by far the biggest compression boost, with size being being reduced by up to 45% of Snappy size. + +The "better" compression mode sees a good improvement in all cases, but usually at a performance cost. + +Incompressible content (`sharnd.out.2gb`, 2GB random data) sees the smallest speedup. +This is likely dominated by synchronization overhead, which is confirmed by the fact that single threaded performance is higher (see above). + +## Decompression + +S2 attempts to create content that is also fast to decompress, except in "better" mode where the smallest representation is used. + +S2 vs Snappy **decompression** speed. Both operating on single core: + +| File | S2 Throughput | vs. Snappy | Better Throughput | vs. Snappy | +|-----------------------------------------------------------------------------------------------------|---------------|------------|-------------------|------------| +| [rawstudio-mint14.tar](https://files.klauspost.com/compress/rawstudio-mint14.7z) | 2117 MB/s | 1.14x | 1738 MB/s | 0.94x | +| [github-june-2days-2019.json](https://files.klauspost.com/compress/github-june-2days-2019.json.zst) | 2401 MB/s | 1.25x | 2307 MB/s | 1.20x | +| [github-ranks-backup.bin](https://files.klauspost.com/compress/github-ranks-backup.bin.zst) | 2075 MB/s | 0.98x | 1764 MB/s | 0.83x | +| [consensus.db.10gb](https://files.klauspost.com/compress/consensus.db.10gb.zst) | 2967 MB/s | 1.05x | 2885 MB/s | 1.02x | +| [adresser.json](https://files.klauspost.com/compress/adresser.json.zst) | 4141 MB/s | 1.07x | 4184 MB/s | 1.08x | +| [gob-stream](https://files.klauspost.com/compress/gob-stream.7z) | 2264 MB/s | 1.12x | 2185 MB/s | 1.08x | +| [10gb.tar](http://mattmahoney.net/dc/10gb.html) | 1525 MB/s | 1.03x | 1347 MB/s | 0.91x | +| sharnd.out.2gb | 3813 MB/s | 0.79x | 3900 MB/s | 0.81x | +| [enwik9](http://mattmahoney.net/dc/textdata.html) | 1246 MB/s | 1.29x | 967 MB/s | 1.00x | +| [silesia.tar](http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip) | 1433 MB/s | 1.12x | 1203 MB/s | 0.94x | +| [enwik10](https://encode.su/threads/3315-enwik10-benchmark-results) | 1284 MB/s | 1.32x | 1010 MB/s | 1.04x | + +### Legend + +* `S2 Throughput`: Decompression speed of S2 encoded content. +* `Better Throughput`: Decompression speed of S2 "better" encoded content. +* `vs Snappy`: Decompression speed of S2 "better" mode compared to Snappy and absolute speed. + + +While the decompression code hasn't changed, there is a significant speedup in decompression speed. +S2 prefers longer matches and will typically only find matches that are 6 bytes or longer. +While this reduces compression a bit, it improves decompression speed. + +The "better" compression mode will actively look for shorter matches, which is why it has a decompression speed quite similar to Snappy. + +Without assembly decompression is also very fast; single goroutine decompression speed. No assembly: + +| File | S2 Throughput | S2 throughput | +|--------------------------------|--------------|---------------| +| consensus.db.10gb.s2 | 1.84x | 2289.8 MB/s | +| 10gb.tar.s2 | 1.30x | 867.07 MB/s | +| rawstudio-mint14.tar.s2 | 1.66x | 1329.65 MB/s | +| github-june-2days-2019.json.s2 | 2.36x | 1831.59 MB/s | +| github-ranks-backup.bin.s2 | 1.73x | 1390.7 MB/s | +| enwik9.s2 | 1.67x | 681.53 MB/s | +| adresser.json.s2 | 3.41x | 4230.53 MB/s | +| silesia.tar.s2 | 1.52x | 811.58 | + +Even though S2 typically compresses better than Snappy, decompression speed is always better. + +### Concurrent Stream Decompression + +For full stream decompression S2 offers a [DecodeConcurrent](https://pkg.go.dev/github.com/klauspost/compress/s2#Reader.DecodeConcurrent) +that will decode a full stream using multiple goroutines. + +Example scaling, AMD Ryzen 3950X, 16 cores, decompression using `s2d -bench=3 `, best of 3: + +| Input | `-cpu=1` | `-cpu=2` | `-cpu=4` | `-cpu=8` | `-cpu=16` | +|-------------------------------------------|------------|------------|------------|------------|-------------| +| enwik10.snappy | 1098.6MB/s | 1819.8MB/s | 3625.6MB/s | 6910.6MB/s | 10818.2MB/s | +| enwik10.s2 | 1303.5MB/s | 2606.1MB/s | 4847.9MB/s | 8878.4MB/s | 9592.1MB/s | +| sofia-air-quality-dataset.tar.snappy | 1302.0MB/s | 2165.0MB/s | 4244.5MB/s | 8241.0MB/s | 12920.5MB/s | +| sofia-air-quality-dataset.tar.s2 | 1399.2MB/s | 2463.2MB/s | 5196.5MB/s | 9639.8MB/s | 11439.5MB/s | +| sofia-air-quality-dataset.tar.s2 (no asm) | 837.5MB/s | 1652.6MB/s | 3183.6MB/s | 5945.0MB/s | 9620.7MB/s | + +Scaling can be expected to be pretty linear until memory bandwidth is saturated. + +For now the DecodeConcurrent can only be used for full streams without seeking or combining with regular reads. + +## Block compression + + +When compressing blocks no concurrent compression is performed just as Snappy. +This is because blocks are for smaller payloads and generally will not benefit from concurrent compression. + +An important change is that incompressible blocks will not be more than at most 10 bytes bigger than the input. +In rare, worst case scenario Snappy blocks could be significantly bigger than the input. + +### Mixed content blocks + +The most reliable is a wide dataset. +For this we use [`webdevdata.org-2015-01-07-subset`](https://files.klauspost.com/compress/webdevdata.org-2015-01-07-4GB-subset.7z), +53927 files, total input size: 4,014,735,833 bytes. Single goroutine used. + +| * | Input | Output | Reduction | MB/s | +|-------------------|------------|------------|-----------|--------| +| S2 | 4014735833 | 1059723369 | 73.60% | **934.34** | +| S2 Better | 4014735833 | 969670507 | 75.85% | 532.70 | +| S2 Best | 4014735833 | 906625668 | **77.85%** | 46.84 | +| Snappy | 4014735833 | 1128706759 | 71.89% | 762.59 | +| S2, Snappy Output | 4014735833 | 1093821420 | 72.75% | 908.60 | +| LZ4 | 4014735833 | 1079259294 | 73.12% | 526.94 | + +S2 delivers both the best single threaded throughput with regular mode and the best compression rate with "best". +"Better" mode provides the same compression speed as LZ4 with better compression ratio. + +When outputting Snappy compatible output it still delivers better throughput (150MB/s more) and better compression. + +As can be seen from the other benchmarks decompression should also be easier on the S2 generated output. + +Though they cannot be compared due to different decompression speeds here are the speed/size comparisons for +other Go compressors: + +| * | Input | Output | Reduction | MB/s | +|-------------------|------------|------------|-----------|--------| +| Zstd Fastest (Go) | 4014735833 | 794608518 | 80.21% | 236.04 | +| Zstd Best (Go) | 4014735833 | 704603356 | 82.45% | 35.63 | +| Deflate (Go) l1 | 4014735833 | 871294239 | 78.30% | 214.04 | +| Deflate (Go) l9 | 4014735833 | 730389060 | 81.81% | 41.17 | + +### Standard block compression + +Benchmarking single block performance is subject to a lot more variation since it only tests a limited number of file patterns. +So individual benchmarks should only be seen as a guideline and the overall picture is more important. + +These micro-benchmarks are with data in cache and trained branch predictors. For a more realistic benchmark see the mixed content above. + +Block compression. Parallel benchmark running on 16 cores, 16 goroutines. + +AMD64 assembly is use for both S2 and Snappy. + +| Absolute Perf | Snappy size | S2 Size | Snappy Speed | S2 Speed | Snappy dec | S2 dec | +|-----------------------|-------------|---------|--------------|-------------|-------------|-------------| +| html | 22843 | 21111 | 16246 MB/s | 17438 MB/s | 40972 MB/s | 49263 MB/s | +| urls.10K | 335492 | 287326 | 7943 MB/s | 9693 MB/s | 22523 MB/s | 26484 MB/s | +| fireworks.jpeg | 123034 | 123100 | 349544 MB/s | 273889 MB/s | 718321 MB/s | 827552 MB/s | +| fireworks.jpeg (200B) | 146 | 155 | 8869 MB/s | 17773 MB/s | 33691 MB/s | 52421 MB/s | +| paper-100k.pdf | 85304 | 84459 | 167546 MB/s | 101263 MB/s | 326905 MB/s | 291944 MB/s | +| html_x_4 | 92234 | 21113 | 15194 MB/s | 50670 MB/s | 30843 MB/s | 32217 MB/s | +| alice29.txt | 88034 | 85975 | 5936 MB/s | 6139 MB/s | 12882 MB/s | 20044 MB/s | +| asyoulik.txt | 77503 | 79650 | 5517 MB/s | 6366 MB/s | 12735 MB/s | 22806 MB/s | +| lcet10.txt | 234661 | 220670 | 6235 MB/s | 6067 MB/s | 14519 MB/s | 18697 MB/s | +| plrabn12.txt | 319267 | 317985 | 5159 MB/s | 5726 MB/s | 11923 MB/s | 19901 MB/s | +| geo.protodata | 23335 | 18690 | 21220 MB/s | 26529 MB/s | 56271 MB/s | 62540 MB/s | +| kppkn.gtb | 69526 | 65312 | 9732 MB/s | 8559 MB/s | 18491 MB/s | 18969 MB/s | +| alice29.txt (128B) | 80 | 82 | 6691 MB/s | 15489 MB/s | 31883 MB/s | 38874 MB/s | +| alice29.txt (1000B) | 774 | 774 | 12204 MB/s | 13000 MB/s | 48056 MB/s | 52341 MB/s | +| alice29.txt (10000B) | 6648 | 6933 | 10044 MB/s | 12806 MB/s | 32378 MB/s | 46322 MB/s | +| alice29.txt (20000B) | 12686 | 13574 | 7733 MB/s | 11210 MB/s | 30566 MB/s | 58969 MB/s | + + +| Relative Perf | Snappy size | S2 size improved | S2 Speed | S2 Dec Speed | +|-----------------------|-------------|------------------|----------|--------------| +| html | 22.31% | 7.58% | 1.07x | 1.20x | +| urls.10K | 47.78% | 14.36% | 1.22x | 1.18x | +| fireworks.jpeg | 99.95% | -0.05% | 0.78x | 1.15x | +| fireworks.jpeg (200B) | 73.00% | -6.16% | 2.00x | 1.56x | +| paper-100k.pdf | 83.30% | 0.99% | 0.60x | 0.89x | +| html_x_4 | 22.52% | 77.11% | 3.33x | 1.04x | +| alice29.txt | 57.88% | 2.34% | 1.03x | 1.56x | +| asyoulik.txt | 61.91% | -2.77% | 1.15x | 1.79x | +| lcet10.txt | 54.99% | 5.96% | 0.97x | 1.29x | +| plrabn12.txt | 66.26% | 0.40% | 1.11x | 1.67x | +| geo.protodata | 19.68% | 19.91% | 1.25x | 1.11x | +| kppkn.gtb | 37.72% | 6.06% | 0.88x | 1.03x | +| alice29.txt (128B) | 62.50% | -2.50% | 2.31x | 1.22x | +| alice29.txt (1000B) | 77.40% | 0.00% | 1.07x | 1.09x | +| alice29.txt (10000B) | 66.48% | -4.29% | 1.27x | 1.43x | +| alice29.txt (20000B) | 63.43% | -7.00% | 1.45x | 1.93x | + +Speed is generally at or above Snappy. Small blocks gets a significant speedup, although at the expense of size. + +Decompression speed is better than Snappy, except in one case. + +Since payloads are very small the variance in terms of size is rather big, so they should only be seen as a general guideline. + +Size is on average around Snappy, but varies on content type. +In cases where compression is worse, it usually is compensated by a speed boost. + + +### Better compression + +Benchmarking single block performance is subject to a lot more variation since it only tests a limited number of file patterns. +So individual benchmarks should only be seen as a guideline and the overall picture is more important. + +| Absolute Perf | Snappy size | Better Size | Snappy Speed | Better Speed | Snappy dec | Better dec | +|-----------------------|-------------|-------------|--------------|--------------|-------------|-------------| +| html | 22843 | 19833 | 16246 MB/s | 7731 MB/s | 40972 MB/s | 40292 MB/s | +| urls.10K | 335492 | 253529 | 7943 MB/s | 3980 MB/s | 22523 MB/s | 20981 MB/s | +| fireworks.jpeg | 123034 | 123100 | 349544 MB/s | 9760 MB/s | 718321 MB/s | 823698 MB/s | +| fireworks.jpeg (200B) | 146 | 142 | 8869 MB/s | 594 MB/s | 33691 MB/s | 30101 MB/s | +| paper-100k.pdf | 85304 | 82915 | 167546 MB/s | 7470 MB/s | 326905 MB/s | 198869 MB/s | +| html_x_4 | 92234 | 19841 | 15194 MB/s | 23403 MB/s | 30843 MB/s | 30937 MB/s | +| alice29.txt | 88034 | 73218 | 5936 MB/s | 2945 MB/s | 12882 MB/s | 16611 MB/s | +| asyoulik.txt | 77503 | 66844 | 5517 MB/s | 2739 MB/s | 12735 MB/s | 14975 MB/s | +| lcet10.txt | 234661 | 190589 | 6235 MB/s | 3099 MB/s | 14519 MB/s | 16634 MB/s | +| plrabn12.txt | 319267 | 270828 | 5159 MB/s | 2600 MB/s | 11923 MB/s | 13382 MB/s | +| geo.protodata | 23335 | 18278 | 21220 MB/s | 11208 MB/s | 56271 MB/s | 57961 MB/s | +| kppkn.gtb | 69526 | 61851 | 9732 MB/s | 4556 MB/s | 18491 MB/s | 16524 MB/s | +| alice29.txt (128B) | 80 | 81 | 6691 MB/s | 529 MB/s | 31883 MB/s | 34225 MB/s | +| alice29.txt (1000B) | 774 | 748 | 12204 MB/s | 1943 MB/s | 48056 MB/s | 42068 MB/s | +| alice29.txt (10000B) | 6648 | 6234 | 10044 MB/s | 2949 MB/s | 32378 MB/s | 28813 MB/s | +| alice29.txt (20000B) | 12686 | 11584 | 7733 MB/s | 2822 MB/s | 30566 MB/s | 27315 MB/s | + + +| Relative Perf | Snappy size | Better size | Better Speed | Better dec | +|-----------------------|-------------|-------------|--------------|------------| +| html | 22.31% | 13.18% | 0.48x | 0.98x | +| urls.10K | 47.78% | 24.43% | 0.50x | 0.93x | +| fireworks.jpeg | 99.95% | -0.05% | 0.03x | 1.15x | +| fireworks.jpeg (200B) | 73.00% | 2.74% | 0.07x | 0.89x | +| paper-100k.pdf | 83.30% | 2.80% | 0.07x | 0.61x | +| html_x_4 | 22.52% | 78.49% | 0.04x | 1.00x | +| alice29.txt | 57.88% | 16.83% | 1.54x | 1.29x | +| asyoulik.txt | 61.91% | 13.75% | 0.50x | 1.18x | +| lcet10.txt | 54.99% | 18.78% | 0.50x | 1.15x | +| plrabn12.txt | 66.26% | 15.17% | 0.50x | 1.12x | +| geo.protodata | 19.68% | 21.67% | 0.50x | 1.03x | +| kppkn.gtb | 37.72% | 11.04% | 0.53x | 0.89x | +| alice29.txt (128B) | 62.50% | -1.25% | 0.47x | 1.07x | +| alice29.txt (1000B) | 77.40% | 3.36% | 0.08x | 0.88x | +| alice29.txt (10000B) | 66.48% | 6.23% | 0.16x | 0.89x | +| alice29.txt (20000B) | 63.43% | 8.69% | 0.29x | 0.89x | + +Except for the mostly incompressible JPEG image compression is better and usually in the +double digits in terms of percentage reduction over Snappy. + +The PDF sample shows a significant slowdown compared to Snappy, as this mode tries harder +to compress the data. Very small blocks are also not favorable for better compression, so throughput is way down. + +This mode aims to provide better compression at the expense of performance and achieves that +without a huge performance penalty, except on very small blocks. + +Decompression speed suffers a little compared to the regular S2 mode, +but still manages to be close to Snappy in spite of increased compression. + +# Best compression mode + +S2 offers a "best" compression mode. + +This will compress as much as possible with little regard to CPU usage. + +Mainly for offline compression, but where decompression speed should still +be high and compatible with other S2 compressed data. + +Some examples compared on 16 core CPU, amd64 assembly used: + +``` +* enwik10 +Default... 10000000000 -> 4761467548 [47.61%]; 1.098s, 8685.6MB/s +Better... 10000000000 -> 4219438251 [42.19%]; 1.925s, 4954.2MB/s +Best... 10000000000 -> 3627364337 [36.27%]; 43.051s, 221.5MB/s + +* github-june-2days-2019.json +Default... 6273951764 -> 1043196283 [16.63%]; 431ms, 13882.3MB/s +Better... 6273951764 -> 949146808 [15.13%]; 547ms, 10938.4MB/s +Best... 6273951764 -> 832855506 [13.27%]; 9.455s, 632.8MB/s + +* nyc-taxi-data-10M.csv +Default... 3325605752 -> 1095998837 [32.96%]; 324ms, 9788.7MB/s +Better... 3325605752 -> 954776589 [28.71%]; 491ms, 6459.4MB/s +Best... 3325605752 -> 779098746 [23.43%]; 8.29s, 382.6MB/s + +* 10gb.tar +Default... 10065157632 -> 5916578242 [58.78%]; 1.028s, 9337.4MB/s +Better... 10065157632 -> 5649207485 [56.13%]; 1.597s, 6010.6MB/s +Best... 10065157632 -> 5208719802 [51.75%]; 32.78s, 292.8MB/ + +* consensus.db.10gb +Default... 10737418240 -> 4562648848 [42.49%]; 882ms, 11610.0MB/s +Better... 10737418240 -> 4542428129 [42.30%]; 1.533s, 6679.7MB/s +Best... 10737418240 -> 4244773384 [39.53%]; 42.96s, 238.4MB/s +``` + +Decompression speed should be around the same as using the 'better' compression mode. + +# Snappy Compatibility + +S2 now offers full compatibility with Snappy. + +This means that the efficient encoders of S2 can be used to generate fully Snappy compatible output. + +There is a [snappy](https://github.com/klauspost/compress/tree/master/snappy) package that can be used by +simply changing imports from `github.com/golang/snappy` to `github.com/klauspost/compress/snappy`. +This uses "better" mode for all operations. +If you would like more control, you can use the s2 package as described below: + +## Blocks + +Snappy compatible blocks can be generated with the S2 encoder. +Compression and speed is typically a bit better `MaxEncodedLen` is also smaller for smaller memory usage. Replace + +| Snappy | S2 replacement | +|----------------------------|-------------------------| +| snappy.Encode(...) | s2.EncodeSnappy(...) | +| snappy.MaxEncodedLen(...) | s2.MaxEncodedLen(...) | + +`s2.EncodeSnappy` can be replaced with `s2.EncodeSnappyBetter` or `s2.EncodeSnappyBest` to get more efficiently compressed snappy compatible output. + +`s2.ConcatBlocks` is compatible with snappy blocks. + +Comparison of [`webdevdata.org-2015-01-07-subset`](https://files.klauspost.com/compress/webdevdata.org-2015-01-07-4GB-subset.7z), +53927 files, total input size: 4,014,735,833 bytes. amd64, single goroutine used: + +| Encoder | Size | MB/s | Reduction | +|-----------------------|------------|------------|------------ +| snappy.Encode | 1128706759 | 725.59 | 71.89% | +| s2.EncodeSnappy | 1093823291 | **899.16** | 72.75% | +| s2.EncodeSnappyBetter | 1001158548 | 578.49 | 75.06% | +| s2.EncodeSnappyBest | 944507998 | 66.00 | **76.47%**| + +## Streams + +For streams, replace `enc = snappy.NewBufferedWriter(w)` with `enc = s2.NewWriter(w, s2.WriterSnappyCompat())`. +All other options are available, but note that block size limit is different for snappy. + +Comparison of different streams, AMD Ryzen 3950x, 16 cores. Size and throughput: + +| File | snappy.NewWriter | S2 Snappy | S2 Snappy, Better | S2 Snappy, Best | +|-----------------------------|--------------------------|---------------------------|--------------------------|-------------------------| +| nyc-taxi-data-10M.csv | 1316042016 - 539.47MB/s | 1307003093 - 10132.73MB/s | 1174534014 - 5002.44MB/s | 1115904679 - 177.97MB/s | +| enwik10 (xml) | 5088294643 - 451.13MB/s | 5175840939 - 9440.69MB/s | 4560784526 - 4487.21MB/s | 4340299103 - 158.92MB/s | +| 10gb.tar (mixed) | 6056946612 - 729.73MB/s | 6208571995 - 9978.05MB/s | 5741646126 - 4919.98MB/s | 5548973895 - 180.44MB/s | +| github-june-2days-2019.json | 1525176492 - 933.00MB/s | 1476519054 - 13150.12MB/s | 1400547532 - 5803.40MB/s | 1321887137 - 204.29MB/s | +| consensus.db.10gb (db) | 5412897703 - 1102.14MB/s | 5354073487 - 13562.91MB/s | 5335069899 - 5294.73MB/s | 5201000954 - 175.72MB/s | + +# Decompression + +All decompression functions map directly to equivalent s2 functions. + +| Snappy | S2 replacement | +|------------------------|--------------------| +| snappy.Decode(...) | s2.Decode(...) | +| snappy.DecodedLen(...) | s2.DecodedLen(...) | +| snappy.NewReader(...) | s2.NewReader(...) | + +Features like [quick forward skipping without decompression](https://pkg.go.dev/github.com/klauspost/compress/s2#Reader.Skip) +are also available for Snappy streams. + +If you know you are only decompressing snappy streams, setting [`ReaderMaxBlockSize(64<<10)`](https://pkg.go.dev/github.com/klauspost/compress/s2#ReaderMaxBlockSize) +on your Reader will reduce memory consumption. + +# Concatenating blocks and streams. + +Concatenating streams will concatenate the output of both without recompressing them. +While this is inefficient in terms of compression it might be usable in certain scenarios. +The 10 byte 'stream identifier' of the second stream can optionally be stripped, but it is not a requirement. + +Blocks can be concatenated using the `ConcatBlocks` function. + +Snappy blocks/streams can safely be concatenated with S2 blocks and streams. +Streams with indexes (see below) will currently not work on concatenated streams. + +# Stream Seek Index + +S2 and Snappy streams can have indexes. These indexes will allow random seeking within the compressed data. + +The index can either be appended to the stream as a skippable block or returned for separate storage. + +When the index is appended to a stream it will be skipped by regular decoders, +so the output remains compatible with other decoders. + +## Creating an Index + +To automatically add an index to a stream, add `WriterAddIndex()` option to your writer. +Then the index will be added to the stream when `Close()` is called. + +``` + // Add Index to stream... + enc := s2.NewWriter(w, s2.WriterAddIndex()) + io.Copy(enc, r) + enc.Close() +``` + +If you want to store the index separately, you can use `CloseIndex()` instead of the regular `Close()`. +This will return the index. Note that `CloseIndex()` should only be called once, and you shouldn't call `Close()`. + +``` + // Get index for separate storage... + enc := s2.NewWriter(w) + io.Copy(enc, r) + index, err := enc.CloseIndex() +``` + +The `index` can then be used needing to read from the stream. +This means the index can be used without needing to seek to the end of the stream +or for manually forwarding streams. See below. + +Finally, an existing S2/Snappy stream can be indexed using the `s2.IndexStream(r io.Reader)` function. + +## Using Indexes + +To use indexes there is a `ReadSeeker(random bool, index []byte) (*ReadSeeker, error)` function available. + +Calling ReadSeeker will return an [io.ReadSeeker](https://pkg.go.dev/io#ReadSeeker) compatible version of the reader. + +If 'random' is specified the returned io.Seeker can be used for random seeking, otherwise only forward seeking is supported. +Enabling random seeking requires the original input to support the [io.Seeker](https://pkg.go.dev/io#Seeker) interface. + +``` + dec := s2.NewReader(r) + rs, err := dec.ReadSeeker(false, nil) + rs.Seek(wantOffset, io.SeekStart) +``` + +Get a seeker to seek forward. Since no index is provided, the index is read from the stream. +This requires that an index was added and that `r` supports the [io.Seeker](https://pkg.go.dev/io#Seeker) interface. + +A custom index can be specified which will be used if supplied. +When using a custom index, it will not be read from the input stream. + +``` + dec := s2.NewReader(r) + rs, err := dec.ReadSeeker(false, index) + rs.Seek(wantOffset, io.SeekStart) +``` + +This will read the index from `index`. Since we specify non-random (forward only) seeking `r` does not have to be an io.Seeker + +``` + dec := s2.NewReader(r) + rs, err := dec.ReadSeeker(true, index) + rs.Seek(wantOffset, io.SeekStart) +``` + +Finally, since we specify that we want to do random seeking `r` must be an io.Seeker. + +The returned [ReadSeeker](https://pkg.go.dev/github.com/klauspost/compress/s2#ReadSeeker) contains a shallow reference to the existing Reader, +meaning changes performed to one is reflected in the other. + +To check if a stream contains an index at the end, the `(*Index).LoadStream(rs io.ReadSeeker) error` can be used. + +## Manually Forwarding Streams + +Indexes can also be read outside the decoder using the [Index](https://pkg.go.dev/github.com/klauspost/compress/s2#Index) type. +This can be used for parsing indexes, either separate or in streams. + +In some cases it may not be possible to serve a seekable stream. +This can for instance be an HTTP stream, where the Range request +is sent at the start of the stream. + +With a little bit of extra code it is still possible to use indexes +to forward to specific offset with a single forward skip. + +It is possible to load the index manually like this: +``` + var index s2.Index + _, err = index.Load(idxBytes) +``` + +This can be used to figure out how much to offset the compressed stream: + +``` + compressedOffset, uncompressedOffset, err := index.Find(wantOffset) +``` + +The `compressedOffset` is the number of bytes that should be skipped +from the beginning of the compressed file. + +The `uncompressedOffset` will then be offset of the uncompressed bytes returned +when decoding from that position. This will always be <= wantOffset. + +When creating a decoder it must be specified that it should *not* expect a stream identifier +at the beginning of the stream. Assuming the io.Reader `r` has been forwarded to `compressedOffset` +we create the decoder like this: + +``` + dec := s2.NewReader(r, s2.ReaderIgnoreStreamIdentifier()) +``` + +We are not completely done. We still need to forward the stream the uncompressed bytes we didn't want. +This is done using the regular "Skip" function: + +``` + err = dec.Skip(wantOffset - uncompressedOffset) +``` + +This will ensure that we are at exactly the offset we want, and reading from `dec` will start at the requested offset. + +## Index Format: + +Each block is structured as a snappy skippable block, with the chunk ID 0x99. + +The block can be read from the front, but contains information so it can be read from the back as well. + +Numbers are stored as fixed size little endian values or [zigzag encoded](https://developers.google.com/protocol-buffers/docs/encoding#signed_integers) [base 128 varints](https://developers.google.com/protocol-buffers/docs/encoding), +with un-encoded value length of 64 bits, unless other limits are specified. + +| Content | Format | +|---------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------| +| ID, `[1]byte` | Always 0x99. | +| Data Length, `[3]byte` | 3 byte little-endian length of the chunk in bytes, following this. | +| Header `[6]byte` | Header, must be `[115, 50, 105, 100, 120, 0]` or in text: "s2idx\x00". | +| UncompressedSize, Varint | Total Uncompressed size. | +| CompressedSize, Varint | Total Compressed size if known. Should be -1 if unknown. | +| EstBlockSize, Varint | Block Size, used for guessing uncompressed offsets. Must be >= 0. | +| Entries, Varint | Number of Entries in index, must be < 65536 and >=0. | +| HasUncompressedOffsets `byte` | 0 if no uncompressed offsets are present, 1 if present. Other values are invalid. | +| UncompressedOffsets, [Entries]VarInt | Uncompressed offsets. See below how to decode. | +| CompressedOffsets, [Entries]VarInt | Compressed offsets. See below how to decode. | +| Block Size, `[4]byte` | Little Endian total encoded size (including header and trailer). Can be used for searching backwards to start of block. | +| Trailer `[6]byte` | Trailer, must be `[0, 120, 100, 105, 50, 115]` or in text: "\x00xdi2s". Can be used for identifying block from end of stream. | + +For regular streams the uncompressed offsets are fully predictable, +so `HasUncompressedOffsets` allows to specify that compressed blocks all have +exactly `EstBlockSize` bytes of uncompressed content. + +Entries *must* be in order, starting with the lowest offset, +and there *must* be no uncompressed offset duplicates. +Entries *may* point to the start of a skippable block, +but it is then not allowed to also have an entry for the next block since +that would give an uncompressed offset duplicate. + +There is no requirement for all blocks to be represented in the index. +In fact there is a maximum of 65536 block entries in an index. + +The writer can use any method to reduce the number of entries. +An implicit block start at 0,0 can be assumed. + +### Decoding entries: + +``` +// Read Uncompressed entries. +// Each assumes EstBlockSize delta from previous. +for each entry { + uOff = 0 + if HasUncompressedOffsets == 1 { + uOff = ReadVarInt // Read value from stream + } + + // Except for the first entry, use previous values. + if entryNum == 0 { + entry[entryNum].UncompressedOffset = uOff + continue + } + + // Uncompressed uses previous offset and adds EstBlockSize + entry[entryNum].UncompressedOffset = entry[entryNum-1].UncompressedOffset + EstBlockSize + uOff +} + + +// Guess that the first block will be 50% of uncompressed size. +// Integer truncating division must be used. +CompressGuess := EstBlockSize / 2 + +// Read Compressed entries. +// Each assumes CompressGuess delta from previous. +// CompressGuess is adjusted for each value. +for each entry { + cOff = ReadVarInt // Read value from stream + + // Except for the first entry, use previous values. + if entryNum == 0 { + entry[entryNum].CompressedOffset = cOff + continue + } + + // Compressed uses previous and our estimate. + entry[entryNum].CompressedOffset = entry[entryNum-1].CompressedOffset + CompressGuess + cOff + + // Adjust compressed offset for next loop, integer truncating division must be used. + CompressGuess += cOff/2 +} +``` + +To decode from any given uncompressed offset `(wantOffset)`: + +* Iterate entries until `entry[n].UncompressedOffset > wantOffset`. +* Start decoding from `entry[n-1].CompressedOffset`. +* Discard `entry[n-1].UncompressedOffset - wantOffset` bytes from the decoded stream. + +See [using indexes](https://github.com/klauspost/compress/tree/master/s2#using-indexes) for functions that perform the operations with a simpler interface. + +# Format Extensions + +* Frame [Stream identifier](https://github.com/google/snappy/blob/master/framing_format.txt#L68) changed from `sNaPpY` to `S2sTwO`. +* [Framed compressed blocks](https://github.com/google/snappy/blob/master/format_description.txt) can be up to 4MB (up from 64KB). +* Compressed blocks can have an offset of `0`, which indicates to repeat the last seen offset. + +Repeat offsets must be encoded as a [2.2.1. Copy with 1-byte offset (01)](https://github.com/google/snappy/blob/master/format_description.txt#L89), where the offset is 0. + +The length is specified by reading the 3-bit length specified in the tag and decode using this table: + +| Length | Actual Length | +|--------|----------------------| +| 0 | 4 | +| 1 | 5 | +| 2 | 6 | +| 3 | 7 | +| 4 | 8 | +| 5 | 8 + read 1 byte | +| 6 | 260 + read 2 bytes | +| 7 | 65540 + read 3 bytes | + +This allows any repeat offset + length to be represented by 2 to 5 bytes. + +Lengths are stored as little endian values. + +The first copy of a block cannot be a repeat offset and the offset is not carried across blocks in streams. + +Default streaming block size is 1MB. + +# LICENSE + +This code is based on the [Snappy-Go](https://github.com/golang/snappy) implementation. + +Use of this source code is governed by a BSD-style license that can be found in the LICENSE file. diff --git a/vendor/github.com/klauspost/compress/s2/decode.go b/vendor/github.com/klauspost/compress/s2/decode.go new file mode 100644 index 0000000..27c0f3c --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/decode.go @@ -0,0 +1,1046 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package s2 + +import ( + "encoding/binary" + "errors" + "fmt" + "io" + "io/ioutil" + "math" + "runtime" + "sync" +) + +var ( + // ErrCorrupt reports that the input is invalid. + ErrCorrupt = errors.New("s2: corrupt input") + // ErrCRC reports that the input failed CRC validation (streams only) + ErrCRC = errors.New("s2: corrupt input, crc mismatch") + // ErrTooLarge reports that the uncompressed length is too large. + ErrTooLarge = errors.New("s2: decoded block is too large") + // ErrUnsupported reports that the input isn't supported. + ErrUnsupported = errors.New("s2: unsupported input") +) + +// ErrCantSeek is returned if the stream cannot be seeked. +type ErrCantSeek struct { + Reason string +} + +// Error returns the error as string. +func (e ErrCantSeek) Error() string { + return fmt.Sprintf("s2: Can't seek because %s", e.Reason) +} + +// DecodedLen returns the length of the decoded block. +func DecodedLen(src []byte) (int, error) { + v, _, err := decodedLen(src) + return v, err +} + +// decodedLen returns the length of the decoded block and the number of bytes +// that the length header occupied. +func decodedLen(src []byte) (blockLen, headerLen int, err error) { + v, n := binary.Uvarint(src) + if n <= 0 || v > 0xffffffff { + return 0, 0, ErrCorrupt + } + + const wordSize = 32 << (^uint(0) >> 32 & 1) + if wordSize == 32 && v > 0x7fffffff { + return 0, 0, ErrTooLarge + } + return int(v), n, nil +} + +const ( + decodeErrCodeCorrupt = 1 +) + +// Decode returns the decoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire decoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +func Decode(dst, src []byte) ([]byte, error) { + dLen, s, err := decodedLen(src) + if err != nil { + return nil, err + } + if dLen <= cap(dst) { + dst = dst[:dLen] + } else { + dst = make([]byte, dLen) + } + if s2Decode(dst, src[s:]) != 0 { + return nil, ErrCorrupt + } + return dst, nil +} + +// NewReader returns a new Reader that decompresses from r, using the framing +// format described at +// https://github.com/google/snappy/blob/master/framing_format.txt with S2 changes. +func NewReader(r io.Reader, opts ...ReaderOption) *Reader { + nr := Reader{ + r: r, + maxBlock: maxBlockSize, + } + for _, opt := range opts { + if err := opt(&nr); err != nil { + nr.err = err + return &nr + } + } + nr.maxBufSize = MaxEncodedLen(nr.maxBlock) + checksumSize + if nr.lazyBuf > 0 { + nr.buf = make([]byte, MaxEncodedLen(nr.lazyBuf)+checksumSize) + } else { + nr.buf = make([]byte, MaxEncodedLen(defaultBlockSize)+checksumSize) + } + nr.readHeader = nr.ignoreStreamID + nr.paramsOK = true + return &nr +} + +// ReaderOption is an option for creating a decoder. +type ReaderOption func(*Reader) error + +// ReaderMaxBlockSize allows to control allocations if the stream +// has been compressed with a smaller WriterBlockSize, or with the default 1MB. +// Blocks must be this size or smaller to decompress, +// otherwise the decoder will return ErrUnsupported. +// +// For streams compressed with Snappy this can safely be set to 64KB (64 << 10). +// +// Default is the maximum limit of 4MB. +func ReaderMaxBlockSize(blockSize int) ReaderOption { + return func(r *Reader) error { + if blockSize > maxBlockSize || blockSize <= 0 { + return errors.New("s2: block size too large. Must be <= 4MB and > 0") + } + if r.lazyBuf == 0 && blockSize < defaultBlockSize { + r.lazyBuf = blockSize + } + r.maxBlock = blockSize + return nil + } +} + +// ReaderAllocBlock allows to control upfront stream allocations +// and not allocate for frames bigger than this initially. +// If frames bigger than this is seen a bigger buffer will be allocated. +// +// Default is 1MB, which is default output size. +func ReaderAllocBlock(blockSize int) ReaderOption { + return func(r *Reader) error { + if blockSize > maxBlockSize || blockSize < 1024 { + return errors.New("s2: invalid ReaderAllocBlock. Must be <= 4MB and >= 1024") + } + r.lazyBuf = blockSize + return nil + } +} + +// ReaderIgnoreStreamIdentifier will make the reader skip the expected +// stream identifier at the beginning of the stream. +// This can be used when serving a stream that has been forwarded to a specific point. +func ReaderIgnoreStreamIdentifier() ReaderOption { + return func(r *Reader) error { + r.ignoreStreamID = true + return nil + } +} + +// ReaderSkippableCB will register a callback for chuncks with the specified ID. +// ID must be a Reserved skippable chunks ID, 0x80-0xfd (inclusive). +// For each chunk with the ID, the callback is called with the content. +// Any returned non-nil error will abort decompression. +// Only one callback per ID is supported, latest sent will be used. +func ReaderSkippableCB(id uint8, fn func(r io.Reader) error) ReaderOption { + return func(r *Reader) error { + if id < 0x80 || id > 0xfd { + return fmt.Errorf("ReaderSkippableCB: Invalid id provided, must be 0x80-0xfd (inclusive)") + } + r.skippableCB[id] = fn + return nil + } +} + +// ReaderIgnoreCRC will make the reader skip CRC calculation and checks. +func ReaderIgnoreCRC() ReaderOption { + return func(r *Reader) error { + r.ignoreCRC = true + return nil + } +} + +// Reader is an io.Reader that can read Snappy-compressed bytes. +type Reader struct { + r io.Reader + err error + decoded []byte + buf []byte + skippableCB [0x80]func(r io.Reader) error + blockStart int64 // Uncompressed offset at start of current. + index *Index + + // decoded[i:j] contains decoded bytes that have not yet been passed on. + i, j int + // maximum block size allowed. + maxBlock int + // maximum expected buffer size. + maxBufSize int + // alloc a buffer this size if > 0. + lazyBuf int + readHeader bool + paramsOK bool + snappyFrame bool + ignoreStreamID bool + ignoreCRC bool +} + +// ensureBufferSize will ensure that the buffer can take at least n bytes. +// If false is returned the buffer exceeds maximum allowed size. +func (r *Reader) ensureBufferSize(n int) bool { + if n > r.maxBufSize { + r.err = ErrCorrupt + return false + } + if cap(r.buf) >= n { + return true + } + // Realloc buffer. + r.buf = make([]byte, n) + return true +} + +// Reset discards any buffered data, resets all state, and switches the Snappy +// reader to read from r. This permits reusing a Reader rather than allocating +// a new one. +func (r *Reader) Reset(reader io.Reader) { + if !r.paramsOK { + return + } + r.index = nil + r.r = reader + r.err = nil + r.i = 0 + r.j = 0 + r.blockStart = 0 + r.readHeader = r.ignoreStreamID +} + +func (r *Reader) readFull(p []byte, allowEOF bool) (ok bool) { + if _, r.err = io.ReadFull(r.r, p); r.err != nil { + if r.err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) { + r.err = ErrCorrupt + } + return false + } + return true +} + +// skippable will skip n bytes. +// If the supplied reader supports seeking that is used. +// tmp is used as a temporary buffer for reading. +// The supplied slice does not need to be the size of the read. +func (r *Reader) skippable(tmp []byte, n int, allowEOF bool, id uint8) (ok bool) { + if id < 0x80 { + r.err = fmt.Errorf("interbal error: skippable id < 0x80") + return false + } + if fn := r.skippableCB[id-0x80]; fn != nil { + rd := io.LimitReader(r.r, int64(n)) + r.err = fn(rd) + if r.err != nil { + return false + } + _, r.err = io.CopyBuffer(ioutil.Discard, rd, tmp) + return r.err == nil + } + if rs, ok := r.r.(io.ReadSeeker); ok { + _, err := rs.Seek(int64(n), io.SeekCurrent) + if err == nil { + return true + } + if err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) { + r.err = ErrCorrupt + return false + } + } + for n > 0 { + if n < len(tmp) { + tmp = tmp[:n] + } + if _, r.err = io.ReadFull(r.r, tmp); r.err != nil { + if r.err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) { + r.err = ErrCorrupt + } + return false + } + n -= len(tmp) + } + return true +} + +// Read satisfies the io.Reader interface. +func (r *Reader) Read(p []byte) (int, error) { + if r.err != nil { + return 0, r.err + } + for { + if r.i < r.j { + n := copy(p, r.decoded[r.i:r.j]) + r.i += n + return n, nil + } + if !r.readFull(r.buf[:4], true) { + return 0, r.err + } + chunkType := r.buf[0] + if !r.readHeader { + if chunkType != chunkTypeStreamIdentifier { + r.err = ErrCorrupt + return 0, r.err + } + r.readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + r.blockStart += int64(r.j) + // Section 4.2. Compressed data (chunk type 0x00). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return 0, r.err + } + if !r.ensureBufferSize(chunkLen) { + if r.err == nil { + r.err = ErrUnsupported + } + return 0, r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return 0, r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[checksumSize:] + + n, err := DecodedLen(buf) + if err != nil { + r.err = err + return 0, r.err + } + if r.snappyFrame && n > maxSnappyBlockSize { + r.err = ErrCorrupt + return 0, r.err + } + + if n > len(r.decoded) { + if n > r.maxBlock { + r.err = ErrCorrupt + return 0, r.err + } + r.decoded = make([]byte, n) + } + if _, err := Decode(r.decoded, buf); err != nil { + r.err = err + return 0, r.err + } + if !r.ignoreCRC && crc(r.decoded[:n]) != checksum { + r.err = ErrCRC + return 0, r.err + } + r.i, r.j = 0, n + continue + + case chunkTypeUncompressedData: + r.blockStart += int64(r.j) + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return 0, r.err + } + if !r.ensureBufferSize(chunkLen) { + if r.err == nil { + r.err = ErrUnsupported + } + return 0, r.err + } + buf := r.buf[:checksumSize] + if !r.readFull(buf, false) { + return 0, r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + // Read directly into r.decoded instead of via r.buf. + n := chunkLen - checksumSize + if r.snappyFrame && n > maxSnappyBlockSize { + r.err = ErrCorrupt + return 0, r.err + } + if n > len(r.decoded) { + if n > r.maxBlock { + r.err = ErrCorrupt + return 0, r.err + } + r.decoded = make([]byte, n) + } + if !r.readFull(r.decoded[:n], false) { + return 0, r.err + } + if !r.ignoreCRC && crc(r.decoded[:n]) != checksum { + r.err = ErrCRC + return 0, r.err + } + r.i, r.j = 0, n + continue + + case chunkTypeStreamIdentifier: + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(magicBody) { + r.err = ErrCorrupt + return 0, r.err + } + if !r.readFull(r.buf[:len(magicBody)], false) { + return 0, r.err + } + if string(r.buf[:len(magicBody)]) != magicBody { + if string(r.buf[:len(magicBody)]) != magicBodySnappy { + r.err = ErrCorrupt + return 0, r.err + } else { + r.snappyFrame = true + } + } else { + r.snappyFrame = false + } + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + // fmt.Printf("ERR chunktype: 0x%x\n", chunkType) + r.err = ErrUnsupported + return 0, r.err + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + if chunkLen > maxChunkSize { + // fmt.Printf("ERR chunkLen: 0x%x\n", chunkLen) + r.err = ErrUnsupported + return 0, r.err + } + + // fmt.Printf("skippable: ID: 0x%x, len: 0x%x\n", chunkType, chunkLen) + if !r.skippable(r.buf, chunkLen, false, chunkType) { + return 0, r.err + } + } +} + +// DecodeConcurrent will decode the full stream to w. +// This function should not be combined with reading, seeking or other operations. +// Up to 'concurrent' goroutines will be used. +// If <= 0, runtime.NumCPU will be used. +// On success the number of bytes decompressed nil and is returned. +// This is mainly intended for bigger streams. +func (r *Reader) DecodeConcurrent(w io.Writer, concurrent int) (written int64, err error) { + if r.i > 0 || r.j > 0 || r.blockStart > 0 { + return 0, errors.New("DecodeConcurrent called after ") + } + if concurrent <= 0 { + concurrent = runtime.NumCPU() + } + + // Write to output + var errMu sync.Mutex + var aErr error + setErr := func(e error) (ok bool) { + errMu.Lock() + defer errMu.Unlock() + if e == nil { + return aErr == nil + } + if aErr == nil { + aErr = e + } + return false + } + hasErr := func() (ok bool) { + errMu.Lock() + v := aErr != nil + errMu.Unlock() + return v + } + + var aWritten int64 + toRead := make(chan []byte, concurrent) + writtenBlocks := make(chan []byte, concurrent) + queue := make(chan chan []byte, concurrent) + reUse := make(chan chan []byte, concurrent) + for i := 0; i < concurrent; i++ { + toRead <- make([]byte, 0, r.maxBufSize) + writtenBlocks <- make([]byte, 0, r.maxBufSize) + reUse <- make(chan []byte, 1) + } + // Writer + var wg sync.WaitGroup + wg.Add(1) + go func() { + defer wg.Done() + for toWrite := range queue { + entry := <-toWrite + reUse <- toWrite + if hasErr() { + writtenBlocks <- entry + continue + } + n, err := w.Write(entry) + want := len(entry) + writtenBlocks <- entry + if err != nil { + setErr(err) + continue + } + if n != want { + setErr(io.ErrShortWrite) + continue + } + aWritten += int64(n) + } + }() + + // Reader + defer func() { + close(queue) + if r.err != nil { + err = r.err + setErr(r.err) + } + wg.Wait() + if err == nil { + err = aErr + } + written = aWritten + }() + + for !hasErr() { + if !r.readFull(r.buf[:4], true) { + if r.err == io.EOF { + r.err = nil + } + return 0, r.err + } + chunkType := r.buf[0] + if !r.readHeader { + if chunkType != chunkTypeStreamIdentifier { + r.err = ErrCorrupt + return 0, r.err + } + r.readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + r.blockStart += int64(r.j) + // Section 4.2. Compressed data (chunk type 0x00). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return 0, r.err + } + if chunkLen > r.maxBufSize { + r.err = ErrCorrupt + return 0, r.err + } + orgBuf := <-toRead + buf := orgBuf[:chunkLen] + + if !r.readFull(buf, false) { + return 0, r.err + } + + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[checksumSize:] + + n, err := DecodedLen(buf) + if err != nil { + r.err = err + return 0, r.err + } + if r.snappyFrame && n > maxSnappyBlockSize { + r.err = ErrCorrupt + return 0, r.err + } + + if n > r.maxBlock { + r.err = ErrCorrupt + return 0, r.err + } + wg.Add(1) + + decoded := <-writtenBlocks + entry := <-reUse + queue <- entry + go func() { + defer wg.Done() + decoded = decoded[:n] + _, err := Decode(decoded, buf) + toRead <- orgBuf + if err != nil { + writtenBlocks <- decoded + setErr(err) + return + } + if !r.ignoreCRC && crc(decoded) != checksum { + writtenBlocks <- decoded + setErr(ErrCRC) + return + } + entry <- decoded + }() + continue + + case chunkTypeUncompressedData: + + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return 0, r.err + } + if chunkLen > r.maxBufSize { + r.err = ErrCorrupt + return 0, r.err + } + // Grab write buffer + orgBuf := <-writtenBlocks + buf := orgBuf[:checksumSize] + if !r.readFull(buf, false) { + return 0, r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + // Read content. + n := chunkLen - checksumSize + + if r.snappyFrame && n > maxSnappyBlockSize { + r.err = ErrCorrupt + return 0, r.err + } + if n > r.maxBlock { + r.err = ErrCorrupt + return 0, r.err + } + // Read uncompressed + buf = orgBuf[:n] + if !r.readFull(buf, false) { + return 0, r.err + } + + if !r.ignoreCRC && crc(buf) != checksum { + r.err = ErrCRC + return 0, r.err + } + entry := <-reUse + queue <- entry + entry <- buf + continue + + case chunkTypeStreamIdentifier: + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(magicBody) { + r.err = ErrCorrupt + return 0, r.err + } + if !r.readFull(r.buf[:len(magicBody)], false) { + return 0, r.err + } + if string(r.buf[:len(magicBody)]) != magicBody { + if string(r.buf[:len(magicBody)]) != magicBodySnappy { + r.err = ErrCorrupt + return 0, r.err + } else { + r.snappyFrame = true + } + } else { + r.snappyFrame = false + } + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + // fmt.Printf("ERR chunktype: 0x%x\n", chunkType) + r.err = ErrUnsupported + return 0, r.err + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + if chunkLen > maxChunkSize { + // fmt.Printf("ERR chunkLen: 0x%x\n", chunkLen) + r.err = ErrUnsupported + return 0, r.err + } + + // fmt.Printf("skippable: ID: 0x%x, len: 0x%x\n", chunkType, chunkLen) + if !r.skippable(r.buf, chunkLen, false, chunkType) { + return 0, r.err + } + } + return 0, r.err +} + +// Skip will skip n bytes forward in the decompressed output. +// For larger skips this consumes less CPU and is faster than reading output and discarding it. +// CRC is not checked on skipped blocks. +// io.ErrUnexpectedEOF is returned if the stream ends before all bytes have been skipped. +// If a decoding error is encountered subsequent calls to Read will also fail. +func (r *Reader) Skip(n int64) error { + if n < 0 { + return errors.New("attempted negative skip") + } + if r.err != nil { + return r.err + } + + for n > 0 { + if r.i < r.j { + // Skip in buffer. + // decoded[i:j] contains decoded bytes that have not yet been passed on. + left := int64(r.j - r.i) + if left >= n { + tmp := int64(r.i) + n + if tmp > math.MaxInt32 { + return errors.New("s2: internal overflow in skip") + } + r.i = int(tmp) + return nil + } + n -= int64(r.j - r.i) + r.i = r.j + } + + // Buffer empty; read blocks until we have content. + if !r.readFull(r.buf[:4], true) { + if r.err == io.EOF { + r.err = io.ErrUnexpectedEOF + } + return r.err + } + chunkType := r.buf[0] + if !r.readHeader { + if chunkType != chunkTypeStreamIdentifier { + r.err = ErrCorrupt + return r.err + } + r.readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + r.blockStart += int64(r.j) + // Section 4.2. Compressed data (chunk type 0x00). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return r.err + } + if !r.ensureBufferSize(chunkLen) { + if r.err == nil { + r.err = ErrUnsupported + } + return r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[checksumSize:] + + dLen, err := DecodedLen(buf) + if err != nil { + r.err = err + return r.err + } + if dLen > r.maxBlock { + r.err = ErrCorrupt + return r.err + } + // Check if destination is within this block + if int64(dLen) > n { + if len(r.decoded) < dLen { + r.decoded = make([]byte, dLen) + } + if _, err := Decode(r.decoded, buf); err != nil { + r.err = err + return r.err + } + if crc(r.decoded[:dLen]) != checksum { + r.err = ErrCorrupt + return r.err + } + } else { + // Skip block completely + n -= int64(dLen) + r.blockStart += int64(dLen) + dLen = 0 + } + r.i, r.j = 0, dLen + continue + case chunkTypeUncompressedData: + r.blockStart += int64(r.j) + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return r.err + } + if !r.ensureBufferSize(chunkLen) { + if r.err != nil { + r.err = ErrUnsupported + } + return r.err + } + buf := r.buf[:checksumSize] + if !r.readFull(buf, false) { + return r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + // Read directly into r.decoded instead of via r.buf. + n2 := chunkLen - checksumSize + if n2 > len(r.decoded) { + if n2 > r.maxBlock { + r.err = ErrCorrupt + return r.err + } + r.decoded = make([]byte, n2) + } + if !r.readFull(r.decoded[:n2], false) { + return r.err + } + if int64(n2) < n { + if crc(r.decoded[:n2]) != checksum { + r.err = ErrCorrupt + return r.err + } + } + r.i, r.j = 0, n2 + continue + case chunkTypeStreamIdentifier: + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(magicBody) { + r.err = ErrCorrupt + return r.err + } + if !r.readFull(r.buf[:len(magicBody)], false) { + return r.err + } + if string(r.buf[:len(magicBody)]) != magicBody { + if string(r.buf[:len(magicBody)]) != magicBodySnappy { + r.err = ErrCorrupt + return r.err + } + } + + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + r.err = ErrUnsupported + return r.err + } + if chunkLen > maxChunkSize { + r.err = ErrUnsupported + return r.err + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + if !r.skippable(r.buf, chunkLen, false, chunkType) { + return r.err + } + } + return nil +} + +// ReadSeeker provides random or forward seeking in compressed content. +// See Reader.ReadSeeker +type ReadSeeker struct { + *Reader +} + +// ReadSeeker will return an io.ReadSeeker compatible version of the reader. +// If 'random' is specified the returned io.Seeker can be used for +// random seeking, otherwise only forward seeking is supported. +// Enabling random seeking requires the original input to support +// the io.Seeker interface. +// A custom index can be specified which will be used if supplied. +// When using a custom index, it will not be read from the input stream. +// The returned ReadSeeker contains a shallow reference to the existing Reader, +// meaning changes performed to one is reflected in the other. +func (r *Reader) ReadSeeker(random bool, index []byte) (*ReadSeeker, error) { + // Read index if provided. + if len(index) != 0 { + if r.index == nil { + r.index = &Index{} + } + if _, err := r.index.Load(index); err != nil { + return nil, ErrCantSeek{Reason: "loading index returned: " + err.Error()} + } + } + + // Check if input is seekable + rs, ok := r.r.(io.ReadSeeker) + if !ok { + if !random { + return &ReadSeeker{Reader: r}, nil + } + return nil, ErrCantSeek{Reason: "input stream isn't seekable"} + } + + if r.index != nil { + // Seekable and index, ok... + return &ReadSeeker{Reader: r}, nil + } + + // Load from stream. + r.index = &Index{} + + // Read current position. + pos, err := rs.Seek(0, io.SeekCurrent) + if err != nil { + return nil, ErrCantSeek{Reason: "seeking input returned: " + err.Error()} + } + err = r.index.LoadStream(rs) + if err != nil { + if err == ErrUnsupported { + // If we don't require random seeking, reset input and return. + if !random { + _, err = rs.Seek(pos, io.SeekStart) + if err != nil { + return nil, ErrCantSeek{Reason: "resetting stream returned: " + err.Error()} + } + r.index = nil + return &ReadSeeker{Reader: r}, nil + } + return nil, ErrCantSeek{Reason: "input stream does not contain an index"} + } + return nil, ErrCantSeek{Reason: "reading index returned: " + err.Error()} + } + + // reset position. + _, err = rs.Seek(pos, io.SeekStart) + if err != nil { + return nil, ErrCantSeek{Reason: "seeking input returned: " + err.Error()} + } + return &ReadSeeker{Reader: r}, nil +} + +// Seek allows seeking in compressed data. +func (r *ReadSeeker) Seek(offset int64, whence int) (int64, error) { + if r.err != nil { + return 0, r.err + } + if offset == 0 && whence == io.SeekCurrent { + return r.blockStart + int64(r.i), nil + } + if !r.readHeader { + // Make sure we read the header. + _, r.err = r.Read([]byte{}) + } + rs, ok := r.r.(io.ReadSeeker) + if r.index == nil || !ok { + if whence == io.SeekCurrent && offset >= 0 { + err := r.Skip(offset) + return r.blockStart + int64(r.i), err + } + if whence == io.SeekStart && offset >= r.blockStart+int64(r.i) { + err := r.Skip(offset - r.blockStart - int64(r.i)) + return r.blockStart + int64(r.i), err + } + return 0, ErrUnsupported + + } + + switch whence { + case io.SeekCurrent: + offset += r.blockStart + int64(r.i) + case io.SeekEnd: + if offset > 0 { + return 0, errors.New("seek after end of file") + } + offset = r.index.TotalUncompressed + offset + } + + if offset < 0 { + return 0, errors.New("seek before start of file") + } + + c, u, err := r.index.Find(offset) + if err != nil { + return r.blockStart + int64(r.i), err + } + + // Seek to next block + _, err = rs.Seek(c, io.SeekStart) + if err != nil { + return 0, err + } + + r.i = r.j // Remove rest of current block. + if u < offset { + // Forward inside block + return offset, r.Skip(offset - u) + } + return offset, nil +} + +// ReadByte satisfies the io.ByteReader interface. +func (r *Reader) ReadByte() (byte, error) { + if r.err != nil { + return 0, r.err + } + if r.i < r.j { + c := r.decoded[r.i] + r.i++ + return c, nil + } + var tmp [1]byte + for i := 0; i < 10; i++ { + n, err := r.Read(tmp[:]) + if err != nil { + return 0, err + } + if n == 1 { + return tmp[0], nil + } + } + return 0, io.ErrNoProgress +} + +// SkippableCB will register a callback for chunks with the specified ID. +// ID must be a Reserved skippable chunks ID, 0x80-0xfe (inclusive). +// For each chunk with the ID, the callback is called with the content. +// Any returned non-nil error will abort decompression. +// Only one callback per ID is supported, latest sent will be used. +// Sending a nil function will disable previous callbacks. +func (r *Reader) SkippableCB(id uint8, fn func(r io.Reader) error) error { + if id < 0x80 || id > chunkTypePadding { + return fmt.Errorf("ReaderSkippableCB: Invalid id provided, must be 0x80-0xfe (inclusive)") + } + r.skippableCB[id] = fn + return nil +} diff --git a/vendor/github.com/klauspost/compress/s2/decode_amd64.s b/vendor/github.com/klauspost/compress/s2/decode_amd64.s new file mode 100644 index 0000000..9b105e0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/decode_amd64.s @@ -0,0 +1,568 @@ +// Copyright 2016 The Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build !appengine +// +build gc +// +build !noasm + +#include "textflag.h" + +#define R_TMP0 AX +#define R_TMP1 BX +#define R_LEN CX +#define R_OFF DX +#define R_SRC SI +#define R_DST DI +#define R_DBASE R8 +#define R_DLEN R9 +#define R_DEND R10 +#define R_SBASE R11 +#define R_SLEN R12 +#define R_SEND R13 +#define R_TMP2 R14 +#define R_TMP3 R15 + +// The asm code generally follows the pure Go code in decode_other.go, except +// where marked with a "!!!". + +// func decode(dst, src []byte) int +// +// All local variables fit into registers. The non-zero stack size is only to +// spill registers and push args when issuing a CALL. The register allocation: +// - R_TMP0 scratch +// - R_TMP1 scratch +// - R_LEN length or x (shared) +// - R_OFF offset +// - R_SRC &src[s] +// - R_DST &dst[d] +// + R_DBASE dst_base +// + R_DLEN dst_len +// + R_DEND dst_base + dst_len +// + R_SBASE src_base +// + R_SLEN src_len +// + R_SEND src_base + src_len +// - R_TMP2 used by doCopy +// - R_TMP3 used by doCopy +// +// The registers R_DBASE-R_SEND (marked with a "+") are set at the start of the +// function, and after a CALL returns, and are not otherwise modified. +// +// The d variable is implicitly R_DST - R_DBASE, and len(dst)-d is R_DEND - R_DST. +// The s variable is implicitly R_SRC - R_SBASE, and len(src)-s is R_SEND - R_SRC. +TEXT ·s2Decode(SB), NOSPLIT, $48-56 + // Initialize R_SRC, R_DST and R_DBASE-R_SEND. + MOVQ dst_base+0(FP), R_DBASE + MOVQ dst_len+8(FP), R_DLEN + MOVQ R_DBASE, R_DST + MOVQ R_DBASE, R_DEND + ADDQ R_DLEN, R_DEND + MOVQ src_base+24(FP), R_SBASE + MOVQ src_len+32(FP), R_SLEN + MOVQ R_SBASE, R_SRC + MOVQ R_SBASE, R_SEND + ADDQ R_SLEN, R_SEND + XORQ R_OFF, R_OFF + +loop: + // for s < len(src) + CMPQ R_SRC, R_SEND + JEQ end + + // R_LEN = uint32(src[s]) + // + // switch src[s] & 0x03 + MOVBLZX (R_SRC), R_LEN + MOVL R_LEN, R_TMP1 + ANDL $3, R_TMP1 + CMPL R_TMP1, $1 + JAE tagCopy + + // ---------------------------------------- + // The code below handles literal tags. + + // case tagLiteral: + // x := uint32(src[s] >> 2) + // switch + SHRL $2, R_LEN + CMPL R_LEN, $60 + JAE tagLit60Plus + + // case x < 60: + // s++ + INCQ R_SRC + +doLit: + // This is the end of the inner "switch", when we have a literal tag. + // + // We assume that R_LEN == x and x fits in a uint32, where x is the variable + // used in the pure Go decode_other.go code. + + // length = int(x) + 1 + // + // Unlike the pure Go code, we don't need to check if length <= 0 because + // R_LEN can hold 64 bits, so the increment cannot overflow. + INCQ R_LEN + + // Prepare to check if copying length bytes will run past the end of dst or + // src. + // + // R_TMP0 = len(dst) - d + // R_TMP1 = len(src) - s + MOVQ R_DEND, R_TMP0 + SUBQ R_DST, R_TMP0 + MOVQ R_SEND, R_TMP1 + SUBQ R_SRC, R_TMP1 + + // !!! Try a faster technique for short (16 or fewer bytes) copies. + // + // if length > 16 || len(dst)-d < 16 || len(src)-s < 16 { + // goto callMemmove // Fall back on calling runtime·memmove. + // } + // + // The C++ snappy code calls this TryFastAppend. It also checks len(src)-s + // against 21 instead of 16, because it cannot assume that all of its input + // is contiguous in memory and so it needs to leave enough source bytes to + // read the next tag without refilling buffers, but Go's Decode assumes + // contiguousness (the src argument is a []byte). + CMPQ R_LEN, $16 + JGT callMemmove + CMPQ R_TMP0, $16 + JLT callMemmove + CMPQ R_TMP1, $16 + JLT callMemmove + + // !!! Implement the copy from src to dst as a 16-byte load and store. + // (Decode's documentation says that dst and src must not overlap.) + // + // This always copies 16 bytes, instead of only length bytes, but that's + // OK. If the input is a valid Snappy encoding then subsequent iterations + // will fix up the overrun. Otherwise, Decode returns a nil []byte (and a + // non-nil error), so the overrun will be ignored. + // + // Note that on amd64, it is legal and cheap to issue unaligned 8-byte or + // 16-byte loads and stores. This technique probably wouldn't be as + // effective on architectures that are fussier about alignment. + MOVOU 0(R_SRC), X0 + MOVOU X0, 0(R_DST) + + // d += length + // s += length + ADDQ R_LEN, R_DST + ADDQ R_LEN, R_SRC + JMP loop + +callMemmove: + // if length > len(dst)-d || length > len(src)-s { etc } + CMPQ R_LEN, R_TMP0 + JGT errCorrupt + CMPQ R_LEN, R_TMP1 + JGT errCorrupt + + // copy(dst[d:], src[s:s+length]) + // + // This means calling runtime·memmove(&dst[d], &src[s], length), so we push + // R_DST, R_SRC and R_LEN as arguments. Coincidentally, we also need to spill those + // three registers to the stack, to save local variables across the CALL. + MOVQ R_DST, 0(SP) + MOVQ R_SRC, 8(SP) + MOVQ R_LEN, 16(SP) + MOVQ R_DST, 24(SP) + MOVQ R_SRC, 32(SP) + MOVQ R_LEN, 40(SP) + MOVQ R_OFF, 48(SP) + CALL runtime·memmove(SB) + + // Restore local variables: unspill registers from the stack and + // re-calculate R_DBASE-R_SEND. + MOVQ 24(SP), R_DST + MOVQ 32(SP), R_SRC + MOVQ 40(SP), R_LEN + MOVQ 48(SP), R_OFF + MOVQ dst_base+0(FP), R_DBASE + MOVQ dst_len+8(FP), R_DLEN + MOVQ R_DBASE, R_DEND + ADDQ R_DLEN, R_DEND + MOVQ src_base+24(FP), R_SBASE + MOVQ src_len+32(FP), R_SLEN + MOVQ R_SBASE, R_SEND + ADDQ R_SLEN, R_SEND + + // d += length + // s += length + ADDQ R_LEN, R_DST + ADDQ R_LEN, R_SRC + JMP loop + +tagLit60Plus: + // !!! This fragment does the + // + // s += x - 58; if uint(s) > uint(len(src)) { etc } + // + // checks. In the asm version, we code it once instead of once per switch case. + ADDQ R_LEN, R_SRC + SUBQ $58, R_SRC + CMPQ R_SRC, R_SEND + JA errCorrupt + + // case x == 60: + CMPL R_LEN, $61 + JEQ tagLit61 + JA tagLit62Plus + + // x = uint32(src[s-1]) + MOVBLZX -1(R_SRC), R_LEN + JMP doLit + +tagLit61: + // case x == 61: + // x = uint32(src[s-2]) | uint32(src[s-1])<<8 + MOVWLZX -2(R_SRC), R_LEN + JMP doLit + +tagLit62Plus: + CMPL R_LEN, $62 + JA tagLit63 + + // case x == 62: + // x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + // We read one byte, safe to read one back, since we are just reading tag. + // x = binary.LittleEndian.Uint32(src[s-1:]) >> 8 + MOVL -4(R_SRC), R_LEN + SHRL $8, R_LEN + JMP doLit + +tagLit63: + // case x == 63: + // x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + MOVL -4(R_SRC), R_LEN + JMP doLit + +// The code above handles literal tags. +// ---------------------------------------- +// The code below handles copy tags. + +tagCopy4: + // case tagCopy4: + // s += 5 + ADDQ $5, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = 1 + int(src[s-5])>>2 + SHRQ $2, R_LEN + INCQ R_LEN + + // offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + MOVLQZX -4(R_SRC), R_OFF + JMP doCopy + +tagCopy2: + // case tagCopy2: + // s += 3 + ADDQ $3, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = 1 + int(src[s-3])>>2 + SHRQ $2, R_LEN + INCQ R_LEN + + // offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + MOVWQZX -2(R_SRC), R_OFF + JMP doCopy + +tagCopy: + // We have a copy tag. We assume that: + // - R_TMP1 == src[s] & 0x03 + // - R_LEN == src[s] + CMPQ R_TMP1, $2 + JEQ tagCopy2 + JA tagCopy4 + + // case tagCopy1: + // s += 2 + ADDQ $2, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + // length = 4 + int(src[s-2])>>2&0x7 + MOVBQZX -1(R_SRC), R_TMP1 + MOVQ R_LEN, R_TMP0 + SHRQ $2, R_LEN + ANDQ $0xe0, R_TMP0 + ANDQ $7, R_LEN + SHLQ $3, R_TMP0 + ADDQ $4, R_LEN + ORQ R_TMP1, R_TMP0 + + // check if repeat code, ZF set by ORQ. + JZ repeatCode + + // This is a regular copy, transfer our temporary value to R_OFF (length) + MOVQ R_TMP0, R_OFF + JMP doCopy + +// This is a repeat code. +repeatCode: + // If length < 9, reuse last offset, with the length already calculated. + CMPQ R_LEN, $9 + JL doCopyRepeat + + // Read additional bytes for length. + JE repeatLen1 + + // Rare, so the extra branch shouldn't hurt too much. + CMPQ R_LEN, $10 + JE repeatLen2 + JMP repeatLen3 + +// Read repeat lengths. +repeatLen1: + // s ++ + ADDQ $1, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = src[s-1] + 8 + MOVBQZX -1(R_SRC), R_LEN + ADDL $8, R_LEN + JMP doCopyRepeat + +repeatLen2: + // s +=2 + ADDQ $2, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = uint32(src[s-2]) | (uint32(src[s-1])<<8) + (1 << 8) + MOVWQZX -2(R_SRC), R_LEN + ADDL $260, R_LEN + JMP doCopyRepeat + +repeatLen3: + // s +=3 + ADDQ $3, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = uint32(src[s-3]) | (uint32(src[s-2])<<8) | (uint32(src[s-1])<<16) + (1 << 16) + // Read one byte further back (just part of the tag, shifted out) + MOVL -4(R_SRC), R_LEN + SHRL $8, R_LEN + ADDL $65540, R_LEN + JMP doCopyRepeat + +doCopy: + // This is the end of the outer "switch", when we have a copy tag. + // + // We assume that: + // - R_LEN == length && R_LEN > 0 + // - R_OFF == offset + + // if d < offset { etc } + MOVQ R_DST, R_TMP1 + SUBQ R_DBASE, R_TMP1 + CMPQ R_TMP1, R_OFF + JLT errCorrupt + + // Repeat values can skip the test above, since any offset > 0 will be in dst. +doCopyRepeat: + // if offset <= 0 { etc } + CMPQ R_OFF, $0 + JLE errCorrupt + + // if length > len(dst)-d { etc } + MOVQ R_DEND, R_TMP1 + SUBQ R_DST, R_TMP1 + CMPQ R_LEN, R_TMP1 + JGT errCorrupt + + // forwardCopy(dst[d:d+length], dst[d-offset:]); d += length + // + // Set: + // - R_TMP2 = len(dst)-d + // - R_TMP3 = &dst[d-offset] + MOVQ R_DEND, R_TMP2 + SUBQ R_DST, R_TMP2 + MOVQ R_DST, R_TMP3 + SUBQ R_OFF, R_TMP3 + + // !!! Try a faster technique for short (16 or fewer bytes) forward copies. + // + // First, try using two 8-byte load/stores, similar to the doLit technique + // above. Even if dst[d:d+length] and dst[d-offset:] can overlap, this is + // still OK if offset >= 8. Note that this has to be two 8-byte load/stores + // and not one 16-byte load/store, and the first store has to be before the + // second load, due to the overlap if offset is in the range [8, 16). + // + // if length > 16 || offset < 8 || len(dst)-d < 16 { + // goto slowForwardCopy + // } + // copy 16 bytes + // d += length + CMPQ R_LEN, $16 + JGT slowForwardCopy + CMPQ R_OFF, $8 + JLT slowForwardCopy + CMPQ R_TMP2, $16 + JLT slowForwardCopy + MOVQ 0(R_TMP3), R_TMP0 + MOVQ R_TMP0, 0(R_DST) + MOVQ 8(R_TMP3), R_TMP1 + MOVQ R_TMP1, 8(R_DST) + ADDQ R_LEN, R_DST + JMP loop + +slowForwardCopy: + // !!! If the forward copy is longer than 16 bytes, or if offset < 8, we + // can still try 8-byte load stores, provided we can overrun up to 10 extra + // bytes. As above, the overrun will be fixed up by subsequent iterations + // of the outermost loop. + // + // The C++ snappy code calls this technique IncrementalCopyFastPath. Its + // commentary says: + // + // ---- + // + // The main part of this loop is a simple copy of eight bytes at a time + // until we've copied (at least) the requested amount of bytes. However, + // if d and d-offset are less than eight bytes apart (indicating a + // repeating pattern of length < 8), we first need to expand the pattern in + // order to get the correct results. For instance, if the buffer looks like + // this, with the eight-byte and patterns marked as + // intervals: + // + // abxxxxxxxxxxxx + // [------] d-offset + // [------] d + // + // a single eight-byte copy from to will repeat the pattern + // once, after which we can move two bytes without moving : + // + // ababxxxxxxxxxx + // [------] d-offset + // [------] d + // + // and repeat the exercise until the two no longer overlap. + // + // This allows us to do very well in the special case of one single byte + // repeated many times, without taking a big hit for more general cases. + // + // The worst case of extra writing past the end of the match occurs when + // offset == 1 and length == 1; the last copy will read from byte positions + // [0..7] and write to [4..11], whereas it was only supposed to write to + // position 1. Thus, ten excess bytes. + // + // ---- + // + // That "10 byte overrun" worst case is confirmed by Go's + // TestSlowForwardCopyOverrun, which also tests the fixUpSlowForwardCopy + // and finishSlowForwardCopy algorithm. + // + // if length > len(dst)-d-10 { + // goto verySlowForwardCopy + // } + SUBQ $10, R_TMP2 + CMPQ R_LEN, R_TMP2 + JGT verySlowForwardCopy + + // We want to keep the offset, so we use R_TMP2 from here. + MOVQ R_OFF, R_TMP2 + +makeOffsetAtLeast8: + // !!! As above, expand the pattern so that offset >= 8 and we can use + // 8-byte load/stores. + // + // for offset < 8 { + // copy 8 bytes from dst[d-offset:] to dst[d:] + // length -= offset + // d += offset + // offset += offset + // // The two previous lines together means that d-offset, and therefore + // // R_TMP3, is unchanged. + // } + CMPQ R_TMP2, $8 + JGE fixUpSlowForwardCopy + MOVQ (R_TMP3), R_TMP1 + MOVQ R_TMP1, (R_DST) + SUBQ R_TMP2, R_LEN + ADDQ R_TMP2, R_DST + ADDQ R_TMP2, R_TMP2 + JMP makeOffsetAtLeast8 + +fixUpSlowForwardCopy: + // !!! Add length (which might be negative now) to d (implied by R_DST being + // &dst[d]) so that d ends up at the right place when we jump back to the + // top of the loop. Before we do that, though, we save R_DST to R_TMP0 so that, if + // length is positive, copying the remaining length bytes will write to the + // right place. + MOVQ R_DST, R_TMP0 + ADDQ R_LEN, R_DST + +finishSlowForwardCopy: + // !!! Repeat 8-byte load/stores until length <= 0. Ending with a negative + // length means that we overrun, but as above, that will be fixed up by + // subsequent iterations of the outermost loop. + CMPQ R_LEN, $0 + JLE loop + MOVQ (R_TMP3), R_TMP1 + MOVQ R_TMP1, (R_TMP0) + ADDQ $8, R_TMP3 + ADDQ $8, R_TMP0 + SUBQ $8, R_LEN + JMP finishSlowForwardCopy + +verySlowForwardCopy: + // verySlowForwardCopy is a simple implementation of forward copy. In C + // parlance, this is a do/while loop instead of a while loop, since we know + // that length > 0. In Go syntax: + // + // for { + // dst[d] = dst[d - offset] + // d++ + // length-- + // if length == 0 { + // break + // } + // } + MOVB (R_TMP3), R_TMP1 + MOVB R_TMP1, (R_DST) + INCQ R_TMP3 + INCQ R_DST + DECQ R_LEN + JNZ verySlowForwardCopy + JMP loop + +// The code above handles copy tags. +// ---------------------------------------- + +end: + // This is the end of the "for s < len(src)". + // + // if d != len(dst) { etc } + CMPQ R_DST, R_DEND + JNE errCorrupt + + // return 0 + MOVQ $0, ret+48(FP) + RET + +errCorrupt: + // return decodeErrCodeCorrupt + MOVQ $1, ret+48(FP) + RET diff --git a/vendor/github.com/klauspost/compress/s2/decode_arm64.s b/vendor/github.com/klauspost/compress/s2/decode_arm64.s new file mode 100644 index 0000000..4b63d50 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/decode_arm64.s @@ -0,0 +1,574 @@ +// Copyright 2020 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// +build !appengine +// +build gc +// +build !noasm + +#include "textflag.h" + +#define R_TMP0 R2 +#define R_TMP1 R3 +#define R_LEN R4 +#define R_OFF R5 +#define R_SRC R6 +#define R_DST R7 +#define R_DBASE R8 +#define R_DLEN R9 +#define R_DEND R10 +#define R_SBASE R11 +#define R_SLEN R12 +#define R_SEND R13 +#define R_TMP2 R14 +#define R_TMP3 R15 + +// TEST_SRC will check if R_SRC is <= SRC_END +#define TEST_SRC() \ + CMP R_SEND, R_SRC \ + BGT errCorrupt + +// MOVD R_SRC, R_TMP1 +// SUB R_SBASE, R_TMP1, R_TMP1 +// CMP R_SLEN, R_TMP1 +// BGT errCorrupt + +// The asm code generally follows the pure Go code in decode_other.go, except +// where marked with a "!!!". + +// func decode(dst, src []byte) int +// +// All local variables fit into registers. The non-zero stack size is only to +// spill registers and push args when issuing a CALL. The register allocation: +// - R_TMP0 scratch +// - R_TMP1 scratch +// - R_LEN length or x +// - R_OFF offset +// - R_SRC &src[s] +// - R_DST &dst[d] +// + R_DBASE dst_base +// + R_DLEN dst_len +// + R_DEND dst_base + dst_len +// + R_SBASE src_base +// + R_SLEN src_len +// + R_SEND src_base + src_len +// - R_TMP2 used by doCopy +// - R_TMP3 used by doCopy +// +// The registers R_DBASE-R_SEND (marked with a "+") are set at the start of the +// function, and after a CALL returns, and are not otherwise modified. +// +// The d variable is implicitly R_DST - R_DBASE, and len(dst)-d is R_DEND - R_DST. +// The s variable is implicitly R_SRC - R_SBASE, and len(src)-s is R_SEND - R_SRC. +TEXT ·s2Decode(SB), NOSPLIT, $56-64 + // Initialize R_SRC, R_DST and R_DBASE-R_SEND. + MOVD dst_base+0(FP), R_DBASE + MOVD dst_len+8(FP), R_DLEN + MOVD R_DBASE, R_DST + MOVD R_DBASE, R_DEND + ADD R_DLEN, R_DEND, R_DEND + MOVD src_base+24(FP), R_SBASE + MOVD src_len+32(FP), R_SLEN + MOVD R_SBASE, R_SRC + MOVD R_SBASE, R_SEND + ADD R_SLEN, R_SEND, R_SEND + MOVD $0, R_OFF + +loop: + // for s < len(src) + CMP R_SEND, R_SRC + BEQ end + + // R_LEN = uint32(src[s]) + // + // switch src[s] & 0x03 + MOVBU (R_SRC), R_LEN + MOVW R_LEN, R_TMP1 + ANDW $3, R_TMP1 + MOVW $1, R1 + CMPW R1, R_TMP1 + BGE tagCopy + + // ---------------------------------------- + // The code below handles literal tags. + + // case tagLiteral: + // x := uint32(src[s] >> 2) + // switch + MOVW $60, R1 + LSRW $2, R_LEN, R_LEN + CMPW R_LEN, R1 + BLS tagLit60Plus + + // case x < 60: + // s++ + ADD $1, R_SRC, R_SRC + +doLit: + // This is the end of the inner "switch", when we have a literal tag. + // + // We assume that R_LEN == x and x fits in a uint32, where x is the variable + // used in the pure Go decode_other.go code. + + // length = int(x) + 1 + // + // Unlike the pure Go code, we don't need to check if length <= 0 because + // R_LEN can hold 64 bits, so the increment cannot overflow. + ADD $1, R_LEN, R_LEN + + // Prepare to check if copying length bytes will run past the end of dst or + // src. + // + // R_TMP0 = len(dst) - d + // R_TMP1 = len(src) - s + MOVD R_DEND, R_TMP0 + SUB R_DST, R_TMP0, R_TMP0 + MOVD R_SEND, R_TMP1 + SUB R_SRC, R_TMP1, R_TMP1 + + // !!! Try a faster technique for short (16 or fewer bytes) copies. + // + // if length > 16 || len(dst)-d < 16 || len(src)-s < 16 { + // goto callMemmove // Fall back on calling runtime·memmove. + // } + // + // The C++ snappy code calls this TryFastAppend. It also checks len(src)-s + // against 21 instead of 16, because it cannot assume that all of its input + // is contiguous in memory and so it needs to leave enough source bytes to + // read the next tag without refilling buffers, but Go's Decode assumes + // contiguousness (the src argument is a []byte). + CMP $16, R_LEN + BGT callMemmove + CMP $16, R_TMP0 + BLT callMemmove + CMP $16, R_TMP1 + BLT callMemmove + + // !!! Implement the copy from src to dst as a 16-byte load and store. + // (Decode's documentation says that dst and src must not overlap.) + // + // This always copies 16 bytes, instead of only length bytes, but that's + // OK. If the input is a valid Snappy encoding then subsequent iterations + // will fix up the overrun. Otherwise, Decode returns a nil []byte (and a + // non-nil error), so the overrun will be ignored. + // + // Note that on arm64, it is legal and cheap to issue unaligned 8-byte or + // 16-byte loads and stores. This technique probably wouldn't be as + // effective on architectures that are fussier about alignment. + LDP 0(R_SRC), (R_TMP2, R_TMP3) + STP (R_TMP2, R_TMP3), 0(R_DST) + + // d += length + // s += length + ADD R_LEN, R_DST, R_DST + ADD R_LEN, R_SRC, R_SRC + B loop + +callMemmove: + // if length > len(dst)-d || length > len(src)-s { etc } + CMP R_TMP0, R_LEN + BGT errCorrupt + CMP R_TMP1, R_LEN + BGT errCorrupt + + // copy(dst[d:], src[s:s+length]) + // + // This means calling runtime·memmove(&dst[d], &src[s], length), so we push + // R_DST, R_SRC and R_LEN as arguments. Coincidentally, we also need to spill those + // three registers to the stack, to save local variables across the CALL. + MOVD R_DST, 8(RSP) + MOVD R_SRC, 16(RSP) + MOVD R_LEN, 24(RSP) + MOVD R_DST, 32(RSP) + MOVD R_SRC, 40(RSP) + MOVD R_LEN, 48(RSP) + MOVD R_OFF, 56(RSP) + CALL runtime·memmove(SB) + + // Restore local variables: unspill registers from the stack and + // re-calculate R_DBASE-R_SEND. + MOVD 32(RSP), R_DST + MOVD 40(RSP), R_SRC + MOVD 48(RSP), R_LEN + MOVD 56(RSP), R_OFF + MOVD dst_base+0(FP), R_DBASE + MOVD dst_len+8(FP), R_DLEN + MOVD R_DBASE, R_DEND + ADD R_DLEN, R_DEND, R_DEND + MOVD src_base+24(FP), R_SBASE + MOVD src_len+32(FP), R_SLEN + MOVD R_SBASE, R_SEND + ADD R_SLEN, R_SEND, R_SEND + + // d += length + // s += length + ADD R_LEN, R_DST, R_DST + ADD R_LEN, R_SRC, R_SRC + B loop + +tagLit60Plus: + // !!! This fragment does the + // + // s += x - 58; if uint(s) > uint(len(src)) { etc } + // + // checks. In the asm version, we code it once instead of once per switch case. + ADD R_LEN, R_SRC, R_SRC + SUB $58, R_SRC, R_SRC + TEST_SRC() + + // case x == 60: + MOVW $61, R1 + CMPW R1, R_LEN + BEQ tagLit61 + BGT tagLit62Plus + + // x = uint32(src[s-1]) + MOVBU -1(R_SRC), R_LEN + B doLit + +tagLit61: + // case x == 61: + // x = uint32(src[s-2]) | uint32(src[s-1])<<8 + MOVHU -2(R_SRC), R_LEN + B doLit + +tagLit62Plus: + CMPW $62, R_LEN + BHI tagLit63 + + // case x == 62: + // x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + MOVHU -3(R_SRC), R_LEN + MOVBU -1(R_SRC), R_TMP1 + ORR R_TMP1<<16, R_LEN + B doLit + +tagLit63: + // case x == 63: + // x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + MOVWU -4(R_SRC), R_LEN + B doLit + + // The code above handles literal tags. + // ---------------------------------------- + // The code below handles copy tags. + +tagCopy4: + // case tagCopy4: + // s += 5 + ADD $5, R_SRC, R_SRC + + // if uint(s) > uint(len(src)) { etc } + MOVD R_SRC, R_TMP1 + SUB R_SBASE, R_TMP1, R_TMP1 + CMP R_SLEN, R_TMP1 + BGT errCorrupt + + // length = 1 + int(src[s-5])>>2 + MOVD $1, R1 + ADD R_LEN>>2, R1, R_LEN + + // offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + MOVWU -4(R_SRC), R_OFF + B doCopy + +tagCopy2: + // case tagCopy2: + // s += 3 + ADD $3, R_SRC, R_SRC + + // if uint(s) > uint(len(src)) { etc } + TEST_SRC() + + // length = 1 + int(src[s-3])>>2 + MOVD $1, R1 + ADD R_LEN>>2, R1, R_LEN + + // offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + MOVHU -2(R_SRC), R_OFF + B doCopy + +tagCopy: + // We have a copy tag. We assume that: + // - R_TMP1 == src[s] & 0x03 + // - R_LEN == src[s] + CMP $2, R_TMP1 + BEQ tagCopy2 + BGT tagCopy4 + + // case tagCopy1: + // s += 2 + ADD $2, R_SRC, R_SRC + + // if uint(s) > uint(len(src)) { etc } + TEST_SRC() + + // offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + // Calculate offset in R_TMP0 in case it is a repeat. + MOVD R_LEN, R_TMP0 + AND $0xe0, R_TMP0 + MOVBU -1(R_SRC), R_TMP1 + ORR R_TMP0<<3, R_TMP1, R_TMP0 + + // length = 4 + int(src[s-2])>>2&0x7 + MOVD $7, R1 + AND R_LEN>>2, R1, R_LEN + ADD $4, R_LEN, R_LEN + + // check if repeat code with offset 0. + CMP $0, R_TMP0 + BEQ repeatCode + + // This is a regular copy, transfer our temporary value to R_OFF (offset) + MOVD R_TMP0, R_OFF + B doCopy + + // This is a repeat code. +repeatCode: + // If length < 9, reuse last offset, with the length already calculated. + CMP $9, R_LEN + BLT doCopyRepeat + BEQ repeatLen1 + CMP $10, R_LEN + BEQ repeatLen2 + +repeatLen3: + // s +=3 + ADD $3, R_SRC, R_SRC + + // if uint(s) > uint(len(src)) { etc } + TEST_SRC() + + // length = uint32(src[s-3]) | (uint32(src[s-2])<<8) | (uint32(src[s-1])<<16) + 65540 + MOVBU -1(R_SRC), R_TMP0 + MOVHU -3(R_SRC), R_LEN + ORR R_TMP0<<16, R_LEN, R_LEN + ADD $65540, R_LEN, R_LEN + B doCopyRepeat + +repeatLen2: + // s +=2 + ADD $2, R_SRC, R_SRC + + // if uint(s) > uint(len(src)) { etc } + TEST_SRC() + + // length = uint32(src[s-2]) | (uint32(src[s-1])<<8) + 260 + MOVHU -2(R_SRC), R_LEN + ADD $260, R_LEN, R_LEN + B doCopyRepeat + +repeatLen1: + // s +=1 + ADD $1, R_SRC, R_SRC + + // if uint(s) > uint(len(src)) { etc } + TEST_SRC() + + // length = src[s-1] + 8 + MOVBU -1(R_SRC), R_LEN + ADD $8, R_LEN, R_LEN + B doCopyRepeat + +doCopy: + // This is the end of the outer "switch", when we have a copy tag. + // + // We assume that: + // - R_LEN == length && R_LEN > 0 + // - R_OFF == offset + + // if d < offset { etc } + MOVD R_DST, R_TMP1 + SUB R_DBASE, R_TMP1, R_TMP1 + CMP R_OFF, R_TMP1 + BLT errCorrupt + + // Repeat values can skip the test above, since any offset > 0 will be in dst. +doCopyRepeat: + + // if offset <= 0 { etc } + CMP $0, R_OFF + BLE errCorrupt + + // if length > len(dst)-d { etc } + MOVD R_DEND, R_TMP1 + SUB R_DST, R_TMP1, R_TMP1 + CMP R_TMP1, R_LEN + BGT errCorrupt + + // forwardCopy(dst[d:d+length], dst[d-offset:]); d += length + // + // Set: + // - R_TMP2 = len(dst)-d + // - R_TMP3 = &dst[d-offset] + MOVD R_DEND, R_TMP2 + SUB R_DST, R_TMP2, R_TMP2 + MOVD R_DST, R_TMP3 + SUB R_OFF, R_TMP3, R_TMP3 + + // !!! Try a faster technique for short (16 or fewer bytes) forward copies. + // + // First, try using two 8-byte load/stores, similar to the doLit technique + // above. Even if dst[d:d+length] and dst[d-offset:] can overlap, this is + // still OK if offset >= 8. Note that this has to be two 8-byte load/stores + // and not one 16-byte load/store, and the first store has to be before the + // second load, due to the overlap if offset is in the range [8, 16). + // + // if length > 16 || offset < 8 || len(dst)-d < 16 { + // goto slowForwardCopy + // } + // copy 16 bytes + // d += length + CMP $16, R_LEN + BGT slowForwardCopy + CMP $8, R_OFF + BLT slowForwardCopy + CMP $16, R_TMP2 + BLT slowForwardCopy + MOVD 0(R_TMP3), R_TMP0 + MOVD R_TMP0, 0(R_DST) + MOVD 8(R_TMP3), R_TMP1 + MOVD R_TMP1, 8(R_DST) + ADD R_LEN, R_DST, R_DST + B loop + +slowForwardCopy: + // !!! If the forward copy is longer than 16 bytes, or if offset < 8, we + // can still try 8-byte load stores, provided we can overrun up to 10 extra + // bytes. As above, the overrun will be fixed up by subsequent iterations + // of the outermost loop. + // + // The C++ snappy code calls this technique IncrementalCopyFastPath. Its + // commentary says: + // + // ---- + // + // The main part of this loop is a simple copy of eight bytes at a time + // until we've copied (at least) the requested amount of bytes. However, + // if d and d-offset are less than eight bytes apart (indicating a + // repeating pattern of length < 8), we first need to expand the pattern in + // order to get the correct results. For instance, if the buffer looks like + // this, with the eight-byte and patterns marked as + // intervals: + // + // abxxxxxxxxxxxx + // [------] d-offset + // [------] d + // + // a single eight-byte copy from to will repeat the pattern + // once, after which we can move two bytes without moving : + // + // ababxxxxxxxxxx + // [------] d-offset + // [------] d + // + // and repeat the exercise until the two no longer overlap. + // + // This allows us to do very well in the special case of one single byte + // repeated many times, without taking a big hit for more general cases. + // + // The worst case of extra writing past the end of the match occurs when + // offset == 1 and length == 1; the last copy will read from byte positions + // [0..7] and write to [4..11], whereas it was only supposed to write to + // position 1. Thus, ten excess bytes. + // + // ---- + // + // That "10 byte overrun" worst case is confirmed by Go's + // TestSlowForwardCopyOverrun, which also tests the fixUpSlowForwardCopy + // and finishSlowForwardCopy algorithm. + // + // if length > len(dst)-d-10 { + // goto verySlowForwardCopy + // } + SUB $10, R_TMP2, R_TMP2 + CMP R_TMP2, R_LEN + BGT verySlowForwardCopy + + // We want to keep the offset, so we use R_TMP2 from here. + MOVD R_OFF, R_TMP2 + +makeOffsetAtLeast8: + // !!! As above, expand the pattern so that offset >= 8 and we can use + // 8-byte load/stores. + // + // for offset < 8 { + // copy 8 bytes from dst[d-offset:] to dst[d:] + // length -= offset + // d += offset + // offset += offset + // // The two previous lines together means that d-offset, and therefore + // // R_TMP3, is unchanged. + // } + CMP $8, R_TMP2 + BGE fixUpSlowForwardCopy + MOVD (R_TMP3), R_TMP1 + MOVD R_TMP1, (R_DST) + SUB R_TMP2, R_LEN, R_LEN + ADD R_TMP2, R_DST, R_DST + ADD R_TMP2, R_TMP2, R_TMP2 + B makeOffsetAtLeast8 + +fixUpSlowForwardCopy: + // !!! Add length (which might be negative now) to d (implied by R_DST being + // &dst[d]) so that d ends up at the right place when we jump back to the + // top of the loop. Before we do that, though, we save R_DST to R_TMP0 so that, if + // length is positive, copying the remaining length bytes will write to the + // right place. + MOVD R_DST, R_TMP0 + ADD R_LEN, R_DST, R_DST + +finishSlowForwardCopy: + // !!! Repeat 8-byte load/stores until length <= 0. Ending with a negative + // length means that we overrun, but as above, that will be fixed up by + // subsequent iterations of the outermost loop. + MOVD $0, R1 + CMP R1, R_LEN + BLE loop + MOVD (R_TMP3), R_TMP1 + MOVD R_TMP1, (R_TMP0) + ADD $8, R_TMP3, R_TMP3 + ADD $8, R_TMP0, R_TMP0 + SUB $8, R_LEN, R_LEN + B finishSlowForwardCopy + +verySlowForwardCopy: + // verySlowForwardCopy is a simple implementation of forward copy. In C + // parlance, this is a do/while loop instead of a while loop, since we know + // that length > 0. In Go syntax: + // + // for { + // dst[d] = dst[d - offset] + // d++ + // length-- + // if length == 0 { + // break + // } + // } + MOVB (R_TMP3), R_TMP1 + MOVB R_TMP1, (R_DST) + ADD $1, R_TMP3, R_TMP3 + ADD $1, R_DST, R_DST + SUB $1, R_LEN, R_LEN + CBNZ R_LEN, verySlowForwardCopy + B loop + + // The code above handles copy tags. + // ---------------------------------------- + +end: + // This is the end of the "for s < len(src)". + // + // if d != len(dst) { etc } + CMP R_DEND, R_DST + BNE errCorrupt + + // return 0 + MOVD $0, ret+48(FP) + RET + +errCorrupt: + // return decodeErrCodeCorrupt + MOVD $1, R_TMP0 + MOVD R_TMP0, ret+48(FP) + RET diff --git a/vendor/github.com/klauspost/compress/s2/decode_asm.go b/vendor/github.com/klauspost/compress/s2/decode_asm.go new file mode 100644 index 0000000..cb3576e --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/decode_asm.go @@ -0,0 +1,17 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +//go:build (amd64 || arm64) && !appengine && gc && !noasm +// +build amd64 arm64 +// +build !appengine +// +build gc +// +build !noasm + +package s2 + +// decode has the same semantics as in decode_other.go. +// +//go:noescape +func s2Decode(dst, src []byte) int diff --git a/vendor/github.com/klauspost/compress/s2/decode_other.go b/vendor/github.com/klauspost/compress/s2/decode_other.go new file mode 100644 index 0000000..1074ebd --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/decode_other.go @@ -0,0 +1,267 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +//go:build (!amd64 && !arm64) || appengine || !gc || noasm +// +build !amd64,!arm64 appengine !gc noasm + +package s2 + +import ( + "fmt" + "strconv" +) + +// decode writes the decoding of src to dst. It assumes that the varint-encoded +// length of the decompressed bytes has already been read, and that len(dst) +// equals that length. +// +// It returns 0 on success or a decodeErrCodeXxx error code on failure. +func s2Decode(dst, src []byte) int { + const debug = false + if debug { + fmt.Println("Starting decode, dst len:", len(dst)) + } + var d, s, length int + offset := 0 + + // As long as we can read at least 5 bytes... + for s < len(src)-5 { + switch src[s] & 0x03 { + case tagLiteral: + x := uint32(src[s] >> 2) + switch { + case x < 60: + s++ + case x == 60: + s += 2 + x = uint32(src[s-1]) + case x == 61: + s += 3 + x = uint32(src[s-2]) | uint32(src[s-1])<<8 + case x == 62: + s += 4 + x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + case x == 63: + s += 5 + x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + length = int(x) + 1 + if length > len(dst)-d || length > len(src)-s || (strconv.IntSize == 32 && length <= 0) { + return decodeErrCodeCorrupt + } + if debug { + fmt.Println("literals, length:", length, "d-after:", d+length) + } + + copy(dst[d:], src[s:s+length]) + d += length + s += length + continue + + case tagCopy1: + s += 2 + length = int(src[s-2]) >> 2 & 0x7 + toffset := int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + if toffset == 0 { + if debug { + fmt.Print("(repeat) ") + } + // keep last offset + switch length { + case 5: + s += 1 + length = int(uint32(src[s-1])) + 4 + case 6: + s += 2 + length = int(uint32(src[s-2])|(uint32(src[s-1])<<8)) + (1 << 8) + case 7: + s += 3 + length = int(uint32(src[s-3])|(uint32(src[s-2])<<8)|(uint32(src[s-1])<<16)) + (1 << 16) + default: // 0-> 4 + } + } else { + offset = toffset + } + length += 4 + case tagCopy2: + s += 3 + length = 1 + int(src[s-3])>>2 + offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + + case tagCopy4: + s += 5 + length = 1 + int(src[s-5])>>2 + offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + } + + if offset <= 0 || d < offset || length > len(dst)-d { + return decodeErrCodeCorrupt + } + + if debug { + fmt.Println("copy, length:", length, "offset:", offset, "d-after:", d+length) + } + + // Copy from an earlier sub-slice of dst to a later sub-slice. + // If no overlap, use the built-in copy: + if offset > length { + copy(dst[d:d+length], dst[d-offset:]) + d += length + continue + } + + // Unlike the built-in copy function, this byte-by-byte copy always runs + // forwards, even if the slices overlap. Conceptually, this is: + // + // d += forwardCopy(dst[d:d+length], dst[d-offset:]) + // + // We align the slices into a and b and show the compiler they are the same size. + // This allows the loop to run without bounds checks. + a := dst[d : d+length] + b := dst[d-offset:] + b = b[:len(a)] + for i := range a { + a[i] = b[i] + } + d += length + } + + // Remaining with extra checks... + for s < len(src) { + switch src[s] & 0x03 { + case tagLiteral: + x := uint32(src[s] >> 2) + switch { + case x < 60: + s++ + case x == 60: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-1]) + case x == 61: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-2]) | uint32(src[s-1])<<8 + case x == 62: + s += 4 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + case x == 63: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + length = int(x) + 1 + if length > len(dst)-d || length > len(src)-s || (strconv.IntSize == 32 && length <= 0) { + return decodeErrCodeCorrupt + } + if debug { + fmt.Println("literals, length:", length, "d-after:", d+length) + } + + copy(dst[d:], src[s:s+length]) + d += length + s += length + continue + + case tagCopy1: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(src[s-2]) >> 2 & 0x7 + toffset := int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + if toffset == 0 { + if debug { + fmt.Print("(repeat) ") + } + // keep last offset + switch length { + case 5: + s += 1 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(uint32(src[s-1])) + 4 + case 6: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(uint32(src[s-2])|(uint32(src[s-1])<<8)) + (1 << 8) + case 7: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(uint32(src[s-3])|(uint32(src[s-2])<<8)|(uint32(src[s-1])<<16)) + (1 << 16) + default: // 0-> 4 + } + } else { + offset = toffset + } + length += 4 + case tagCopy2: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = 1 + int(src[s-3])>>2 + offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + + case tagCopy4: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = 1 + int(src[s-5])>>2 + offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + } + + if offset <= 0 || d < offset || length > len(dst)-d { + return decodeErrCodeCorrupt + } + + if debug { + fmt.Println("copy, length:", length, "offset:", offset, "d-after:", d+length) + } + + // Copy from an earlier sub-slice of dst to a later sub-slice. + // If no overlap, use the built-in copy: + if offset > length { + copy(dst[d:d+length], dst[d-offset:]) + d += length + continue + } + + // Unlike the built-in copy function, this byte-by-byte copy always runs + // forwards, even if the slices overlap. Conceptually, this is: + // + // d += forwardCopy(dst[d:d+length], dst[d-offset:]) + // + // We align the slices into a and b and show the compiler they are the same size. + // This allows the loop to run without bounds checks. + a := dst[d : d+length] + b := dst[d-offset:] + b = b[:len(a)] + for i := range a { + a[i] = b[i] + } + d += length + } + + if d != len(dst) { + return decodeErrCodeCorrupt + } + return 0 +} diff --git a/vendor/github.com/klauspost/compress/s2/encode.go b/vendor/github.com/klauspost/compress/s2/encode.go new file mode 100644 index 0000000..1aefabf --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encode.go @@ -0,0 +1,1341 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package s2 + +import ( + "crypto/rand" + "encoding/binary" + "errors" + "fmt" + "io" + "math" + "math/bits" + "runtime" + "sync" +) + +// Encode returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func Encode(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if cap(dst) < n { + dst = make([]byte, n) + } else { + dst = dst[:n] + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlock(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeBetter returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// EncodeBetter compresses better than Encode but typically with a +// 10-40% speed decrease on both compression and decompression. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeBetter(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if len(dst) < n { + dst = make([]byte, n) + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlockBetter(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeBest returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// EncodeBest compresses as good as reasonably possible but with a +// big speed decrease. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeBest(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if len(dst) < n { + dst = make([]byte, n) + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlockBest(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeSnappy returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The output is Snappy compatible and will likely decompress faster. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeSnappy(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if cap(dst) < n { + dst = make([]byte, n) + } else { + dst = dst[:n] + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + + n := encodeBlockSnappy(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeSnappyBetter returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The output is Snappy compatible and will likely decompress faster. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeSnappyBetter(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if cap(dst) < n { + dst = make([]byte, n) + } else { + dst = dst[:n] + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + + n := encodeBlockBetterSnappy(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeSnappyBest returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The output is Snappy compatible and will likely decompress faster. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeSnappyBest(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if cap(dst) < n { + dst = make([]byte, n) + } else { + dst = dst[:n] + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + + n := encodeBlockBestSnappy(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// ConcatBlocks will concatenate the supplied blocks and append them to the supplied destination. +// If the destination is nil or too small, a new will be allocated. +// The blocks are not validated, so garbage in = garbage out. +// dst may not overlap block data. +// Any data in dst is preserved as is, so it will not be considered a block. +func ConcatBlocks(dst []byte, blocks ...[]byte) ([]byte, error) { + totalSize := uint64(0) + compSize := 0 + for _, b := range blocks { + l, hdr, err := decodedLen(b) + if err != nil { + return nil, err + } + totalSize += uint64(l) + compSize += len(b) - hdr + } + if totalSize == 0 { + dst = append(dst, 0) + return dst, nil + } + if totalSize > math.MaxUint32 { + return nil, ErrTooLarge + } + var tmp [binary.MaxVarintLen32]byte + hdrSize := binary.PutUvarint(tmp[:], totalSize) + wantSize := hdrSize + compSize + + if cap(dst)-len(dst) < wantSize { + dst = append(make([]byte, 0, wantSize+len(dst)), dst...) + } + dst = append(dst, tmp[:hdrSize]...) + for _, b := range blocks { + _, hdr, err := decodedLen(b) + if err != nil { + return nil, err + } + dst = append(dst, b[hdr:]...) + } + return dst, nil +} + +// inputMargin is the minimum number of extra input bytes to keep, inside +// encodeBlock's inner loop. On some architectures, this margin lets us +// implement a fast path for emitLiteral, where the copy of short (<= 16 byte) +// literals can be implemented as a single load to and store from a 16-byte +// register. That literal's actual length can be as short as 1 byte, so this +// can copy up to 15 bytes too much, but that's OK as subsequent iterations of +// the encoding loop will fix up the copy overrun, and this inputMargin ensures +// that we don't overrun the dst and src buffers. +const inputMargin = 8 + +// minNonLiteralBlockSize is the minimum size of the input to encodeBlock that +// will be accepted by the encoder. +const minNonLiteralBlockSize = 32 + +// MaxBlockSize is the maximum value where MaxEncodedLen will return a valid block size. +// Blocks this big are highly discouraged, though. +const MaxBlockSize = math.MaxUint32 - binary.MaxVarintLen32 - 5 + +// MaxEncodedLen returns the maximum length of a snappy block, given its +// uncompressed length. +// +// It will return a negative value if srcLen is too large to encode. +// 32 bit platforms will have lower thresholds for rejecting big content. +func MaxEncodedLen(srcLen int) int { + n := uint64(srcLen) + if n > 0xffffffff { + // Also includes negative. + return -1 + } + // Size of the varint encoded block size. + n = n + uint64((bits.Len64(n)+7)/7) + + // Add maximum size of encoding block as literals. + n += uint64(literalExtraSize(int64(srcLen))) + if n > 0xffffffff { + return -1 + } + return int(n) +} + +var errClosed = errors.New("s2: Writer is closed") + +// NewWriter returns a new Writer that compresses to w, using the +// framing format described at +// https://github.com/google/snappy/blob/master/framing_format.txt +// +// Users must call Close to guarantee all data has been forwarded to +// the underlying io.Writer and that resources are released. +// They may also call Flush zero or more times before calling Close. +func NewWriter(w io.Writer, opts ...WriterOption) *Writer { + w2 := Writer{ + blockSize: defaultBlockSize, + concurrency: runtime.GOMAXPROCS(0), + randSrc: rand.Reader, + level: levelFast, + } + for _, opt := range opts { + if err := opt(&w2); err != nil { + w2.errState = err + return &w2 + } + } + w2.obufLen = obufHeaderLen + MaxEncodedLen(w2.blockSize) + w2.paramsOK = true + w2.ibuf = make([]byte, 0, w2.blockSize) + w2.buffers.New = func() interface{} { + return make([]byte, w2.obufLen) + } + w2.Reset(w) + return &w2 +} + +// Writer is an io.Writer that can write Snappy-compressed bytes. +type Writer struct { + errMu sync.Mutex + errState error + + // ibuf is a buffer for the incoming (uncompressed) bytes. + ibuf []byte + + blockSize int + obufLen int + concurrency int + written int64 + uncompWritten int64 // Bytes sent to compression + output chan chan result + buffers sync.Pool + pad int + + writer io.Writer + randSrc io.Reader + writerWg sync.WaitGroup + index Index + + // wroteStreamHeader is whether we have written the stream header. + wroteStreamHeader bool + paramsOK bool + snappy bool + flushOnWrite bool + appendIndex bool + level uint8 +} + +const ( + levelUncompressed = iota + 1 + levelFast + levelBetter + levelBest +) + +type result struct { + b []byte + // Uncompressed start offset + startOffset int64 +} + +// err returns the previously set error. +// If no error has been set it is set to err if not nil. +func (w *Writer) err(err error) error { + w.errMu.Lock() + errSet := w.errState + if errSet == nil && err != nil { + w.errState = err + errSet = err + } + w.errMu.Unlock() + return errSet +} + +// Reset discards the writer's state and switches the Snappy writer to write to w. +// This permits reusing a Writer rather than allocating a new one. +func (w *Writer) Reset(writer io.Writer) { + if !w.paramsOK { + return + } + // Close previous writer, if any. + if w.output != nil { + close(w.output) + w.writerWg.Wait() + w.output = nil + } + w.errState = nil + w.ibuf = w.ibuf[:0] + w.wroteStreamHeader = false + w.written = 0 + w.writer = writer + w.uncompWritten = 0 + w.index.reset(w.blockSize) + + // If we didn't get a writer, stop here. + if writer == nil { + return + } + // If no concurrency requested, don't spin up writer goroutine. + if w.concurrency == 1 { + return + } + + toWrite := make(chan chan result, w.concurrency) + w.output = toWrite + w.writerWg.Add(1) + + // Start a writer goroutine that will write all output in order. + go func() { + defer w.writerWg.Done() + + // Get a queued write. + for write := range toWrite { + // Wait for the data to be available. + input := <-write + in := input.b + if len(in) > 0 { + if w.err(nil) == nil { + // Don't expose data from previous buffers. + toWrite := in[:len(in):len(in)] + // Write to output. + n, err := writer.Write(toWrite) + if err == nil && n != len(toWrite) { + err = io.ErrShortBuffer + } + _ = w.err(err) + w.err(w.index.add(w.written, input.startOffset)) + w.written += int64(n) + } + } + if cap(in) >= w.obufLen { + w.buffers.Put(in) + } + // close the incoming write request. + // This can be used for synchronizing flushes. + close(write) + } + }() +} + +// Write satisfies the io.Writer interface. +func (w *Writer) Write(p []byte) (nRet int, errRet error) { + if err := w.err(nil); err != nil { + return 0, err + } + if w.flushOnWrite { + return w.write(p) + } + // If we exceed the input buffer size, start writing + for len(p) > (cap(w.ibuf)-len(w.ibuf)) && w.err(nil) == nil { + var n int + if len(w.ibuf) == 0 { + // Large write, empty buffer. + // Write directly from p to avoid copy. + n, _ = w.write(p) + } else { + n = copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p) + w.ibuf = w.ibuf[:len(w.ibuf)+n] + w.write(w.ibuf) + w.ibuf = w.ibuf[:0] + } + nRet += n + p = p[n:] + } + if err := w.err(nil); err != nil { + return nRet, err + } + // p should always be able to fit into w.ibuf now. + n := copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p) + w.ibuf = w.ibuf[:len(w.ibuf)+n] + nRet += n + return nRet, nil +} + +// ReadFrom implements the io.ReaderFrom interface. +// Using this is typically more efficient since it avoids a memory copy. +// ReadFrom reads data from r until EOF or error. +// The return value n is the number of bytes read. +// Any error except io.EOF encountered during the read is also returned. +func (w *Writer) ReadFrom(r io.Reader) (n int64, err error) { + if err := w.err(nil); err != nil { + return 0, err + } + if len(w.ibuf) > 0 { + err := w.Flush() + if err != nil { + return 0, err + } + } + if br, ok := r.(byter); ok { + buf := br.Bytes() + if err := w.EncodeBuffer(buf); err != nil { + return 0, err + } + return int64(len(buf)), w.Flush() + } + for { + inbuf := w.buffers.Get().([]byte)[:w.blockSize+obufHeaderLen] + n2, err := io.ReadFull(r, inbuf[obufHeaderLen:]) + if err != nil { + if err == io.ErrUnexpectedEOF { + err = io.EOF + } + if err != io.EOF { + return n, w.err(err) + } + } + if n2 == 0 { + break + } + n += int64(n2) + err2 := w.writeFull(inbuf[:n2+obufHeaderLen]) + if w.err(err2) != nil { + break + } + + if err != nil { + // We got EOF and wrote everything + break + } + } + + return n, w.err(nil) +} + +// AddSkippableBlock will add a skippable block to the stream. +// The ID must be 0x80-0xfe (inclusive). +// Length of the skippable block must be <= 16777215 bytes. +func (w *Writer) AddSkippableBlock(id uint8, data []byte) (err error) { + if err := w.err(nil); err != nil { + return err + } + if len(data) == 0 { + return nil + } + if id < 0x80 || id > chunkTypePadding { + return fmt.Errorf("invalid skippable block id %x", id) + } + if len(data) > maxChunkSize { + return fmt.Errorf("skippable block excessed maximum size") + } + var header [4]byte + chunkLen := 4 + len(data) + header[0] = id + header[1] = uint8(chunkLen >> 0) + header[2] = uint8(chunkLen >> 8) + header[3] = uint8(chunkLen >> 16) + if w.concurrency == 1 { + write := func(b []byte) error { + n, err := w.writer.Write(b) + if err = w.err(err); err != nil { + return err + } + if n != len(data) { + return w.err(io.ErrShortWrite) + } + w.written += int64(n) + return w.err(nil) + } + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + if w.snappy { + if err := write([]byte(magicChunkSnappy)); err != nil { + return err + } + } else { + if err := write([]byte(magicChunk)); err != nil { + return err + } + } + } + if err := write(header[:]); err != nil { + return err + } + if err := write(data); err != nil { + return err + } + } + + // Create output... + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + if w.snappy { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunkSnappy)} + } else { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunk)} + } + } + + // Copy input. + inbuf := w.buffers.Get().([]byte)[:4] + copy(inbuf, header[:]) + inbuf = append(inbuf, data...) + + output := make(chan result, 1) + // Queue output. + w.output <- output + output <- result{startOffset: w.uncompWritten, b: inbuf} + + return nil +} + +// EncodeBuffer will add a buffer to the stream. +// This is the fastest way to encode a stream, +// but the input buffer cannot be written to by the caller +// until Flush or Close has been called when concurrency != 1. +// +// If you cannot control that, use the regular Write function. +// +// Note that input is not buffered. +// This means that each write will result in discrete blocks being created. +// For buffered writes, use the regular Write function. +func (w *Writer) EncodeBuffer(buf []byte) (err error) { + if err := w.err(nil); err != nil { + return err + } + + if w.flushOnWrite { + _, err := w.write(buf) + return err + } + // Flush queued data first. + if len(w.ibuf) > 0 { + err := w.Flush() + if err != nil { + return err + } + } + if w.concurrency == 1 { + _, err := w.writeSync(buf) + return err + } + + // Spawn goroutine and write block to output channel. + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + if w.snappy { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunkSnappy)} + } else { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunk)} + } + } + + for len(buf) > 0 { + // Cut input. + uncompressed := buf + if len(uncompressed) > w.blockSize { + uncompressed = uncompressed[:w.blockSize] + } + buf = buf[len(uncompressed):] + // Get an output buffer. + obuf := w.buffers.Get().([]byte)[:len(uncompressed)+obufHeaderLen] + output := make(chan result) + // Queue output now, so we keep order. + w.output <- output + res := result{ + startOffset: w.uncompWritten, + } + w.uncompWritten += int64(len(uncompressed)) + go func() { + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + n2 := w.encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + + // Check if we should use this, or store as uncompressed instead. + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + // copy uncompressed + copy(obuf[obufHeaderLen:], uncompressed) + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + // Queue final output. + res.b = obuf + output <- res + }() + } + return nil +} + +func (w *Writer) encodeBlock(obuf, uncompressed []byte) int { + if w.snappy { + switch w.level { + case levelFast: + return encodeBlockSnappy(obuf, uncompressed) + case levelBetter: + return encodeBlockBetterSnappy(obuf, uncompressed) + case levelBest: + return encodeBlockBestSnappy(obuf, uncompressed) + } + return 0 + } + switch w.level { + case levelFast: + return encodeBlock(obuf, uncompressed) + case levelBetter: + return encodeBlockBetter(obuf, uncompressed) + case levelBest: + return encodeBlockBest(obuf, uncompressed) + } + return 0 +} + +func (w *Writer) write(p []byte) (nRet int, errRet error) { + if err := w.err(nil); err != nil { + return 0, err + } + if w.concurrency == 1 { + return w.writeSync(p) + } + + // Spawn goroutine and write block to output channel. + for len(p) > 0 { + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + if w.snappy { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunkSnappy)} + } else { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunk)} + } + } + + var uncompressed []byte + if len(p) > w.blockSize { + uncompressed, p = p[:w.blockSize], p[w.blockSize:] + } else { + uncompressed, p = p, nil + } + + // Copy input. + // If the block is incompressible, this is used for the result. + inbuf := w.buffers.Get().([]byte)[:len(uncompressed)+obufHeaderLen] + obuf := w.buffers.Get().([]byte)[:w.obufLen] + copy(inbuf[obufHeaderLen:], uncompressed) + uncompressed = inbuf[obufHeaderLen:] + + output := make(chan result) + // Queue output now, so we keep order. + w.output <- output + res := result{ + startOffset: w.uncompWritten, + } + w.uncompWritten += int64(len(uncompressed)) + + go func() { + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + n2 := w.encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + + // Check if we should use this, or store as uncompressed instead. + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + // Use input as output. + obuf, inbuf = inbuf, obuf + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + // Queue final output. + res.b = obuf + output <- res + + // Put unused buffer back in pool. + w.buffers.Put(inbuf) + }() + nRet += len(uncompressed) + } + return nRet, nil +} + +// writeFull is a special version of write that will always write the full buffer. +// Data to be compressed should start at offset obufHeaderLen and fill the remainder of the buffer. +// The data will be written as a single block. +// The caller is not allowed to use inbuf after this function has been called. +func (w *Writer) writeFull(inbuf []byte) (errRet error) { + if err := w.err(nil); err != nil { + return err + } + + if w.concurrency == 1 { + _, err := w.writeSync(inbuf[obufHeaderLen:]) + return err + } + + // Spawn goroutine and write block to output channel. + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + if w.snappy { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunkSnappy)} + } else { + hWriter <- result{startOffset: w.uncompWritten, b: []byte(magicChunk)} + } + } + + // Get an output buffer. + obuf := w.buffers.Get().([]byte)[:w.obufLen] + uncompressed := inbuf[obufHeaderLen:] + + output := make(chan result) + // Queue output now, so we keep order. + w.output <- output + res := result{ + startOffset: w.uncompWritten, + } + w.uncompWritten += int64(len(uncompressed)) + + go func() { + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + n2 := w.encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + + // Check if we should use this, or store as uncompressed instead. + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + // Use input as output. + obuf, inbuf = inbuf, obuf + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + // Queue final output. + res.b = obuf + output <- res + + // Put unused buffer back in pool. + w.buffers.Put(inbuf) + }() + return nil +} + +func (w *Writer) writeSync(p []byte) (nRet int, errRet error) { + if err := w.err(nil); err != nil { + return 0, err + } + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + var n int + var err error + if w.snappy { + n, err = w.writer.Write([]byte(magicChunkSnappy)) + } else { + n, err = w.writer.Write([]byte(magicChunk)) + } + if err != nil { + return 0, w.err(err) + } + if n != len(magicChunk) { + return 0, w.err(io.ErrShortWrite) + } + w.written += int64(n) + } + + for len(p) > 0 { + var uncompressed []byte + if len(p) > w.blockSize { + uncompressed, p = p[:w.blockSize], p[w.blockSize:] + } else { + uncompressed, p = p, nil + } + + obuf := w.buffers.Get().([]byte)[:w.obufLen] + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + n2 := w.encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + obuf = obuf[:8] + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + n, err := w.writer.Write(obuf) + if err != nil { + return 0, w.err(err) + } + if n != len(obuf) { + return 0, w.err(io.ErrShortWrite) + } + w.err(w.index.add(w.written, w.uncompWritten)) + w.written += int64(n) + w.uncompWritten += int64(len(uncompressed)) + + if chunkType == chunkTypeUncompressedData { + // Write uncompressed data. + n, err := w.writer.Write(uncompressed) + if err != nil { + return 0, w.err(err) + } + if n != len(uncompressed) { + return 0, w.err(io.ErrShortWrite) + } + w.written += int64(n) + } + w.buffers.Put(obuf) + // Queue final output. + nRet += len(uncompressed) + } + return nRet, nil +} + +// Flush flushes the Writer to its underlying io.Writer. +// This does not apply padding. +func (w *Writer) Flush() error { + if err := w.err(nil); err != nil { + return err + } + + // Queue any data still in input buffer. + if len(w.ibuf) != 0 { + if !w.wroteStreamHeader { + _, err := w.writeSync(w.ibuf) + w.ibuf = w.ibuf[:0] + return w.err(err) + } else { + _, err := w.write(w.ibuf) + w.ibuf = w.ibuf[:0] + err = w.err(err) + if err != nil { + return err + } + } + } + if w.output == nil { + return w.err(nil) + } + + // Send empty buffer + res := make(chan result) + w.output <- res + // Block until this has been picked up. + res <- result{b: nil, startOffset: w.uncompWritten} + // When it is closed, we have flushed. + <-res + return w.err(nil) +} + +// Close calls Flush and then closes the Writer. +// Calling Close multiple times is ok, +// but calling CloseIndex after this will make it not return the index. +func (w *Writer) Close() error { + _, err := w.closeIndex(w.appendIndex) + return err +} + +// CloseIndex calls Close and returns an index on first call. +// This is not required if you are only adding index to a stream. +func (w *Writer) CloseIndex() ([]byte, error) { + return w.closeIndex(true) +} + +func (w *Writer) closeIndex(idx bool) ([]byte, error) { + err := w.Flush() + if w.output != nil { + close(w.output) + w.writerWg.Wait() + w.output = nil + } + + var index []byte + if w.err(nil) == nil && w.writer != nil { + // Create index. + if idx { + compSize := int64(-1) + if w.pad <= 1 { + compSize = w.written + } + index = w.index.appendTo(w.ibuf[:0], w.uncompWritten, compSize) + // Count as written for padding. + if w.appendIndex { + w.written += int64(len(index)) + } + } + + if w.pad > 1 { + tmp := w.ibuf[:0] + if len(index) > 0 { + // Allocate another buffer. + tmp = w.buffers.Get().([]byte)[:0] + defer w.buffers.Put(tmp) + } + add := calcSkippableFrame(w.written, int64(w.pad)) + frame, err := skippableFrame(tmp, add, w.randSrc) + if err = w.err(err); err != nil { + return nil, err + } + n, err2 := w.writer.Write(frame) + if err2 == nil && n != len(frame) { + err2 = io.ErrShortWrite + } + _ = w.err(err2) + } + if len(index) > 0 && w.appendIndex { + n, err2 := w.writer.Write(index) + if err2 == nil && n != len(index) { + err2 = io.ErrShortWrite + } + _ = w.err(err2) + } + } + err = w.err(errClosed) + if err == errClosed { + return index, nil + } + return nil, err +} + +// calcSkippableFrame will return a total size to be added for written +// to be divisible by multiple. +// The value will always be > skippableFrameHeader. +// The function will panic if written < 0 or wantMultiple <= 0. +func calcSkippableFrame(written, wantMultiple int64) int { + if wantMultiple <= 0 { + panic("wantMultiple <= 0") + } + if written < 0 { + panic("written < 0") + } + leftOver := written % wantMultiple + if leftOver == 0 { + return 0 + } + toAdd := wantMultiple - leftOver + for toAdd < skippableFrameHeader { + toAdd += wantMultiple + } + return int(toAdd) +} + +// skippableFrame will add a skippable frame with a total size of bytes. +// total should be >= skippableFrameHeader and < maxBlockSize + skippableFrameHeader +func skippableFrame(dst []byte, total int, r io.Reader) ([]byte, error) { + if total == 0 { + return dst, nil + } + if total < skippableFrameHeader { + return dst, fmt.Errorf("s2: requested skippable frame (%d) < 4", total) + } + if int64(total) >= maxBlockSize+skippableFrameHeader { + return dst, fmt.Errorf("s2: requested skippable frame (%d) >= max 1<<24", total) + } + // Chunk type 0xfe "Section 4.4 Padding (chunk type 0xfe)" + dst = append(dst, chunkTypePadding) + f := uint32(total - skippableFrameHeader) + // Add chunk length. + dst = append(dst, uint8(f), uint8(f>>8), uint8(f>>16)) + // Add data + start := len(dst) + dst = append(dst, make([]byte, f)...) + _, err := io.ReadFull(r, dst[start:]) + return dst, err +} + +// WriterOption is an option for creating a encoder. +type WriterOption func(*Writer) error + +// WriterConcurrency will set the concurrency, +// meaning the maximum number of decoders to run concurrently. +// The value supplied must be at least 1. +// By default this will be set to GOMAXPROCS. +func WriterConcurrency(n int) WriterOption { + return func(w *Writer) error { + if n <= 0 { + return errors.New("concurrency must be at least 1") + } + w.concurrency = n + return nil + } +} + +// WriterAddIndex will append an index to the end of a stream +// when it is closed. +func WriterAddIndex() WriterOption { + return func(w *Writer) error { + w.appendIndex = true + return nil + } +} + +// WriterBetterCompression will enable better compression. +// EncodeBetter compresses better than Encode but typically with a +// 10-40% speed decrease on both compression and decompression. +func WriterBetterCompression() WriterOption { + return func(w *Writer) error { + w.level = levelBetter + return nil + } +} + +// WriterBestCompression will enable better compression. +// EncodeBetter compresses better than Encode but typically with a +// big speed decrease on compression. +func WriterBestCompression() WriterOption { + return func(w *Writer) error { + w.level = levelBest + return nil + } +} + +// WriterUncompressed will bypass compression. +// The stream will be written as uncompressed blocks only. +// If concurrency is > 1 CRC and output will still be done async. +func WriterUncompressed() WriterOption { + return func(w *Writer) error { + w.level = levelUncompressed + return nil + } +} + +// WriterBlockSize allows to override the default block size. +// Blocks will be this size or smaller. +// Minimum size is 4KB and and maximum size is 4MB. +// +// Bigger blocks may give bigger throughput on systems with many cores, +// and will increase compression slightly, but it will limit the possible +// concurrency for smaller payloads for both encoding and decoding. +// Default block size is 1MB. +// +// When writing Snappy compatible output using WriterSnappyCompat, +// the maximum block size is 64KB. +func WriterBlockSize(n int) WriterOption { + return func(w *Writer) error { + if w.snappy && n > maxSnappyBlockSize || n < minBlockSize { + return errors.New("s2: block size too large. Must be <= 64K and >=4KB on for snappy compatible output") + } + if n > maxBlockSize || n < minBlockSize { + return errors.New("s2: block size too large. Must be <= 4MB and >=4KB") + } + w.blockSize = n + return nil + } +} + +// WriterPadding will add padding to all output so the size will be a multiple of n. +// This can be used to obfuscate the exact output size or make blocks of a certain size. +// The contents will be a skippable frame, so it will be invisible by the decoder. +// n must be > 0 and <= 4MB. +// The padded area will be filled with data from crypto/rand.Reader. +// The padding will be applied whenever Close is called on the writer. +func WriterPadding(n int) WriterOption { + return func(w *Writer) error { + if n <= 0 { + return fmt.Errorf("s2: padding must be at least 1") + } + // No need to waste our time. + if n == 1 { + w.pad = 0 + } + if n > maxBlockSize { + return fmt.Errorf("s2: padding must less than 4MB") + } + w.pad = n + return nil + } +} + +// WriterPaddingSrc will get random data for padding from the supplied source. +// By default crypto/rand is used. +func WriterPaddingSrc(reader io.Reader) WriterOption { + return func(w *Writer) error { + w.randSrc = reader + return nil + } +} + +// WriterSnappyCompat will write snappy compatible output. +// The output can be decompressed using either snappy or s2. +// If block size is more than 64KB it is set to that. +func WriterSnappyCompat() WriterOption { + return func(w *Writer) error { + w.snappy = true + if w.blockSize > 64<<10 { + // We choose 8 bytes less than 64K, since that will make literal emits slightly more effective. + // And allows us to skip some size checks. + w.blockSize = (64 << 10) - 8 + } + return nil + } +} + +// WriterFlushOnWrite will compress blocks on each call to the Write function. +// +// This is quite inefficient as blocks size will depend on the write size. +// +// Use WriterConcurrency(1) to also make sure that output is flushed. +// When Write calls return, otherwise they will be written when compression is done. +func WriterFlushOnWrite() WriterOption { + return func(w *Writer) error { + w.flushOnWrite = true + return nil + } +} diff --git a/vendor/github.com/klauspost/compress/s2/encode_all.go b/vendor/github.com/klauspost/compress/s2/encode_all.go new file mode 100644 index 0000000..8b16c38 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encode_all.go @@ -0,0 +1,456 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package s2 + +import ( + "bytes" + "encoding/binary" + "math/bits" +) + +func load32(b []byte, i int) uint32 { + return binary.LittleEndian.Uint32(b[i:]) +} + +func load64(b []byte, i int) uint64 { + return binary.LittleEndian.Uint64(b[i:]) +} + +// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash6(u uint64, h uint8) uint32 { + const prime6bytes = 227718039650203 + return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & 63)) +} + +func encodeGo(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if len(dst) < n { + dst = make([]byte, n) + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlockGo(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// encodeBlockGo encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockGo(dst, src []byte) (d int) { + // Initialize the hash table. + const ( + tableBits = 14 + maxTableSize = 1 << tableBits + + debug = false + ) + + var table [maxTableSize]uint32 + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + + // Bail if we can't compress to at least this. + dstLimit := len(src) - len(src)>>5 - 5 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + cv := load64(src, s) + + // We search for a repeat at -1, but don't output repeats when nextEmit == 0 + repeat := 1 + + for { + candidate := 0 + for { + // Next src position to check + nextS := s + (s-nextEmit)>>6 + 4 + if nextS > sLimit { + goto emitRemainder + } + hash0 := hash6(cv, tableBits) + hash1 := hash6(cv>>8, tableBits) + candidate = int(table[hash0]) + candidate2 := int(table[hash1]) + table[hash0] = uint32(s) + table[hash1] = uint32(s + 1) + hash2 := hash6(cv>>16, tableBits) + + // Check repeat at offset checkRep. + const checkRep = 1 + if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) { + base := s + checkRep + // Extend back + for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; { + i-- + base-- + } + d += emitLiteral(dst[d:], src[nextEmit:base]) + + // Extend forward + candidate := s - repeat + 4 + checkRep + s += 4 + checkRep + for s <= sLimit { + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + if debug { + // Validate match. + if s <= candidate { + panic("s <= candidate") + } + a := src[base:s] + b := src[base-repeat : base-repeat+(s-base)] + if !bytes.Equal(a, b) { + panic("mismatch") + } + } + if nextEmit > 0 { + // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset. + d += emitRepeat(dst[d:], repeat, s-base) + } else { + // First match, cannot be repeat. + d += emitCopy(dst[d:], repeat, s-base) + } + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + cv = load64(src, s) + continue + } + + if uint32(cv) == load32(src, candidate) { + break + } + candidate = int(table[hash2]) + if uint32(cv>>8) == load32(src, candidate2) { + table[hash2] = uint32(s + 2) + candidate = candidate2 + s++ + break + } + table[hash2] = uint32(s + 2) + if uint32(cv>>16) == load32(src, candidate) { + s += 2 + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards. + // The top bytes will be rechecked to get the full match. + for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] { + candidate-- + s-- + } + + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + + d += emitLiteral(dst[d:], src[nextEmit:s]) + + // Call emitCopy, and then see if another emitCopy could be our next + // move. Repeat until we find no match for the input immediately after + // what was consumed by the last emitCopy call. + // + // If we exit this loop normally then we need to call emitLiteral next, + // though we don't yet know how big the literal will be. We handle that + // by proceeding to the next iteration of the main loop. We also can + // exit this loop via goto if we get close to exhausting the input. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + base := s + repeat = base - candidate + + // Extend the 4-byte match as long as possible. + s += 4 + candidate += 4 + for s <= len(src)-8 { + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + + d += emitCopy(dst[d:], repeat, s-base) + if debug { + // Validate match. + if s <= candidate { + panic("s <= candidate") + } + a := src[base:s] + b := src[base-repeat : base-repeat+(s-base)] + if !bytes.Equal(a, b) { + panic("mismatch") + } + } + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Check for an immediate match, otherwise start search at s+1 + x := load64(src, s-2) + m2Hash := hash6(x, tableBits) + currHash := hash6(x>>16, tableBits) + candidate = int(table[currHash]) + table[m2Hash] = uint32(s - 2) + table[currHash] = uint32(s) + if debug && s == candidate { + panic("s == candidate") + } + if uint32(x>>16) != load32(src, candidate) { + cv = load64(src, s+1) + s++ + break + } + } + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} + +func encodeBlockSnappyGo(dst, src []byte) (d int) { + // Initialize the hash table. + const ( + tableBits = 14 + maxTableSize = 1 << tableBits + ) + + var table [maxTableSize]uint32 + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + + // Bail if we can't compress to at least this. + dstLimit := len(src) - len(src)>>5 - 5 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + cv := load64(src, s) + + // We search for a repeat at -1, but don't output repeats when nextEmit == 0 + repeat := 1 + + for { + candidate := 0 + for { + // Next src position to check + nextS := s + (s-nextEmit)>>6 + 4 + if nextS > sLimit { + goto emitRemainder + } + hash0 := hash6(cv, tableBits) + hash1 := hash6(cv>>8, tableBits) + candidate = int(table[hash0]) + candidate2 := int(table[hash1]) + table[hash0] = uint32(s) + table[hash1] = uint32(s + 1) + hash2 := hash6(cv>>16, tableBits) + + // Check repeat at offset checkRep. + const checkRep = 1 + if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) { + base := s + checkRep + // Extend back + for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; { + i-- + base-- + } + d += emitLiteral(dst[d:], src[nextEmit:base]) + + // Extend forward + candidate := s - repeat + 4 + checkRep + s += 4 + checkRep + for s <= sLimit { + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + + d += emitCopyNoRepeat(dst[d:], repeat, s-base) + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + cv = load64(src, s) + continue + } + + if uint32(cv) == load32(src, candidate) { + break + } + candidate = int(table[hash2]) + if uint32(cv>>8) == load32(src, candidate2) { + table[hash2] = uint32(s + 2) + candidate = candidate2 + s++ + break + } + table[hash2] = uint32(s + 2) + if uint32(cv>>16) == load32(src, candidate) { + s += 2 + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards + for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] { + candidate-- + s-- + } + + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + // A 4-byte match has been found. We'll later see if more than 4 bytes + // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit + // them as literal bytes. + + d += emitLiteral(dst[d:], src[nextEmit:s]) + + // Call emitCopy, and then see if another emitCopy could be our next + // move. Repeat until we find no match for the input immediately after + // what was consumed by the last emitCopy call. + // + // If we exit this loop normally then we need to call emitLiteral next, + // though we don't yet know how big the literal will be. We handle that + // by proceeding to the next iteration of the main loop. We also can + // exit this loop via goto if we get close to exhausting the input. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + base := s + repeat = base - candidate + + // Extend the 4-byte match as long as possible. + s += 4 + candidate += 4 + for s <= len(src)-8 { + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + + d += emitCopyNoRepeat(dst[d:], repeat, s-base) + if false { + // Validate match. + a := src[base:s] + b := src[base-repeat : base-repeat+(s-base)] + if !bytes.Equal(a, b) { + panic("mismatch") + } + } + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Check for an immediate match, otherwise start search at s+1 + x := load64(src, s-2) + m2Hash := hash6(x, tableBits) + currHash := hash6(x>>16, tableBits) + candidate = int(table[currHash]) + table[m2Hash] = uint32(s - 2) + table[currHash] = uint32(s) + if uint32(x>>16) != load32(src, candidate) { + cv = load64(src, s+1) + s++ + break + } + } + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} diff --git a/vendor/github.com/klauspost/compress/s2/encode_amd64.go b/vendor/github.com/klauspost/compress/s2/encode_amd64.go new file mode 100644 index 0000000..e612225 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encode_amd64.go @@ -0,0 +1,142 @@ +//go:build !appengine && !noasm && gc +// +build !appengine,!noasm,gc + +package s2 + +// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlock(dst, src []byte) (d int) { + const ( + // Use 12 bit table when less than... + limit12B = 16 << 10 + // Use 10 bit table when less than... + limit10B = 4 << 10 + // Use 8 bit table when less than... + limit8B = 512 + ) + + if len(src) >= 4<<20 { + return encodeBlockAsm(dst, src) + } + if len(src) >= limit12B { + return encodeBlockAsm4MB(dst, src) + } + if len(src) >= limit10B { + return encodeBlockAsm12B(dst, src) + } + if len(src) >= limit8B { + return encodeBlockAsm10B(dst, src) + } + if len(src) < minNonLiteralBlockSize { + return 0 + } + return encodeBlockAsm8B(dst, src) +} + +// encodeBlockBetter encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockBetter(dst, src []byte) (d int) { + const ( + // Use 12 bit table when less than... + limit12B = 16 << 10 + // Use 10 bit table when less than... + limit10B = 4 << 10 + // Use 8 bit table when less than... + limit8B = 512 + ) + + if len(src) > 4<<20 { + return encodeBetterBlockAsm(dst, src) + } + if len(src) >= limit12B { + return encodeBetterBlockAsm4MB(dst, src) + } + if len(src) >= limit10B { + return encodeBetterBlockAsm12B(dst, src) + } + if len(src) >= limit8B { + return encodeBetterBlockAsm10B(dst, src) + } + if len(src) < minNonLiteralBlockSize { + return 0 + } + return encodeBetterBlockAsm8B(dst, src) +} + +// encodeBlockSnappy encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockSnappy(dst, src []byte) (d int) { + const ( + // Use 12 bit table when less than... + limit12B = 16 << 10 + // Use 10 bit table when less than... + limit10B = 4 << 10 + // Use 8 bit table when less than... + limit8B = 512 + ) + if len(src) >= 64<<10 { + return encodeSnappyBlockAsm(dst, src) + } + if len(src) >= limit12B { + return encodeSnappyBlockAsm64K(dst, src) + } + if len(src) >= limit10B { + return encodeSnappyBlockAsm12B(dst, src) + } + if len(src) >= limit8B { + return encodeSnappyBlockAsm10B(dst, src) + } + if len(src) < minNonLiteralBlockSize { + return 0 + } + return encodeSnappyBlockAsm8B(dst, src) +} + +// encodeBlockSnappy encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockBetterSnappy(dst, src []byte) (d int) { + const ( + // Use 12 bit table when less than... + limit12B = 16 << 10 + // Use 10 bit table when less than... + limit10B = 4 << 10 + // Use 8 bit table when less than... + limit8B = 512 + ) + if len(src) >= 64<<10 { + return encodeSnappyBetterBlockAsm(dst, src) + } + if len(src) >= limit12B { + return encodeSnappyBetterBlockAsm64K(dst, src) + } + if len(src) >= limit10B { + return encodeSnappyBetterBlockAsm12B(dst, src) + } + if len(src) >= limit8B { + return encodeSnappyBetterBlockAsm10B(dst, src) + } + if len(src) < minNonLiteralBlockSize { + return 0 + } + return encodeSnappyBetterBlockAsm8B(dst, src) +} diff --git a/vendor/github.com/klauspost/compress/s2/encode_best.go b/vendor/github.com/klauspost/compress/s2/encode_best.go new file mode 100644 index 0000000..4bc80bc --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encode_best.go @@ -0,0 +1,630 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package s2 + +import ( + "fmt" + "math/bits" +) + +// encodeBlockBest encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockBest(dst, src []byte) (d int) { + // Initialize the hash tables. + const ( + // Long hash matches. + lTableBits = 19 + maxLTableSize = 1 << lTableBits + + // Short hash matches. + sTableBits = 16 + maxSTableSize = 1 << sTableBits + + inputMargin = 8 + 2 + ) + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + if len(src) < minNonLiteralBlockSize { + return 0 + } + + var lTable [maxLTableSize]uint64 + var sTable [maxSTableSize]uint64 + + // Bail if we can't compress to at least this. + dstLimit := len(src) - 5 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + cv := load64(src, s) + + // We search for a repeat at -1, but don't output repeats when nextEmit == 0 + repeat := 1 + const lowbitMask = 0xffffffff + getCur := func(x uint64) int { + return int(x & lowbitMask) + } + getPrev := func(x uint64) int { + return int(x >> 32) + } + const maxSkip = 64 + + for { + type match struct { + offset int + s int + length int + score int + rep bool + } + var best match + for { + // Next src position to check + nextS := (s-nextEmit)>>8 + 1 + if nextS > maxSkip { + nextS = s + maxSkip + } else { + nextS += s + } + if nextS > sLimit { + goto emitRemainder + } + hashL := hash8(cv, lTableBits) + hashS := hash4(cv, sTableBits) + candidateL := lTable[hashL] + candidateS := sTable[hashS] + + score := func(m match) int { + // Matches that are longer forward are penalized since we must emit it as a literal. + score := m.length - m.s + if nextEmit == m.s { + // If we do not have to emit literals, we save 1 byte + score++ + } + offset := m.s - m.offset + if m.rep { + return score - emitRepeatSize(offset, m.length) + } + return score - emitCopySize(offset, m.length) + } + + matchAt := func(offset, s int, first uint32, rep bool) match { + if best.length != 0 && best.s-best.offset == s-offset { + // Don't retest if we have the same offset. + return match{offset: offset, s: s} + } + if load32(src, offset) != first { + return match{offset: offset, s: s} + } + m := match{offset: offset, s: s, length: 4 + offset, rep: rep} + s += 4 + for s <= sLimit { + if diff := load64(src, s) ^ load64(src, m.length); diff != 0 { + m.length += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + m.length += 8 + } + m.length -= offset + m.score = score(m) + if m.score <= -m.s { + // Eliminate if no savings, we might find a better one. + m.length = 0 + } + return m + } + + bestOf := func(a, b match) match { + if b.length == 0 { + return a + } + if a.length == 0 { + return b + } + as := a.score + b.s + bs := b.score + a.s + if as >= bs { + return a + } + return b + } + + best = bestOf(matchAt(getCur(candidateL), s, uint32(cv), false), matchAt(getPrev(candidateL), s, uint32(cv), false)) + best = bestOf(best, matchAt(getCur(candidateS), s, uint32(cv), false)) + best = bestOf(best, matchAt(getPrev(candidateS), s, uint32(cv), false)) + + { + best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8), true)) + if best.length > 0 { + // s+1 + nextShort := sTable[hash4(cv>>8, sTableBits)] + s := s + 1 + cv := load64(src, s) + nextLong := lTable[hash8(cv, lTableBits)] + best = bestOf(best, matchAt(getCur(nextShort), s, uint32(cv), false)) + best = bestOf(best, matchAt(getPrev(nextShort), s, uint32(cv), false)) + best = bestOf(best, matchAt(getCur(nextLong), s, uint32(cv), false)) + best = bestOf(best, matchAt(getPrev(nextLong), s, uint32(cv), false)) + // Repeat at + 2 + best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8), true)) + + // s+2 + if true { + nextShort = sTable[hash4(cv>>8, sTableBits)] + s++ + cv = load64(src, s) + nextLong = lTable[hash8(cv, lTableBits)] + best = bestOf(best, matchAt(getCur(nextShort), s, uint32(cv), false)) + best = bestOf(best, matchAt(getPrev(nextShort), s, uint32(cv), false)) + best = bestOf(best, matchAt(getCur(nextLong), s, uint32(cv), false)) + best = bestOf(best, matchAt(getPrev(nextLong), s, uint32(cv), false)) + } + // Search for a match at best match end, see if that is better. + if sAt := best.s + best.length; sAt < sLimit { + sBack := best.s + backL := best.length + // Load initial values + cv = load64(src, sBack) + // Search for mismatch + next := lTable[hash8(load64(src, sAt), lTableBits)] + //next := sTable[hash4(load64(src, sAt), sTableBits)] + + if checkAt := getCur(next) - backL; checkAt > 0 { + best = bestOf(best, matchAt(checkAt, sBack, uint32(cv), false)) + } + if checkAt := getPrev(next) - backL; checkAt > 0 { + best = bestOf(best, matchAt(checkAt, sBack, uint32(cv), false)) + } + } + } + } + + // Update table + lTable[hashL] = uint64(s) | candidateL<<32 + sTable[hashS] = uint64(s) | candidateS<<32 + + if best.length > 0 { + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards, not needed for repeats... + s = best.s + if !best.rep { + for best.offset > 0 && s > nextEmit && src[best.offset-1] == src[s-1] { + best.offset-- + best.length++ + s-- + } + } + if false && best.offset >= s { + panic(fmt.Errorf("t %d >= s %d", best.offset, s)) + } + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + base := s + offset := s - best.offset + + s += best.length + + if offset > 65535 && s-base <= 5 && !best.rep { + // Bail if the match is equal or worse to the encoding. + s = best.s + 1 + if s >= sLimit { + goto emitRemainder + } + cv = load64(src, s) + continue + } + d += emitLiteral(dst[d:], src[nextEmit:base]) + if best.rep { + if nextEmit > 0 { + // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset. + d += emitRepeat(dst[d:], offset, best.length) + } else { + // First match, cannot be repeat. + d += emitCopy(dst[d:], offset, best.length) + } + } else { + d += emitCopy(dst[d:], offset, best.length) + } + repeat = offset + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Fill tables... + for i := best.s + 1; i < s; i++ { + cv0 := load64(src, i) + long0 := hash8(cv0, lTableBits) + short0 := hash4(cv0, sTableBits) + lTable[long0] = uint64(i) | lTable[long0]<<32 + sTable[short0] = uint64(i) | sTable[short0]<<32 + } + cv = load64(src, s) + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} + +// encodeBlockBestSnappy encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockBestSnappy(dst, src []byte) (d int) { + // Initialize the hash tables. + const ( + // Long hash matches. + lTableBits = 19 + maxLTableSize = 1 << lTableBits + + // Short hash matches. + sTableBits = 16 + maxSTableSize = 1 << sTableBits + + inputMargin = 8 + 2 + ) + + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + if len(src) < minNonLiteralBlockSize { + return 0 + } + + var lTable [maxLTableSize]uint64 + var sTable [maxSTableSize]uint64 + + // Bail if we can't compress to at least this. + dstLimit := len(src) - 5 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + cv := load64(src, s) + + // We search for a repeat at -1, but don't output repeats when nextEmit == 0 + repeat := 1 + const lowbitMask = 0xffffffff + getCur := func(x uint64) int { + return int(x & lowbitMask) + } + getPrev := func(x uint64) int { + return int(x >> 32) + } + const maxSkip = 64 + + for { + type match struct { + offset int + s int + length int + score int + } + var best match + for { + // Next src position to check + nextS := (s-nextEmit)>>8 + 1 + if nextS > maxSkip { + nextS = s + maxSkip + } else { + nextS += s + } + if nextS > sLimit { + goto emitRemainder + } + hashL := hash8(cv, lTableBits) + hashS := hash4(cv, sTableBits) + candidateL := lTable[hashL] + candidateS := sTable[hashS] + + score := func(m match) int { + // Matches that are longer forward are penalized since we must emit it as a literal. + score := m.length - m.s + if nextEmit == m.s { + // If we do not have to emit literals, we save 1 byte + score++ + } + offset := m.s - m.offset + + return score - emitCopyNoRepeatSize(offset, m.length) + } + + matchAt := func(offset, s int, first uint32) match { + if best.length != 0 && best.s-best.offset == s-offset { + // Don't retest if we have the same offset. + return match{offset: offset, s: s} + } + if load32(src, offset) != first { + return match{offset: offset, s: s} + } + m := match{offset: offset, s: s, length: 4 + offset} + s += 4 + for s <= sLimit { + if diff := load64(src, s) ^ load64(src, m.length); diff != 0 { + m.length += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + m.length += 8 + } + m.length -= offset + m.score = score(m) + if m.score <= -m.s { + // Eliminate if no savings, we might find a better one. + m.length = 0 + } + return m + } + + bestOf := func(a, b match) match { + if b.length == 0 { + return a + } + if a.length == 0 { + return b + } + as := a.score + b.s + bs := b.score + a.s + if as >= bs { + return a + } + return b + } + + best = bestOf(matchAt(getCur(candidateL), s, uint32(cv)), matchAt(getPrev(candidateL), s, uint32(cv))) + best = bestOf(best, matchAt(getCur(candidateS), s, uint32(cv))) + best = bestOf(best, matchAt(getPrev(candidateS), s, uint32(cv))) + + { + best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8))) + if best.length > 0 { + // s+1 + nextShort := sTable[hash4(cv>>8, sTableBits)] + s := s + 1 + cv := load64(src, s) + nextLong := lTable[hash8(cv, lTableBits)] + best = bestOf(best, matchAt(getCur(nextShort), s, uint32(cv))) + best = bestOf(best, matchAt(getPrev(nextShort), s, uint32(cv))) + best = bestOf(best, matchAt(getCur(nextLong), s, uint32(cv))) + best = bestOf(best, matchAt(getPrev(nextLong), s, uint32(cv))) + // Repeat at + 2 + best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8))) + + // s+2 + if true { + nextShort = sTable[hash4(cv>>8, sTableBits)] + s++ + cv = load64(src, s) + nextLong = lTable[hash8(cv, lTableBits)] + best = bestOf(best, matchAt(getCur(nextShort), s, uint32(cv))) + best = bestOf(best, matchAt(getPrev(nextShort), s, uint32(cv))) + best = bestOf(best, matchAt(getCur(nextLong), s, uint32(cv))) + best = bestOf(best, matchAt(getPrev(nextLong), s, uint32(cv))) + } + // Search for a match at best match end, see if that is better. + if sAt := best.s + best.length; sAt < sLimit { + sBack := best.s + backL := best.length + // Load initial values + cv = load64(src, sBack) + // Search for mismatch + next := lTable[hash8(load64(src, sAt), lTableBits)] + //next := sTable[hash4(load64(src, sAt), sTableBits)] + + if checkAt := getCur(next) - backL; checkAt > 0 { + best = bestOf(best, matchAt(checkAt, sBack, uint32(cv))) + } + if checkAt := getPrev(next) - backL; checkAt > 0 { + best = bestOf(best, matchAt(checkAt, sBack, uint32(cv))) + } + } + } + } + + // Update table + lTable[hashL] = uint64(s) | candidateL<<32 + sTable[hashS] = uint64(s) | candidateS<<32 + + if best.length > 0 { + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards, not needed for repeats... + s = best.s + if true { + for best.offset > 0 && s > nextEmit && src[best.offset-1] == src[s-1] { + best.offset-- + best.length++ + s-- + } + } + if false && best.offset >= s { + panic(fmt.Errorf("t %d >= s %d", best.offset, s)) + } + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + base := s + offset := s - best.offset + + s += best.length + + if offset > 65535 && s-base <= 5 { + // Bail if the match is equal or worse to the encoding. + s = best.s + 1 + if s >= sLimit { + goto emitRemainder + } + cv = load64(src, s) + continue + } + d += emitLiteral(dst[d:], src[nextEmit:base]) + d += emitCopyNoRepeat(dst[d:], offset, best.length) + repeat = offset + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Fill tables... + for i := best.s + 1; i < s; i++ { + cv0 := load64(src, i) + long0 := hash8(cv0, lTableBits) + short0 := hash4(cv0, sTableBits) + lTable[long0] = uint64(i) | lTable[long0]<<32 + sTable[short0] = uint64(i) | sTable[short0]<<32 + } + cv = load64(src, s) + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} + +// emitCopySize returns the size to encode the offset+length +// +// It assumes that: +// 1 <= offset && offset <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 +func emitCopySize(offset, length int) int { + if offset >= 65536 { + i := 0 + if length > 64 { + length -= 64 + if length >= 4 { + // Emit remaining as repeats + return 5 + emitRepeatSize(offset, length) + } + i = 5 + } + if length == 0 { + return i + } + return i + 5 + } + + // Offset no more than 2 bytes. + if length > 64 { + if offset < 2048 { + // Emit 8 bytes, then rest as repeats... + return 2 + emitRepeatSize(offset, length-8) + } + // Emit remaining as repeats, at least 4 bytes remain. + return 3 + emitRepeatSize(offset, length-60) + } + if length >= 12 || offset >= 2048 { + return 3 + } + // Emit the remaining copy, encoded as 2 bytes. + return 2 +} + +// emitCopyNoRepeatSize returns the size to encode the offset+length +// +// It assumes that: +// 1 <= offset && offset <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 +func emitCopyNoRepeatSize(offset, length int) int { + if offset >= 65536 { + return 5 + 5*(length/64) + } + + // Offset no more than 2 bytes. + if length > 64 { + // Emit remaining as repeats, at least 4 bytes remain. + return 3 + 3*(length/60) + } + if length >= 12 || offset >= 2048 { + return 3 + } + // Emit the remaining copy, encoded as 2 bytes. + return 2 +} + +// emitRepeatSize returns the number of bytes required to encode a repeat. +// Length must be at least 4 and < 1<<24 +func emitRepeatSize(offset, length int) int { + // Repeat offset, make length cheaper + if length <= 4+4 || (length < 8+4 && offset < 2048) { + return 2 + } + if length < (1<<8)+4+4 { + return 3 + } + if length < (1<<16)+(1<<8)+4 { + return 4 + } + const maxRepeat = (1 << 24) - 1 + length -= (1 << 16) - 4 + left := 0 + if length > maxRepeat { + left = length - maxRepeat + 4 + length = maxRepeat - 4 + } + if left > 0 { + return 5 + emitRepeatSize(offset, left) + } + return 5 +} diff --git a/vendor/github.com/klauspost/compress/s2/encode_better.go b/vendor/github.com/klauspost/compress/s2/encode_better.go new file mode 100644 index 0000000..943215b --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encode_better.go @@ -0,0 +1,431 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package s2 + +import ( + "math/bits" +) + +// hash4 returns the hash of the lowest 4 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <32. +func hash4(u uint64, h uint8) uint32 { + const prime4bytes = 2654435761 + return (uint32(u) * prime4bytes) >> ((32 - h) & 31) +} + +// hash5 returns the hash of the lowest 5 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash5(u uint64, h uint8) uint32 { + const prime5bytes = 889523592379 + return uint32(((u << (64 - 40)) * prime5bytes) >> ((64 - h) & 63)) +} + +// hash7 returns the hash of the lowest 7 bytes of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash7(u uint64, h uint8) uint32 { + const prime7bytes = 58295818150454627 + return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & 63)) +} + +// hash8 returns the hash of u to fit in a hash table with h bits. +// Preferably h should be a constant and should always be <64. +func hash8(u uint64, h uint8) uint32 { + const prime8bytes = 0xcf1bbcdcb7a56463 + return uint32((u * prime8bytes) >> ((64 - h) & 63)) +} + +// encodeBlockBetter encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockBetterGo(dst, src []byte) (d int) { + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + if len(src) < minNonLiteralBlockSize { + return 0 + } + + // Initialize the hash tables. + const ( + // Long hash matches. + lTableBits = 16 + maxLTableSize = 1 << lTableBits + + // Short hash matches. + sTableBits = 14 + maxSTableSize = 1 << sTableBits + ) + + var lTable [maxLTableSize]uint32 + var sTable [maxSTableSize]uint32 + + // Bail if we can't compress to at least this. + dstLimit := len(src) - len(src)>>5 - 6 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + cv := load64(src, s) + + // We initialize repeat to 0, so we never match on first attempt + repeat := 0 + + for { + candidateL := 0 + nextS := 0 + for { + // Next src position to check + nextS = s + (s-nextEmit)>>7 + 1 + if nextS > sLimit { + goto emitRemainder + } + hashL := hash7(cv, lTableBits) + hashS := hash4(cv, sTableBits) + candidateL = int(lTable[hashL]) + candidateS := int(sTable[hashS]) + lTable[hashL] = uint32(s) + sTable[hashS] = uint32(s) + + // Check repeat at offset checkRep. + const checkRep = 1 + if false && uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) { + base := s + checkRep + // Extend back + for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; { + i-- + base-- + } + d += emitLiteral(dst[d:], src[nextEmit:base]) + + // Extend forward + candidate := s - repeat + 4 + checkRep + s += 4 + checkRep + for s < len(src) { + if len(src)-s < 8 { + if src[s] == src[candidate] { + s++ + candidate++ + continue + } + break + } + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + if nextEmit > 0 { + // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset. + d += emitRepeat(dst[d:], repeat, s-base) + } else { + // First match, cannot be repeat. + d += emitCopy(dst[d:], repeat, s-base) + } + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + cv = load64(src, s) + continue + } + + if uint32(cv) == load32(src, candidateL) { + break + } + + // Check our short candidate + if uint32(cv) == load32(src, candidateS) { + // Try a long candidate at s+1 + hashL = hash7(cv>>8, lTableBits) + candidateL = int(lTable[hashL]) + lTable[hashL] = uint32(s + 1) + if uint32(cv>>8) == load32(src, candidateL) { + s++ + break + } + // Use our short candidate. + candidateL = candidateS + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards + for candidateL > 0 && s > nextEmit && src[candidateL-1] == src[s-1] { + candidateL-- + s-- + } + + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + base := s + offset := base - candidateL + + // Extend the 4-byte match as long as possible. + s += 4 + candidateL += 4 + for s < len(src) { + if len(src)-s < 8 { + if src[s] == src[candidateL] { + s++ + candidateL++ + continue + } + break + } + if diff := load64(src, s) ^ load64(src, candidateL); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidateL += 8 + } + + if offset > 65535 && s-base <= 5 && repeat != offset { + // Bail if the match is equal or worse to the encoding. + s = nextS + 1 + if s >= sLimit { + goto emitRemainder + } + cv = load64(src, s) + continue + } + + d += emitLiteral(dst[d:], src[nextEmit:base]) + if repeat == offset { + d += emitRepeat(dst[d:], offset, s-base) + } else { + d += emitCopy(dst[d:], offset, s-base) + repeat = offset + } + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Index match start+1 (long) and start+2 (short) + index0 := base + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load64(src, index0) + cv1 := load64(src, index1) + cv = load64(src, s) + lTable[hash7(cv0, lTableBits)] = uint32(index0) + lTable[hash7(cv0>>8, lTableBits)] = uint32(index0 + 1) + lTable[hash7(cv1, lTableBits)] = uint32(index1) + lTable[hash7(cv1>>8, lTableBits)] = uint32(index1 + 1) + sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1) + sTable[hash4(cv0>>16, sTableBits)] = uint32(index0 + 2) + sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1) + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} + +// encodeBlockBetterSnappyGo encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) && +// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize +func encodeBlockBetterSnappyGo(dst, src []byte) (d int) { + // sLimit is when to stop looking for offset/length copies. The inputMargin + // lets us use a fast path for emitLiteral in the main loop, while we are + // looking for copies. + sLimit := len(src) - inputMargin + if len(src) < minNonLiteralBlockSize { + return 0 + } + + // Initialize the hash tables. + const ( + // Long hash matches. + lTableBits = 16 + maxLTableSize = 1 << lTableBits + + // Short hash matches. + sTableBits = 14 + maxSTableSize = 1 << sTableBits + ) + + var lTable [maxLTableSize]uint32 + var sTable [maxSTableSize]uint32 + + // Bail if we can't compress to at least this. + dstLimit := len(src) - len(src)>>5 - 6 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := 0 + + // The encoded form must start with a literal, as there are no previous + // bytes to copy, so we start looking for hash matches at s == 1. + s := 1 + cv := load64(src, s) + + // We initialize repeat to 0, so we never match on first attempt + repeat := 0 + const maxSkip = 100 + + for { + candidateL := 0 + nextS := 0 + for { + // Next src position to check + nextS = (s-nextEmit)>>7 + 1 + if nextS > maxSkip { + nextS = s + maxSkip + } else { + nextS += s + } + + if nextS > sLimit { + goto emitRemainder + } + hashL := hash7(cv, lTableBits) + hashS := hash4(cv, sTableBits) + candidateL = int(lTable[hashL]) + candidateS := int(sTable[hashS]) + lTable[hashL] = uint32(s) + sTable[hashS] = uint32(s) + + if uint32(cv) == load32(src, candidateL) { + break + } + + // Check our short candidate + if uint32(cv) == load32(src, candidateS) { + // Try a long candidate at s+1 + hashL = hash7(cv>>8, lTableBits) + candidateL = int(lTable[hashL]) + lTable[hashL] = uint32(s + 1) + if uint32(cv>>8) == load32(src, candidateL) { + s++ + break + } + // Use our short candidate. + candidateL = candidateS + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards + for candidateL > 0 && s > nextEmit && src[candidateL-1] == src[s-1] { + candidateL-- + s-- + } + + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + base := s + offset := base - candidateL + + // Extend the 4-byte match as long as possible. + s += 4 + candidateL += 4 + for s < len(src) { + if len(src)-s < 8 { + if src[s] == src[candidateL] { + s++ + candidateL++ + continue + } + break + } + if diff := load64(src, s) ^ load64(src, candidateL); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidateL += 8 + } + + if offset > 65535 && s-base <= 5 && repeat != offset { + // Bail if the match is equal or worse to the encoding. + s = nextS + 1 + if s >= sLimit { + goto emitRemainder + } + cv = load64(src, s) + continue + } + + d += emitLiteral(dst[d:], src[nextEmit:base]) + d += emitCopyNoRepeat(dst[d:], offset, s-base) + repeat = offset + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Index match start+1 (long) and start+2 (short) + index0 := base + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load64(src, index0) + cv1 := load64(src, index1) + cv = load64(src, s) + lTable[hash7(cv0, lTableBits)] = uint32(index0) + lTable[hash7(cv0>>8, lTableBits)] = uint32(index0 + 1) + lTable[hash7(cv1, lTableBits)] = uint32(index1) + lTable[hash7(cv1>>8, lTableBits)] = uint32(index1 + 1) + sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1) + sTable[hash4(cv0>>16, sTableBits)] = uint32(index0 + 2) + sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1) + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} diff --git a/vendor/github.com/klauspost/compress/s2/encode_go.go b/vendor/github.com/klauspost/compress/s2/encode_go.go new file mode 100644 index 0000000..94784b8 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encode_go.go @@ -0,0 +1,307 @@ +//go:build !amd64 || appengine || !gc || noasm +// +build !amd64 appengine !gc noasm + +package s2 + +import ( + "math/bits" +) + +// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) +func encodeBlock(dst, src []byte) (d int) { + if len(src) < minNonLiteralBlockSize { + return 0 + } + return encodeBlockGo(dst, src) +} + +// encodeBlockBetter encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) +func encodeBlockBetter(dst, src []byte) (d int) { + return encodeBlockBetterGo(dst, src) +} + +// encodeBlockBetter encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) +func encodeBlockBetterSnappy(dst, src []byte) (d int) { + return encodeBlockBetterSnappyGo(dst, src) +} + +// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It +// assumes that the varint-encoded length of the decompressed bytes has already +// been written. +// +// It also assumes that: +// len(dst) >= MaxEncodedLen(len(src)) +func encodeBlockSnappy(dst, src []byte) (d int) { + if len(src) < minNonLiteralBlockSize { + return 0 + } + return encodeBlockSnappyGo(dst, src) +} + +// emitLiteral writes a literal chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 0 <= len(lit) && len(lit) <= math.MaxUint32 +func emitLiteral(dst, lit []byte) int { + if len(lit) == 0 { + return 0 + } + const num = 63<<2 | tagLiteral + i, n := 0, uint(len(lit)-1) + switch { + case n < 60: + dst[0] = uint8(n)<<2 | tagLiteral + i = 1 + case n < 1<<8: + dst[1] = uint8(n) + dst[0] = 60<<2 | tagLiteral + i = 2 + case n < 1<<16: + dst[2] = uint8(n >> 8) + dst[1] = uint8(n) + dst[0] = 61<<2 | tagLiteral + i = 3 + case n < 1<<24: + dst[3] = uint8(n >> 16) + dst[2] = uint8(n >> 8) + dst[1] = uint8(n) + dst[0] = 62<<2 | tagLiteral + i = 4 + default: + dst[4] = uint8(n >> 24) + dst[3] = uint8(n >> 16) + dst[2] = uint8(n >> 8) + dst[1] = uint8(n) + dst[0] = 63<<2 | tagLiteral + i = 5 + } + return i + copy(dst[i:], lit) +} + +// emitRepeat writes a repeat chunk and returns the number of bytes written. +// Length must be at least 4 and < 1<<24 +func emitRepeat(dst []byte, offset, length int) int { + // Repeat offset, make length cheaper + length -= 4 + if length <= 4 { + dst[0] = uint8(length)<<2 | tagCopy1 + dst[1] = 0 + return 2 + } + if length < 8 && offset < 2048 { + // Encode WITH offset + dst[1] = uint8(offset) + dst[0] = uint8(offset>>8)<<5 | uint8(length)<<2 | tagCopy1 + return 2 + } + if length < (1<<8)+4 { + length -= 4 + dst[2] = uint8(length) + dst[1] = 0 + dst[0] = 5<<2 | tagCopy1 + return 3 + } + if length < (1<<16)+(1<<8) { + length -= 1 << 8 + dst[3] = uint8(length >> 8) + dst[2] = uint8(length >> 0) + dst[1] = 0 + dst[0] = 6<<2 | tagCopy1 + return 4 + } + const maxRepeat = (1 << 24) - 1 + length -= 1 << 16 + left := 0 + if length > maxRepeat { + left = length - maxRepeat + 4 + length = maxRepeat - 4 + } + dst[4] = uint8(length >> 16) + dst[3] = uint8(length >> 8) + dst[2] = uint8(length >> 0) + dst[1] = 0 + dst[0] = 7<<2 | tagCopy1 + if left > 0 { + return 5 + emitRepeat(dst[5:], offset, left) + } + return 5 +} + +// emitCopy writes a copy chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 1 <= offset && offset <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 +func emitCopy(dst []byte, offset, length int) int { + if offset >= 65536 { + i := 0 + if length > 64 { + // Emit a length 64 copy, encoded as 5 bytes. + dst[4] = uint8(offset >> 24) + dst[3] = uint8(offset >> 16) + dst[2] = uint8(offset >> 8) + dst[1] = uint8(offset) + dst[0] = 63<<2 | tagCopy4 + length -= 64 + if length >= 4 { + // Emit remaining as repeats + return 5 + emitRepeat(dst[5:], offset, length) + } + i = 5 + } + if length == 0 { + return i + } + // Emit a copy, offset encoded as 4 bytes. + dst[i+0] = uint8(length-1)<<2 | tagCopy4 + dst[i+1] = uint8(offset) + dst[i+2] = uint8(offset >> 8) + dst[i+3] = uint8(offset >> 16) + dst[i+4] = uint8(offset >> 24) + return i + 5 + } + + // Offset no more than 2 bytes. + if length > 64 { + off := 3 + if offset < 2048 { + // emit 8 bytes as tagCopy1, rest as repeats. + dst[1] = uint8(offset) + dst[0] = uint8(offset>>8)<<5 | uint8(8-4)<<2 | tagCopy1 + length -= 8 + off = 2 + } else { + // Emit a length 60 copy, encoded as 3 bytes. + // Emit remaining as repeat value (minimum 4 bytes). + dst[2] = uint8(offset >> 8) + dst[1] = uint8(offset) + dst[0] = 59<<2 | tagCopy2 + length -= 60 + } + // Emit remaining as repeats, at least 4 bytes remain. + return off + emitRepeat(dst[off:], offset, length) + } + if length >= 12 || offset >= 2048 { + // Emit the remaining copy, encoded as 3 bytes. + dst[2] = uint8(offset >> 8) + dst[1] = uint8(offset) + dst[0] = uint8(length-1)<<2 | tagCopy2 + return 3 + } + // Emit the remaining copy, encoded as 2 bytes. + dst[1] = uint8(offset) + dst[0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1 + return 2 +} + +// emitCopyNoRepeat writes a copy chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 1 <= offset && offset <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 +func emitCopyNoRepeat(dst []byte, offset, length int) int { + if offset >= 65536 { + i := 0 + if length > 64 { + // Emit a length 64 copy, encoded as 5 bytes. + dst[4] = uint8(offset >> 24) + dst[3] = uint8(offset >> 16) + dst[2] = uint8(offset >> 8) + dst[1] = uint8(offset) + dst[0] = 63<<2 | tagCopy4 + length -= 64 + if length >= 4 { + // Emit remaining as repeats + return 5 + emitCopyNoRepeat(dst[5:], offset, length) + } + i = 5 + } + if length == 0 { + return i + } + // Emit a copy, offset encoded as 4 bytes. + dst[i+0] = uint8(length-1)<<2 | tagCopy4 + dst[i+1] = uint8(offset) + dst[i+2] = uint8(offset >> 8) + dst[i+3] = uint8(offset >> 16) + dst[i+4] = uint8(offset >> 24) + return i + 5 + } + + // Offset no more than 2 bytes. + if length > 64 { + // Emit a length 60 copy, encoded as 3 bytes. + // Emit remaining as repeat value (minimum 4 bytes). + dst[2] = uint8(offset >> 8) + dst[1] = uint8(offset) + dst[0] = 59<<2 | tagCopy2 + length -= 60 + // Emit remaining as repeats, at least 4 bytes remain. + return 3 + emitCopyNoRepeat(dst[3:], offset, length) + } + if length >= 12 || offset >= 2048 { + // Emit the remaining copy, encoded as 3 bytes. + dst[2] = uint8(offset >> 8) + dst[1] = uint8(offset) + dst[0] = uint8(length-1)<<2 | tagCopy2 + return 3 + } + // Emit the remaining copy, encoded as 2 bytes. + dst[1] = uint8(offset) + dst[0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1 + return 2 +} + +// matchLen returns how many bytes match in a and b +// +// It assumes that: +// len(a) <= len(b) +// +func matchLen(a []byte, b []byte) int { + b = b[:len(a)] + var checked int + if len(a) > 4 { + // Try 4 bytes first + if diff := load32(a, 0) ^ load32(b, 0); diff != 0 { + return bits.TrailingZeros32(diff) >> 3 + } + // Switch to 8 byte matching. + checked = 4 + a = a[4:] + b = b[4:] + for len(a) >= 8 { + b = b[:len(a)] + if diff := load64(a, 0) ^ load64(b, 0); diff != 0 { + return checked + (bits.TrailingZeros64(diff) >> 3) + } + checked += 8 + a = a[8:] + b = b[8:] + } + } + b = b[:len(a)] + for i := range a { + if a[i] != b[i] { + return int(i) + checked + } + } + return len(a) + checked +} diff --git a/vendor/github.com/klauspost/compress/s2/encodeblock_amd64.go b/vendor/github.com/klauspost/compress/s2/encodeblock_amd64.go new file mode 100644 index 0000000..88f27c0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encodeblock_amd64.go @@ -0,0 +1,191 @@ +// Code generated by command: go run gen.go -out ../encodeblock_amd64.s -stubs ../encodeblock_amd64.go -pkg=s2. DO NOT EDIT. + +//go:build !appengine && !noasm && gc && !noasm +// +build !appengine,!noasm,gc,!noasm + +package s2 + +func _dummy_() + +// encodeBlockAsm encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4294967295 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBlockAsm(dst []byte, src []byte) int + +// encodeBlockAsm4MB encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4194304 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBlockAsm4MB(dst []byte, src []byte) int + +// encodeBlockAsm12B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 16383 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBlockAsm12B(dst []byte, src []byte) int + +// encodeBlockAsm10B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4095 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBlockAsm10B(dst []byte, src []byte) int + +// encodeBlockAsm8B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 511 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBlockAsm8B(dst []byte, src []byte) int + +// encodeBetterBlockAsm encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4294967295 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBetterBlockAsm(dst []byte, src []byte) int + +// encodeBetterBlockAsm4MB encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4194304 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBetterBlockAsm4MB(dst []byte, src []byte) int + +// encodeBetterBlockAsm12B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 16383 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBetterBlockAsm12B(dst []byte, src []byte) int + +// encodeBetterBlockAsm10B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4095 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBetterBlockAsm10B(dst []byte, src []byte) int + +// encodeBetterBlockAsm8B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 511 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeBetterBlockAsm8B(dst []byte, src []byte) int + +// encodeSnappyBlockAsm encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4294967295 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBlockAsm(dst []byte, src []byte) int + +// encodeSnappyBlockAsm64K encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 65535 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBlockAsm64K(dst []byte, src []byte) int + +// encodeSnappyBlockAsm12B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 16383 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBlockAsm12B(dst []byte, src []byte) int + +// encodeSnappyBlockAsm10B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4095 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBlockAsm10B(dst []byte, src []byte) int + +// encodeSnappyBlockAsm8B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 511 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBlockAsm8B(dst []byte, src []byte) int + +// encodeSnappyBetterBlockAsm encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4294967295 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBetterBlockAsm(dst []byte, src []byte) int + +// encodeSnappyBetterBlockAsm64K encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 65535 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBetterBlockAsm64K(dst []byte, src []byte) int + +// encodeSnappyBetterBlockAsm12B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 16383 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBetterBlockAsm12B(dst []byte, src []byte) int + +// encodeSnappyBetterBlockAsm10B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 4095 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBetterBlockAsm10B(dst []byte, src []byte) int + +// encodeSnappyBetterBlockAsm8B encodes a non-empty src to a guaranteed-large-enough dst. +// Maximum input 511 bytes. +// It assumes that the varint-encoded length of the decompressed bytes has already been written. +// +//go:noescape +func encodeSnappyBetterBlockAsm8B(dst []byte, src []byte) int + +// emitLiteral writes a literal chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes with margin of 0 bytes +// 0 <= len(lit) && len(lit) <= math.MaxUint32 +// +//go:noescape +func emitLiteral(dst []byte, lit []byte) int + +// emitRepeat writes a repeat chunk and returns the number of bytes written. +// Length must be at least 4 and < 1<<32 +// +//go:noescape +func emitRepeat(dst []byte, offset int, length int) int + +// emitCopy writes a copy chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 1 <= offset && offset <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 +// +//go:noescape +func emitCopy(dst []byte, offset int, length int) int + +// emitCopyNoRepeat writes a copy chunk and returns the number of bytes written. +// +// It assumes that: +// dst is long enough to hold the encoded bytes +// 1 <= offset && offset <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 +// +//go:noescape +func emitCopyNoRepeat(dst []byte, offset int, length int) int + +// matchLen returns how many bytes match in a and b +// +// It assumes that: +// len(a) <= len(b) +// +//go:noescape +func matchLen(a []byte, b []byte) int diff --git a/vendor/github.com/klauspost/compress/s2/encodeblock_amd64.s b/vendor/github.com/klauspost/compress/s2/encodeblock_amd64.s new file mode 100644 index 0000000..36915d9 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/encodeblock_amd64.s @@ -0,0 +1,17779 @@ +// Code generated by command: go run gen.go -out ../encodeblock_amd64.s -stubs ../encodeblock_amd64.go -pkg=s2. DO NOT EDIT. + +//go:build !appengine && !noasm && gc && !noasm +// +build !appengine,!noasm,gc,!noasm + +#include "textflag.h" + +// func _dummy_() +TEXT ·_dummy_(SB), $0 +#ifdef GOAMD64_v4 +#ifndef GOAMD64_v3 +#define GOAMD64_v3 +#endif +#endif + RET + +// func encodeBlockAsm(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBlockAsm(SB), $65560-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000200, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBlockAsm: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBlockAsm + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBlockAsm: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x06, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBlockAsm + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + SHLQ $0x10, R11 + IMULQ R9, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeBlockAsm + LEAL 1(CX), DI + MOVL 12(SP), R8 + MOVL DI, SI + SUBL 16(SP), SI + JZ repeat_extend_back_end_encodeBlockAsm + +repeat_extend_back_loop_encodeBlockAsm: + CMPL DI, R8 + JLE repeat_extend_back_end_encodeBlockAsm + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeBlockAsm + LEAL -1(DI), DI + DECL SI + JNZ repeat_extend_back_loop_encodeBlockAsm + +repeat_extend_back_end_encodeBlockAsm: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeBlockAsm + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeBlockAsm + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeBlockAsm + CMPL SI, $0x00010000 + JLT three_bytes_repeat_emit_encodeBlockAsm + CMPL SI, $0x01000000 + JLT four_bytes_repeat_emit_encodeBlockAsm + MOVB $0xfc, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_repeat_emit_encodeBlockAsm + +four_bytes_repeat_emit_encodeBlockAsm: + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_repeat_emit_encodeBlockAsm + +three_bytes_repeat_emit_encodeBlockAsm: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeBlockAsm + +two_bytes_repeat_emit_encodeBlockAsm: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeBlockAsm + JMP memmove_long_repeat_emit_encodeBlockAsm + +one_byte_repeat_emit_encodeBlockAsm: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm + +emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm + +emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm + +emit_lit_memmove_repeat_emit_encodeBlockAsm_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_repeat_emit_encodeBlockAsm: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeBlockAsm + +memmove_long_repeat_emit_encodeBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R12 + SHRQ $0x05, R12 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R13 + SUBQ R11, R13 + DECQ R12 + JA emit_lit_memmove_long_repeat_emit_encodeBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(R10)(R13*1), R11 + LEAQ -32(AX)(R13*1), R14 + +emit_lit_memmove_long_repeat_emit_encodeBlockAsmlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R14) + MOVOA X5, 16(R14) + ADDQ $0x20, R14 + ADDQ $0x20, R11 + ADDQ $0x20, R13 + DECQ R12 + JNA emit_lit_memmove_long_repeat_emit_encodeBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(R10)(R13*1), X4 + MOVOU -16(R10)(R13*1), X5 + MOVOA X4, -32(AX)(R13*1) + MOVOA X5, -16(AX)(R13*1) + ADDQ $0x20, R13 + CMPQ R9, R13 + JAE emit_lit_memmove_long_repeat_emit_encodeBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeBlockAsm: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R9 + SUBL CX, R9 + LEAQ (DX)(CX*1), R10 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R12, R12 + CMPL R9, $0x08 + JL matchlen_match4_repeat_extend_encodeBlockAsm + +matchlen_loopback_repeat_extend_encodeBlockAsm: + MOVQ (R10)(R12*1), R11 + XORQ (SI)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_repeat_extend_encodeBlockAsm + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP repeat_extend_forward_end_encodeBlockAsm + +matchlen_loop_repeat_extend_encodeBlockAsm: + LEAL -8(R9), R9 + LEAL 8(R12), R12 + CMPL R9, $0x08 + JGE matchlen_loopback_repeat_extend_encodeBlockAsm + JZ repeat_extend_forward_end_encodeBlockAsm + +matchlen_match4_repeat_extend_encodeBlockAsm: + CMPL R9, $0x04 + JL matchlen_match2_repeat_extend_encodeBlockAsm + MOVL (R10)(R12*1), R11 + CMPL (SI)(R12*1), R11 + JNE matchlen_match2_repeat_extend_encodeBlockAsm + SUBL $0x04, R9 + LEAL 4(R12), R12 + +matchlen_match2_repeat_extend_encodeBlockAsm: + CMPL R9, $0x02 + JL matchlen_match1_repeat_extend_encodeBlockAsm + MOVW (R10)(R12*1), R11 + CMPW (SI)(R12*1), R11 + JNE matchlen_match1_repeat_extend_encodeBlockAsm + SUBL $0x02, R9 + LEAL 2(R12), R12 + +matchlen_match1_repeat_extend_encodeBlockAsm: + CMPL R9, $0x01 + JL repeat_extend_forward_end_encodeBlockAsm + MOVB (R10)(R12*1), R11 + CMPB (SI)(R12*1), R11 + JNE repeat_extend_forward_end_encodeBlockAsm + LEAL 1(R12), R12 + +repeat_extend_forward_end_encodeBlockAsm: + ADDL R12, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + TESTL R8, R8 + JZ repeat_as_copy_encodeBlockAsm + + // emitRepeat +emit_repeat_again_match_repeat_encodeBlockAsm: + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_match_repeat_encodeBlockAsm + CMPL R8, $0x0c + JGE cant_repeat_two_offset_match_repeat_encodeBlockAsm + CMPL DI, $0x00000800 + JLT repeat_two_offset_match_repeat_encodeBlockAsm + +cant_repeat_two_offset_match_repeat_encodeBlockAsm: + CMPL SI, $0x00000104 + JLT repeat_three_match_repeat_encodeBlockAsm + CMPL SI, $0x00010100 + JLT repeat_four_match_repeat_encodeBlockAsm + CMPL SI, $0x0100ffff + JLT repeat_five_match_repeat_encodeBlockAsm + LEAL -16842747(SI), SI + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_repeat_encodeBlockAsm + +repeat_five_match_repeat_encodeBlockAsm: + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_four_match_repeat_encodeBlockAsm: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_three_match_repeat_encodeBlockAsm: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_match_repeat_encodeBlockAsm: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_offset_match_repeat_encodeBlockAsm: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_as_copy_encodeBlockAsm: + // emitCopy + CMPL DI, $0x00010000 + JL two_byte_offset_repeat_as_copy_encodeBlockAsm + +four_bytes_loop_back_repeat_as_copy_encodeBlockAsm: + CMPL SI, $0x40 + JLE four_bytes_remain_repeat_as_copy_encodeBlockAsm + MOVB $0xff, (AX) + MOVL DI, 1(AX) + LEAL -64(SI), SI + ADDQ $0x05, AX + CMPL SI, $0x04 + JL four_bytes_remain_repeat_as_copy_encodeBlockAsm + + // emitRepeat +emit_repeat_again_repeat_as_copy_encodeBlockAsm_emit_copy: + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm_emit_copy + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm_emit_copy + CMPL SI, $0x00010100 + JLT repeat_four_repeat_as_copy_encodeBlockAsm_emit_copy + CMPL SI, $0x0100ffff + JLT repeat_five_repeat_as_copy_encodeBlockAsm_emit_copy + LEAL -16842747(SI), SI + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_repeat_as_copy_encodeBlockAsm_emit_copy + +repeat_five_repeat_as_copy_encodeBlockAsm_emit_copy: + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_four_repeat_as_copy_encodeBlockAsm_emit_copy: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_three_repeat_as_copy_encodeBlockAsm_emit_copy: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_repeat_as_copy_encodeBlockAsm_emit_copy: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + JMP four_bytes_loop_back_repeat_as_copy_encodeBlockAsm + +four_bytes_remain_repeat_as_copy_encodeBlockAsm: + TESTL SI, SI + JZ repeat_end_emit_encodeBlockAsm + MOVB $0x03, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVL DI, 1(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm + +two_byte_offset_repeat_as_copy_encodeBlockAsm: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeBlockAsm + CMPL DI, $0x00000800 + JAE long_offset_short_repeat_as_copy_encodeBlockAsm + MOVL $0x00000001, R8 + LEAL 16(R8), R8 + MOVB DI, 1(AX) + MOVL DI, R9 + SHRL $0x08, R9 + SHLL $0x05, R9 + ORL R9, R8 + MOVB R8, (AX) + ADDQ $0x02, AX + SUBL $0x08, SI + + // emitRepeat + LEAL -4(SI), SI + JMP cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + +emit_repeat_again_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + CMPL SI, $0x00010100 + JLT repeat_four_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + CMPL SI, $0x0100ffff + JLT repeat_five_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + LEAL -16842747(SI), SI + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b + +repeat_five_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_four_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_three_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short_2b: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +long_offset_short_repeat_as_copy_encodeBlockAsm: + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + + // emitRepeat +emit_repeat_again_repeat_as_copy_encodeBlockAsm_emit_copy_short: + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm_emit_copy_short + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm_emit_copy_short + CMPL SI, $0x00010100 + JLT repeat_four_repeat_as_copy_encodeBlockAsm_emit_copy_short + CMPL SI, $0x0100ffff + JLT repeat_five_repeat_as_copy_encodeBlockAsm_emit_copy_short + LEAL -16842747(SI), SI + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_repeat_as_copy_encodeBlockAsm_emit_copy_short + +repeat_five_repeat_as_copy_encodeBlockAsm_emit_copy_short: + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_four_repeat_as_copy_encodeBlockAsm_emit_copy_short: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_three_repeat_as_copy_encodeBlockAsm_emit_copy_short: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_repeat_as_copy_encodeBlockAsm_emit_copy_short: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +repeat_two_offset_repeat_as_copy_encodeBlockAsm_emit_copy_short: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + JMP two_byte_offset_repeat_as_copy_encodeBlockAsm + +two_byte_offset_short_repeat_as_copy_encodeBlockAsm: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm + +emit_copy_three_repeat_as_copy_encodeBlockAsm: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeBlockAsm: + MOVL CX, 12(SP) + JMP search_loop_encodeBlockAsm + +no_repeat_found_encodeBlockAsm: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBlockAsm + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeBlockAsm + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeBlockAsm + MOVL 20(SP), CX + JMP search_loop_encodeBlockAsm + +candidate3_match_encodeBlockAsm: + ADDL $0x02, CX + JMP candidate_match_encodeBlockAsm + +candidate2_match_encodeBlockAsm: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeBlockAsm: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBlockAsm + +match_extend_back_loop_encodeBlockAsm: + CMPL CX, DI + JLE match_extend_back_end_encodeBlockAsm + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBlockAsm + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBlockAsm + JMP match_extend_back_loop_encodeBlockAsm + +match_extend_back_end_encodeBlockAsm: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 5(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBlockAsm: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeBlockAsm + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeBlockAsm + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeBlockAsm + CMPL R8, $0x00010000 + JLT three_bytes_match_emit_encodeBlockAsm + CMPL R8, $0x01000000 + JLT four_bytes_match_emit_encodeBlockAsm + MOVB $0xfc, (AX) + MOVL R8, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_match_emit_encodeBlockAsm + +four_bytes_match_emit_encodeBlockAsm: + MOVL R8, R10 + SHRL $0x10, R10 + MOVB $0xf8, (AX) + MOVW R8, 1(AX) + MOVB R10, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_encodeBlockAsm + +three_bytes_match_emit_encodeBlockAsm: + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBlockAsm + +two_bytes_match_emit_encodeBlockAsm: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeBlockAsm + JMP memmove_long_match_emit_encodeBlockAsm + +one_byte_match_emit_encodeBlockAsm: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBlockAsm: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeBlockAsm + +emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm + +emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm + +emit_lit_memmove_match_emit_encodeBlockAsm_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBlockAsm: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeBlockAsm + +memmove_long_match_emit_encodeBlockAsm: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeBlockAsmlarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeBlockAsm: +match_nolit_loop_encodeBlockAsm: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeBlockAsm + +matchlen_loopback_match_nolit_encodeBlockAsm: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeBlockAsm + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeBlockAsm + +matchlen_loop_match_nolit_encodeBlockAsm: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeBlockAsm + JZ match_nolit_end_encodeBlockAsm + +matchlen_match4_match_nolit_encodeBlockAsm: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeBlockAsm + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeBlockAsm + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeBlockAsm: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeBlockAsm + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeBlockAsm + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeBlockAsm: + CMPL DI, $0x01 + JL match_nolit_end_encodeBlockAsm + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeBlockAsm + LEAL 1(R10), R10 + +match_nolit_end_encodeBlockAsm: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy + CMPL SI, $0x00010000 + JL two_byte_offset_match_nolit_encodeBlockAsm + +four_bytes_loop_back_match_nolit_encodeBlockAsm: + CMPL R10, $0x40 + JLE four_bytes_remain_match_nolit_encodeBlockAsm + MOVB $0xff, (AX) + MOVL SI, 1(AX) + LEAL -64(R10), R10 + ADDQ $0x05, AX + CMPL R10, $0x04 + JL four_bytes_remain_match_nolit_encodeBlockAsm + + // emitRepeat +emit_repeat_again_match_nolit_encodeBlockAsm_emit_copy: + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm_emit_copy + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy + +cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm_emit_copy + CMPL R10, $0x00010100 + JLT repeat_four_match_nolit_encodeBlockAsm_emit_copy + CMPL R10, $0x0100ffff + JLT repeat_five_match_nolit_encodeBlockAsm_emit_copy + LEAL -16842747(R10), R10 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_encodeBlockAsm_emit_copy + +repeat_five_match_nolit_encodeBlockAsm_emit_copy: + LEAL -65536(R10), R10 + MOVL R10, SI + MOVW $0x001d, (AX) + MOVW R10, 2(AX) + SARL $0x10, SI + MOVB SI, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_four_match_nolit_encodeBlockAsm_emit_copy: + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_three_match_nolit_encodeBlockAsm_emit_copy: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_two_match_nolit_encodeBlockAsm_emit_copy: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + JMP four_bytes_loop_back_match_nolit_encodeBlockAsm + +four_bytes_remain_match_nolit_encodeBlockAsm: + TESTL R10, R10 + JZ match_nolit_emitcopy_end_encodeBlockAsm + MOVB $0x03, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +two_byte_offset_match_nolit_encodeBlockAsm: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBlockAsm + CMPL SI, $0x00000800 + JAE long_offset_short_match_nolit_encodeBlockAsm + MOVL $0x00000001, DI + LEAL 16(DI), DI + MOVB SI, 1(AX) + MOVL SI, R8 + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, DI + MOVB DI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R10 + + // emitRepeat + LEAL -4(R10), R10 + JMP cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short_2b + +emit_repeat_again_match_nolit_encodeBlockAsm_emit_copy_short_2b: + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm_emit_copy_short_2b + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short_2b + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short_2b: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm_emit_copy_short_2b + CMPL R10, $0x00010100 + JLT repeat_four_match_nolit_encodeBlockAsm_emit_copy_short_2b + CMPL R10, $0x0100ffff + JLT repeat_five_match_nolit_encodeBlockAsm_emit_copy_short_2b + LEAL -16842747(R10), R10 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_encodeBlockAsm_emit_copy_short_2b + +repeat_five_match_nolit_encodeBlockAsm_emit_copy_short_2b: + LEAL -65536(R10), R10 + MOVL R10, SI + MOVW $0x001d, (AX) + MOVW R10, 2(AX) + SARL $0x10, SI + MOVB SI, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_four_match_nolit_encodeBlockAsm_emit_copy_short_2b: + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_three_match_nolit_encodeBlockAsm_emit_copy_short_2b: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_two_match_nolit_encodeBlockAsm_emit_copy_short_2b: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short_2b: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +long_offset_short_match_nolit_encodeBlockAsm: + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + + // emitRepeat +emit_repeat_again_match_nolit_encodeBlockAsm_emit_copy_short: + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm_emit_copy_short + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm_emit_copy_short + CMPL R10, $0x00010100 + JLT repeat_four_match_nolit_encodeBlockAsm_emit_copy_short + CMPL R10, $0x0100ffff + JLT repeat_five_match_nolit_encodeBlockAsm_emit_copy_short + LEAL -16842747(R10), R10 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_encodeBlockAsm_emit_copy_short + +repeat_five_match_nolit_encodeBlockAsm_emit_copy_short: + LEAL -65536(R10), R10 + MOVL R10, SI + MOVW $0x001d, (AX) + MOVW R10, 2(AX) + SARL $0x10, SI + MOVB SI, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_four_match_nolit_encodeBlockAsm_emit_copy_short: + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_three_match_nolit_encodeBlockAsm_emit_copy_short: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_two_match_nolit_encodeBlockAsm_emit_copy_short: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +repeat_two_offset_match_nolit_encodeBlockAsm_emit_copy_short: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + JMP two_byte_offset_match_nolit_encodeBlockAsm + +two_byte_offset_short_match_nolit_encodeBlockAsm: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeBlockAsm + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBlockAsm + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm + +emit_copy_three_match_nolit_encodeBlockAsm: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeBlockAsm: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBlockAsm + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBlockAsm: + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x10, R8 + IMULQ R9, R8 + SHRQ $0x32, R8 + SHLQ $0x10, SI + IMULQ R9, SI + SHRQ $0x32, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeBlockAsm + INCL CX + JMP search_loop_encodeBlockAsm + +emit_remainder_encodeBlockAsm: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 5(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBlockAsm: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBlockAsm + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBlockAsm + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBlockAsm + CMPL DX, $0x00010000 + JLT three_bytes_emit_remainder_encodeBlockAsm + CMPL DX, $0x01000000 + JLT four_bytes_emit_remainder_encodeBlockAsm + MOVB $0xfc, (AX) + MOVL DX, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_emit_remainder_encodeBlockAsm + +four_bytes_emit_remainder_encodeBlockAsm: + MOVL DX, BX + SHRL $0x10, BX + MOVB $0xf8, (AX) + MOVW DX, 1(AX) + MOVB BL, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_emit_remainder_encodeBlockAsm + +three_bytes_emit_remainder_encodeBlockAsm: + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBlockAsm + +two_bytes_emit_remainder_encodeBlockAsm: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBlockAsm + JMP memmove_long_emit_remainder_encodeBlockAsm + +one_byte_emit_remainder_encodeBlockAsm: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm + +emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm + +emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm + +emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm + +emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm + +emit_lit_memmove_emit_remainder_encodeBlockAsm_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBlockAsm: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBlockAsm + +memmove_long_emit_remainder_encodeBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBlockAsmlarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBlockAsm: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBlockAsm4MB(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBlockAsm4MB(SB), $65560-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000200, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBlockAsm4MB: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBlockAsm4MB + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBlockAsm4MB: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x06, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBlockAsm4MB + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + SHLQ $0x10, R11 + IMULQ R9, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeBlockAsm4MB + LEAL 1(CX), DI + MOVL 12(SP), R8 + MOVL DI, SI + SUBL 16(SP), SI + JZ repeat_extend_back_end_encodeBlockAsm4MB + +repeat_extend_back_loop_encodeBlockAsm4MB: + CMPL DI, R8 + JLE repeat_extend_back_end_encodeBlockAsm4MB + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeBlockAsm4MB + LEAL -1(DI), DI + DECL SI + JNZ repeat_extend_back_loop_encodeBlockAsm4MB + +repeat_extend_back_end_encodeBlockAsm4MB: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeBlockAsm4MB + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeBlockAsm4MB + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeBlockAsm4MB + CMPL SI, $0x00010000 + JLT three_bytes_repeat_emit_encodeBlockAsm4MB + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_repeat_emit_encodeBlockAsm4MB + +three_bytes_repeat_emit_encodeBlockAsm4MB: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeBlockAsm4MB + +two_bytes_repeat_emit_encodeBlockAsm4MB: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeBlockAsm4MB + JMP memmove_long_repeat_emit_encodeBlockAsm4MB + +one_byte_repeat_emit_encodeBlockAsm4MB: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeBlockAsm4MB: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm4MB + +emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm4MB + +emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm4MB + +emit_lit_memmove_repeat_emit_encodeBlockAsm4MB_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_repeat_emit_encodeBlockAsm4MB: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeBlockAsm4MB + +memmove_long_repeat_emit_encodeBlockAsm4MB: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R12 + SHRQ $0x05, R12 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R13 + SUBQ R11, R13 + DECQ R12 + JA emit_lit_memmove_long_repeat_emit_encodeBlockAsm4MBlarge_forward_sse_loop_32 + LEAQ -32(R10)(R13*1), R11 + LEAQ -32(AX)(R13*1), R14 + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm4MBlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R14) + MOVOA X5, 16(R14) + ADDQ $0x20, R14 + ADDQ $0x20, R11 + ADDQ $0x20, R13 + DECQ R12 + JNA emit_lit_memmove_long_repeat_emit_encodeBlockAsm4MBlarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm4MBlarge_forward_sse_loop_32: + MOVOU -32(R10)(R13*1), X4 + MOVOU -16(R10)(R13*1), X5 + MOVOA X4, -32(AX)(R13*1) + MOVOA X5, -16(AX)(R13*1) + ADDQ $0x20, R13 + CMPQ R9, R13 + JAE emit_lit_memmove_long_repeat_emit_encodeBlockAsm4MBlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeBlockAsm4MB: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R9 + SUBL CX, R9 + LEAQ (DX)(CX*1), R10 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R12, R12 + CMPL R9, $0x08 + JL matchlen_match4_repeat_extend_encodeBlockAsm4MB + +matchlen_loopback_repeat_extend_encodeBlockAsm4MB: + MOVQ (R10)(R12*1), R11 + XORQ (SI)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_repeat_extend_encodeBlockAsm4MB + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP repeat_extend_forward_end_encodeBlockAsm4MB + +matchlen_loop_repeat_extend_encodeBlockAsm4MB: + LEAL -8(R9), R9 + LEAL 8(R12), R12 + CMPL R9, $0x08 + JGE matchlen_loopback_repeat_extend_encodeBlockAsm4MB + JZ repeat_extend_forward_end_encodeBlockAsm4MB + +matchlen_match4_repeat_extend_encodeBlockAsm4MB: + CMPL R9, $0x04 + JL matchlen_match2_repeat_extend_encodeBlockAsm4MB + MOVL (R10)(R12*1), R11 + CMPL (SI)(R12*1), R11 + JNE matchlen_match2_repeat_extend_encodeBlockAsm4MB + SUBL $0x04, R9 + LEAL 4(R12), R12 + +matchlen_match2_repeat_extend_encodeBlockAsm4MB: + CMPL R9, $0x02 + JL matchlen_match1_repeat_extend_encodeBlockAsm4MB + MOVW (R10)(R12*1), R11 + CMPW (SI)(R12*1), R11 + JNE matchlen_match1_repeat_extend_encodeBlockAsm4MB + SUBL $0x02, R9 + LEAL 2(R12), R12 + +matchlen_match1_repeat_extend_encodeBlockAsm4MB: + CMPL R9, $0x01 + JL repeat_extend_forward_end_encodeBlockAsm4MB + MOVB (R10)(R12*1), R11 + CMPB (SI)(R12*1), R11 + JNE repeat_extend_forward_end_encodeBlockAsm4MB + LEAL 1(R12), R12 + +repeat_extend_forward_end_encodeBlockAsm4MB: + ADDL R12, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + TESTL R8, R8 + JZ repeat_as_copy_encodeBlockAsm4MB + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_match_repeat_encodeBlockAsm4MB + CMPL R8, $0x0c + JGE cant_repeat_two_offset_match_repeat_encodeBlockAsm4MB + CMPL DI, $0x00000800 + JLT repeat_two_offset_match_repeat_encodeBlockAsm4MB + +cant_repeat_two_offset_match_repeat_encodeBlockAsm4MB: + CMPL SI, $0x00000104 + JLT repeat_three_match_repeat_encodeBlockAsm4MB + CMPL SI, $0x00010100 + JLT repeat_four_match_repeat_encodeBlockAsm4MB + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_four_match_repeat_encodeBlockAsm4MB: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_three_match_repeat_encodeBlockAsm4MB: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_match_repeat_encodeBlockAsm4MB: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_offset_match_repeat_encodeBlockAsm4MB: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_as_copy_encodeBlockAsm4MB: + // emitCopy + CMPL DI, $0x00010000 + JL two_byte_offset_repeat_as_copy_encodeBlockAsm4MB + +four_bytes_loop_back_repeat_as_copy_encodeBlockAsm4MB: + CMPL SI, $0x40 + JLE four_bytes_remain_repeat_as_copy_encodeBlockAsm4MB + MOVB $0xff, (AX) + MOVL DI, 1(AX) + LEAL -64(SI), SI + ADDQ $0x05, AX + CMPL SI, $0x04 + JL four_bytes_remain_repeat_as_copy_encodeBlockAsm4MB + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm4MB_emit_copy + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm4MB_emit_copy + CMPL SI, $0x00010100 + JLT repeat_four_repeat_as_copy_encodeBlockAsm4MB_emit_copy + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_four_repeat_as_copy_encodeBlockAsm4MB_emit_copy: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_three_repeat_as_copy_encodeBlockAsm4MB_emit_copy: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_repeat_as_copy_encodeBlockAsm4MB_emit_copy: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + JMP four_bytes_loop_back_repeat_as_copy_encodeBlockAsm4MB + +four_bytes_remain_repeat_as_copy_encodeBlockAsm4MB: + TESTL SI, SI + JZ repeat_end_emit_encodeBlockAsm4MB + MOVB $0x03, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVL DI, 1(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +two_byte_offset_repeat_as_copy_encodeBlockAsm4MB: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeBlockAsm4MB + CMPL DI, $0x00000800 + JAE long_offset_short_repeat_as_copy_encodeBlockAsm4MB + MOVL $0x00000001, R8 + LEAL 16(R8), R8 + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, R8 + MOVB R8, (AX) + ADDQ $0x02, AX + SUBL $0x08, SI + + // emitRepeat + LEAL -4(SI), SI + JMP cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b + CMPL SI, $0x00010100 + JLT repeat_four_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_four_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_three_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short_2b: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +long_offset_short_repeat_as_copy_encodeBlockAsm4MB: + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short + CMPL SI, $0x00010100 + JLT repeat_four_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short + LEAL -65536(SI), SI + MOVL SI, DI + MOVW $0x001d, (AX) + MOVW SI, 2(AX) + SARL $0x10, DI + MOVB DI, 4(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_four_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short: + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_three_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +repeat_two_offset_repeat_as_copy_encodeBlockAsm4MB_emit_copy_short: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + JMP two_byte_offset_repeat_as_copy_encodeBlockAsm4MB + +two_byte_offset_short_repeat_as_copy_encodeBlockAsm4MB: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm4MB + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm4MB + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm4MB + +emit_copy_three_repeat_as_copy_encodeBlockAsm4MB: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeBlockAsm4MB: + MOVL CX, 12(SP) + JMP search_loop_encodeBlockAsm4MB + +no_repeat_found_encodeBlockAsm4MB: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBlockAsm4MB + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeBlockAsm4MB + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeBlockAsm4MB + MOVL 20(SP), CX + JMP search_loop_encodeBlockAsm4MB + +candidate3_match_encodeBlockAsm4MB: + ADDL $0x02, CX + JMP candidate_match_encodeBlockAsm4MB + +candidate2_match_encodeBlockAsm4MB: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeBlockAsm4MB: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBlockAsm4MB + +match_extend_back_loop_encodeBlockAsm4MB: + CMPL CX, DI + JLE match_extend_back_end_encodeBlockAsm4MB + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBlockAsm4MB + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBlockAsm4MB + JMP match_extend_back_loop_encodeBlockAsm4MB + +match_extend_back_end_encodeBlockAsm4MB: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 4(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBlockAsm4MB + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBlockAsm4MB: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeBlockAsm4MB + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeBlockAsm4MB + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeBlockAsm4MB + CMPL R8, $0x00010000 + JLT three_bytes_match_emit_encodeBlockAsm4MB + MOVL R8, R10 + SHRL $0x10, R10 + MOVB $0xf8, (AX) + MOVW R8, 1(AX) + MOVB R10, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_encodeBlockAsm4MB + +three_bytes_match_emit_encodeBlockAsm4MB: + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBlockAsm4MB + +two_bytes_match_emit_encodeBlockAsm4MB: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeBlockAsm4MB + JMP memmove_long_match_emit_encodeBlockAsm4MB + +one_byte_match_emit_encodeBlockAsm4MB: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBlockAsm4MB: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBlockAsm4MB_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBlockAsm4MB: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeBlockAsm4MB + +memmove_long_match_emit_encodeBlockAsm4MB: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeBlockAsm4MBlarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeBlockAsm4MBlarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeBlockAsm4MBlarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBlockAsm4MBlarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeBlockAsm4MBlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeBlockAsm4MB: +match_nolit_loop_encodeBlockAsm4MB: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeBlockAsm4MB + +matchlen_loopback_match_nolit_encodeBlockAsm4MB: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeBlockAsm4MB + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeBlockAsm4MB + +matchlen_loop_match_nolit_encodeBlockAsm4MB: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeBlockAsm4MB + JZ match_nolit_end_encodeBlockAsm4MB + +matchlen_match4_match_nolit_encodeBlockAsm4MB: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeBlockAsm4MB + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeBlockAsm4MB + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeBlockAsm4MB: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeBlockAsm4MB + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeBlockAsm4MB + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeBlockAsm4MB: + CMPL DI, $0x01 + JL match_nolit_end_encodeBlockAsm4MB + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeBlockAsm4MB + LEAL 1(R10), R10 + +match_nolit_end_encodeBlockAsm4MB: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy + CMPL SI, $0x00010000 + JL two_byte_offset_match_nolit_encodeBlockAsm4MB + +four_bytes_loop_back_match_nolit_encodeBlockAsm4MB: + CMPL R10, $0x40 + JLE four_bytes_remain_match_nolit_encodeBlockAsm4MB + MOVB $0xff, (AX) + MOVL SI, 1(AX) + LEAL -64(R10), R10 + ADDQ $0x05, AX + CMPL R10, $0x04 + JL four_bytes_remain_match_nolit_encodeBlockAsm4MB + + // emitRepeat + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm4MB_emit_copy + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy + +cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm4MB_emit_copy + CMPL R10, $0x00010100 + JLT repeat_four_match_nolit_encodeBlockAsm4MB_emit_copy + LEAL -65536(R10), R10 + MOVL R10, SI + MOVW $0x001d, (AX) + MOVW R10, 2(AX) + SARL $0x10, SI + MOVB SI, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_four_match_nolit_encodeBlockAsm4MB_emit_copy: + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_three_match_nolit_encodeBlockAsm4MB_emit_copy: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_two_match_nolit_encodeBlockAsm4MB_emit_copy: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + JMP four_bytes_loop_back_match_nolit_encodeBlockAsm4MB + +four_bytes_remain_match_nolit_encodeBlockAsm4MB: + TESTL R10, R10 + JZ match_nolit_emitcopy_end_encodeBlockAsm4MB + MOVB $0x03, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +two_byte_offset_match_nolit_encodeBlockAsm4MB: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBlockAsm4MB + CMPL SI, $0x00000800 + JAE long_offset_short_match_nolit_encodeBlockAsm4MB + MOVL $0x00000001, DI + LEAL 16(DI), DI + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, DI + MOVB DI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R10 + + // emitRepeat + LEAL -4(R10), R10 + JMP cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b + CMPL R10, $0x00010100 + JLT repeat_four_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b + LEAL -65536(R10), R10 + MOVL R10, SI + MOVW $0x001d, (AX) + MOVW R10, 2(AX) + SARL $0x10, SI + MOVB SI, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_four_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b: + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_three_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_two_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short_2b: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +long_offset_short_match_nolit_encodeBlockAsm4MB: + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + + // emitRepeat + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm4MB_emit_copy_short + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm4MB_emit_copy_short + CMPL R10, $0x00010100 + JLT repeat_four_match_nolit_encodeBlockAsm4MB_emit_copy_short + LEAL -65536(R10), R10 + MOVL R10, SI + MOVW $0x001d, (AX) + MOVW R10, 2(AX) + SARL $0x10, SI + MOVB SI, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_four_match_nolit_encodeBlockAsm4MB_emit_copy_short: + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_three_match_nolit_encodeBlockAsm4MB_emit_copy_short: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_two_match_nolit_encodeBlockAsm4MB_emit_copy_short: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +repeat_two_offset_match_nolit_encodeBlockAsm4MB_emit_copy_short: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + JMP two_byte_offset_match_nolit_encodeBlockAsm4MB + +two_byte_offset_short_match_nolit_encodeBlockAsm4MB: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeBlockAsm4MB + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBlockAsm4MB + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm4MB + +emit_copy_three_match_nolit_encodeBlockAsm4MB: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeBlockAsm4MB: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBlockAsm4MB + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBlockAsm4MB + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBlockAsm4MB: + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x10, R8 + IMULQ R9, R8 + SHRQ $0x32, R8 + SHLQ $0x10, SI + IMULQ R9, SI + SHRQ $0x32, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeBlockAsm4MB + INCL CX + JMP search_loop_encodeBlockAsm4MB + +emit_remainder_encodeBlockAsm4MB: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 4(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBlockAsm4MB + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBlockAsm4MB: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBlockAsm4MB + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBlockAsm4MB + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBlockAsm4MB + CMPL DX, $0x00010000 + JLT three_bytes_emit_remainder_encodeBlockAsm4MB + MOVL DX, BX + SHRL $0x10, BX + MOVB $0xf8, (AX) + MOVW DX, 1(AX) + MOVB BL, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_emit_remainder_encodeBlockAsm4MB + +three_bytes_emit_remainder_encodeBlockAsm4MB: + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBlockAsm4MB + +two_bytes_emit_remainder_encodeBlockAsm4MB: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBlockAsm4MB + JMP memmove_long_emit_remainder_encodeBlockAsm4MB + +one_byte_emit_remainder_encodeBlockAsm4MB: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBlockAsm4MB: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBlockAsm4MB_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBlockAsm4MB: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBlockAsm4MB + +memmove_long_emit_remainder_encodeBlockAsm4MB: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBlockAsm4MBlarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm4MBlarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBlockAsm4MBlarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm4MBlarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBlockAsm4MBlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBlockAsm4MB: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBlockAsm12B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBlockAsm12B(SB), $16408-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000080, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBlockAsm12B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBlockAsm12B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBlockAsm12B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x05, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBlockAsm12B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x000000cf1bbcdcbb, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x18, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + SHLQ $0x18, R11 + IMULQ R9, R11 + SHRQ $0x34, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x18, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeBlockAsm12B + LEAL 1(CX), DI + MOVL 12(SP), R8 + MOVL DI, SI + SUBL 16(SP), SI + JZ repeat_extend_back_end_encodeBlockAsm12B + +repeat_extend_back_loop_encodeBlockAsm12B: + CMPL DI, R8 + JLE repeat_extend_back_end_encodeBlockAsm12B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeBlockAsm12B + LEAL -1(DI), DI + DECL SI + JNZ repeat_extend_back_loop_encodeBlockAsm12B + +repeat_extend_back_end_encodeBlockAsm12B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeBlockAsm12B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeBlockAsm12B + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeBlockAsm12B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeBlockAsm12B + +two_bytes_repeat_emit_encodeBlockAsm12B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeBlockAsm12B + JMP memmove_long_repeat_emit_encodeBlockAsm12B + +one_byte_repeat_emit_encodeBlockAsm12B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm12B + +emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm12B + +emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm12B + +emit_lit_memmove_repeat_emit_encodeBlockAsm12B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_repeat_emit_encodeBlockAsm12B: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeBlockAsm12B + +memmove_long_repeat_emit_encodeBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R12 + SHRQ $0x05, R12 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R13 + SUBQ R11, R13 + DECQ R12 + JA emit_lit_memmove_long_repeat_emit_encodeBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R13*1), R11 + LEAQ -32(AX)(R13*1), R14 + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm12Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R14) + MOVOA X5, 16(R14) + ADDQ $0x20, R14 + ADDQ $0x20, R11 + ADDQ $0x20, R13 + DECQ R12 + JNA emit_lit_memmove_long_repeat_emit_encodeBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R13*1), X4 + MOVOU -16(R10)(R13*1), X5 + MOVOA X4, -32(AX)(R13*1) + MOVOA X5, -16(AX)(R13*1) + ADDQ $0x20, R13 + CMPQ R9, R13 + JAE emit_lit_memmove_long_repeat_emit_encodeBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeBlockAsm12B: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R9 + SUBL CX, R9 + LEAQ (DX)(CX*1), R10 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R12, R12 + CMPL R9, $0x08 + JL matchlen_match4_repeat_extend_encodeBlockAsm12B + +matchlen_loopback_repeat_extend_encodeBlockAsm12B: + MOVQ (R10)(R12*1), R11 + XORQ (SI)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_repeat_extend_encodeBlockAsm12B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP repeat_extend_forward_end_encodeBlockAsm12B + +matchlen_loop_repeat_extend_encodeBlockAsm12B: + LEAL -8(R9), R9 + LEAL 8(R12), R12 + CMPL R9, $0x08 + JGE matchlen_loopback_repeat_extend_encodeBlockAsm12B + JZ repeat_extend_forward_end_encodeBlockAsm12B + +matchlen_match4_repeat_extend_encodeBlockAsm12B: + CMPL R9, $0x04 + JL matchlen_match2_repeat_extend_encodeBlockAsm12B + MOVL (R10)(R12*1), R11 + CMPL (SI)(R12*1), R11 + JNE matchlen_match2_repeat_extend_encodeBlockAsm12B + SUBL $0x04, R9 + LEAL 4(R12), R12 + +matchlen_match2_repeat_extend_encodeBlockAsm12B: + CMPL R9, $0x02 + JL matchlen_match1_repeat_extend_encodeBlockAsm12B + MOVW (R10)(R12*1), R11 + CMPW (SI)(R12*1), R11 + JNE matchlen_match1_repeat_extend_encodeBlockAsm12B + SUBL $0x02, R9 + LEAL 2(R12), R12 + +matchlen_match1_repeat_extend_encodeBlockAsm12B: + CMPL R9, $0x01 + JL repeat_extend_forward_end_encodeBlockAsm12B + MOVB (R10)(R12*1), R11 + CMPB (SI)(R12*1), R11 + JNE repeat_extend_forward_end_encodeBlockAsm12B + LEAL 1(R12), R12 + +repeat_extend_forward_end_encodeBlockAsm12B: + ADDL R12, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + TESTL R8, R8 + JZ repeat_as_copy_encodeBlockAsm12B + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_match_repeat_encodeBlockAsm12B + CMPL R8, $0x0c + JGE cant_repeat_two_offset_match_repeat_encodeBlockAsm12B + CMPL DI, $0x00000800 + JLT repeat_two_offset_match_repeat_encodeBlockAsm12B + +cant_repeat_two_offset_match_repeat_encodeBlockAsm12B: + CMPL SI, $0x00000104 + JLT repeat_three_match_repeat_encodeBlockAsm12B + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_three_match_repeat_encodeBlockAsm12B: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_two_match_repeat_encodeBlockAsm12B: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_two_offset_match_repeat_encodeBlockAsm12B: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_as_copy_encodeBlockAsm12B: + // emitCopy +two_byte_offset_repeat_as_copy_encodeBlockAsm12B: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeBlockAsm12B + CMPL DI, $0x00000800 + JAE long_offset_short_repeat_as_copy_encodeBlockAsm12B + MOVL $0x00000001, R8 + LEAL 16(R8), R8 + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, R8 + MOVB R8, (AX) + ADDQ $0x02, AX + SUBL $0x08, SI + + // emitRepeat + LEAL -4(SI), SI + JMP cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_three_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_two_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short_2b: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + +long_offset_short_repeat_as_copy_encodeBlockAsm12B: + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm12B_emit_copy_short + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm12B_emit_copy_short + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_three_repeat_as_copy_encodeBlockAsm12B_emit_copy_short: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_two_repeat_as_copy_encodeBlockAsm12B_emit_copy_short: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + +repeat_two_offset_repeat_as_copy_encodeBlockAsm12B_emit_copy_short: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + JMP two_byte_offset_repeat_as_copy_encodeBlockAsm12B + +two_byte_offset_short_repeat_as_copy_encodeBlockAsm12B: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm12B + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm12B + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm12B + +emit_copy_three_repeat_as_copy_encodeBlockAsm12B: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeBlockAsm12B: + MOVL CX, 12(SP) + JMP search_loop_encodeBlockAsm12B + +no_repeat_found_encodeBlockAsm12B: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBlockAsm12B + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeBlockAsm12B + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeBlockAsm12B + MOVL 20(SP), CX + JMP search_loop_encodeBlockAsm12B + +candidate3_match_encodeBlockAsm12B: + ADDL $0x02, CX + JMP candidate_match_encodeBlockAsm12B + +candidate2_match_encodeBlockAsm12B: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeBlockAsm12B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBlockAsm12B + +match_extend_back_loop_encodeBlockAsm12B: + CMPL CX, DI + JLE match_extend_back_end_encodeBlockAsm12B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBlockAsm12B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBlockAsm12B + JMP match_extend_back_loop_encodeBlockAsm12B + +match_extend_back_end_encodeBlockAsm12B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBlockAsm12B: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeBlockAsm12B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeBlockAsm12B + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeBlockAsm12B + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBlockAsm12B + +two_bytes_match_emit_encodeBlockAsm12B: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeBlockAsm12B + JMP memmove_long_match_emit_encodeBlockAsm12B + +one_byte_match_emit_encodeBlockAsm12B: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBlockAsm12B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeBlockAsm12B + +emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm12B + +emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm12B + +emit_lit_memmove_match_emit_encodeBlockAsm12B_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBlockAsm12B: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeBlockAsm12B + +memmove_long_match_emit_encodeBlockAsm12B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeBlockAsm12Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeBlockAsm12B: +match_nolit_loop_encodeBlockAsm12B: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeBlockAsm12B + +matchlen_loopback_match_nolit_encodeBlockAsm12B: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeBlockAsm12B + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeBlockAsm12B + +matchlen_loop_match_nolit_encodeBlockAsm12B: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeBlockAsm12B + JZ match_nolit_end_encodeBlockAsm12B + +matchlen_match4_match_nolit_encodeBlockAsm12B: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeBlockAsm12B + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeBlockAsm12B + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeBlockAsm12B: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeBlockAsm12B + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeBlockAsm12B + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeBlockAsm12B: + CMPL DI, $0x01 + JL match_nolit_end_encodeBlockAsm12B + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeBlockAsm12B + LEAL 1(R10), R10 + +match_nolit_end_encodeBlockAsm12B: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeBlockAsm12B: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBlockAsm12B + CMPL SI, $0x00000800 + JAE long_offset_short_match_nolit_encodeBlockAsm12B + MOVL $0x00000001, DI + LEAL 16(DI), DI + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, DI + MOVB DI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R10 + + // emitRepeat + LEAL -4(R10), R10 + JMP cant_repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short_2b + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm12B_emit_copy_short_2b + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short_2b + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short_2b: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm12B_emit_copy_short_2b + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +repeat_three_match_nolit_encodeBlockAsm12B_emit_copy_short_2b: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +repeat_two_match_nolit_encodeBlockAsm12B_emit_copy_short_2b: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short_2b: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +long_offset_short_match_nolit_encodeBlockAsm12B: + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + + // emitRepeat + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm12B_emit_copy_short + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm12B_emit_copy_short + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +repeat_three_match_nolit_encodeBlockAsm12B_emit_copy_short: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +repeat_two_match_nolit_encodeBlockAsm12B_emit_copy_short: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +repeat_two_offset_match_nolit_encodeBlockAsm12B_emit_copy_short: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + JMP two_byte_offset_match_nolit_encodeBlockAsm12B + +two_byte_offset_short_match_nolit_encodeBlockAsm12B: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeBlockAsm12B + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBlockAsm12B + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm12B + +emit_copy_three_match_nolit_encodeBlockAsm12B: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeBlockAsm12B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBlockAsm12B + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBlockAsm12B: + MOVQ $0x000000cf1bbcdcbb, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x18, R8 + IMULQ R9, R8 + SHRQ $0x34, R8 + SHLQ $0x18, SI + IMULQ R9, SI + SHRQ $0x34, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeBlockAsm12B + INCL CX + JMP search_loop_encodeBlockAsm12B + +emit_remainder_encodeBlockAsm12B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBlockAsm12B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBlockAsm12B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBlockAsm12B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBlockAsm12B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBlockAsm12B + +two_bytes_emit_remainder_encodeBlockAsm12B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBlockAsm12B + JMP memmove_long_emit_remainder_encodeBlockAsm12B + +one_byte_emit_remainder_encodeBlockAsm12B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBlockAsm12B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBlockAsm12B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBlockAsm12B + +memmove_long_emit_remainder_encodeBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm12Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBlockAsm12B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBlockAsm10B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBlockAsm10B(SB), $4120-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000020, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBlockAsm10B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBlockAsm10B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBlockAsm10B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x05, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBlockAsm10B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x9e3779b1, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + SHLQ $0x20, R11 + IMULQ R9, R11 + SHRQ $0x36, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeBlockAsm10B + LEAL 1(CX), DI + MOVL 12(SP), R8 + MOVL DI, SI + SUBL 16(SP), SI + JZ repeat_extend_back_end_encodeBlockAsm10B + +repeat_extend_back_loop_encodeBlockAsm10B: + CMPL DI, R8 + JLE repeat_extend_back_end_encodeBlockAsm10B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeBlockAsm10B + LEAL -1(DI), DI + DECL SI + JNZ repeat_extend_back_loop_encodeBlockAsm10B + +repeat_extend_back_end_encodeBlockAsm10B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeBlockAsm10B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeBlockAsm10B + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeBlockAsm10B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeBlockAsm10B + +two_bytes_repeat_emit_encodeBlockAsm10B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeBlockAsm10B + JMP memmove_long_repeat_emit_encodeBlockAsm10B + +one_byte_repeat_emit_encodeBlockAsm10B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm10B + +emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm10B + +emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm10B + +emit_lit_memmove_repeat_emit_encodeBlockAsm10B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_repeat_emit_encodeBlockAsm10B: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeBlockAsm10B + +memmove_long_repeat_emit_encodeBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R12 + SHRQ $0x05, R12 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R13 + SUBQ R11, R13 + DECQ R12 + JA emit_lit_memmove_long_repeat_emit_encodeBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R13*1), R11 + LEAQ -32(AX)(R13*1), R14 + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm10Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R14) + MOVOA X5, 16(R14) + ADDQ $0x20, R14 + ADDQ $0x20, R11 + ADDQ $0x20, R13 + DECQ R12 + JNA emit_lit_memmove_long_repeat_emit_encodeBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R13*1), X4 + MOVOU -16(R10)(R13*1), X5 + MOVOA X4, -32(AX)(R13*1) + MOVOA X5, -16(AX)(R13*1) + ADDQ $0x20, R13 + CMPQ R9, R13 + JAE emit_lit_memmove_long_repeat_emit_encodeBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeBlockAsm10B: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R9 + SUBL CX, R9 + LEAQ (DX)(CX*1), R10 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R12, R12 + CMPL R9, $0x08 + JL matchlen_match4_repeat_extend_encodeBlockAsm10B + +matchlen_loopback_repeat_extend_encodeBlockAsm10B: + MOVQ (R10)(R12*1), R11 + XORQ (SI)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_repeat_extend_encodeBlockAsm10B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP repeat_extend_forward_end_encodeBlockAsm10B + +matchlen_loop_repeat_extend_encodeBlockAsm10B: + LEAL -8(R9), R9 + LEAL 8(R12), R12 + CMPL R9, $0x08 + JGE matchlen_loopback_repeat_extend_encodeBlockAsm10B + JZ repeat_extend_forward_end_encodeBlockAsm10B + +matchlen_match4_repeat_extend_encodeBlockAsm10B: + CMPL R9, $0x04 + JL matchlen_match2_repeat_extend_encodeBlockAsm10B + MOVL (R10)(R12*1), R11 + CMPL (SI)(R12*1), R11 + JNE matchlen_match2_repeat_extend_encodeBlockAsm10B + SUBL $0x04, R9 + LEAL 4(R12), R12 + +matchlen_match2_repeat_extend_encodeBlockAsm10B: + CMPL R9, $0x02 + JL matchlen_match1_repeat_extend_encodeBlockAsm10B + MOVW (R10)(R12*1), R11 + CMPW (SI)(R12*1), R11 + JNE matchlen_match1_repeat_extend_encodeBlockAsm10B + SUBL $0x02, R9 + LEAL 2(R12), R12 + +matchlen_match1_repeat_extend_encodeBlockAsm10B: + CMPL R9, $0x01 + JL repeat_extend_forward_end_encodeBlockAsm10B + MOVB (R10)(R12*1), R11 + CMPB (SI)(R12*1), R11 + JNE repeat_extend_forward_end_encodeBlockAsm10B + LEAL 1(R12), R12 + +repeat_extend_forward_end_encodeBlockAsm10B: + ADDL R12, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + TESTL R8, R8 + JZ repeat_as_copy_encodeBlockAsm10B + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_match_repeat_encodeBlockAsm10B + CMPL R8, $0x0c + JGE cant_repeat_two_offset_match_repeat_encodeBlockAsm10B + CMPL DI, $0x00000800 + JLT repeat_two_offset_match_repeat_encodeBlockAsm10B + +cant_repeat_two_offset_match_repeat_encodeBlockAsm10B: + CMPL SI, $0x00000104 + JLT repeat_three_match_repeat_encodeBlockAsm10B + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_three_match_repeat_encodeBlockAsm10B: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_two_match_repeat_encodeBlockAsm10B: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_two_offset_match_repeat_encodeBlockAsm10B: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_as_copy_encodeBlockAsm10B: + // emitCopy +two_byte_offset_repeat_as_copy_encodeBlockAsm10B: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeBlockAsm10B + CMPL DI, $0x00000800 + JAE long_offset_short_repeat_as_copy_encodeBlockAsm10B + MOVL $0x00000001, R8 + LEAL 16(R8), R8 + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, R8 + MOVB R8, (AX) + ADDQ $0x02, AX + SUBL $0x08, SI + + // emitRepeat + LEAL -4(SI), SI + JMP cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_three_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_two_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short_2b: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + +long_offset_short_repeat_as_copy_encodeBlockAsm10B: + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + + // emitRepeat + MOVL SI, R8 + LEAL -4(SI), SI + CMPL R8, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm10B_emit_copy_short + CMPL R8, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short + CMPL DI, $0x00000800 + JLT repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm10B_emit_copy_short + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_three_repeat_as_copy_encodeBlockAsm10B_emit_copy_short: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_two_repeat_as_copy_encodeBlockAsm10B_emit_copy_short: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + +repeat_two_offset_repeat_as_copy_encodeBlockAsm10B_emit_copy_short: + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + JMP two_byte_offset_repeat_as_copy_encodeBlockAsm10B + +two_byte_offset_short_repeat_as_copy_encodeBlockAsm10B: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm10B + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm10B + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm10B + +emit_copy_three_repeat_as_copy_encodeBlockAsm10B: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeBlockAsm10B: + MOVL CX, 12(SP) + JMP search_loop_encodeBlockAsm10B + +no_repeat_found_encodeBlockAsm10B: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBlockAsm10B + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeBlockAsm10B + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeBlockAsm10B + MOVL 20(SP), CX + JMP search_loop_encodeBlockAsm10B + +candidate3_match_encodeBlockAsm10B: + ADDL $0x02, CX + JMP candidate_match_encodeBlockAsm10B + +candidate2_match_encodeBlockAsm10B: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeBlockAsm10B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBlockAsm10B + +match_extend_back_loop_encodeBlockAsm10B: + CMPL CX, DI + JLE match_extend_back_end_encodeBlockAsm10B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBlockAsm10B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBlockAsm10B + JMP match_extend_back_loop_encodeBlockAsm10B + +match_extend_back_end_encodeBlockAsm10B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBlockAsm10B: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeBlockAsm10B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeBlockAsm10B + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeBlockAsm10B + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBlockAsm10B + +two_bytes_match_emit_encodeBlockAsm10B: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeBlockAsm10B + JMP memmove_long_match_emit_encodeBlockAsm10B + +one_byte_match_emit_encodeBlockAsm10B: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBlockAsm10B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeBlockAsm10B + +emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm10B + +emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm10B + +emit_lit_memmove_match_emit_encodeBlockAsm10B_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBlockAsm10B: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeBlockAsm10B + +memmove_long_match_emit_encodeBlockAsm10B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeBlockAsm10Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeBlockAsm10B: +match_nolit_loop_encodeBlockAsm10B: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeBlockAsm10B + +matchlen_loopback_match_nolit_encodeBlockAsm10B: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeBlockAsm10B + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeBlockAsm10B + +matchlen_loop_match_nolit_encodeBlockAsm10B: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeBlockAsm10B + JZ match_nolit_end_encodeBlockAsm10B + +matchlen_match4_match_nolit_encodeBlockAsm10B: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeBlockAsm10B + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeBlockAsm10B + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeBlockAsm10B: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeBlockAsm10B + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeBlockAsm10B + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeBlockAsm10B: + CMPL DI, $0x01 + JL match_nolit_end_encodeBlockAsm10B + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeBlockAsm10B + LEAL 1(R10), R10 + +match_nolit_end_encodeBlockAsm10B: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeBlockAsm10B: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBlockAsm10B + CMPL SI, $0x00000800 + JAE long_offset_short_match_nolit_encodeBlockAsm10B + MOVL $0x00000001, DI + LEAL 16(DI), DI + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, DI + MOVB DI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R10 + + // emitRepeat + LEAL -4(R10), R10 + JMP cant_repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short_2b + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm10B_emit_copy_short_2b + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short_2b + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short_2b: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm10B_emit_copy_short_2b + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +repeat_three_match_nolit_encodeBlockAsm10B_emit_copy_short_2b: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +repeat_two_match_nolit_encodeBlockAsm10B_emit_copy_short_2b: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short_2b: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +long_offset_short_match_nolit_encodeBlockAsm10B: + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + + // emitRepeat + MOVL R10, DI + LEAL -4(R10), R10 + CMPL DI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm10B_emit_copy_short + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short + CMPL SI, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm10B_emit_copy_short + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +repeat_three_match_nolit_encodeBlockAsm10B_emit_copy_short: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +repeat_two_match_nolit_encodeBlockAsm10B_emit_copy_short: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +repeat_two_offset_match_nolit_encodeBlockAsm10B_emit_copy_short: + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + JMP two_byte_offset_match_nolit_encodeBlockAsm10B + +two_byte_offset_short_match_nolit_encodeBlockAsm10B: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeBlockAsm10B + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBlockAsm10B + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm10B + +emit_copy_three_match_nolit_encodeBlockAsm10B: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeBlockAsm10B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBlockAsm10B + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBlockAsm10B: + MOVQ $0x9e3779b1, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x20, R8 + IMULQ R9, R8 + SHRQ $0x36, R8 + SHLQ $0x20, SI + IMULQ R9, SI + SHRQ $0x36, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeBlockAsm10B + INCL CX + JMP search_loop_encodeBlockAsm10B + +emit_remainder_encodeBlockAsm10B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBlockAsm10B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBlockAsm10B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBlockAsm10B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBlockAsm10B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBlockAsm10B + +two_bytes_emit_remainder_encodeBlockAsm10B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBlockAsm10B + JMP memmove_long_emit_remainder_encodeBlockAsm10B + +one_byte_emit_remainder_encodeBlockAsm10B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBlockAsm10B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBlockAsm10B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBlockAsm10B + +memmove_long_emit_remainder_encodeBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm10Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBlockAsm10B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBlockAsm8B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBlockAsm8B(SB), $1048-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000008, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBlockAsm8B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBlockAsm8B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBlockAsm8B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x04, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBlockAsm8B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x9e3779b1, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x38, R10 + SHLQ $0x20, R11 + IMULQ R9, R11 + SHRQ $0x38, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x38, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeBlockAsm8B + LEAL 1(CX), DI + MOVL 12(SP), R8 + MOVL DI, SI + SUBL 16(SP), SI + JZ repeat_extend_back_end_encodeBlockAsm8B + +repeat_extend_back_loop_encodeBlockAsm8B: + CMPL DI, R8 + JLE repeat_extend_back_end_encodeBlockAsm8B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeBlockAsm8B + LEAL -1(DI), DI + DECL SI + JNZ repeat_extend_back_loop_encodeBlockAsm8B + +repeat_extend_back_end_encodeBlockAsm8B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeBlockAsm8B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeBlockAsm8B + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeBlockAsm8B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeBlockAsm8B + +two_bytes_repeat_emit_encodeBlockAsm8B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeBlockAsm8B + JMP memmove_long_repeat_emit_encodeBlockAsm8B + +one_byte_repeat_emit_encodeBlockAsm8B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeBlockAsm8B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm8B + +emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm8B + +emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_repeat_emit_encodeBlockAsm8B + +emit_lit_memmove_repeat_emit_encodeBlockAsm8B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_repeat_emit_encodeBlockAsm8B: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeBlockAsm8B + +memmove_long_repeat_emit_encodeBlockAsm8B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R12 + SHRQ $0x05, R12 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R13 + SUBQ R11, R13 + DECQ R12 + JA emit_lit_memmove_long_repeat_emit_encodeBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R13*1), R11 + LEAQ -32(AX)(R13*1), R14 + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm8Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R14) + MOVOA X5, 16(R14) + ADDQ $0x20, R14 + ADDQ $0x20, R11 + ADDQ $0x20, R13 + DECQ R12 + JNA emit_lit_memmove_long_repeat_emit_encodeBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R13*1), X4 + MOVOU -16(R10)(R13*1), X5 + MOVOA X4, -32(AX)(R13*1) + MOVOA X5, -16(AX)(R13*1) + ADDQ $0x20, R13 + CMPQ R9, R13 + JAE emit_lit_memmove_long_repeat_emit_encodeBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeBlockAsm8B: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R9 + SUBL CX, R9 + LEAQ (DX)(CX*1), R10 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R12, R12 + CMPL R9, $0x08 + JL matchlen_match4_repeat_extend_encodeBlockAsm8B + +matchlen_loopback_repeat_extend_encodeBlockAsm8B: + MOVQ (R10)(R12*1), R11 + XORQ (SI)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_repeat_extend_encodeBlockAsm8B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP repeat_extend_forward_end_encodeBlockAsm8B + +matchlen_loop_repeat_extend_encodeBlockAsm8B: + LEAL -8(R9), R9 + LEAL 8(R12), R12 + CMPL R9, $0x08 + JGE matchlen_loopback_repeat_extend_encodeBlockAsm8B + JZ repeat_extend_forward_end_encodeBlockAsm8B + +matchlen_match4_repeat_extend_encodeBlockAsm8B: + CMPL R9, $0x04 + JL matchlen_match2_repeat_extend_encodeBlockAsm8B + MOVL (R10)(R12*1), R11 + CMPL (SI)(R12*1), R11 + JNE matchlen_match2_repeat_extend_encodeBlockAsm8B + SUBL $0x04, R9 + LEAL 4(R12), R12 + +matchlen_match2_repeat_extend_encodeBlockAsm8B: + CMPL R9, $0x02 + JL matchlen_match1_repeat_extend_encodeBlockAsm8B + MOVW (R10)(R12*1), R11 + CMPW (SI)(R12*1), R11 + JNE matchlen_match1_repeat_extend_encodeBlockAsm8B + SUBL $0x02, R9 + LEAL 2(R12), R12 + +matchlen_match1_repeat_extend_encodeBlockAsm8B: + CMPL R9, $0x01 + JL repeat_extend_forward_end_encodeBlockAsm8B + MOVB (R10)(R12*1), R11 + CMPB (SI)(R12*1), R11 + JNE repeat_extend_forward_end_encodeBlockAsm8B + LEAL 1(R12), R12 + +repeat_extend_forward_end_encodeBlockAsm8B: + ADDL R12, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + TESTL R8, R8 + JZ repeat_as_copy_encodeBlockAsm8B + + // emitRepeat + MOVL SI, DI + LEAL -4(SI), SI + CMPL DI, $0x08 + JLE repeat_two_match_repeat_encodeBlockAsm8B + CMPL DI, $0x0c + JGE cant_repeat_two_offset_match_repeat_encodeBlockAsm8B + +cant_repeat_two_offset_match_repeat_encodeBlockAsm8B: + CMPL SI, $0x00000104 + JLT repeat_three_match_repeat_encodeBlockAsm8B + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_three_match_repeat_encodeBlockAsm8B: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_two_match_repeat_encodeBlockAsm8B: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_as_copy_encodeBlockAsm8B: + // emitCopy +two_byte_offset_repeat_as_copy_encodeBlockAsm8B: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeBlockAsm8B + CMPL DI, $0x00000800 + JAE long_offset_short_repeat_as_copy_encodeBlockAsm8B + MOVL $0x00000001, R8 + LEAL 16(R8), R8 + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, R8 + MOVB R8, (AX) + ADDQ $0x02, AX + SUBL $0x08, SI + + // emitRepeat + LEAL -4(SI), SI + JMP cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b + MOVL SI, DI + LEAL -4(SI), SI + CMPL DI, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b + CMPL DI, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_three_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_two_repeat_as_copy_encodeBlockAsm8B_emit_copy_short_2b: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + +long_offset_short_repeat_as_copy_encodeBlockAsm8B: + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + + // emitRepeat + MOVL SI, DI + LEAL -4(SI), SI + CMPL DI, $0x08 + JLE repeat_two_repeat_as_copy_encodeBlockAsm8B_emit_copy_short + CMPL DI, $0x0c + JGE cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm8B_emit_copy_short + +cant_repeat_two_offset_repeat_as_copy_encodeBlockAsm8B_emit_copy_short: + CMPL SI, $0x00000104 + JLT repeat_three_repeat_as_copy_encodeBlockAsm8B_emit_copy_short + LEAL -256(SI), SI + MOVW $0x0019, (AX) + MOVW SI, 2(AX) + ADDQ $0x04, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_three_repeat_as_copy_encodeBlockAsm8B_emit_copy_short: + LEAL -4(SI), SI + MOVW $0x0015, (AX) + MOVB SI, 2(AX) + ADDQ $0x03, AX + JMP repeat_end_emit_encodeBlockAsm8B + +repeat_two_repeat_as_copy_encodeBlockAsm8B_emit_copy_short: + SHLL $0x02, SI + ORL $0x01, SI + MOVW SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + XORQ R8, R8 + LEAL 1(R8)(SI*4), SI + MOVB DI, 1(AX) + SARL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + JMP two_byte_offset_repeat_as_copy_encodeBlockAsm8B + +two_byte_offset_short_repeat_as_copy_encodeBlockAsm8B: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeBlockAsm8B + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeBlockAsm8B + +emit_copy_three_repeat_as_copy_encodeBlockAsm8B: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeBlockAsm8B: + MOVL CX, 12(SP) + JMP search_loop_encodeBlockAsm8B + +no_repeat_found_encodeBlockAsm8B: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBlockAsm8B + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeBlockAsm8B + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeBlockAsm8B + MOVL 20(SP), CX + JMP search_loop_encodeBlockAsm8B + +candidate3_match_encodeBlockAsm8B: + ADDL $0x02, CX + JMP candidate_match_encodeBlockAsm8B + +candidate2_match_encodeBlockAsm8B: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeBlockAsm8B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBlockAsm8B + +match_extend_back_loop_encodeBlockAsm8B: + CMPL CX, DI + JLE match_extend_back_end_encodeBlockAsm8B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBlockAsm8B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBlockAsm8B + JMP match_extend_back_loop_encodeBlockAsm8B + +match_extend_back_end_encodeBlockAsm8B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBlockAsm8B: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeBlockAsm8B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeBlockAsm8B + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeBlockAsm8B + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBlockAsm8B + +two_bytes_match_emit_encodeBlockAsm8B: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeBlockAsm8B + JMP memmove_long_match_emit_encodeBlockAsm8B + +one_byte_match_emit_encodeBlockAsm8B: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBlockAsm8B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeBlockAsm8B + +emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm8B + +emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBlockAsm8B + +emit_lit_memmove_match_emit_encodeBlockAsm8B_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBlockAsm8B: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeBlockAsm8B + +memmove_long_match_emit_encodeBlockAsm8B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeBlockAsm8Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeBlockAsm8B: +match_nolit_loop_encodeBlockAsm8B: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeBlockAsm8B + +matchlen_loopback_match_nolit_encodeBlockAsm8B: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeBlockAsm8B + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeBlockAsm8B + +matchlen_loop_match_nolit_encodeBlockAsm8B: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeBlockAsm8B + JZ match_nolit_end_encodeBlockAsm8B + +matchlen_match4_match_nolit_encodeBlockAsm8B: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeBlockAsm8B + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeBlockAsm8B + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeBlockAsm8B: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeBlockAsm8B + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeBlockAsm8B + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeBlockAsm8B: + CMPL DI, $0x01 + JL match_nolit_end_encodeBlockAsm8B + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeBlockAsm8B + LEAL 1(R10), R10 + +match_nolit_end_encodeBlockAsm8B: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeBlockAsm8B: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBlockAsm8B + CMPL SI, $0x00000800 + JAE long_offset_short_match_nolit_encodeBlockAsm8B + MOVL $0x00000001, DI + LEAL 16(DI), DI + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, DI + MOVB DI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R10 + + // emitRepeat + LEAL -4(R10), R10 + JMP cant_repeat_two_offset_match_nolit_encodeBlockAsm8B_emit_copy_short_2b + MOVL R10, SI + LEAL -4(R10), R10 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm8B_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm8B_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBlockAsm8B_emit_copy_short_2b: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm8B_emit_copy_short_2b + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + +repeat_three_match_nolit_encodeBlockAsm8B_emit_copy_short_2b: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + +repeat_two_match_nolit_encodeBlockAsm8B_emit_copy_short_2b: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + +long_offset_short_match_nolit_encodeBlockAsm8B: + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + + // emitRepeat + MOVL R10, SI + LEAL -4(R10), R10 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBlockAsm8B_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBlockAsm8B_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBlockAsm8B_emit_copy_short: + CMPL R10, $0x00000104 + JLT repeat_three_match_nolit_encodeBlockAsm8B_emit_copy_short + LEAL -256(R10), R10 + MOVW $0x0019, (AX) + MOVW R10, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + +repeat_three_match_nolit_encodeBlockAsm8B_emit_copy_short: + LEAL -4(R10), R10 + MOVW $0x0015, (AX) + MOVB R10, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + +repeat_two_match_nolit_encodeBlockAsm8B_emit_copy_short: + SHLL $0x02, R10 + ORL $0x01, R10 + MOVW R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + XORQ DI, DI + LEAL 1(DI)(R10*4), R10 + MOVB SI, 1(AX) + SARL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + JMP two_byte_offset_match_nolit_encodeBlockAsm8B + +two_byte_offset_short_match_nolit_encodeBlockAsm8B: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeBlockAsm8B + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBlockAsm8B + +emit_copy_three_match_nolit_encodeBlockAsm8B: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeBlockAsm8B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBlockAsm8B + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBlockAsm8B: + MOVQ $0x9e3779b1, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x20, R8 + IMULQ R9, R8 + SHRQ $0x38, R8 + SHLQ $0x20, SI + IMULQ R9, SI + SHRQ $0x38, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeBlockAsm8B + INCL CX + JMP search_loop_encodeBlockAsm8B + +emit_remainder_encodeBlockAsm8B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBlockAsm8B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBlockAsm8B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBlockAsm8B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBlockAsm8B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBlockAsm8B + +two_bytes_emit_remainder_encodeBlockAsm8B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBlockAsm8B + JMP memmove_long_emit_remainder_encodeBlockAsm8B + +one_byte_emit_remainder_encodeBlockAsm8B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBlockAsm8B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBlockAsm8B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBlockAsm8B + +memmove_long_emit_remainder_encodeBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm8Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBlockAsm8B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBetterBlockAsm(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBetterBlockAsm(SB), $327704-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000a00, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBetterBlockAsm: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBetterBlockAsm + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -6(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBetterBlockAsm: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x07, SI + CMPL SI, $0x63 + JLE check_maxskip_ok_encodeBetterBlockAsm + LEAL 100(CX), SI + JMP check_maxskip_cont_encodeBetterBlockAsm + +check_maxskip_ok_encodeBetterBlockAsm: + LEAL 1(CX)(SI*1), SI + +check_maxskip_cont_encodeBetterBlockAsm: + CMPL SI, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x00cf1bbcdcbfa563, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 262168(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 262168(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeBetterBlockAsm + MOVL 20(SP), CX + JMP search_loop_encodeBetterBlockAsm + +candidateS_match_encodeBetterBlockAsm: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm + DECL CX + MOVL R8, SI + +candidate_match_encodeBetterBlockAsm: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBetterBlockAsm + +match_extend_back_loop_encodeBetterBlockAsm: + CMPL CX, DI + JLE match_extend_back_end_encodeBetterBlockAsm + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBetterBlockAsm + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBetterBlockAsm + JMP match_extend_back_loop_encodeBetterBlockAsm + +match_extend_back_end_encodeBetterBlockAsm: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 5(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBetterBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBetterBlockAsm: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeBetterBlockAsm + +matchlen_loopback_match_nolit_encodeBetterBlockAsm: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeBetterBlockAsm + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeBetterBlockAsm + +matchlen_loop_match_nolit_encodeBetterBlockAsm: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeBetterBlockAsm + JZ match_nolit_end_encodeBetterBlockAsm + +matchlen_match4_match_nolit_encodeBetterBlockAsm: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeBetterBlockAsm + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeBetterBlockAsm + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeBetterBlockAsm: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeBetterBlockAsm + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeBetterBlockAsm + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeBetterBlockAsm: + CMPL R8, $0x01 + JL match_nolit_end_encodeBetterBlockAsm + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeBetterBlockAsm + LEAL 1(R12), R12 + +match_nolit_end_encodeBetterBlockAsm: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + CMPL 16(SP), R8 + JEQ match_is_repeat_encodeBetterBlockAsm + CMPL R12, $0x01 + JG match_length_ok_encodeBetterBlockAsm + CMPL R8, $0x0000ffff + JLE match_length_ok_encodeBetterBlockAsm + MOVL 20(SP), CX + INCL CX + JMP search_loop_encodeBetterBlockAsm + +match_length_ok_encodeBetterBlockAsm: + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeBetterBlockAsm + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeBetterBlockAsm + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeBetterBlockAsm + CMPL SI, $0x00010000 + JLT three_bytes_match_emit_encodeBetterBlockAsm + CMPL SI, $0x01000000 + JLT four_bytes_match_emit_encodeBetterBlockAsm + MOVB $0xfc, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm + +four_bytes_match_emit_encodeBetterBlockAsm: + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm + +three_bytes_match_emit_encodeBetterBlockAsm: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm + +two_bytes_match_emit_encodeBetterBlockAsm: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeBetterBlockAsm + JMP memmove_long_match_emit_encodeBetterBlockAsm + +one_byte_match_emit_encodeBetterBlockAsm: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBetterBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_encodeBetterBlockAsm_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBetterBlockAsm: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeBetterBlockAsm + +memmove_long_match_emit_encodeBetterBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeBetterBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsmlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeBetterBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeBetterBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeBetterBlockAsm: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy + CMPL R8, $0x00010000 + JL two_byte_offset_match_nolit_encodeBetterBlockAsm + +four_bytes_loop_back_match_nolit_encodeBetterBlockAsm: + CMPL R12, $0x40 + JLE four_bytes_remain_match_nolit_encodeBetterBlockAsm + MOVB $0xff, (AX) + MOVL R8, 1(AX) + LEAL -64(R12), R12 + ADDQ $0x05, AX + CMPL R12, $0x04 + JL four_bytes_remain_match_nolit_encodeBetterBlockAsm + + // emitRepeat +emit_repeat_again_match_nolit_encodeBetterBlockAsm_emit_copy: + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm_emit_copy + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm_emit_copy + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_encodeBetterBlockAsm_emit_copy + CMPL R12, $0x0100ffff + JLT repeat_five_match_nolit_encodeBetterBlockAsm_emit_copy + LEAL -16842747(R12), R12 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_encodeBetterBlockAsm_emit_copy + +repeat_five_match_nolit_encodeBetterBlockAsm_emit_copy: + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_four_match_nolit_encodeBetterBlockAsm_emit_copy: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_three_match_nolit_encodeBetterBlockAsm_emit_copy: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_match_nolit_encodeBetterBlockAsm_emit_copy: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + JMP four_bytes_loop_back_match_nolit_encodeBetterBlockAsm + +four_bytes_remain_match_nolit_encodeBetterBlockAsm: + TESTL R12, R12 + JZ match_nolit_emitcopy_end_encodeBetterBlockAsm + MOVB $0x03, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVL R8, 1(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +two_byte_offset_match_nolit_encodeBetterBlockAsm: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBetterBlockAsm + CMPL R8, $0x00000800 + JAE long_offset_short_match_nolit_encodeBetterBlockAsm + MOVL $0x00000001, SI + LEAL 16(SI), SI + MOVB R8, 1(AX) + MOVL R8, R9 + SHRL $0x08, R9 + SHLL $0x05, R9 + ORL R9, SI + MOVB SI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R12 + + // emitRepeat + LEAL -4(R12), R12 + JMP cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + +emit_repeat_again_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + CMPL R12, $0x0100ffff + JLT repeat_five_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + LEAL -16842747(R12), R12 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b + +repeat_five_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_four_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_three_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short_2b: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +long_offset_short_match_nolit_encodeBetterBlockAsm: + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + + // emitRepeat +emit_repeat_again_match_nolit_encodeBetterBlockAsm_emit_copy_short: + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm_emit_copy_short + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_encodeBetterBlockAsm_emit_copy_short + CMPL R12, $0x0100ffff + JLT repeat_five_match_nolit_encodeBetterBlockAsm_emit_copy_short + LEAL -16842747(R12), R12 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_encodeBetterBlockAsm_emit_copy_short + +repeat_five_match_nolit_encodeBetterBlockAsm_emit_copy_short: + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_four_match_nolit_encodeBetterBlockAsm_emit_copy_short: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_three_match_nolit_encodeBetterBlockAsm_emit_copy_short: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_match_nolit_encodeBetterBlockAsm_emit_copy_short: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_offset_match_nolit_encodeBetterBlockAsm_emit_copy_short: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + JMP two_byte_offset_match_nolit_encodeBetterBlockAsm + +two_byte_offset_short_match_nolit_encodeBetterBlockAsm: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +emit_copy_three_match_nolit_encodeBetterBlockAsm: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +match_is_repeat_encodeBetterBlockAsm: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_repeat_encodeBetterBlockAsm + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_repeat_encodeBetterBlockAsm + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_repeat_encodeBetterBlockAsm + CMPL SI, $0x00010000 + JLT three_bytes_match_emit_repeat_encodeBetterBlockAsm + CMPL SI, $0x01000000 + JLT four_bytes_match_emit_repeat_encodeBetterBlockAsm + MOVB $0xfc, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm + +four_bytes_match_emit_repeat_encodeBetterBlockAsm: + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm + +three_bytes_match_emit_repeat_encodeBetterBlockAsm: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm + +two_bytes_match_emit_repeat_encodeBetterBlockAsm: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_repeat_encodeBetterBlockAsm + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm + +one_byte_match_emit_repeat_encodeBetterBlockAsm: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_repeat_encodeBetterBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_33through64 + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm: + MOVQ SI, AX + JMP emit_literal_done_match_emit_repeat_encodeBetterBlockAsm + +memmove_long_match_emit_repeat_encodeBetterBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsmlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_repeat_encodeBetterBlockAsm: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitRepeat +emit_repeat_again_match_nolit_repeat_encodeBetterBlockAsm: + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_repeat_encodeBetterBlockAsm + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm + +cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_repeat_encodeBetterBlockAsm + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_repeat_encodeBetterBlockAsm + CMPL R12, $0x0100ffff + JLT repeat_five_match_nolit_repeat_encodeBetterBlockAsm + LEAL -16842747(R12), R12 + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + JMP emit_repeat_again_match_nolit_repeat_encodeBetterBlockAsm + +repeat_five_match_nolit_repeat_encodeBetterBlockAsm: + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_four_match_nolit_repeat_encodeBetterBlockAsm: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_three_match_nolit_repeat_encodeBetterBlockAsm: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_match_nolit_repeat_encodeBetterBlockAsm: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm + +repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + +match_nolit_emitcopy_end_encodeBetterBlockAsm: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBetterBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBetterBlockAsm: + MOVQ $0x00cf1bbcdcbfa563, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x32, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 262168(SP)(R11*4) + MOVL R15, 262168(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 262168(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeBetterBlockAsm + +emit_remainder_encodeBetterBlockAsm: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 5(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBetterBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBetterBlockAsm: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBetterBlockAsm + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBetterBlockAsm + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBetterBlockAsm + CMPL DX, $0x00010000 + JLT three_bytes_emit_remainder_encodeBetterBlockAsm + CMPL DX, $0x01000000 + JLT four_bytes_emit_remainder_encodeBetterBlockAsm + MOVB $0xfc, (AX) + MOVL DX, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm + +four_bytes_emit_remainder_encodeBetterBlockAsm: + MOVL DX, BX + SHRL $0x10, BX + MOVB $0xf8, (AX) + MOVW DX, 1(AX) + MOVB BL, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm + +three_bytes_emit_remainder_encodeBetterBlockAsm: + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm + +two_bytes_emit_remainder_encodeBetterBlockAsm: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBetterBlockAsm + JMP memmove_long_emit_remainder_encodeBetterBlockAsm + +one_byte_emit_remainder_encodeBetterBlockAsm: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBetterBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBetterBlockAsm: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBetterBlockAsm + +memmove_long_emit_remainder_encodeBetterBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsmlarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBetterBlockAsm: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBetterBlockAsm4MB(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBetterBlockAsm4MB(SB), $327704-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000a00, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBetterBlockAsm4MB: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBetterBlockAsm4MB + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -6(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBetterBlockAsm4MB: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x07, SI + CMPL SI, $0x63 + JLE check_maxskip_ok_encodeBetterBlockAsm4MB + LEAL 100(CX), SI + JMP check_maxskip_cont_encodeBetterBlockAsm4MB + +check_maxskip_ok_encodeBetterBlockAsm4MB: + LEAL 1(CX)(SI*1), SI + +check_maxskip_cont_encodeBetterBlockAsm4MB: + CMPL SI, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm4MB + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x00cf1bbcdcbfa563, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 262168(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 262168(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm4MB + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeBetterBlockAsm4MB + MOVL 20(SP), CX + JMP search_loop_encodeBetterBlockAsm4MB + +candidateS_match_encodeBetterBlockAsm4MB: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm4MB + DECL CX + MOVL R8, SI + +candidate_match_encodeBetterBlockAsm4MB: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBetterBlockAsm4MB + +match_extend_back_loop_encodeBetterBlockAsm4MB: + CMPL CX, DI + JLE match_extend_back_end_encodeBetterBlockAsm4MB + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBetterBlockAsm4MB + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBetterBlockAsm4MB + JMP match_extend_back_loop_encodeBetterBlockAsm4MB + +match_extend_back_end_encodeBetterBlockAsm4MB: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 4(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBetterBlockAsm4MB + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBetterBlockAsm4MB: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeBetterBlockAsm4MB + +matchlen_loopback_match_nolit_encodeBetterBlockAsm4MB: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeBetterBlockAsm4MB + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeBetterBlockAsm4MB + +matchlen_loop_match_nolit_encodeBetterBlockAsm4MB: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeBetterBlockAsm4MB + JZ match_nolit_end_encodeBetterBlockAsm4MB + +matchlen_match4_match_nolit_encodeBetterBlockAsm4MB: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeBetterBlockAsm4MB + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeBetterBlockAsm4MB + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeBetterBlockAsm4MB: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeBetterBlockAsm4MB + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeBetterBlockAsm4MB + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeBetterBlockAsm4MB: + CMPL R8, $0x01 + JL match_nolit_end_encodeBetterBlockAsm4MB + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeBetterBlockAsm4MB + LEAL 1(R12), R12 + +match_nolit_end_encodeBetterBlockAsm4MB: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + CMPL 16(SP), R8 + JEQ match_is_repeat_encodeBetterBlockAsm4MB + CMPL R12, $0x01 + JG match_length_ok_encodeBetterBlockAsm4MB + CMPL R8, $0x0000ffff + JLE match_length_ok_encodeBetterBlockAsm4MB + MOVL 20(SP), CX + INCL CX + JMP search_loop_encodeBetterBlockAsm4MB + +match_length_ok_encodeBetterBlockAsm4MB: + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeBetterBlockAsm4MB + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeBetterBlockAsm4MB + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeBetterBlockAsm4MB + CMPL SI, $0x00010000 + JLT three_bytes_match_emit_encodeBetterBlockAsm4MB + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm4MB + +three_bytes_match_emit_encodeBetterBlockAsm4MB: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm4MB + +two_bytes_match_emit_encodeBetterBlockAsm4MB: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeBetterBlockAsm4MB + JMP memmove_long_match_emit_encodeBetterBlockAsm4MB + +one_byte_match_emit_encodeBetterBlockAsm4MB: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBetterBlockAsm4MB: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_encodeBetterBlockAsm4MB_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBetterBlockAsm4MB: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeBetterBlockAsm4MB + +memmove_long_match_emit_encodeBetterBlockAsm4MB: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm4MBlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm4MBlarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeBetterBlockAsm4MB: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy + CMPL R8, $0x00010000 + JL two_byte_offset_match_nolit_encodeBetterBlockAsm4MB + +four_bytes_loop_back_match_nolit_encodeBetterBlockAsm4MB: + CMPL R12, $0x40 + JLE four_bytes_remain_match_nolit_encodeBetterBlockAsm4MB + MOVB $0xff, (AX) + MOVL R8, 1(AX) + LEAL -64(R12), R12 + ADDQ $0x05, AX + CMPL R12, $0x04 + JL four_bytes_remain_match_nolit_encodeBetterBlockAsm4MB + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm4MB_emit_copy + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm4MB_emit_copy + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_encodeBetterBlockAsm4MB_emit_copy + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_four_match_nolit_encodeBetterBlockAsm4MB_emit_copy: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_three_match_nolit_encodeBetterBlockAsm4MB_emit_copy: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_match_nolit_encodeBetterBlockAsm4MB_emit_copy: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + JMP four_bytes_loop_back_match_nolit_encodeBetterBlockAsm4MB + +four_bytes_remain_match_nolit_encodeBetterBlockAsm4MB: + TESTL R12, R12 + JZ match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + MOVB $0x03, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVL R8, 1(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +two_byte_offset_match_nolit_encodeBetterBlockAsm4MB: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBetterBlockAsm4MB + CMPL R8, $0x00000800 + JAE long_offset_short_match_nolit_encodeBetterBlockAsm4MB + MOVL $0x00000001, SI + LEAL 16(SI), SI + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, SI + MOVB SI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R12 + + // emitRepeat + LEAL -4(R12), R12 + JMP cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_four_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_three_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short_2b: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +long_offset_short_match_nolit_encodeBetterBlockAsm4MB: + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_four_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_three_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_offset_match_nolit_encodeBetterBlockAsm4MB_emit_copy_short: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + JMP two_byte_offset_match_nolit_encodeBetterBlockAsm4MB + +two_byte_offset_short_match_nolit_encodeBetterBlockAsm4MB: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm4MB + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm4MB + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +emit_copy_three_match_nolit_encodeBetterBlockAsm4MB: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +match_is_repeat_encodeBetterBlockAsm4MB: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_repeat_encodeBetterBlockAsm4MB + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_repeat_encodeBetterBlockAsm4MB + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_repeat_encodeBetterBlockAsm4MB + CMPL SI, $0x00010000 + JLT three_bytes_match_emit_repeat_encodeBetterBlockAsm4MB + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm4MB + +three_bytes_match_emit_repeat_encodeBetterBlockAsm4MB: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm4MB + +two_bytes_match_emit_repeat_encodeBetterBlockAsm4MB: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_repeat_encodeBetterBlockAsm4MB + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm4MB + +one_byte_match_emit_repeat_encodeBetterBlockAsm4MB: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_repeat_encodeBetterBlockAsm4MB: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_33through64 + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm4MB + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm4MB_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm4MB: + MOVQ SI, AX + JMP emit_literal_done_match_emit_repeat_encodeBetterBlockAsm4MB + +memmove_long_match_emit_repeat_encodeBetterBlockAsm4MB: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm4MBlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm4MBlarge_big_loop_back + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_repeat_encodeBetterBlockAsm4MB: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_repeat_encodeBetterBlockAsm4MB + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm4MB + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm4MB + +cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm4MB: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_repeat_encodeBetterBlockAsm4MB + CMPL R12, $0x00010100 + JLT repeat_four_match_nolit_repeat_encodeBetterBlockAsm4MB + LEAL -65536(R12), R12 + MOVL R12, R8 + MOVW $0x001d, (AX) + MOVW R12, 2(AX) + SARL $0x10, R8 + MOVB R8, 4(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_four_match_nolit_repeat_encodeBetterBlockAsm4MB: + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_three_match_nolit_repeat_encodeBetterBlockAsm4MB: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_match_nolit_repeat_encodeBetterBlockAsm4MB: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm4MB + +repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm4MB: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + +match_nolit_emitcopy_end_encodeBetterBlockAsm4MB: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm4MB + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBetterBlockAsm4MB + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBetterBlockAsm4MB: + MOVQ $0x00cf1bbcdcbfa563, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x32, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 262168(SP)(R11*4) + MOVL R15, 262168(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 262168(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeBetterBlockAsm4MB + +emit_remainder_encodeBetterBlockAsm4MB: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 4(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBetterBlockAsm4MB + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBetterBlockAsm4MB: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBetterBlockAsm4MB + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBetterBlockAsm4MB + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBetterBlockAsm4MB + CMPL DX, $0x00010000 + JLT three_bytes_emit_remainder_encodeBetterBlockAsm4MB + MOVL DX, BX + SHRL $0x10, BX + MOVB $0xf8, (AX) + MOVW DX, 1(AX) + MOVB BL, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm4MB + +three_bytes_emit_remainder_encodeBetterBlockAsm4MB: + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm4MB + +two_bytes_emit_remainder_encodeBetterBlockAsm4MB: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBetterBlockAsm4MB + JMP memmove_long_emit_remainder_encodeBetterBlockAsm4MB + +one_byte_emit_remainder_encodeBetterBlockAsm4MB: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBetterBlockAsm4MB: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm4MB + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm4MB_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBetterBlockAsm4MB: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBetterBlockAsm4MB + +memmove_long_emit_remainder_encodeBetterBlockAsm4MB: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm4MBlarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm4MBlarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm4MBlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBetterBlockAsm4MB: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBetterBlockAsm12B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBetterBlockAsm12B(SB), $81944-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000280, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBetterBlockAsm12B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBetterBlockAsm12B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -6(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBetterBlockAsm12B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x06, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm12B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x34, R11 + MOVL 24(SP)(R10*4), SI + MOVL 65560(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 65560(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm12B + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeBetterBlockAsm12B + MOVL 20(SP), CX + JMP search_loop_encodeBetterBlockAsm12B + +candidateS_match_encodeBetterBlockAsm12B: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm12B + DECL CX + MOVL R8, SI + +candidate_match_encodeBetterBlockAsm12B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBetterBlockAsm12B + +match_extend_back_loop_encodeBetterBlockAsm12B: + CMPL CX, DI + JLE match_extend_back_end_encodeBetterBlockAsm12B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBetterBlockAsm12B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBetterBlockAsm12B + JMP match_extend_back_loop_encodeBetterBlockAsm12B + +match_extend_back_end_encodeBetterBlockAsm12B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBetterBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBetterBlockAsm12B: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeBetterBlockAsm12B + +matchlen_loopback_match_nolit_encodeBetterBlockAsm12B: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeBetterBlockAsm12B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeBetterBlockAsm12B + +matchlen_loop_match_nolit_encodeBetterBlockAsm12B: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeBetterBlockAsm12B + JZ match_nolit_end_encodeBetterBlockAsm12B + +matchlen_match4_match_nolit_encodeBetterBlockAsm12B: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeBetterBlockAsm12B + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeBetterBlockAsm12B + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeBetterBlockAsm12B: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeBetterBlockAsm12B + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeBetterBlockAsm12B + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeBetterBlockAsm12B: + CMPL R8, $0x01 + JL match_nolit_end_encodeBetterBlockAsm12B + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeBetterBlockAsm12B + LEAL 1(R12), R12 + +match_nolit_end_encodeBetterBlockAsm12B: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + CMPL 16(SP), R8 + JEQ match_is_repeat_encodeBetterBlockAsm12B + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeBetterBlockAsm12B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeBetterBlockAsm12B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeBetterBlockAsm12B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm12B + +two_bytes_match_emit_encodeBetterBlockAsm12B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeBetterBlockAsm12B + JMP memmove_long_match_emit_encodeBetterBlockAsm12B + +one_byte_match_emit_encodeBetterBlockAsm12B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBetterBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm12B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBetterBlockAsm12B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeBetterBlockAsm12B + +memmove_long_match_emit_encodeBetterBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm12Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeBetterBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeBetterBlockAsm12B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeBetterBlockAsm12B: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBetterBlockAsm12B + CMPL R8, $0x00000800 + JAE long_offset_short_match_nolit_encodeBetterBlockAsm12B + MOVL $0x00000001, SI + LEAL 16(SI), SI + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, SI + MOVB SI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R12 + + // emitRepeat + LEAL -4(R12), R12 + JMP cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_three_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_two_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short_2b: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +long_offset_short_match_nolit_encodeBetterBlockAsm12B: + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm12B_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm12B_emit_copy_short + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_three_match_nolit_encodeBetterBlockAsm12B_emit_copy_short: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_two_match_nolit_encodeBetterBlockAsm12B_emit_copy_short: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_two_offset_match_nolit_encodeBetterBlockAsm12B_emit_copy_short: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + JMP two_byte_offset_match_nolit_encodeBetterBlockAsm12B + +two_byte_offset_short_match_nolit_encodeBetterBlockAsm12B: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm12B + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm12B + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +emit_copy_three_match_nolit_encodeBetterBlockAsm12B: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +match_is_repeat_encodeBetterBlockAsm12B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_repeat_encodeBetterBlockAsm12B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_repeat_encodeBetterBlockAsm12B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_repeat_encodeBetterBlockAsm12B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm12B + +two_bytes_match_emit_repeat_encodeBetterBlockAsm12B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_repeat_encodeBetterBlockAsm12B + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm12B + +one_byte_match_emit_repeat_encodeBetterBlockAsm12B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_repeat_encodeBetterBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm12B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm12B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm12B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_repeat_encodeBetterBlockAsm12B + +memmove_long_match_emit_repeat_encodeBetterBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm12Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_repeat_encodeBetterBlockAsm12B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_repeat_encodeBetterBlockAsm12B + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm12B + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm12B + +cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm12B: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_repeat_encodeBetterBlockAsm12B + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_three_match_nolit_repeat_encodeBetterBlockAsm12B: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_two_match_nolit_repeat_encodeBetterBlockAsm12B: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm12B + +repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm12B: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + +match_nolit_emitcopy_end_encodeBetterBlockAsm12B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm12B + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBetterBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBetterBlockAsm12B: + MOVQ $0x0000cf1bbcdcbf9b, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x32, R10 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x32, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x34, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x34, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 65560(SP)(R11*4) + MOVL R15, 65560(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x32, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x34, R11 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x32, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 65560(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeBetterBlockAsm12B + +emit_remainder_encodeBetterBlockAsm12B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBetterBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBetterBlockAsm12B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBetterBlockAsm12B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBetterBlockAsm12B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBetterBlockAsm12B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm12B + +two_bytes_emit_remainder_encodeBetterBlockAsm12B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBetterBlockAsm12B + JMP memmove_long_emit_remainder_encodeBetterBlockAsm12B + +one_byte_emit_remainder_encodeBetterBlockAsm12B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBetterBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm12B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBetterBlockAsm12B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBetterBlockAsm12B + +memmove_long_emit_remainder_encodeBetterBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm12Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBetterBlockAsm12B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBetterBlockAsm10B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBetterBlockAsm10B(SB), $20504-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x000000a0, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBetterBlockAsm10B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBetterBlockAsm10B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -6(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBetterBlockAsm10B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x05, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm10B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x36, R11 + MOVL 24(SP)(R10*4), SI + MOVL 16408(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 16408(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm10B + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeBetterBlockAsm10B + MOVL 20(SP), CX + JMP search_loop_encodeBetterBlockAsm10B + +candidateS_match_encodeBetterBlockAsm10B: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm10B + DECL CX + MOVL R8, SI + +candidate_match_encodeBetterBlockAsm10B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBetterBlockAsm10B + +match_extend_back_loop_encodeBetterBlockAsm10B: + CMPL CX, DI + JLE match_extend_back_end_encodeBetterBlockAsm10B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBetterBlockAsm10B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBetterBlockAsm10B + JMP match_extend_back_loop_encodeBetterBlockAsm10B + +match_extend_back_end_encodeBetterBlockAsm10B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBetterBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBetterBlockAsm10B: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeBetterBlockAsm10B + +matchlen_loopback_match_nolit_encodeBetterBlockAsm10B: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeBetterBlockAsm10B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeBetterBlockAsm10B + +matchlen_loop_match_nolit_encodeBetterBlockAsm10B: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeBetterBlockAsm10B + JZ match_nolit_end_encodeBetterBlockAsm10B + +matchlen_match4_match_nolit_encodeBetterBlockAsm10B: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeBetterBlockAsm10B + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeBetterBlockAsm10B + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeBetterBlockAsm10B: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeBetterBlockAsm10B + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeBetterBlockAsm10B + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeBetterBlockAsm10B: + CMPL R8, $0x01 + JL match_nolit_end_encodeBetterBlockAsm10B + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeBetterBlockAsm10B + LEAL 1(R12), R12 + +match_nolit_end_encodeBetterBlockAsm10B: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + CMPL 16(SP), R8 + JEQ match_is_repeat_encodeBetterBlockAsm10B + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeBetterBlockAsm10B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeBetterBlockAsm10B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeBetterBlockAsm10B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm10B + +two_bytes_match_emit_encodeBetterBlockAsm10B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeBetterBlockAsm10B + JMP memmove_long_match_emit_encodeBetterBlockAsm10B + +one_byte_match_emit_encodeBetterBlockAsm10B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBetterBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm10B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBetterBlockAsm10B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeBetterBlockAsm10B + +memmove_long_match_emit_encodeBetterBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm10Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeBetterBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeBetterBlockAsm10B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeBetterBlockAsm10B: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBetterBlockAsm10B + CMPL R8, $0x00000800 + JAE long_offset_short_match_nolit_encodeBetterBlockAsm10B + MOVL $0x00000001, SI + LEAL 16(SI), SI + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, SI + MOVB SI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R12 + + // emitRepeat + LEAL -4(R12), R12 + JMP cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_three_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_two_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short_2b: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +long_offset_short_match_nolit_encodeBetterBlockAsm10B: + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm10B_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm10B_emit_copy_short + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_three_match_nolit_encodeBetterBlockAsm10B_emit_copy_short: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_two_match_nolit_encodeBetterBlockAsm10B_emit_copy_short: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_two_offset_match_nolit_encodeBetterBlockAsm10B_emit_copy_short: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + JMP two_byte_offset_match_nolit_encodeBetterBlockAsm10B + +two_byte_offset_short_match_nolit_encodeBetterBlockAsm10B: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm10B + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm10B + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +emit_copy_three_match_nolit_encodeBetterBlockAsm10B: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +match_is_repeat_encodeBetterBlockAsm10B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_repeat_encodeBetterBlockAsm10B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_repeat_encodeBetterBlockAsm10B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_repeat_encodeBetterBlockAsm10B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm10B + +two_bytes_match_emit_repeat_encodeBetterBlockAsm10B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_repeat_encodeBetterBlockAsm10B + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm10B + +one_byte_match_emit_repeat_encodeBetterBlockAsm10B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_repeat_encodeBetterBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm10B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm10B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm10B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_repeat_encodeBetterBlockAsm10B + +memmove_long_match_emit_repeat_encodeBetterBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm10Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_repeat_encodeBetterBlockAsm10B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_repeat_encodeBetterBlockAsm10B + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm10B + CMPL R8, $0x00000800 + JLT repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm10B + +cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm10B: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_repeat_encodeBetterBlockAsm10B + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_three_match_nolit_repeat_encodeBetterBlockAsm10B: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_two_match_nolit_repeat_encodeBetterBlockAsm10B: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm10B + +repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm10B: + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + +match_nolit_emitcopy_end_encodeBetterBlockAsm10B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm10B + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBetterBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBetterBlockAsm10B: + MOVQ $0x0000cf1bbcdcbf9b, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x34, R10 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x34, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x36, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x36, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 16408(SP)(R11*4) + MOVL R15, 16408(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x34, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x36, R11 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x34, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 16408(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeBetterBlockAsm10B + +emit_remainder_encodeBetterBlockAsm10B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBetterBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBetterBlockAsm10B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBetterBlockAsm10B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBetterBlockAsm10B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBetterBlockAsm10B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm10B + +two_bytes_emit_remainder_encodeBetterBlockAsm10B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBetterBlockAsm10B + JMP memmove_long_emit_remainder_encodeBetterBlockAsm10B + +one_byte_emit_remainder_encodeBetterBlockAsm10B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBetterBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm10B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBetterBlockAsm10B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBetterBlockAsm10B + +memmove_long_emit_remainder_encodeBetterBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm10Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBetterBlockAsm10B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeBetterBlockAsm8B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeBetterBlockAsm8B(SB), $5144-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000028, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeBetterBlockAsm8B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeBetterBlockAsm8B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -6(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeBetterBlockAsm8B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x04, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm8B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x38, R11 + MOVL 24(SP)(R10*4), SI + MOVL 4120(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 4120(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm8B + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeBetterBlockAsm8B + MOVL 20(SP), CX + JMP search_loop_encodeBetterBlockAsm8B + +candidateS_match_encodeBetterBlockAsm8B: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeBetterBlockAsm8B + DECL CX + MOVL R8, SI + +candidate_match_encodeBetterBlockAsm8B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeBetterBlockAsm8B + +match_extend_back_loop_encodeBetterBlockAsm8B: + CMPL CX, DI + JLE match_extend_back_end_encodeBetterBlockAsm8B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeBetterBlockAsm8B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeBetterBlockAsm8B + JMP match_extend_back_loop_encodeBetterBlockAsm8B + +match_extend_back_end_encodeBetterBlockAsm8B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeBetterBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeBetterBlockAsm8B: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeBetterBlockAsm8B + +matchlen_loopback_match_nolit_encodeBetterBlockAsm8B: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeBetterBlockAsm8B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeBetterBlockAsm8B + +matchlen_loop_match_nolit_encodeBetterBlockAsm8B: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeBetterBlockAsm8B + JZ match_nolit_end_encodeBetterBlockAsm8B + +matchlen_match4_match_nolit_encodeBetterBlockAsm8B: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeBetterBlockAsm8B + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeBetterBlockAsm8B + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeBetterBlockAsm8B: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeBetterBlockAsm8B + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeBetterBlockAsm8B + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeBetterBlockAsm8B: + CMPL R8, $0x01 + JL match_nolit_end_encodeBetterBlockAsm8B + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeBetterBlockAsm8B + LEAL 1(R12), R12 + +match_nolit_end_encodeBetterBlockAsm8B: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + CMPL 16(SP), R8 + JEQ match_is_repeat_encodeBetterBlockAsm8B + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeBetterBlockAsm8B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeBetterBlockAsm8B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeBetterBlockAsm8B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeBetterBlockAsm8B + +two_bytes_match_emit_encodeBetterBlockAsm8B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeBetterBlockAsm8B + JMP memmove_long_match_emit_encodeBetterBlockAsm8B + +one_byte_match_emit_encodeBetterBlockAsm8B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeBetterBlockAsm8B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x04 + JLE emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_4 + CMPQ R9, $0x08 + JB emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_4through7 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_4: + MOVL (R10), R11 + MOVL R11, (AX) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_4through7: + MOVL (R10), R11 + MOVL -4(R10)(R9*1), R10 + MOVL R11, (AX) + MOVL R10, -4(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeBetterBlockAsm8B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeBetterBlockAsm8B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeBetterBlockAsm8B + +memmove_long_match_emit_encodeBetterBlockAsm8B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm8Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeBetterBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeBetterBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeBetterBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeBetterBlockAsm8B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeBetterBlockAsm8B: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeBetterBlockAsm8B + CMPL R8, $0x00000800 + JAE long_offset_short_match_nolit_encodeBetterBlockAsm8B + MOVL $0x00000001, SI + LEAL 16(SI), SI + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, SI + MOVB SI, (AX) + ADDQ $0x02, AX + SUBL $0x08, R12 + + // emitRepeat + LEAL -4(R12), R12 + JMP cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +repeat_three_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +repeat_two_match_nolit_encodeBetterBlockAsm8B_emit_copy_short_2b: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +long_offset_short_match_nolit_encodeBetterBlockAsm8B: + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_encodeBetterBlockAsm8B_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm8B_emit_copy_short + +cant_repeat_two_offset_match_nolit_encodeBetterBlockAsm8B_emit_copy_short: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_encodeBetterBlockAsm8B_emit_copy_short + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +repeat_three_match_nolit_encodeBetterBlockAsm8B_emit_copy_short: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +repeat_two_match_nolit_encodeBetterBlockAsm8B_emit_copy_short: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + JMP two_byte_offset_match_nolit_encodeBetterBlockAsm8B + +two_byte_offset_short_match_nolit_encodeBetterBlockAsm8B: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeBetterBlockAsm8B + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +emit_copy_three_match_nolit_encodeBetterBlockAsm8B: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +match_is_repeat_encodeBetterBlockAsm8B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_repeat_encodeBetterBlockAsm8B + MOVL DI, R8 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R9 + SUBL SI, R8 + LEAL -1(R8), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_repeat_encodeBetterBlockAsm8B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_repeat_encodeBetterBlockAsm8B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm8B + +two_bytes_match_emit_repeat_encodeBetterBlockAsm8B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_repeat_encodeBetterBlockAsm8B + JMP memmove_long_match_emit_repeat_encodeBetterBlockAsm8B + +one_byte_match_emit_repeat_encodeBetterBlockAsm8B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_repeat_encodeBetterBlockAsm8B: + LEAQ (AX)(R8*1), SI + + // genMemMoveShort + CMPQ R8, $0x04 + JLE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_4 + CMPQ R8, $0x08 + JB emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_4through7 + CMPQ R8, $0x10 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_8through16 + CMPQ R8, $0x20 + JBE emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_4: + MOVL (R9), R10 + MOVL R10, (AX) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_4through7: + MOVL (R9), R10 + MOVL -4(R9)(R8*1), R9 + MOVL R10, (AX) + MOVL R9, -4(AX)(R8*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_8through16: + MOVQ (R9), R10 + MOVQ -8(R9)(R8*1), R9 + MOVQ R10, (AX) + MOVQ R9, -8(AX)(R8*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_17through32: + MOVOU (R9), X0 + MOVOU -16(R9)(R8*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R8*1) + JMP memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm8B + +emit_lit_memmove_match_emit_repeat_encodeBetterBlockAsm8B_memmove_move_33through64: + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + +memmove_end_copy_match_emit_repeat_encodeBetterBlockAsm8B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_repeat_encodeBetterBlockAsm8B + +memmove_long_match_emit_repeat_encodeBetterBlockAsm8B: + LEAQ (AX)(R8*1), SI + + // genMemMoveLong + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVQ R8, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R13 + SUBQ R10, R13 + DECQ R11 + JA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(R9)(R13*1), R10 + LEAQ -32(AX)(R13*1), R14 + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm8Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R14) + MOVOA X5, 16(R14) + ADDQ $0x20, R14 + ADDQ $0x20, R10 + ADDQ $0x20, R13 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(R9)(R13*1), X4 + MOVOU -16(R9)(R13*1), X5 + MOVOA X4, -32(AX)(R13*1) + MOVOA X5, -16(AX)(R13*1) + ADDQ $0x20, R13 + CMPQ R8, R13 + JAE emit_lit_memmove_long_match_emit_repeat_encodeBetterBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + MOVQ SI, AX + +emit_literal_done_match_emit_repeat_encodeBetterBlockAsm8B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitRepeat + MOVL R12, SI + LEAL -4(R12), R12 + CMPL SI, $0x08 + JLE repeat_two_match_nolit_repeat_encodeBetterBlockAsm8B + CMPL SI, $0x0c + JGE cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm8B + +cant_repeat_two_offset_match_nolit_repeat_encodeBetterBlockAsm8B: + CMPL R12, $0x00000104 + JLT repeat_three_match_nolit_repeat_encodeBetterBlockAsm8B + LEAL -256(R12), R12 + MOVW $0x0019, (AX) + MOVW R12, 2(AX) + ADDQ $0x04, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +repeat_three_match_nolit_repeat_encodeBetterBlockAsm8B: + LEAL -4(R12), R12 + MOVW $0x0015, (AX) + MOVB R12, 2(AX) + ADDQ $0x03, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + +repeat_two_match_nolit_repeat_encodeBetterBlockAsm8B: + SHLL $0x02, R12 + ORL $0x01, R12 + MOVW R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeBetterBlockAsm8B + XORQ SI, SI + LEAL 1(SI)(R12*4), R12 + MOVB R8, 1(AX) + SARL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + +match_nolit_emitcopy_end_encodeBetterBlockAsm8B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeBetterBlockAsm8B + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeBetterBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeBetterBlockAsm8B: + MOVQ $0x0000cf1bbcdcbf9b, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x36, R10 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x36, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x38, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x38, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 4120(SP)(R11*4) + MOVL R15, 4120(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x36, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x38, R11 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x36, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 4120(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeBetterBlockAsm8B + +emit_remainder_encodeBetterBlockAsm8B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeBetterBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeBetterBlockAsm8B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeBetterBlockAsm8B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeBetterBlockAsm8B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeBetterBlockAsm8B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeBetterBlockAsm8B + +two_bytes_emit_remainder_encodeBetterBlockAsm8B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeBetterBlockAsm8B + JMP memmove_long_emit_remainder_encodeBetterBlockAsm8B + +one_byte_emit_remainder_encodeBetterBlockAsm8B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeBetterBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeBetterBlockAsm8B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeBetterBlockAsm8B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeBetterBlockAsm8B + +memmove_long_emit_remainder_encodeBetterBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm8Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeBetterBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeBetterBlockAsm8B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBlockAsm(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBlockAsm(SB), $65560-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000200, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBlockAsm: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBlockAsm + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBlockAsm: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x06, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + SHLQ $0x10, R11 + IMULQ R9, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeSnappyBlockAsm + LEAL 1(CX), DI + MOVL 12(SP), SI + MOVL DI, R8 + SUBL 16(SP), R8 + JZ repeat_extend_back_end_encodeSnappyBlockAsm + +repeat_extend_back_loop_encodeSnappyBlockAsm: + CMPL DI, SI + JLE repeat_extend_back_end_encodeSnappyBlockAsm + MOVB -1(DX)(R8*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeSnappyBlockAsm + LEAL -1(DI), DI + DECL R8 + JNZ repeat_extend_back_loop_encodeSnappyBlockAsm + +repeat_extend_back_end_encodeSnappyBlockAsm: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeSnappyBlockAsm + MOVL DI, R8 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R9 + SUBL SI, R8 + LEAL -1(R8), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeSnappyBlockAsm + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeSnappyBlockAsm + CMPL SI, $0x00010000 + JLT three_bytes_repeat_emit_encodeSnappyBlockAsm + CMPL SI, $0x01000000 + JLT four_bytes_repeat_emit_encodeSnappyBlockAsm + MOVB $0xfc, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm + +four_bytes_repeat_emit_encodeSnappyBlockAsm: + MOVL SI, R10 + SHRL $0x10, R10 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R10, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm + +three_bytes_repeat_emit_encodeSnappyBlockAsm: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm + +two_bytes_repeat_emit_encodeSnappyBlockAsm: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeSnappyBlockAsm + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm + +one_byte_repeat_emit_encodeSnappyBlockAsm: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeSnappyBlockAsm: + LEAQ (AX)(R8*1), SI + + // genMemMoveShort + CMPQ R8, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_8 + CMPQ R8, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_8through16 + CMPQ R8, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_8: + MOVQ (R9), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_8through16: + MOVQ (R9), R10 + MOVQ -8(R9)(R8*1), R9 + MOVQ R10, (AX) + MOVQ R9, -8(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_17through32: + MOVOU (R9), X0 + MOVOU -16(R9)(R8*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm_memmove_move_33through64: + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + +memmove_end_copy_repeat_emit_encodeSnappyBlockAsm: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeSnappyBlockAsm + +memmove_long_repeat_emit_encodeSnappyBlockAsm: + LEAQ (AX)(R8*1), SI + + // genMemMoveLong + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVQ R8, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(R9)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsmlarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(R9)(R12*1), X4 + MOVOU -16(R9)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R8, R12 + JAE emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeSnappyBlockAsm: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R11, R11 + CMPL R8, $0x08 + JL matchlen_match4_repeat_extend_encodeSnappyBlockAsm + +matchlen_loopback_repeat_extend_encodeSnappyBlockAsm: + MOVQ (R9)(R11*1), R10 + XORQ (SI)(R11*1), R10 + TESTQ R10, R10 + JZ matchlen_loop_repeat_extend_encodeSnappyBlockAsm + +#ifdef GOAMD64_v3 + TZCNTQ R10, R10 + +#else + BSFQ R10, R10 + +#endif + SARQ $0x03, R10 + LEAL (R11)(R10*1), R11 + JMP repeat_extend_forward_end_encodeSnappyBlockAsm + +matchlen_loop_repeat_extend_encodeSnappyBlockAsm: + LEAL -8(R8), R8 + LEAL 8(R11), R11 + CMPL R8, $0x08 + JGE matchlen_loopback_repeat_extend_encodeSnappyBlockAsm + JZ repeat_extend_forward_end_encodeSnappyBlockAsm + +matchlen_match4_repeat_extend_encodeSnappyBlockAsm: + CMPL R8, $0x04 + JL matchlen_match2_repeat_extend_encodeSnappyBlockAsm + MOVL (R9)(R11*1), R10 + CMPL (SI)(R11*1), R10 + JNE matchlen_match2_repeat_extend_encodeSnappyBlockAsm + SUBL $0x04, R8 + LEAL 4(R11), R11 + +matchlen_match2_repeat_extend_encodeSnappyBlockAsm: + CMPL R8, $0x02 + JL matchlen_match1_repeat_extend_encodeSnappyBlockAsm + MOVW (R9)(R11*1), R10 + CMPW (SI)(R11*1), R10 + JNE matchlen_match1_repeat_extend_encodeSnappyBlockAsm + SUBL $0x02, R8 + LEAL 2(R11), R11 + +matchlen_match1_repeat_extend_encodeSnappyBlockAsm: + CMPL R8, $0x01 + JL repeat_extend_forward_end_encodeSnappyBlockAsm + MOVB (R9)(R11*1), R10 + CMPB (SI)(R11*1), R10 + JNE repeat_extend_forward_end_encodeSnappyBlockAsm + LEAL 1(R11), R11 + +repeat_extend_forward_end_encodeSnappyBlockAsm: + ADDL R11, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + + // emitCopy + CMPL DI, $0x00010000 + JL two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm + +four_bytes_loop_back_repeat_as_copy_encodeSnappyBlockAsm: + CMPL SI, $0x40 + JLE four_bytes_remain_repeat_as_copy_encodeSnappyBlockAsm + MOVB $0xff, (AX) + MOVL DI, 1(AX) + LEAL -64(SI), SI + ADDQ $0x05, AX + CMPL SI, $0x04 + JL four_bytes_remain_repeat_as_copy_encodeSnappyBlockAsm + JMP four_bytes_loop_back_repeat_as_copy_encodeSnappyBlockAsm + +four_bytes_remain_repeat_as_copy_encodeSnappyBlockAsm: + TESTL SI, SI + JZ repeat_end_emit_encodeSnappyBlockAsm + MOVB $0x03, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVL DI, 1(AX) + ADDQ $0x05, AX + JMP repeat_end_emit_encodeSnappyBlockAsm + +two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + JMP two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm + +two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeSnappyBlockAsm + +emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeSnappyBlockAsm: + MOVL CX, 12(SP) + JMP search_loop_encodeSnappyBlockAsm + +no_repeat_found_encodeSnappyBlockAsm: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBlockAsm + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeSnappyBlockAsm + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeSnappyBlockAsm + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBlockAsm + +candidate3_match_encodeSnappyBlockAsm: + ADDL $0x02, CX + JMP candidate_match_encodeSnappyBlockAsm + +candidate2_match_encodeSnappyBlockAsm: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeSnappyBlockAsm: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBlockAsm + +match_extend_back_loop_encodeSnappyBlockAsm: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBlockAsm + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBlockAsm + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBlockAsm + JMP match_extend_back_loop_encodeSnappyBlockAsm + +match_extend_back_end_encodeSnappyBlockAsm: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 5(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBlockAsm: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeSnappyBlockAsm + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeSnappyBlockAsm + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBlockAsm + CMPL R8, $0x00010000 + JLT three_bytes_match_emit_encodeSnappyBlockAsm + CMPL R8, $0x01000000 + JLT four_bytes_match_emit_encodeSnappyBlockAsm + MOVB $0xfc, (AX) + MOVL R8, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm + +four_bytes_match_emit_encodeSnappyBlockAsm: + MOVL R8, R10 + SHRL $0x10, R10 + MOVB $0xf8, (AX) + MOVW R8, 1(AX) + MOVB R10, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm + +three_bytes_match_emit_encodeSnappyBlockAsm: + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm + +two_bytes_match_emit_encodeSnappyBlockAsm: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeSnappyBlockAsm + JMP memmove_long_match_emit_encodeSnappyBlockAsm + +one_byte_match_emit_encodeSnappyBlockAsm: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBlockAsm: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBlockAsm: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeSnappyBlockAsm + +memmove_long_match_emit_encodeSnappyBlockAsm: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsmlarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeSnappyBlockAsm: +match_nolit_loop_encodeSnappyBlockAsm: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBlockAsm + +matchlen_loopback_match_nolit_encodeSnappyBlockAsm: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeSnappyBlockAsm + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeSnappyBlockAsm + +matchlen_loop_match_nolit_encodeSnappyBlockAsm: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBlockAsm + JZ match_nolit_end_encodeSnappyBlockAsm + +matchlen_match4_match_nolit_encodeSnappyBlockAsm: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBlockAsm + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeSnappyBlockAsm + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeSnappyBlockAsm: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBlockAsm + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeSnappyBlockAsm + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeSnappyBlockAsm: + CMPL DI, $0x01 + JL match_nolit_end_encodeSnappyBlockAsm + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeSnappyBlockAsm + LEAL 1(R10), R10 + +match_nolit_end_encodeSnappyBlockAsm: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy + CMPL SI, $0x00010000 + JL two_byte_offset_match_nolit_encodeSnappyBlockAsm + +four_bytes_loop_back_match_nolit_encodeSnappyBlockAsm: + CMPL R10, $0x40 + JLE four_bytes_remain_match_nolit_encodeSnappyBlockAsm + MOVB $0xff, (AX) + MOVL SI, 1(AX) + LEAL -64(R10), R10 + ADDQ $0x05, AX + CMPL R10, $0x04 + JL four_bytes_remain_match_nolit_encodeSnappyBlockAsm + JMP four_bytes_loop_back_match_nolit_encodeSnappyBlockAsm + +four_bytes_remain_match_nolit_encodeSnappyBlockAsm: + TESTL R10, R10 + JZ match_nolit_emitcopy_end_encodeSnappyBlockAsm + MOVB $0x03, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeSnappyBlockAsm + +two_byte_offset_match_nolit_encodeSnappyBlockAsm: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBlockAsm + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBlockAsm + +two_byte_offset_short_match_nolit_encodeSnappyBlockAsm: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBlockAsm + +emit_copy_three_match_nolit_encodeSnappyBlockAsm: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBlockAsm: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBlockAsm: + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x10, R8 + IMULQ R9, R8 + SHRQ $0x32, R8 + SHLQ $0x10, SI + IMULQ R9, SI + SHRQ $0x32, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeSnappyBlockAsm + INCL CX + JMP search_loop_encodeSnappyBlockAsm + +emit_remainder_encodeSnappyBlockAsm: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 5(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBlockAsm: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBlockAsm + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBlockAsm + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBlockAsm + CMPL DX, $0x00010000 + JLT three_bytes_emit_remainder_encodeSnappyBlockAsm + CMPL DX, $0x01000000 + JLT four_bytes_emit_remainder_encodeSnappyBlockAsm + MOVB $0xfc, (AX) + MOVL DX, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm + +four_bytes_emit_remainder_encodeSnappyBlockAsm: + MOVL DX, BX + SHRL $0x10, BX + MOVB $0xf8, (AX) + MOVW DX, 1(AX) + MOVB BL, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm + +three_bytes_emit_remainder_encodeSnappyBlockAsm: + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm + +two_bytes_emit_remainder_encodeSnappyBlockAsm: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBlockAsm + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm + +one_byte_emit_remainder_encodeSnappyBlockAsm: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBlockAsm: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBlockAsm + +memmove_long_emit_remainder_encodeSnappyBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsmlarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBlockAsm: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBlockAsm64K(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBlockAsm64K(SB), $65560-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000200, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBlockAsm64K: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBlockAsm64K + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBlockAsm64K: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x06, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm64K + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + SHLQ $0x10, R11 + IMULQ R9, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeSnappyBlockAsm64K + LEAL 1(CX), DI + MOVL 12(SP), SI + MOVL DI, R8 + SUBL 16(SP), R8 + JZ repeat_extend_back_end_encodeSnappyBlockAsm64K + +repeat_extend_back_loop_encodeSnappyBlockAsm64K: + CMPL DI, SI + JLE repeat_extend_back_end_encodeSnappyBlockAsm64K + MOVB -1(DX)(R8*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeSnappyBlockAsm64K + LEAL -1(DI), DI + DECL R8 + JNZ repeat_extend_back_loop_encodeSnappyBlockAsm64K + +repeat_extend_back_end_encodeSnappyBlockAsm64K: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeSnappyBlockAsm64K + MOVL DI, R8 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R9 + SUBL SI, R8 + LEAL -1(R8), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeSnappyBlockAsm64K + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeSnappyBlockAsm64K + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm64K + +two_bytes_repeat_emit_encodeSnappyBlockAsm64K: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeSnappyBlockAsm64K + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm64K + +one_byte_repeat_emit_encodeSnappyBlockAsm64K: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeSnappyBlockAsm64K: + LEAQ (AX)(R8*1), SI + + // genMemMoveShort + CMPQ R8, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_8 + CMPQ R8, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_8through16 + CMPQ R8, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_8: + MOVQ (R9), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm64K + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_8through16: + MOVQ (R9), R10 + MOVQ -8(R9)(R8*1), R9 + MOVQ R10, (AX) + MOVQ R9, -8(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm64K + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_17through32: + MOVOU (R9), X0 + MOVOU -16(R9)(R8*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm64K + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm64K_memmove_move_33through64: + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + +memmove_end_copy_repeat_emit_encodeSnappyBlockAsm64K: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeSnappyBlockAsm64K + +memmove_long_repeat_emit_encodeSnappyBlockAsm64K: + LEAQ (AX)(R8*1), SI + + // genMemMoveLong + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVQ R8, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32 + LEAQ -32(R9)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm64Klarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm64Klarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32: + MOVOU -32(R9)(R12*1), X4 + MOVOU -16(R9)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R8, R12 + JAE emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeSnappyBlockAsm64K: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R11, R11 + CMPL R8, $0x08 + JL matchlen_match4_repeat_extend_encodeSnappyBlockAsm64K + +matchlen_loopback_repeat_extend_encodeSnappyBlockAsm64K: + MOVQ (R9)(R11*1), R10 + XORQ (SI)(R11*1), R10 + TESTQ R10, R10 + JZ matchlen_loop_repeat_extend_encodeSnappyBlockAsm64K + +#ifdef GOAMD64_v3 + TZCNTQ R10, R10 + +#else + BSFQ R10, R10 + +#endif + SARQ $0x03, R10 + LEAL (R11)(R10*1), R11 + JMP repeat_extend_forward_end_encodeSnappyBlockAsm64K + +matchlen_loop_repeat_extend_encodeSnappyBlockAsm64K: + LEAL -8(R8), R8 + LEAL 8(R11), R11 + CMPL R8, $0x08 + JGE matchlen_loopback_repeat_extend_encodeSnappyBlockAsm64K + JZ repeat_extend_forward_end_encodeSnappyBlockAsm64K + +matchlen_match4_repeat_extend_encodeSnappyBlockAsm64K: + CMPL R8, $0x04 + JL matchlen_match2_repeat_extend_encodeSnappyBlockAsm64K + MOVL (R9)(R11*1), R10 + CMPL (SI)(R11*1), R10 + JNE matchlen_match2_repeat_extend_encodeSnappyBlockAsm64K + SUBL $0x04, R8 + LEAL 4(R11), R11 + +matchlen_match2_repeat_extend_encodeSnappyBlockAsm64K: + CMPL R8, $0x02 + JL matchlen_match1_repeat_extend_encodeSnappyBlockAsm64K + MOVW (R9)(R11*1), R10 + CMPW (SI)(R11*1), R10 + JNE matchlen_match1_repeat_extend_encodeSnappyBlockAsm64K + SUBL $0x02, R8 + LEAL 2(R11), R11 + +matchlen_match1_repeat_extend_encodeSnappyBlockAsm64K: + CMPL R8, $0x01 + JL repeat_extend_forward_end_encodeSnappyBlockAsm64K + MOVB (R9)(R11*1), R10 + CMPB (SI)(R11*1), R10 + JNE repeat_extend_forward_end_encodeSnappyBlockAsm64K + LEAL 1(R11), R11 + +repeat_extend_forward_end_encodeSnappyBlockAsm64K: + ADDL R11, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + + // emitCopy +two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm64K: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm64K + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + JMP two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm64K + +two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm64K: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm64K + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm64K + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeSnappyBlockAsm64K + +emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm64K: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeSnappyBlockAsm64K: + MOVL CX, 12(SP) + JMP search_loop_encodeSnappyBlockAsm64K + +no_repeat_found_encodeSnappyBlockAsm64K: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBlockAsm64K + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeSnappyBlockAsm64K + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeSnappyBlockAsm64K + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBlockAsm64K + +candidate3_match_encodeSnappyBlockAsm64K: + ADDL $0x02, CX + JMP candidate_match_encodeSnappyBlockAsm64K + +candidate2_match_encodeSnappyBlockAsm64K: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeSnappyBlockAsm64K: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBlockAsm64K + +match_extend_back_loop_encodeSnappyBlockAsm64K: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBlockAsm64K + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBlockAsm64K + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBlockAsm64K + JMP match_extend_back_loop_encodeSnappyBlockAsm64K + +match_extend_back_end_encodeSnappyBlockAsm64K: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBlockAsm64K + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBlockAsm64K: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeSnappyBlockAsm64K + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeSnappyBlockAsm64K + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBlockAsm64K + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm64K + +two_bytes_match_emit_encodeSnappyBlockAsm64K: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeSnappyBlockAsm64K + JMP memmove_long_match_emit_encodeSnappyBlockAsm64K + +one_byte_match_emit_encodeSnappyBlockAsm64K: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBlockAsm64K: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm64K + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm64K + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm64K + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm64K_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBlockAsm64K: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeSnappyBlockAsm64K + +memmove_long_match_emit_encodeSnappyBlockAsm64K: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm64Klarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm64Klarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeSnappyBlockAsm64K: +match_nolit_loop_encodeSnappyBlockAsm64K: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBlockAsm64K + +matchlen_loopback_match_nolit_encodeSnappyBlockAsm64K: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeSnappyBlockAsm64K + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeSnappyBlockAsm64K + +matchlen_loop_match_nolit_encodeSnappyBlockAsm64K: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBlockAsm64K + JZ match_nolit_end_encodeSnappyBlockAsm64K + +matchlen_match4_match_nolit_encodeSnappyBlockAsm64K: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBlockAsm64K + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeSnappyBlockAsm64K + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeSnappyBlockAsm64K: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBlockAsm64K + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeSnappyBlockAsm64K + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeSnappyBlockAsm64K: + CMPL DI, $0x01 + JL match_nolit_end_encodeSnappyBlockAsm64K + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeSnappyBlockAsm64K + LEAL 1(R10), R10 + +match_nolit_end_encodeSnappyBlockAsm64K: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBlockAsm64K: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBlockAsm64K + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBlockAsm64K + +two_byte_offset_short_match_nolit_encodeSnappyBlockAsm64K: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm64K + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm64K + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBlockAsm64K + +emit_copy_three_match_nolit_encodeSnappyBlockAsm64K: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBlockAsm64K: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm64K + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBlockAsm64K + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBlockAsm64K: + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x10, R8 + IMULQ R9, R8 + SHRQ $0x32, R8 + SHLQ $0x10, SI + IMULQ R9, SI + SHRQ $0x32, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeSnappyBlockAsm64K + INCL CX + JMP search_loop_encodeSnappyBlockAsm64K + +emit_remainder_encodeSnappyBlockAsm64K: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBlockAsm64K + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBlockAsm64K: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBlockAsm64K + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBlockAsm64K + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBlockAsm64K + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm64K + +two_bytes_emit_remainder_encodeSnappyBlockAsm64K: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBlockAsm64K + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm64K + +one_byte_emit_remainder_encodeSnappyBlockAsm64K: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBlockAsm64K: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm64K_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBlockAsm64K: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBlockAsm64K + +memmove_long_emit_remainder_encodeSnappyBlockAsm64K: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm64Klarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm64Klarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm64Klarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBlockAsm64K: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBlockAsm12B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBlockAsm12B(SB), $16408-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000080, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBlockAsm12B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBlockAsm12B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBlockAsm12B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x05, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm12B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x000000cf1bbcdcbb, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x18, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + SHLQ $0x18, R11 + IMULQ R9, R11 + SHRQ $0x34, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x18, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeSnappyBlockAsm12B + LEAL 1(CX), DI + MOVL 12(SP), SI + MOVL DI, R8 + SUBL 16(SP), R8 + JZ repeat_extend_back_end_encodeSnappyBlockAsm12B + +repeat_extend_back_loop_encodeSnappyBlockAsm12B: + CMPL DI, SI + JLE repeat_extend_back_end_encodeSnappyBlockAsm12B + MOVB -1(DX)(R8*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeSnappyBlockAsm12B + LEAL -1(DI), DI + DECL R8 + JNZ repeat_extend_back_loop_encodeSnappyBlockAsm12B + +repeat_extend_back_end_encodeSnappyBlockAsm12B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeSnappyBlockAsm12B + MOVL DI, R8 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R9 + SUBL SI, R8 + LEAL -1(R8), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeSnappyBlockAsm12B + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeSnappyBlockAsm12B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm12B + +two_bytes_repeat_emit_encodeSnappyBlockAsm12B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeSnappyBlockAsm12B + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm12B + +one_byte_repeat_emit_encodeSnappyBlockAsm12B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeSnappyBlockAsm12B: + LEAQ (AX)(R8*1), SI + + // genMemMoveShort + CMPQ R8, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_8 + CMPQ R8, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_8through16 + CMPQ R8, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_8: + MOVQ (R9), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm12B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_8through16: + MOVQ (R9), R10 + MOVQ -8(R9)(R8*1), R9 + MOVQ R10, (AX) + MOVQ R9, -8(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm12B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_17through32: + MOVOU (R9), X0 + MOVOU -16(R9)(R8*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm12B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm12B_memmove_move_33through64: + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + +memmove_end_copy_repeat_emit_encodeSnappyBlockAsm12B: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeSnappyBlockAsm12B + +memmove_long_repeat_emit_encodeSnappyBlockAsm12B: + LEAQ (AX)(R8*1), SI + + // genMemMoveLong + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVQ R8, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(R9)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm12Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(R9)(R12*1), X4 + MOVOU -16(R9)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R8, R12 + JAE emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeSnappyBlockAsm12B: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R11, R11 + CMPL R8, $0x08 + JL matchlen_match4_repeat_extend_encodeSnappyBlockAsm12B + +matchlen_loopback_repeat_extend_encodeSnappyBlockAsm12B: + MOVQ (R9)(R11*1), R10 + XORQ (SI)(R11*1), R10 + TESTQ R10, R10 + JZ matchlen_loop_repeat_extend_encodeSnappyBlockAsm12B + +#ifdef GOAMD64_v3 + TZCNTQ R10, R10 + +#else + BSFQ R10, R10 + +#endif + SARQ $0x03, R10 + LEAL (R11)(R10*1), R11 + JMP repeat_extend_forward_end_encodeSnappyBlockAsm12B + +matchlen_loop_repeat_extend_encodeSnappyBlockAsm12B: + LEAL -8(R8), R8 + LEAL 8(R11), R11 + CMPL R8, $0x08 + JGE matchlen_loopback_repeat_extend_encodeSnappyBlockAsm12B + JZ repeat_extend_forward_end_encodeSnappyBlockAsm12B + +matchlen_match4_repeat_extend_encodeSnappyBlockAsm12B: + CMPL R8, $0x04 + JL matchlen_match2_repeat_extend_encodeSnappyBlockAsm12B + MOVL (R9)(R11*1), R10 + CMPL (SI)(R11*1), R10 + JNE matchlen_match2_repeat_extend_encodeSnappyBlockAsm12B + SUBL $0x04, R8 + LEAL 4(R11), R11 + +matchlen_match2_repeat_extend_encodeSnappyBlockAsm12B: + CMPL R8, $0x02 + JL matchlen_match1_repeat_extend_encodeSnappyBlockAsm12B + MOVW (R9)(R11*1), R10 + CMPW (SI)(R11*1), R10 + JNE matchlen_match1_repeat_extend_encodeSnappyBlockAsm12B + SUBL $0x02, R8 + LEAL 2(R11), R11 + +matchlen_match1_repeat_extend_encodeSnappyBlockAsm12B: + CMPL R8, $0x01 + JL repeat_extend_forward_end_encodeSnappyBlockAsm12B + MOVB (R9)(R11*1), R10 + CMPB (SI)(R11*1), R10 + JNE repeat_extend_forward_end_encodeSnappyBlockAsm12B + LEAL 1(R11), R11 + +repeat_extend_forward_end_encodeSnappyBlockAsm12B: + ADDL R11, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + + // emitCopy +two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm12B: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm12B + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + JMP two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm12B + +two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm12B: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm12B + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm12B + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeSnappyBlockAsm12B + +emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm12B: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeSnappyBlockAsm12B: + MOVL CX, 12(SP) + JMP search_loop_encodeSnappyBlockAsm12B + +no_repeat_found_encodeSnappyBlockAsm12B: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBlockAsm12B + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeSnappyBlockAsm12B + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeSnappyBlockAsm12B + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBlockAsm12B + +candidate3_match_encodeSnappyBlockAsm12B: + ADDL $0x02, CX + JMP candidate_match_encodeSnappyBlockAsm12B + +candidate2_match_encodeSnappyBlockAsm12B: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeSnappyBlockAsm12B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBlockAsm12B + +match_extend_back_loop_encodeSnappyBlockAsm12B: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBlockAsm12B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBlockAsm12B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBlockAsm12B + JMP match_extend_back_loop_encodeSnappyBlockAsm12B + +match_extend_back_end_encodeSnappyBlockAsm12B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBlockAsm12B: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeSnappyBlockAsm12B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeSnappyBlockAsm12B + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBlockAsm12B + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm12B + +two_bytes_match_emit_encodeSnappyBlockAsm12B: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeSnappyBlockAsm12B + JMP memmove_long_match_emit_encodeSnappyBlockAsm12B + +one_byte_match_emit_encodeSnappyBlockAsm12B: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBlockAsm12B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm12B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm12B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm12B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm12B_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBlockAsm12B: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeSnappyBlockAsm12B + +memmove_long_match_emit_encodeSnappyBlockAsm12B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm12Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeSnappyBlockAsm12B: +match_nolit_loop_encodeSnappyBlockAsm12B: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBlockAsm12B + +matchlen_loopback_match_nolit_encodeSnappyBlockAsm12B: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeSnappyBlockAsm12B + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeSnappyBlockAsm12B + +matchlen_loop_match_nolit_encodeSnappyBlockAsm12B: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBlockAsm12B + JZ match_nolit_end_encodeSnappyBlockAsm12B + +matchlen_match4_match_nolit_encodeSnappyBlockAsm12B: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBlockAsm12B + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeSnappyBlockAsm12B + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeSnappyBlockAsm12B: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBlockAsm12B + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeSnappyBlockAsm12B + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeSnappyBlockAsm12B: + CMPL DI, $0x01 + JL match_nolit_end_encodeSnappyBlockAsm12B + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeSnappyBlockAsm12B + LEAL 1(R10), R10 + +match_nolit_end_encodeSnappyBlockAsm12B: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBlockAsm12B: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBlockAsm12B + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBlockAsm12B + +two_byte_offset_short_match_nolit_encodeSnappyBlockAsm12B: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm12B + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm12B + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBlockAsm12B + +emit_copy_three_match_nolit_encodeSnappyBlockAsm12B: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBlockAsm12B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm12B + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBlockAsm12B: + MOVQ $0x000000cf1bbcdcbb, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x18, R8 + IMULQ R9, R8 + SHRQ $0x34, R8 + SHLQ $0x18, SI + IMULQ R9, SI + SHRQ $0x34, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeSnappyBlockAsm12B + INCL CX + JMP search_loop_encodeSnappyBlockAsm12B + +emit_remainder_encodeSnappyBlockAsm12B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBlockAsm12B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBlockAsm12B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBlockAsm12B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBlockAsm12B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm12B + +two_bytes_emit_remainder_encodeSnappyBlockAsm12B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBlockAsm12B + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm12B + +one_byte_emit_remainder_encodeSnappyBlockAsm12B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm12B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBlockAsm12B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBlockAsm12B + +memmove_long_emit_remainder_encodeSnappyBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm12Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBlockAsm12B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBlockAsm10B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBlockAsm10B(SB), $4120-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000020, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBlockAsm10B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBlockAsm10B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBlockAsm10B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x05, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm10B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x9e3779b1, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + SHLQ $0x20, R11 + IMULQ R9, R11 + SHRQ $0x36, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeSnappyBlockAsm10B + LEAL 1(CX), DI + MOVL 12(SP), SI + MOVL DI, R8 + SUBL 16(SP), R8 + JZ repeat_extend_back_end_encodeSnappyBlockAsm10B + +repeat_extend_back_loop_encodeSnappyBlockAsm10B: + CMPL DI, SI + JLE repeat_extend_back_end_encodeSnappyBlockAsm10B + MOVB -1(DX)(R8*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeSnappyBlockAsm10B + LEAL -1(DI), DI + DECL R8 + JNZ repeat_extend_back_loop_encodeSnappyBlockAsm10B + +repeat_extend_back_end_encodeSnappyBlockAsm10B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeSnappyBlockAsm10B + MOVL DI, R8 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R9 + SUBL SI, R8 + LEAL -1(R8), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeSnappyBlockAsm10B + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeSnappyBlockAsm10B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm10B + +two_bytes_repeat_emit_encodeSnappyBlockAsm10B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeSnappyBlockAsm10B + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm10B + +one_byte_repeat_emit_encodeSnappyBlockAsm10B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeSnappyBlockAsm10B: + LEAQ (AX)(R8*1), SI + + // genMemMoveShort + CMPQ R8, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_8 + CMPQ R8, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_8through16 + CMPQ R8, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_8: + MOVQ (R9), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm10B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_8through16: + MOVQ (R9), R10 + MOVQ -8(R9)(R8*1), R9 + MOVQ R10, (AX) + MOVQ R9, -8(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm10B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_17through32: + MOVOU (R9), X0 + MOVOU -16(R9)(R8*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm10B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm10B_memmove_move_33through64: + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + +memmove_end_copy_repeat_emit_encodeSnappyBlockAsm10B: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeSnappyBlockAsm10B + +memmove_long_repeat_emit_encodeSnappyBlockAsm10B: + LEAQ (AX)(R8*1), SI + + // genMemMoveLong + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVQ R8, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(R9)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm10Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(R9)(R12*1), X4 + MOVOU -16(R9)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R8, R12 + JAE emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeSnappyBlockAsm10B: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R11, R11 + CMPL R8, $0x08 + JL matchlen_match4_repeat_extend_encodeSnappyBlockAsm10B + +matchlen_loopback_repeat_extend_encodeSnappyBlockAsm10B: + MOVQ (R9)(R11*1), R10 + XORQ (SI)(R11*1), R10 + TESTQ R10, R10 + JZ matchlen_loop_repeat_extend_encodeSnappyBlockAsm10B + +#ifdef GOAMD64_v3 + TZCNTQ R10, R10 + +#else + BSFQ R10, R10 + +#endif + SARQ $0x03, R10 + LEAL (R11)(R10*1), R11 + JMP repeat_extend_forward_end_encodeSnappyBlockAsm10B + +matchlen_loop_repeat_extend_encodeSnappyBlockAsm10B: + LEAL -8(R8), R8 + LEAL 8(R11), R11 + CMPL R8, $0x08 + JGE matchlen_loopback_repeat_extend_encodeSnappyBlockAsm10B + JZ repeat_extend_forward_end_encodeSnappyBlockAsm10B + +matchlen_match4_repeat_extend_encodeSnappyBlockAsm10B: + CMPL R8, $0x04 + JL matchlen_match2_repeat_extend_encodeSnappyBlockAsm10B + MOVL (R9)(R11*1), R10 + CMPL (SI)(R11*1), R10 + JNE matchlen_match2_repeat_extend_encodeSnappyBlockAsm10B + SUBL $0x04, R8 + LEAL 4(R11), R11 + +matchlen_match2_repeat_extend_encodeSnappyBlockAsm10B: + CMPL R8, $0x02 + JL matchlen_match1_repeat_extend_encodeSnappyBlockAsm10B + MOVW (R9)(R11*1), R10 + CMPW (SI)(R11*1), R10 + JNE matchlen_match1_repeat_extend_encodeSnappyBlockAsm10B + SUBL $0x02, R8 + LEAL 2(R11), R11 + +matchlen_match1_repeat_extend_encodeSnappyBlockAsm10B: + CMPL R8, $0x01 + JL repeat_extend_forward_end_encodeSnappyBlockAsm10B + MOVB (R9)(R11*1), R10 + CMPB (SI)(R11*1), R10 + JNE repeat_extend_forward_end_encodeSnappyBlockAsm10B + LEAL 1(R11), R11 + +repeat_extend_forward_end_encodeSnappyBlockAsm10B: + ADDL R11, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + + // emitCopy +two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm10B: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm10B + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + JMP two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm10B + +two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm10B: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm10B + CMPL DI, $0x00000800 + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm10B + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeSnappyBlockAsm10B + +emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm10B: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeSnappyBlockAsm10B: + MOVL CX, 12(SP) + JMP search_loop_encodeSnappyBlockAsm10B + +no_repeat_found_encodeSnappyBlockAsm10B: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBlockAsm10B + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeSnappyBlockAsm10B + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeSnappyBlockAsm10B + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBlockAsm10B + +candidate3_match_encodeSnappyBlockAsm10B: + ADDL $0x02, CX + JMP candidate_match_encodeSnappyBlockAsm10B + +candidate2_match_encodeSnappyBlockAsm10B: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeSnappyBlockAsm10B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBlockAsm10B + +match_extend_back_loop_encodeSnappyBlockAsm10B: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBlockAsm10B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBlockAsm10B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBlockAsm10B + JMP match_extend_back_loop_encodeSnappyBlockAsm10B + +match_extend_back_end_encodeSnappyBlockAsm10B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBlockAsm10B: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeSnappyBlockAsm10B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeSnappyBlockAsm10B + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBlockAsm10B + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm10B + +two_bytes_match_emit_encodeSnappyBlockAsm10B: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeSnappyBlockAsm10B + JMP memmove_long_match_emit_encodeSnappyBlockAsm10B + +one_byte_match_emit_encodeSnappyBlockAsm10B: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBlockAsm10B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm10B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm10B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm10B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm10B_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBlockAsm10B: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeSnappyBlockAsm10B + +memmove_long_match_emit_encodeSnappyBlockAsm10B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm10Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeSnappyBlockAsm10B: +match_nolit_loop_encodeSnappyBlockAsm10B: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBlockAsm10B + +matchlen_loopback_match_nolit_encodeSnappyBlockAsm10B: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeSnappyBlockAsm10B + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeSnappyBlockAsm10B + +matchlen_loop_match_nolit_encodeSnappyBlockAsm10B: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBlockAsm10B + JZ match_nolit_end_encodeSnappyBlockAsm10B + +matchlen_match4_match_nolit_encodeSnappyBlockAsm10B: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBlockAsm10B + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeSnappyBlockAsm10B + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeSnappyBlockAsm10B: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBlockAsm10B + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeSnappyBlockAsm10B + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeSnappyBlockAsm10B: + CMPL DI, $0x01 + JL match_nolit_end_encodeSnappyBlockAsm10B + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeSnappyBlockAsm10B + LEAL 1(R10), R10 + +match_nolit_end_encodeSnappyBlockAsm10B: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBlockAsm10B: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBlockAsm10B + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBlockAsm10B + +two_byte_offset_short_match_nolit_encodeSnappyBlockAsm10B: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm10B + CMPL SI, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm10B + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBlockAsm10B + +emit_copy_three_match_nolit_encodeSnappyBlockAsm10B: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBlockAsm10B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm10B + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBlockAsm10B: + MOVQ $0x9e3779b1, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x20, R8 + IMULQ R9, R8 + SHRQ $0x36, R8 + SHLQ $0x20, SI + IMULQ R9, SI + SHRQ $0x36, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeSnappyBlockAsm10B + INCL CX + JMP search_loop_encodeSnappyBlockAsm10B + +emit_remainder_encodeSnappyBlockAsm10B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBlockAsm10B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBlockAsm10B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBlockAsm10B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBlockAsm10B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm10B + +two_bytes_emit_remainder_encodeSnappyBlockAsm10B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBlockAsm10B + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm10B + +one_byte_emit_remainder_encodeSnappyBlockAsm10B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm10B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBlockAsm10B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBlockAsm10B + +memmove_long_emit_remainder_encodeSnappyBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm10Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBlockAsm10B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBlockAsm8B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBlockAsm8B(SB), $1048-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000008, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBlockAsm8B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBlockAsm8B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL CX, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBlockAsm8B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x04, SI + LEAL 4(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm8B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x9e3779b1, R9 + MOVQ DI, R10 + MOVQ DI, R11 + SHRQ $0x08, R11 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x38, R10 + SHLQ $0x20, R11 + IMULQ R9, R11 + SHRQ $0x38, R11 + MOVL 24(SP)(R10*4), SI + MOVL 24(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + LEAL 1(CX), R10 + MOVL R10, 24(SP)(R11*4) + MOVQ DI, R10 + SHRQ $0x10, R10 + SHLQ $0x20, R10 + IMULQ R9, R10 + SHRQ $0x38, R10 + MOVL CX, R9 + SUBL 16(SP), R9 + MOVL 1(DX)(R9*1), R11 + MOVQ DI, R9 + SHRQ $0x08, R9 + CMPL R9, R11 + JNE no_repeat_found_encodeSnappyBlockAsm8B + LEAL 1(CX), DI + MOVL 12(SP), SI + MOVL DI, R8 + SUBL 16(SP), R8 + JZ repeat_extend_back_end_encodeSnappyBlockAsm8B + +repeat_extend_back_loop_encodeSnappyBlockAsm8B: + CMPL DI, SI + JLE repeat_extend_back_end_encodeSnappyBlockAsm8B + MOVB -1(DX)(R8*1), BL + MOVB -1(DX)(DI*1), R9 + CMPB BL, R9 + JNE repeat_extend_back_end_encodeSnappyBlockAsm8B + LEAL -1(DI), DI + DECL R8 + JNZ repeat_extend_back_loop_encodeSnappyBlockAsm8B + +repeat_extend_back_end_encodeSnappyBlockAsm8B: + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_repeat_emit_encodeSnappyBlockAsm8B + MOVL DI, R8 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R9 + SUBL SI, R8 + LEAL -1(R8), SI + CMPL SI, $0x3c + JLT one_byte_repeat_emit_encodeSnappyBlockAsm8B + CMPL SI, $0x00000100 + JLT two_bytes_repeat_emit_encodeSnappyBlockAsm8B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm8B + +two_bytes_repeat_emit_encodeSnappyBlockAsm8B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_repeat_emit_encodeSnappyBlockAsm8B + JMP memmove_long_repeat_emit_encodeSnappyBlockAsm8B + +one_byte_repeat_emit_encodeSnappyBlockAsm8B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_repeat_emit_encodeSnappyBlockAsm8B: + LEAQ (AX)(R8*1), SI + + // genMemMoveShort + CMPQ R8, $0x08 + JLE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_8 + CMPQ R8, $0x10 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_8through16 + CMPQ R8, $0x20 + JBE emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_8: + MOVQ (R9), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm8B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_8through16: + MOVQ (R9), R10 + MOVQ -8(R9)(R8*1), R9 + MOVQ R10, (AX) + MOVQ R9, -8(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm8B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_17through32: + MOVOU (R9), X0 + MOVOU -16(R9)(R8*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R8*1) + JMP memmove_end_copy_repeat_emit_encodeSnappyBlockAsm8B + +emit_lit_memmove_repeat_emit_encodeSnappyBlockAsm8B_memmove_move_33through64: + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + +memmove_end_copy_repeat_emit_encodeSnappyBlockAsm8B: + MOVQ SI, AX + JMP emit_literal_done_repeat_emit_encodeSnappyBlockAsm8B + +memmove_long_repeat_emit_encodeSnappyBlockAsm8B: + LEAQ (AX)(R8*1), SI + + // genMemMoveLong + MOVOU (R9), X0 + MOVOU 16(R9), X1 + MOVOU -32(R9)(R8*1), X2 + MOVOU -16(R9)(R8*1), X3 + MOVQ R8, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(R9)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm8Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(R9)(R12*1), X4 + MOVOU -16(R9)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R8, R12 + JAE emit_lit_memmove_long_repeat_emit_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R8*1) + MOVOU X3, -16(AX)(R8*1) + MOVQ SI, AX + +emit_literal_done_repeat_emit_encodeSnappyBlockAsm8B: + ADDL $0x05, CX + MOVL CX, SI + SUBL 16(SP), SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R11, R11 + CMPL R8, $0x08 + JL matchlen_match4_repeat_extend_encodeSnappyBlockAsm8B + +matchlen_loopback_repeat_extend_encodeSnappyBlockAsm8B: + MOVQ (R9)(R11*1), R10 + XORQ (SI)(R11*1), R10 + TESTQ R10, R10 + JZ matchlen_loop_repeat_extend_encodeSnappyBlockAsm8B + +#ifdef GOAMD64_v3 + TZCNTQ R10, R10 + +#else + BSFQ R10, R10 + +#endif + SARQ $0x03, R10 + LEAL (R11)(R10*1), R11 + JMP repeat_extend_forward_end_encodeSnappyBlockAsm8B + +matchlen_loop_repeat_extend_encodeSnappyBlockAsm8B: + LEAL -8(R8), R8 + LEAL 8(R11), R11 + CMPL R8, $0x08 + JGE matchlen_loopback_repeat_extend_encodeSnappyBlockAsm8B + JZ repeat_extend_forward_end_encodeSnappyBlockAsm8B + +matchlen_match4_repeat_extend_encodeSnappyBlockAsm8B: + CMPL R8, $0x04 + JL matchlen_match2_repeat_extend_encodeSnappyBlockAsm8B + MOVL (R9)(R11*1), R10 + CMPL (SI)(R11*1), R10 + JNE matchlen_match2_repeat_extend_encodeSnappyBlockAsm8B + SUBL $0x04, R8 + LEAL 4(R11), R11 + +matchlen_match2_repeat_extend_encodeSnappyBlockAsm8B: + CMPL R8, $0x02 + JL matchlen_match1_repeat_extend_encodeSnappyBlockAsm8B + MOVW (R9)(R11*1), R10 + CMPW (SI)(R11*1), R10 + JNE matchlen_match1_repeat_extend_encodeSnappyBlockAsm8B + SUBL $0x02, R8 + LEAL 2(R11), R11 + +matchlen_match1_repeat_extend_encodeSnappyBlockAsm8B: + CMPL R8, $0x01 + JL repeat_extend_forward_end_encodeSnappyBlockAsm8B + MOVB (R9)(R11*1), R10 + CMPB (SI)(R11*1), R10 + JNE repeat_extend_forward_end_encodeSnappyBlockAsm8B + LEAL 1(R11), R11 + +repeat_extend_forward_end_encodeSnappyBlockAsm8B: + ADDL R11, CX + MOVL CX, SI + SUBL DI, SI + MOVL 16(SP), DI + + // emitCopy +two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm8B: + CMPL SI, $0x40 + JLE two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm8B + MOVB $0xee, (AX) + MOVW DI, 1(AX) + LEAL -60(SI), SI + ADDQ $0x03, AX + JMP two_byte_offset_repeat_as_copy_encodeSnappyBlockAsm8B + +two_byte_offset_short_repeat_as_copy_encodeSnappyBlockAsm8B: + CMPL SI, $0x0c + JGE emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm8B + MOVB $0x01, BL + LEAL -16(BX)(SI*4), SI + MOVB DI, 1(AX) + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, AX + JMP repeat_end_emit_encodeSnappyBlockAsm8B + +emit_copy_three_repeat_as_copy_encodeSnappyBlockAsm8B: + MOVB $0x02, BL + LEAL -4(BX)(SI*4), SI + MOVB SI, (AX) + MOVW DI, 1(AX) + ADDQ $0x03, AX + +repeat_end_emit_encodeSnappyBlockAsm8B: + MOVL CX, 12(SP) + JMP search_loop_encodeSnappyBlockAsm8B + +no_repeat_found_encodeSnappyBlockAsm8B: + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBlockAsm8B + SHRQ $0x08, DI + MOVL 24(SP)(R10*4), SI + LEAL 2(CX), R9 + CMPL (DX)(R8*1), DI + JEQ candidate2_match_encodeSnappyBlockAsm8B + MOVL R9, 24(SP)(R10*4) + SHRQ $0x08, DI + CMPL (DX)(SI*1), DI + JEQ candidate3_match_encodeSnappyBlockAsm8B + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBlockAsm8B + +candidate3_match_encodeSnappyBlockAsm8B: + ADDL $0x02, CX + JMP candidate_match_encodeSnappyBlockAsm8B + +candidate2_match_encodeSnappyBlockAsm8B: + MOVL R9, 24(SP)(R10*4) + INCL CX + MOVL R8, SI + +candidate_match_encodeSnappyBlockAsm8B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBlockAsm8B + +match_extend_back_loop_encodeSnappyBlockAsm8B: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBlockAsm8B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBlockAsm8B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBlockAsm8B + JMP match_extend_back_loop_encodeSnappyBlockAsm8B + +match_extend_back_end_encodeSnappyBlockAsm8B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBlockAsm8B: + MOVL CX, DI + MOVL 12(SP), R8 + CMPL R8, DI + JEQ emit_literal_done_match_emit_encodeSnappyBlockAsm8B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(R8*1), DI + SUBL R8, R9 + LEAL -1(R9), R8 + CMPL R8, $0x3c + JLT one_byte_match_emit_encodeSnappyBlockAsm8B + CMPL R8, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBlockAsm8B + MOVB $0xf4, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBlockAsm8B + +two_bytes_match_emit_encodeSnappyBlockAsm8B: + MOVB $0xf0, (AX) + MOVB R8, 1(AX) + ADDQ $0x02, AX + CMPL R8, $0x40 + JL memmove_match_emit_encodeSnappyBlockAsm8B + JMP memmove_long_match_emit_encodeSnappyBlockAsm8B + +one_byte_match_emit_encodeSnappyBlockAsm8B: + SHLB $0x02, R8 + MOVB R8, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBlockAsm8B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_8: + MOVQ (DI), R10 + MOVQ R10, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm8B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_8through16: + MOVQ (DI), R10 + MOVQ -8(DI)(R9*1), DI + MOVQ R10, (AX) + MOVQ DI, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm8B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_17through32: + MOVOU (DI), X0 + MOVOU -16(DI)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBlockAsm8B + +emit_lit_memmove_match_emit_encodeSnappyBlockAsm8B_memmove_move_33through64: + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBlockAsm8B: + MOVQ R8, AX + JMP emit_literal_done_match_emit_encodeSnappyBlockAsm8B + +memmove_long_match_emit_encodeSnappyBlockAsm8B: + LEAQ (AX)(R9*1), R8 + + // genMemMoveLong + MOVOU (DI), X0 + MOVOU 16(DI), X1 + MOVOU -32(DI)(R9*1), X2 + MOVOU -16(DI)(R9*1), X3 + MOVQ R9, R11 + SHRQ $0x05, R11 + MOVQ AX, R10 + ANDL $0x0000001f, R10 + MOVQ $0x00000040, R12 + SUBQ R10, R12 + DECQ R11 + JA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(DI)(R12*1), R10 + LEAQ -32(AX)(R12*1), R13 + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm8Blarge_big_loop_back: + MOVOU (R10), X4 + MOVOU 16(R10), X5 + MOVOA X4, (R13) + MOVOA X5, 16(R13) + ADDQ $0x20, R13 + ADDQ $0x20, R10 + ADDQ $0x20, R12 + DECQ R11 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(DI)(R12*1), X4 + MOVOU -16(DI)(R12*1), X5 + MOVOA X4, -32(AX)(R12*1) + MOVOA X5, -16(AX)(R12*1) + ADDQ $0x20, R12 + CMPQ R9, R12 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ R8, AX + +emit_literal_done_match_emit_encodeSnappyBlockAsm8B: +match_nolit_loop_encodeSnappyBlockAsm8B: + MOVL CX, DI + SUBL SI, DI + MOVL DI, 16(SP) + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), DI + SUBL CX, DI + LEAQ (DX)(CX*1), R8 + LEAQ (DX)(SI*1), SI + + // matchLen + XORL R10, R10 + CMPL DI, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBlockAsm8B + +matchlen_loopback_match_nolit_encodeSnappyBlockAsm8B: + MOVQ (R8)(R10*1), R9 + XORQ (SI)(R10*1), R9 + TESTQ R9, R9 + JZ matchlen_loop_match_nolit_encodeSnappyBlockAsm8B + +#ifdef GOAMD64_v3 + TZCNTQ R9, R9 + +#else + BSFQ R9, R9 + +#endif + SARQ $0x03, R9 + LEAL (R10)(R9*1), R10 + JMP match_nolit_end_encodeSnappyBlockAsm8B + +matchlen_loop_match_nolit_encodeSnappyBlockAsm8B: + LEAL -8(DI), DI + LEAL 8(R10), R10 + CMPL DI, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBlockAsm8B + JZ match_nolit_end_encodeSnappyBlockAsm8B + +matchlen_match4_match_nolit_encodeSnappyBlockAsm8B: + CMPL DI, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBlockAsm8B + MOVL (R8)(R10*1), R9 + CMPL (SI)(R10*1), R9 + JNE matchlen_match2_match_nolit_encodeSnappyBlockAsm8B + SUBL $0x04, DI + LEAL 4(R10), R10 + +matchlen_match2_match_nolit_encodeSnappyBlockAsm8B: + CMPL DI, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBlockAsm8B + MOVW (R8)(R10*1), R9 + CMPW (SI)(R10*1), R9 + JNE matchlen_match1_match_nolit_encodeSnappyBlockAsm8B + SUBL $0x02, DI + LEAL 2(R10), R10 + +matchlen_match1_match_nolit_encodeSnappyBlockAsm8B: + CMPL DI, $0x01 + JL match_nolit_end_encodeSnappyBlockAsm8B + MOVB (R8)(R10*1), R9 + CMPB (SI)(R10*1), R9 + JNE match_nolit_end_encodeSnappyBlockAsm8B + LEAL 1(R10), R10 + +match_nolit_end_encodeSnappyBlockAsm8B: + ADDL R10, CX + MOVL 16(SP), SI + ADDL $0x04, R10 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBlockAsm8B: + CMPL R10, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBlockAsm8B + MOVB $0xee, (AX) + MOVW SI, 1(AX) + LEAL -60(R10), R10 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBlockAsm8B + +two_byte_offset_short_match_nolit_encodeSnappyBlockAsm8B: + CMPL R10, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBlockAsm8B + MOVB $0x01, BL + LEAL -16(BX)(R10*4), R10 + MOVB SI, 1(AX) + SHRL $0x08, SI + SHLL $0x05, SI + ORL SI, R10 + MOVB R10, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBlockAsm8B + +emit_copy_three_match_nolit_encodeSnappyBlockAsm8B: + MOVB $0x02, BL + LEAL -4(BX)(R10*4), R10 + MOVB R10, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBlockAsm8B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBlockAsm8B + MOVQ -2(DX)(CX*1), DI + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBlockAsm8B: + MOVQ $0x9e3779b1, R9 + MOVQ DI, R8 + SHRQ $0x10, DI + MOVQ DI, SI + SHLQ $0x20, R8 + IMULQ R9, R8 + SHRQ $0x38, R8 + SHLQ $0x20, SI + IMULQ R9, SI + SHRQ $0x38, SI + LEAL -2(CX), R9 + LEAQ 24(SP)(SI*4), R10 + MOVL (R10), SI + MOVL R9, 24(SP)(R8*4) + MOVL CX, (R10) + CMPL (DX)(SI*1), DI + JEQ match_nolit_loop_encodeSnappyBlockAsm8B + INCL CX + JMP search_loop_encodeSnappyBlockAsm8B + +emit_remainder_encodeSnappyBlockAsm8B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBlockAsm8B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBlockAsm8B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBlockAsm8B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBlockAsm8B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm8B + +two_bytes_emit_remainder_encodeSnappyBlockAsm8B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBlockAsm8B + JMP memmove_long_emit_remainder_encodeSnappyBlockAsm8B + +one_byte_emit_remainder_encodeSnappyBlockAsm8B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBlockAsm8B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBlockAsm8B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBlockAsm8B + +memmove_long_emit_remainder_encodeSnappyBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm8Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBlockAsm8B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBetterBlockAsm(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBetterBlockAsm(SB), $327704-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000a00, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBetterBlockAsm: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBetterBlockAsm + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBetterBlockAsm: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x07, SI + CMPL SI, $0x63 + JLE check_maxskip_ok_encodeSnappyBetterBlockAsm + LEAL 100(CX), SI + JMP check_maxskip_cont_encodeSnappyBetterBlockAsm + +check_maxskip_ok_encodeSnappyBetterBlockAsm: + LEAL 1(CX)(SI*1), SI + +check_maxskip_cont_encodeSnappyBetterBlockAsm: + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x00cf1bbcdcbfa563, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 262168(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 262168(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeSnappyBetterBlockAsm + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBetterBlockAsm + +candidateS_match_encodeSnappyBetterBlockAsm: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm + DECL CX + MOVL R8, SI + +candidate_match_encodeSnappyBetterBlockAsm: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm + +match_extend_back_loop_encodeSnappyBetterBlockAsm: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBetterBlockAsm + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBetterBlockAsm + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm + JMP match_extend_back_loop_encodeSnappyBetterBlockAsm + +match_extend_back_end_encodeSnappyBetterBlockAsm: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 5(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBetterBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBetterBlockAsm: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm + +matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeSnappyBetterBlockAsm + +matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm + JZ match_nolit_end_encodeSnappyBetterBlockAsm + +matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm: + CMPL R8, $0x01 + JL match_nolit_end_encodeSnappyBetterBlockAsm + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeSnappyBetterBlockAsm + LEAL 1(R12), R12 + +match_nolit_end_encodeSnappyBetterBlockAsm: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + CMPL R12, $0x01 + JG match_length_ok_encodeSnappyBetterBlockAsm + CMPL R8, $0x0000ffff + JLE match_length_ok_encodeSnappyBetterBlockAsm + MOVL 20(SP), CX + INCL CX + JMP search_loop_encodeSnappyBetterBlockAsm + +match_length_ok_encodeSnappyBetterBlockAsm: + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeSnappyBetterBlockAsm + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeSnappyBetterBlockAsm + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBetterBlockAsm + CMPL SI, $0x00010000 + JLT three_bytes_match_emit_encodeSnappyBetterBlockAsm + CMPL SI, $0x01000000 + JLT four_bytes_match_emit_encodeSnappyBetterBlockAsm + MOVB $0xfc, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm + +four_bytes_match_emit_encodeSnappyBetterBlockAsm: + MOVL SI, R11 + SHRL $0x10, R11 + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB R11, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm + +three_bytes_match_emit_encodeSnappyBetterBlockAsm: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm + +two_bytes_match_emit_encodeSnappyBetterBlockAsm: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeSnappyBetterBlockAsm + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm + +one_byte_match_emit_encodeSnappyBetterBlockAsm: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBetterBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeSnappyBetterBlockAsm + +memmove_long_match_emit_encodeSnappyBetterBlockAsm: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsmlarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeSnappyBetterBlockAsm: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy + CMPL R8, $0x00010000 + JL two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm + +four_bytes_loop_back_match_nolit_encodeSnappyBetterBlockAsm: + CMPL R12, $0x40 + JLE four_bytes_remain_match_nolit_encodeSnappyBetterBlockAsm + MOVB $0xff, (AX) + MOVL R8, 1(AX) + LEAL -64(R12), R12 + ADDQ $0x05, AX + CMPL R12, $0x04 + JL four_bytes_remain_match_nolit_encodeSnappyBetterBlockAsm + JMP four_bytes_loop_back_match_nolit_encodeSnappyBetterBlockAsm + +four_bytes_remain_match_nolit_encodeSnappyBetterBlockAsm: + TESTL R12, R12 + JZ match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm + MOVB $0x03, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVL R8, 1(AX) + ADDQ $0x05, AX + JMP match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm + +two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm + +two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm + +emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBetterBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBetterBlockAsm: + MOVQ $0x00cf1bbcdcbfa563, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x32, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 262168(SP)(R11*4) + MOVL R15, 262168(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 262168(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeSnappyBetterBlockAsm + +emit_remainder_encodeSnappyBetterBlockAsm: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 5(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBetterBlockAsm + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBetterBlockAsm: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBetterBlockAsm + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBetterBlockAsm + CMPL DX, $0x00010000 + JLT three_bytes_emit_remainder_encodeSnappyBetterBlockAsm + CMPL DX, $0x01000000 + JLT four_bytes_emit_remainder_encodeSnappyBetterBlockAsm + MOVB $0xfc, (AX) + MOVL DX, 1(AX) + ADDQ $0x05, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm + +four_bytes_emit_remainder_encodeSnappyBetterBlockAsm: + MOVL DX, BX + SHRL $0x10, BX + MOVB $0xf8, (AX) + MOVW DX, 1(AX) + MOVB BL, 3(AX) + ADDQ $0x04, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm + +three_bytes_emit_remainder_encodeSnappyBetterBlockAsm: + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm + +two_bytes_emit_remainder_encodeSnappyBetterBlockAsm: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBetterBlockAsm + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm + +one_byte_emit_remainder_encodeSnappyBetterBlockAsm: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBetterBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm + +memmove_long_emit_remainder_encodeSnappyBetterBlockAsm: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsmlarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsmlarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsmlarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsmlarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsmlarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBetterBlockAsm64K(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBetterBlockAsm64K(SB), $327704-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000a00, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBetterBlockAsm64K: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBetterBlockAsm64K + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBetterBlockAsm64K: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x07, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm64K + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x00cf1bbcdcbfa563, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x32, R11 + MOVL 24(SP)(R10*4), SI + MOVL 262168(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 262168(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm64K + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeSnappyBetterBlockAsm64K + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBetterBlockAsm64K + +candidateS_match_encodeSnappyBetterBlockAsm64K: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x08, R10 + IMULQ R9, R10 + SHRQ $0x30, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm64K + DECL CX + MOVL R8, SI + +candidate_match_encodeSnappyBetterBlockAsm64K: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm64K + +match_extend_back_loop_encodeSnappyBetterBlockAsm64K: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBetterBlockAsm64K + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBetterBlockAsm64K + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm64K + JMP match_extend_back_loop_encodeSnappyBetterBlockAsm64K + +match_extend_back_end_encodeSnappyBetterBlockAsm64K: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBetterBlockAsm64K + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBetterBlockAsm64K: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm64K + +matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm64K: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm64K + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeSnappyBetterBlockAsm64K + +matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm64K: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm64K + JZ match_nolit_end_encodeSnappyBetterBlockAsm64K + +matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm64K: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm64K + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm64K + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm64K: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm64K + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm64K + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm64K: + CMPL R8, $0x01 + JL match_nolit_end_encodeSnappyBetterBlockAsm64K + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeSnappyBetterBlockAsm64K + LEAL 1(R12), R12 + +match_nolit_end_encodeSnappyBetterBlockAsm64K: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeSnappyBetterBlockAsm64K + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeSnappyBetterBlockAsm64K + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBetterBlockAsm64K + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm64K + +two_bytes_match_emit_encodeSnappyBetterBlockAsm64K: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeSnappyBetterBlockAsm64K + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm64K + +one_byte_match_emit_encodeSnappyBetterBlockAsm64K: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBetterBlockAsm64K: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm64K_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm64K: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeSnappyBetterBlockAsm64K + +memmove_long_match_emit_encodeSnappyBetterBlockAsm64K: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm64Klarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm64Klarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm64Klarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm64Klarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm64Klarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeSnappyBetterBlockAsm64K: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm64K: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm64K + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm64K + +two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm64K: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm64K + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm64K + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm64K + +emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm64K: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm64K: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm64K + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBetterBlockAsm64K + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBetterBlockAsm64K: + MOVQ $0x00cf1bbcdcbfa563, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x32, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 262168(SP)(R11*4) + MOVL R15, 262168(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x08, R10 + IMULQ SI, R10 + SHRQ $0x30, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x32, R11 + SHLQ $0x08, R13 + IMULQ SI, R13 + SHRQ $0x30, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 262168(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeSnappyBetterBlockAsm64K + +emit_remainder_encodeSnappyBetterBlockAsm64K: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBetterBlockAsm64K + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBetterBlockAsm64K: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm64K + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBetterBlockAsm64K + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBetterBlockAsm64K + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64K + +two_bytes_emit_remainder_encodeSnappyBetterBlockAsm64K: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBetterBlockAsm64K + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64K + +one_byte_emit_remainder_encodeSnappyBetterBlockAsm64K: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBetterBlockAsm64K: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm64K + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm64K_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm64K: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm64K + +memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64K: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64Klarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64Klarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64Klarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64Klarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm64Klarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm64K: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBetterBlockAsm12B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBetterBlockAsm12B(SB), $81944-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000280, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBetterBlockAsm12B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBetterBlockAsm12B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBetterBlockAsm12B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x06, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm12B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x34, R11 + MOVL 24(SP)(R10*4), SI + MOVL 65560(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 65560(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm12B + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeSnappyBetterBlockAsm12B + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBetterBlockAsm12B + +candidateS_match_encodeSnappyBetterBlockAsm12B: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x32, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm12B + DECL CX + MOVL R8, SI + +candidate_match_encodeSnappyBetterBlockAsm12B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm12B + +match_extend_back_loop_encodeSnappyBetterBlockAsm12B: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBetterBlockAsm12B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBetterBlockAsm12B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm12B + JMP match_extend_back_loop_encodeSnappyBetterBlockAsm12B + +match_extend_back_end_encodeSnappyBetterBlockAsm12B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBetterBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBetterBlockAsm12B: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm12B + +matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm12B: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm12B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeSnappyBetterBlockAsm12B + +matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm12B: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm12B + JZ match_nolit_end_encodeSnappyBetterBlockAsm12B + +matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm12B: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm12B + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm12B + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm12B: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm12B + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm12B + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm12B: + CMPL R8, $0x01 + JL match_nolit_end_encodeSnappyBetterBlockAsm12B + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeSnappyBetterBlockAsm12B + LEAL 1(R12), R12 + +match_nolit_end_encodeSnappyBetterBlockAsm12B: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeSnappyBetterBlockAsm12B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeSnappyBetterBlockAsm12B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBetterBlockAsm12B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm12B + +two_bytes_match_emit_encodeSnappyBetterBlockAsm12B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeSnappyBetterBlockAsm12B + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm12B + +one_byte_match_emit_encodeSnappyBetterBlockAsm12B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBetterBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm12B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm12B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeSnappyBetterBlockAsm12B + +memmove_long_match_emit_encodeSnappyBetterBlockAsm12B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm12Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeSnappyBetterBlockAsm12B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm12B: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm12B + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm12B + +two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm12B: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm12B + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm12B + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm12B + +emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm12B: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm12B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm12B + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBetterBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBetterBlockAsm12B: + MOVQ $0x0000cf1bbcdcbf9b, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x32, R10 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x32, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x34, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x34, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 65560(SP)(R11*4) + MOVL R15, 65560(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x32, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x34, R11 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x32, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 65560(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeSnappyBetterBlockAsm12B + +emit_remainder_encodeSnappyBetterBlockAsm12B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBetterBlockAsm12B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBetterBlockAsm12B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm12B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBetterBlockAsm12B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBetterBlockAsm12B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12B + +two_bytes_emit_remainder_encodeSnappyBetterBlockAsm12B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBetterBlockAsm12B + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12B + +one_byte_emit_remainder_encodeSnappyBetterBlockAsm12B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBetterBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm12B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm12B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm12B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm12B + +memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm12Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm12B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBetterBlockAsm10B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBetterBlockAsm10B(SB), $20504-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x000000a0, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBetterBlockAsm10B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBetterBlockAsm10B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBetterBlockAsm10B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x05, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm10B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x36, R11 + MOVL 24(SP)(R10*4), SI + MOVL 16408(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 16408(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm10B + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeSnappyBetterBlockAsm10B + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBetterBlockAsm10B + +candidateS_match_encodeSnappyBetterBlockAsm10B: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x34, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm10B + DECL CX + MOVL R8, SI + +candidate_match_encodeSnappyBetterBlockAsm10B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm10B + +match_extend_back_loop_encodeSnappyBetterBlockAsm10B: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBetterBlockAsm10B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBetterBlockAsm10B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm10B + JMP match_extend_back_loop_encodeSnappyBetterBlockAsm10B + +match_extend_back_end_encodeSnappyBetterBlockAsm10B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBetterBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBetterBlockAsm10B: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm10B + +matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm10B: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm10B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeSnappyBetterBlockAsm10B + +matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm10B: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm10B + JZ match_nolit_end_encodeSnappyBetterBlockAsm10B + +matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm10B: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm10B + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm10B + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm10B: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm10B + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm10B + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm10B: + CMPL R8, $0x01 + JL match_nolit_end_encodeSnappyBetterBlockAsm10B + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeSnappyBetterBlockAsm10B + LEAL 1(R12), R12 + +match_nolit_end_encodeSnappyBetterBlockAsm10B: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeSnappyBetterBlockAsm10B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeSnappyBetterBlockAsm10B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBetterBlockAsm10B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm10B + +two_bytes_match_emit_encodeSnappyBetterBlockAsm10B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeSnappyBetterBlockAsm10B + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm10B + +one_byte_match_emit_encodeSnappyBetterBlockAsm10B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBetterBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm10B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm10B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeSnappyBetterBlockAsm10B + +memmove_long_match_emit_encodeSnappyBetterBlockAsm10B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm10Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeSnappyBetterBlockAsm10B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm10B: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm10B + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm10B + +two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm10B: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm10B + CMPL R8, $0x00000800 + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm10B + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm10B + +emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm10B: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm10B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm10B + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBetterBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBetterBlockAsm10B: + MOVQ $0x0000cf1bbcdcbf9b, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x34, R10 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x34, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x36, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x36, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 16408(SP)(R11*4) + MOVL R15, 16408(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x34, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x36, R11 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x34, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 16408(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeSnappyBetterBlockAsm10B + +emit_remainder_encodeSnappyBetterBlockAsm10B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBetterBlockAsm10B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBetterBlockAsm10B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm10B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBetterBlockAsm10B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBetterBlockAsm10B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10B + +two_bytes_emit_remainder_encodeSnappyBetterBlockAsm10B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBetterBlockAsm10B + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10B + +one_byte_emit_remainder_encodeSnappyBetterBlockAsm10B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBetterBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm10B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm10B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm10B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm10B + +memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm10Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm10B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func encodeSnappyBetterBlockAsm8B(dst []byte, src []byte) int +// Requires: BMI, SSE2 +TEXT ·encodeSnappyBetterBlockAsm8B(SB), $5144-56 + MOVQ dst_base+0(FP), AX + MOVQ $0x00000028, CX + LEAQ 24(SP), DX + PXOR X0, X0 + +zero_loop_encodeSnappyBetterBlockAsm8B: + MOVOU X0, (DX) + MOVOU X0, 16(DX) + MOVOU X0, 32(DX) + MOVOU X0, 48(DX) + MOVOU X0, 64(DX) + MOVOU X0, 80(DX) + MOVOU X0, 96(DX) + MOVOU X0, 112(DX) + ADDQ $0x80, DX + DECQ CX + JNZ zero_loop_encodeSnappyBetterBlockAsm8B + MOVL $0x00000000, 12(SP) + MOVQ src_len+32(FP), CX + LEAQ -9(CX), DX + LEAQ -8(CX), SI + MOVL SI, 8(SP) + SHRQ $0x05, CX + SUBL CX, DX + LEAQ (AX)(DX*1), DX + MOVQ DX, (SP) + MOVL $0x00000001, CX + MOVL $0x00000000, 16(SP) + MOVQ src_base+24(FP), DX + +search_loop_encodeSnappyBetterBlockAsm8B: + MOVL CX, SI + SUBL 12(SP), SI + SHRL $0x04, SI + LEAL 1(CX)(SI*1), SI + CMPL SI, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm8B + MOVQ (DX)(CX*1), DI + MOVL SI, 20(SP) + MOVQ $0x0000cf1bbcdcbf9b, R9 + MOVQ $0x9e3779b1, SI + MOVQ DI, R10 + MOVQ DI, R11 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + SHLQ $0x20, R11 + IMULQ SI, R11 + SHRQ $0x38, R11 + MOVL 24(SP)(R10*4), SI + MOVL 4120(SP)(R11*4), R8 + MOVL CX, 24(SP)(R10*4) + MOVL CX, 4120(SP)(R11*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm8B + CMPL (DX)(R8*1), DI + JEQ candidateS_match_encodeSnappyBetterBlockAsm8B + MOVL 20(SP), CX + JMP search_loop_encodeSnappyBetterBlockAsm8B + +candidateS_match_encodeSnappyBetterBlockAsm8B: + SHRQ $0x08, DI + MOVQ DI, R10 + SHLQ $0x10, R10 + IMULQ R9, R10 + SHRQ $0x36, R10 + MOVL 24(SP)(R10*4), SI + INCL CX + MOVL CX, 24(SP)(R10*4) + CMPL (DX)(SI*1), DI + JEQ candidate_match_encodeSnappyBetterBlockAsm8B + DECL CX + MOVL R8, SI + +candidate_match_encodeSnappyBetterBlockAsm8B: + MOVL 12(SP), DI + TESTL SI, SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm8B + +match_extend_back_loop_encodeSnappyBetterBlockAsm8B: + CMPL CX, DI + JLE match_extend_back_end_encodeSnappyBetterBlockAsm8B + MOVB -1(DX)(SI*1), BL + MOVB -1(DX)(CX*1), R8 + CMPB BL, R8 + JNE match_extend_back_end_encodeSnappyBetterBlockAsm8B + LEAL -1(CX), CX + DECL SI + JZ match_extend_back_end_encodeSnappyBetterBlockAsm8B + JMP match_extend_back_loop_encodeSnappyBetterBlockAsm8B + +match_extend_back_end_encodeSnappyBetterBlockAsm8B: + MOVL CX, DI + SUBL 12(SP), DI + LEAQ 3(AX)(DI*1), DI + CMPQ DI, (SP) + JL match_dst_size_check_encodeSnappyBetterBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_dst_size_check_encodeSnappyBetterBlockAsm8B: + MOVL CX, DI + ADDL $0x04, CX + ADDL $0x04, SI + MOVQ src_len+32(FP), R8 + SUBL CX, R8 + LEAQ (DX)(CX*1), R9 + LEAQ (DX)(SI*1), R10 + + // matchLen + XORL R12, R12 + CMPL R8, $0x08 + JL matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm8B + +matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm8B: + MOVQ (R9)(R12*1), R11 + XORQ (R10)(R12*1), R11 + TESTQ R11, R11 + JZ matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm8B + +#ifdef GOAMD64_v3 + TZCNTQ R11, R11 + +#else + BSFQ R11, R11 + +#endif + SARQ $0x03, R11 + LEAL (R12)(R11*1), R12 + JMP match_nolit_end_encodeSnappyBetterBlockAsm8B + +matchlen_loop_match_nolit_encodeSnappyBetterBlockAsm8B: + LEAL -8(R8), R8 + LEAL 8(R12), R12 + CMPL R8, $0x08 + JGE matchlen_loopback_match_nolit_encodeSnappyBetterBlockAsm8B + JZ match_nolit_end_encodeSnappyBetterBlockAsm8B + +matchlen_match4_match_nolit_encodeSnappyBetterBlockAsm8B: + CMPL R8, $0x04 + JL matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm8B + MOVL (R9)(R12*1), R11 + CMPL (R10)(R12*1), R11 + JNE matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm8B + SUBL $0x04, R8 + LEAL 4(R12), R12 + +matchlen_match2_match_nolit_encodeSnappyBetterBlockAsm8B: + CMPL R8, $0x02 + JL matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm8B + MOVW (R9)(R12*1), R11 + CMPW (R10)(R12*1), R11 + JNE matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm8B + SUBL $0x02, R8 + LEAL 2(R12), R12 + +matchlen_match1_match_nolit_encodeSnappyBetterBlockAsm8B: + CMPL R8, $0x01 + JL match_nolit_end_encodeSnappyBetterBlockAsm8B + MOVB (R9)(R12*1), R11 + CMPB (R10)(R12*1), R11 + JNE match_nolit_end_encodeSnappyBetterBlockAsm8B + LEAL 1(R12), R12 + +match_nolit_end_encodeSnappyBetterBlockAsm8B: + MOVL CX, R8 + SUBL SI, R8 + + // Check if repeat + MOVL R8, 16(SP) + MOVL 12(SP), SI + CMPL SI, DI + JEQ emit_literal_done_match_emit_encodeSnappyBetterBlockAsm8B + MOVL DI, R9 + MOVL DI, 12(SP) + LEAQ (DX)(SI*1), R10 + SUBL SI, R9 + LEAL -1(R9), SI + CMPL SI, $0x3c + JLT one_byte_match_emit_encodeSnappyBetterBlockAsm8B + CMPL SI, $0x00000100 + JLT two_bytes_match_emit_encodeSnappyBetterBlockAsm8B + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm8B + +two_bytes_match_emit_encodeSnappyBetterBlockAsm8B: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_match_emit_encodeSnappyBetterBlockAsm8B + JMP memmove_long_match_emit_encodeSnappyBetterBlockAsm8B + +one_byte_match_emit_encodeSnappyBetterBlockAsm8B: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, AX + +memmove_match_emit_encodeSnappyBetterBlockAsm8B: + LEAQ (AX)(R9*1), SI + + // genMemMoveShort + CMPQ R9, $0x08 + JLE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_8 + CMPQ R9, $0x10 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_8through16 + CMPQ R9, $0x20 + JBE emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_8: + MOVQ (R10), R11 + MOVQ R11, (AX) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_8through16: + MOVQ (R10), R11 + MOVQ -8(R10)(R9*1), R10 + MOVQ R11, (AX) + MOVQ R10, -8(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_17through32: + MOVOU (R10), X0 + MOVOU -16(R10)(R9*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(R9*1) + JMP memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_match_emit_encodeSnappyBetterBlockAsm8B_memmove_move_33through64: + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + +memmove_end_copy_match_emit_encodeSnappyBetterBlockAsm8B: + MOVQ SI, AX + JMP emit_literal_done_match_emit_encodeSnappyBetterBlockAsm8B + +memmove_long_match_emit_encodeSnappyBetterBlockAsm8B: + LEAQ (AX)(R9*1), SI + + // genMemMoveLong + MOVOU (R10), X0 + MOVOU 16(R10), X1 + MOVOU -32(R10)(R9*1), X2 + MOVOU -16(R10)(R9*1), X3 + MOVQ R9, R13 + SHRQ $0x05, R13 + MOVQ AX, R11 + ANDL $0x0000001f, R11 + MOVQ $0x00000040, R14 + SUBQ R11, R14 + DECQ R13 + JA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(R10)(R14*1), R11 + LEAQ -32(AX)(R14*1), R15 + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm8Blarge_big_loop_back: + MOVOU (R11), X4 + MOVOU 16(R11), X5 + MOVOA X4, (R15) + MOVOA X5, 16(R15) + ADDQ $0x20, R15 + ADDQ $0x20, R11 + ADDQ $0x20, R14 + DECQ R13 + JNA emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(R10)(R14*1), X4 + MOVOU -16(R10)(R14*1), X5 + MOVOA X4, -32(AX)(R14*1) + MOVOA X5, -16(AX)(R14*1) + ADDQ $0x20, R14 + CMPQ R9, R14 + JAE emit_lit_memmove_long_match_emit_encodeSnappyBetterBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(R9*1) + MOVOU X3, -16(AX)(R9*1) + MOVQ SI, AX + +emit_literal_done_match_emit_encodeSnappyBetterBlockAsm8B: + ADDL R12, CX + ADDL $0x04, R12 + MOVL CX, 12(SP) + + // emitCopy +two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm8B: + CMPL R12, $0x40 + JLE two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm8B + MOVB $0xee, (AX) + MOVW R8, 1(AX) + LEAL -60(R12), R12 + ADDQ $0x03, AX + JMP two_byte_offset_match_nolit_encodeSnappyBetterBlockAsm8B + +two_byte_offset_short_match_nolit_encodeSnappyBetterBlockAsm8B: + CMPL R12, $0x0c + JGE emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm8B + MOVB $0x01, BL + LEAL -16(BX)(R12*4), R12 + MOVB R8, 1(AX) + SHRL $0x08, R8 + SHLL $0x05, R8 + ORL R8, R12 + MOVB R12, (AX) + ADDQ $0x02, AX + JMP match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm8B + +emit_copy_three_match_nolit_encodeSnappyBetterBlockAsm8B: + MOVB $0x02, BL + LEAL -4(BX)(R12*4), R12 + MOVB R12, (AX) + MOVW R8, 1(AX) + ADDQ $0x03, AX + +match_nolit_emitcopy_end_encodeSnappyBetterBlockAsm8B: + CMPL CX, 8(SP) + JGE emit_remainder_encodeSnappyBetterBlockAsm8B + CMPQ AX, (SP) + JL match_nolit_dst_ok_encodeSnappyBetterBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +match_nolit_dst_ok_encodeSnappyBetterBlockAsm8B: + MOVQ $0x0000cf1bbcdcbf9b, SI + MOVQ $0x9e3779b1, R8 + INCL DI + MOVQ (DX)(DI*1), R9 + MOVQ R9, R10 + MOVQ R9, R11 + MOVQ R9, R12 + SHRQ $0x08, R11 + MOVQ R11, R13 + SHRQ $0x10, R12 + LEAL 1(DI), R14 + LEAL 2(DI), R15 + MOVQ -2(DX)(CX*1), R9 + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x36, R10 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x36, R13 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x38, R11 + SHLQ $0x20, R12 + IMULQ R8, R12 + SHRQ $0x38, R12 + MOVL DI, 24(SP)(R10*4) + MOVL R14, 24(SP)(R13*4) + MOVL R14, 4120(SP)(R11*4) + MOVL R15, 4120(SP)(R12*4) + MOVQ R9, R10 + MOVQ R9, R11 + SHRQ $0x08, R11 + MOVQ R11, R13 + LEAL -2(CX), R9 + LEAL -1(CX), DI + SHLQ $0x10, R10 + IMULQ SI, R10 + SHRQ $0x36, R10 + SHLQ $0x20, R11 + IMULQ R8, R11 + SHRQ $0x38, R11 + SHLQ $0x10, R13 + IMULQ SI, R13 + SHRQ $0x36, R13 + MOVL R9, 24(SP)(R10*4) + MOVL DI, 4120(SP)(R11*4) + MOVL DI, 24(SP)(R13*4) + JMP search_loop_encodeSnappyBetterBlockAsm8B + +emit_remainder_encodeSnappyBetterBlockAsm8B: + MOVQ src_len+32(FP), CX + SUBL 12(SP), CX + LEAQ 3(AX)(CX*1), CX + CMPQ CX, (SP) + JL emit_remainder_ok_encodeSnappyBetterBlockAsm8B + MOVQ $0x00000000, ret+48(FP) + RET + +emit_remainder_ok_encodeSnappyBetterBlockAsm8B: + MOVQ src_len+32(FP), CX + MOVL 12(SP), BX + CMPL BX, CX + JEQ emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm8B + MOVL CX, SI + MOVL CX, 12(SP) + LEAQ (DX)(BX*1), CX + SUBL BX, SI + LEAL -1(SI), DX + CMPL DX, $0x3c + JLT one_byte_emit_remainder_encodeSnappyBetterBlockAsm8B + CMPL DX, $0x00000100 + JLT two_bytes_emit_remainder_encodeSnappyBetterBlockAsm8B + MOVB $0xf4, (AX) + MOVW DX, 1(AX) + ADDQ $0x03, AX + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8B + +two_bytes_emit_remainder_encodeSnappyBetterBlockAsm8B: + MOVB $0xf0, (AX) + MOVB DL, 1(AX) + ADDQ $0x02, AX + CMPL DX, $0x40 + JL memmove_emit_remainder_encodeSnappyBetterBlockAsm8B + JMP memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8B + +one_byte_emit_remainder_encodeSnappyBetterBlockAsm8B: + SHLB $0x02, DL + MOVB DL, (AX) + ADDQ $0x01, AX + +memmove_emit_remainder_encodeSnappyBetterBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveShort + CMPQ BX, $0x03 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_1or2 + JE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_3 + CMPQ BX, $0x08 + JB emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_4through7 + CMPQ BX, $0x10 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_8through16 + CMPQ BX, $0x20 + JBE emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_17through32 + JMP emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_33through64 + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(BX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(BX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(BX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(BX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(BX*1) + JMP memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm8B + +emit_lit_memmove_emit_remainder_encodeSnappyBetterBlockAsm8B_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + +memmove_end_copy_emit_remainder_encodeSnappyBetterBlockAsm8B: + MOVQ DX, AX + JMP emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm8B + +memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8B: + LEAQ (AX)(SI*1), DX + MOVL SI, BX + + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(BX*1), X2 + MOVOU -16(CX)(BX*1), X3 + MOVQ BX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8Blarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8Blarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8Blarge_big_loop_back + +emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8Blarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ BX, R8 + JAE emit_lit_memmove_long_emit_remainder_encodeSnappyBetterBlockAsm8Blarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(BX*1) + MOVOU X3, -16(AX)(BX*1) + MOVQ DX, AX + +emit_literal_done_emit_remainder_encodeSnappyBetterBlockAsm8B: + MOVQ dst_base+0(FP), CX + SUBQ CX, AX + MOVQ AX, ret+48(FP) + RET + +// func emitLiteral(dst []byte, lit []byte) int +// Requires: SSE2 +TEXT ·emitLiteral(SB), NOSPLIT, $0-56 + MOVQ lit_len+32(FP), DX + MOVQ dst_base+0(FP), AX + MOVQ lit_base+24(FP), CX + TESTQ DX, DX + JZ emit_literal_end_standalone_skip + MOVL DX, BX + LEAL -1(DX), SI + CMPL SI, $0x3c + JLT one_byte_standalone + CMPL SI, $0x00000100 + JLT two_bytes_standalone + CMPL SI, $0x00010000 + JLT three_bytes_standalone + CMPL SI, $0x01000000 + JLT four_bytes_standalone + MOVB $0xfc, (AX) + MOVL SI, 1(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP memmove_long_standalone + +four_bytes_standalone: + MOVL SI, DI + SHRL $0x10, DI + MOVB $0xf8, (AX) + MOVW SI, 1(AX) + MOVB DI, 3(AX) + ADDQ $0x04, BX + ADDQ $0x04, AX + JMP memmove_long_standalone + +three_bytes_standalone: + MOVB $0xf4, (AX) + MOVW SI, 1(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + JMP memmove_long_standalone + +two_bytes_standalone: + MOVB $0xf0, (AX) + MOVB SI, 1(AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + CMPL SI, $0x40 + JL memmove_standalone + JMP memmove_long_standalone + +one_byte_standalone: + SHLB $0x02, SI + MOVB SI, (AX) + ADDQ $0x01, BX + ADDQ $0x01, AX + +memmove_standalone: + // genMemMoveShort + CMPQ DX, $0x03 + JB emit_lit_memmove_standalone_memmove_move_1or2 + JE emit_lit_memmove_standalone_memmove_move_3 + CMPQ DX, $0x08 + JB emit_lit_memmove_standalone_memmove_move_4through7 + CMPQ DX, $0x10 + JBE emit_lit_memmove_standalone_memmove_move_8through16 + CMPQ DX, $0x20 + JBE emit_lit_memmove_standalone_memmove_move_17through32 + JMP emit_lit_memmove_standalone_memmove_move_33through64 + +emit_lit_memmove_standalone_memmove_move_1or2: + MOVB (CX), SI + MOVB -1(CX)(DX*1), CL + MOVB SI, (AX) + MOVB CL, -1(AX)(DX*1) + JMP emit_literal_end_standalone + +emit_lit_memmove_standalone_memmove_move_3: + MOVW (CX), SI + MOVB 2(CX), CL + MOVW SI, (AX) + MOVB CL, 2(AX) + JMP emit_literal_end_standalone + +emit_lit_memmove_standalone_memmove_move_4through7: + MOVL (CX), SI + MOVL -4(CX)(DX*1), CX + MOVL SI, (AX) + MOVL CX, -4(AX)(DX*1) + JMP emit_literal_end_standalone + +emit_lit_memmove_standalone_memmove_move_8through16: + MOVQ (CX), SI + MOVQ -8(CX)(DX*1), CX + MOVQ SI, (AX) + MOVQ CX, -8(AX)(DX*1) + JMP emit_literal_end_standalone + +emit_lit_memmove_standalone_memmove_move_17through32: + MOVOU (CX), X0 + MOVOU -16(CX)(DX*1), X1 + MOVOU X0, (AX) + MOVOU X1, -16(AX)(DX*1) + JMP emit_literal_end_standalone + +emit_lit_memmove_standalone_memmove_move_33through64: + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(DX*1), X2 + MOVOU -16(CX)(DX*1), X3 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(DX*1) + MOVOU X3, -16(AX)(DX*1) + JMP emit_literal_end_standalone + JMP emit_literal_end_standalone + +memmove_long_standalone: + // genMemMoveLong + MOVOU (CX), X0 + MOVOU 16(CX), X1 + MOVOU -32(CX)(DX*1), X2 + MOVOU -16(CX)(DX*1), X3 + MOVQ DX, DI + SHRQ $0x05, DI + MOVQ AX, SI + ANDL $0x0000001f, SI + MOVQ $0x00000040, R8 + SUBQ SI, R8 + DECQ DI + JA emit_lit_memmove_long_standalonelarge_forward_sse_loop_32 + LEAQ -32(CX)(R8*1), SI + LEAQ -32(AX)(R8*1), R9 + +emit_lit_memmove_long_standalonelarge_big_loop_back: + MOVOU (SI), X4 + MOVOU 16(SI), X5 + MOVOA X4, (R9) + MOVOA X5, 16(R9) + ADDQ $0x20, R9 + ADDQ $0x20, SI + ADDQ $0x20, R8 + DECQ DI + JNA emit_lit_memmove_long_standalonelarge_big_loop_back + +emit_lit_memmove_long_standalonelarge_forward_sse_loop_32: + MOVOU -32(CX)(R8*1), X4 + MOVOU -16(CX)(R8*1), X5 + MOVOA X4, -32(AX)(R8*1) + MOVOA X5, -16(AX)(R8*1) + ADDQ $0x20, R8 + CMPQ DX, R8 + JAE emit_lit_memmove_long_standalonelarge_forward_sse_loop_32 + MOVOU X0, (AX) + MOVOU X1, 16(AX) + MOVOU X2, -32(AX)(DX*1) + MOVOU X3, -16(AX)(DX*1) + JMP emit_literal_end_standalone + JMP emit_literal_end_standalone + +emit_literal_end_standalone_skip: + XORQ BX, BX + +emit_literal_end_standalone: + MOVQ BX, ret+48(FP) + RET + +// func emitRepeat(dst []byte, offset int, length int) int +TEXT ·emitRepeat(SB), NOSPLIT, $0-48 + XORQ BX, BX + MOVQ dst_base+0(FP), AX + MOVQ offset+24(FP), CX + MOVQ length+32(FP), DX + + // emitRepeat +emit_repeat_again_standalone: + MOVL DX, SI + LEAL -4(DX), DX + CMPL SI, $0x08 + JLE repeat_two_standalone + CMPL SI, $0x0c + JGE cant_repeat_two_offset_standalone + CMPL CX, $0x00000800 + JLT repeat_two_offset_standalone + +cant_repeat_two_offset_standalone: + CMPL DX, $0x00000104 + JLT repeat_three_standalone + CMPL DX, $0x00010100 + JLT repeat_four_standalone + CMPL DX, $0x0100ffff + JLT repeat_five_standalone + LEAL -16842747(DX), DX + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + ADDQ $0x05, BX + JMP emit_repeat_again_standalone + +repeat_five_standalone: + LEAL -65536(DX), DX + MOVL DX, CX + MOVW $0x001d, (AX) + MOVW DX, 2(AX) + SARL $0x10, CX + MOVB CL, 4(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP gen_emit_repeat_end + +repeat_four_standalone: + LEAL -256(DX), DX + MOVW $0x0019, (AX) + MOVW DX, 2(AX) + ADDQ $0x04, BX + ADDQ $0x04, AX + JMP gen_emit_repeat_end + +repeat_three_standalone: + LEAL -4(DX), DX + MOVW $0x0015, (AX) + MOVB DL, 2(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + JMP gen_emit_repeat_end + +repeat_two_standalone: + SHLL $0x02, DX + ORL $0x01, DX + MOVW DX, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_repeat_end + +repeat_two_offset_standalone: + XORQ SI, SI + LEAL 1(SI)(DX*4), DX + MOVB CL, 1(AX) + SARL $0x08, CX + SHLL $0x05, CX + ORL CX, DX + MOVB DL, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + +gen_emit_repeat_end: + MOVQ BX, ret+40(FP) + RET + +// func emitCopy(dst []byte, offset int, length int) int +TEXT ·emitCopy(SB), NOSPLIT, $0-48 + XORQ BX, BX + MOVQ dst_base+0(FP), AX + MOVQ offset+24(FP), CX + MOVQ length+32(FP), DX + + // emitCopy + CMPL CX, $0x00010000 + JL two_byte_offset_standalone + +four_bytes_loop_back_standalone: + CMPL DX, $0x40 + JLE four_bytes_remain_standalone + MOVB $0xff, (AX) + MOVL CX, 1(AX) + LEAL -64(DX), DX + ADDQ $0x05, BX + ADDQ $0x05, AX + CMPL DX, $0x04 + JL four_bytes_remain_standalone + + // emitRepeat +emit_repeat_again_standalone_emit_copy: + MOVL DX, SI + LEAL -4(DX), DX + CMPL SI, $0x08 + JLE repeat_two_standalone_emit_copy + CMPL SI, $0x0c + JGE cant_repeat_two_offset_standalone_emit_copy + CMPL CX, $0x00000800 + JLT repeat_two_offset_standalone_emit_copy + +cant_repeat_two_offset_standalone_emit_copy: + CMPL DX, $0x00000104 + JLT repeat_three_standalone_emit_copy + CMPL DX, $0x00010100 + JLT repeat_four_standalone_emit_copy + CMPL DX, $0x0100ffff + JLT repeat_five_standalone_emit_copy + LEAL -16842747(DX), DX + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + ADDQ $0x05, BX + JMP emit_repeat_again_standalone_emit_copy + +repeat_five_standalone_emit_copy: + LEAL -65536(DX), DX + MOVL DX, CX + MOVW $0x001d, (AX) + MOVW DX, 2(AX) + SARL $0x10, CX + MOVB CL, 4(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP gen_emit_copy_end + +repeat_four_standalone_emit_copy: + LEAL -256(DX), DX + MOVW $0x0019, (AX) + MOVW DX, 2(AX) + ADDQ $0x04, BX + ADDQ $0x04, AX + JMP gen_emit_copy_end + +repeat_three_standalone_emit_copy: + LEAL -4(DX), DX + MOVW $0x0015, (AX) + MOVB DL, 2(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + JMP gen_emit_copy_end + +repeat_two_standalone_emit_copy: + SHLL $0x02, DX + ORL $0x01, DX + MOVW DX, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + +repeat_two_offset_standalone_emit_copy: + XORQ SI, SI + LEAL 1(SI)(DX*4), DX + MOVB CL, 1(AX) + SARL $0x08, CX + SHLL $0x05, CX + ORL CX, DX + MOVB DL, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + JMP four_bytes_loop_back_standalone + +four_bytes_remain_standalone: + TESTL DX, DX + JZ gen_emit_copy_end + MOVB $0x03, SI + LEAL -4(SI)(DX*4), DX + MOVB DL, (AX) + MOVL CX, 1(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP gen_emit_copy_end + +two_byte_offset_standalone: + CMPL DX, $0x40 + JLE two_byte_offset_short_standalone + CMPL CX, $0x00000800 + JAE long_offset_short_standalone + MOVL $0x00000001, SI + LEAL 16(SI), SI + MOVB CL, 1(AX) + MOVL CX, DI + SHRL $0x08, DI + SHLL $0x05, DI + ORL DI, SI + MOVB SI, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + SUBL $0x08, DX + + // emitRepeat + LEAL -4(DX), DX + JMP cant_repeat_two_offset_standalone_emit_copy_short_2b + +emit_repeat_again_standalone_emit_copy_short_2b: + MOVL DX, SI + LEAL -4(DX), DX + CMPL SI, $0x08 + JLE repeat_two_standalone_emit_copy_short_2b + CMPL SI, $0x0c + JGE cant_repeat_two_offset_standalone_emit_copy_short_2b + CMPL CX, $0x00000800 + JLT repeat_two_offset_standalone_emit_copy_short_2b + +cant_repeat_two_offset_standalone_emit_copy_short_2b: + CMPL DX, $0x00000104 + JLT repeat_three_standalone_emit_copy_short_2b + CMPL DX, $0x00010100 + JLT repeat_four_standalone_emit_copy_short_2b + CMPL DX, $0x0100ffff + JLT repeat_five_standalone_emit_copy_short_2b + LEAL -16842747(DX), DX + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + ADDQ $0x05, BX + JMP emit_repeat_again_standalone_emit_copy_short_2b + +repeat_five_standalone_emit_copy_short_2b: + LEAL -65536(DX), DX + MOVL DX, CX + MOVW $0x001d, (AX) + MOVW DX, 2(AX) + SARL $0x10, CX + MOVB CL, 4(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP gen_emit_copy_end + +repeat_four_standalone_emit_copy_short_2b: + LEAL -256(DX), DX + MOVW $0x0019, (AX) + MOVW DX, 2(AX) + ADDQ $0x04, BX + ADDQ $0x04, AX + JMP gen_emit_copy_end + +repeat_three_standalone_emit_copy_short_2b: + LEAL -4(DX), DX + MOVW $0x0015, (AX) + MOVB DL, 2(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + JMP gen_emit_copy_end + +repeat_two_standalone_emit_copy_short_2b: + SHLL $0x02, DX + ORL $0x01, DX + MOVW DX, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + +repeat_two_offset_standalone_emit_copy_short_2b: + XORQ SI, SI + LEAL 1(SI)(DX*4), DX + MOVB CL, 1(AX) + SARL $0x08, CX + SHLL $0x05, CX + ORL CX, DX + MOVB DL, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + +long_offset_short_standalone: + MOVB $0xee, (AX) + MOVW CX, 1(AX) + LEAL -60(DX), DX + ADDQ $0x03, AX + ADDQ $0x03, BX + + // emitRepeat +emit_repeat_again_standalone_emit_copy_short: + MOVL DX, SI + LEAL -4(DX), DX + CMPL SI, $0x08 + JLE repeat_two_standalone_emit_copy_short + CMPL SI, $0x0c + JGE cant_repeat_two_offset_standalone_emit_copy_short + CMPL CX, $0x00000800 + JLT repeat_two_offset_standalone_emit_copy_short + +cant_repeat_two_offset_standalone_emit_copy_short: + CMPL DX, $0x00000104 + JLT repeat_three_standalone_emit_copy_short + CMPL DX, $0x00010100 + JLT repeat_four_standalone_emit_copy_short + CMPL DX, $0x0100ffff + JLT repeat_five_standalone_emit_copy_short + LEAL -16842747(DX), DX + MOVW $0x001d, (AX) + MOVW $0xfffb, 2(AX) + MOVB $0xff, 4(AX) + ADDQ $0x05, AX + ADDQ $0x05, BX + JMP emit_repeat_again_standalone_emit_copy_short + +repeat_five_standalone_emit_copy_short: + LEAL -65536(DX), DX + MOVL DX, CX + MOVW $0x001d, (AX) + MOVW DX, 2(AX) + SARL $0x10, CX + MOVB CL, 4(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP gen_emit_copy_end + +repeat_four_standalone_emit_copy_short: + LEAL -256(DX), DX + MOVW $0x0019, (AX) + MOVW DX, 2(AX) + ADDQ $0x04, BX + ADDQ $0x04, AX + JMP gen_emit_copy_end + +repeat_three_standalone_emit_copy_short: + LEAL -4(DX), DX + MOVW $0x0015, (AX) + MOVB DL, 2(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + JMP gen_emit_copy_end + +repeat_two_standalone_emit_copy_short: + SHLL $0x02, DX + ORL $0x01, DX + MOVW DX, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + +repeat_two_offset_standalone_emit_copy_short: + XORQ SI, SI + LEAL 1(SI)(DX*4), DX + MOVB CL, 1(AX) + SARL $0x08, CX + SHLL $0x05, CX + ORL CX, DX + MOVB DL, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + JMP two_byte_offset_standalone + +two_byte_offset_short_standalone: + CMPL DX, $0x0c + JGE emit_copy_three_standalone + CMPL CX, $0x00000800 + JGE emit_copy_three_standalone + MOVB $0x01, SI + LEAL -16(SI)(DX*4), DX + MOVB CL, 1(AX) + SHRL $0x08, CX + SHLL $0x05, CX + ORL CX, DX + MOVB DL, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end + +emit_copy_three_standalone: + MOVB $0x02, SI + LEAL -4(SI)(DX*4), DX + MOVB DL, (AX) + MOVW CX, 1(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + +gen_emit_copy_end: + MOVQ BX, ret+40(FP) + RET + +// func emitCopyNoRepeat(dst []byte, offset int, length int) int +TEXT ·emitCopyNoRepeat(SB), NOSPLIT, $0-48 + XORQ BX, BX + MOVQ dst_base+0(FP), AX + MOVQ offset+24(FP), CX + MOVQ length+32(FP), DX + + // emitCopy + CMPL CX, $0x00010000 + JL two_byte_offset_standalone_snappy + +four_bytes_loop_back_standalone_snappy: + CMPL DX, $0x40 + JLE four_bytes_remain_standalone_snappy + MOVB $0xff, (AX) + MOVL CX, 1(AX) + LEAL -64(DX), DX + ADDQ $0x05, BX + ADDQ $0x05, AX + CMPL DX, $0x04 + JL four_bytes_remain_standalone_snappy + JMP four_bytes_loop_back_standalone_snappy + +four_bytes_remain_standalone_snappy: + TESTL DX, DX + JZ gen_emit_copy_end_snappy + MOVB $0x03, SI + LEAL -4(SI)(DX*4), DX + MOVB DL, (AX) + MOVL CX, 1(AX) + ADDQ $0x05, BX + ADDQ $0x05, AX + JMP gen_emit_copy_end_snappy + +two_byte_offset_standalone_snappy: + CMPL DX, $0x40 + JLE two_byte_offset_short_standalone_snappy + MOVB $0xee, (AX) + MOVW CX, 1(AX) + LEAL -60(DX), DX + ADDQ $0x03, AX + ADDQ $0x03, BX + JMP two_byte_offset_standalone_snappy + +two_byte_offset_short_standalone_snappy: + CMPL DX, $0x0c + JGE emit_copy_three_standalone_snappy + CMPL CX, $0x00000800 + JGE emit_copy_three_standalone_snappy + MOVB $0x01, SI + LEAL -16(SI)(DX*4), DX + MOVB CL, 1(AX) + SHRL $0x08, CX + SHLL $0x05, CX + ORL CX, DX + MOVB DL, (AX) + ADDQ $0x02, BX + ADDQ $0x02, AX + JMP gen_emit_copy_end_snappy + +emit_copy_three_standalone_snappy: + MOVB $0x02, SI + LEAL -4(SI)(DX*4), DX + MOVB DL, (AX) + MOVW CX, 1(AX) + ADDQ $0x03, BX + ADDQ $0x03, AX + +gen_emit_copy_end_snappy: + MOVQ BX, ret+40(FP) + RET + +// func matchLen(a []byte, b []byte) int +// Requires: BMI +TEXT ·matchLen(SB), NOSPLIT, $0-56 + MOVQ a_base+0(FP), AX + MOVQ b_base+24(FP), CX + MOVQ a_len+8(FP), DX + + // matchLen + XORL SI, SI + CMPL DX, $0x08 + JL matchlen_match4_standalone + +matchlen_loopback_standalone: + MOVQ (AX)(SI*1), BX + XORQ (CX)(SI*1), BX + TESTQ BX, BX + JZ matchlen_loop_standalone + +#ifdef GOAMD64_v3 + TZCNTQ BX, BX + +#else + BSFQ BX, BX + +#endif + SARQ $0x03, BX + LEAL (SI)(BX*1), SI + JMP gen_match_len_end + +matchlen_loop_standalone: + LEAL -8(DX), DX + LEAL 8(SI), SI + CMPL DX, $0x08 + JGE matchlen_loopback_standalone + JZ gen_match_len_end + +matchlen_match4_standalone: + CMPL DX, $0x04 + JL matchlen_match2_standalone + MOVL (AX)(SI*1), BX + CMPL (CX)(SI*1), BX + JNE matchlen_match2_standalone + SUBL $0x04, DX + LEAL 4(SI), SI + +matchlen_match2_standalone: + CMPL DX, $0x02 + JL matchlen_match1_standalone + MOVW (AX)(SI*1), BX + CMPW (CX)(SI*1), BX + JNE matchlen_match1_standalone + SUBL $0x02, DX + LEAL 2(SI), SI + +matchlen_match1_standalone: + CMPL DX, $0x01 + JL gen_match_len_end + MOVB (AX)(SI*1), BL + CMPB (CX)(SI*1), BL + JNE gen_match_len_end + LEAL 1(SI), SI + +gen_match_len_end: + MOVQ SI, ret+48(FP) + RET diff --git a/vendor/github.com/klauspost/compress/s2/index.go b/vendor/github.com/klauspost/compress/s2/index.go new file mode 100644 index 0000000..dd9ecfe --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/index.go @@ -0,0 +1,598 @@ +// Copyright (c) 2022+ Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package s2 + +import ( + "bytes" + "encoding/binary" + "encoding/json" + "fmt" + "io" + "sort" +) + +const ( + S2IndexHeader = "s2idx\x00" + S2IndexTrailer = "\x00xdi2s" + maxIndexEntries = 1 << 16 +) + +// Index represents an S2/Snappy index. +type Index struct { + TotalUncompressed int64 // Total Uncompressed size if known. Will be -1 if unknown. + TotalCompressed int64 // Total Compressed size if known. Will be -1 if unknown. + info []struct { + compressedOffset int64 + uncompressedOffset int64 + } + estBlockUncomp int64 +} + +func (i *Index) reset(maxBlock int) { + i.estBlockUncomp = int64(maxBlock) + i.TotalCompressed = -1 + i.TotalUncompressed = -1 + if len(i.info) > 0 { + i.info = i.info[:0] + } +} + +// allocInfos will allocate an empty slice of infos. +func (i *Index) allocInfos(n int) { + if n > maxIndexEntries { + panic("n > maxIndexEntries") + } + i.info = make([]struct { + compressedOffset int64 + uncompressedOffset int64 + }, 0, n) +} + +// add an uncompressed and compressed pair. +// Entries must be sent in order. +func (i *Index) add(compressedOffset, uncompressedOffset int64) error { + if i == nil { + return nil + } + lastIdx := len(i.info) - 1 + if lastIdx >= 0 { + latest := i.info[lastIdx] + if latest.uncompressedOffset == uncompressedOffset { + // Uncompressed didn't change, don't add entry, + // but update start index. + latest.compressedOffset = compressedOffset + i.info[lastIdx] = latest + return nil + } + if latest.uncompressedOffset > uncompressedOffset { + return fmt.Errorf("internal error: Earlier uncompressed received (%d > %d)", latest.uncompressedOffset, uncompressedOffset) + } + if latest.compressedOffset > compressedOffset { + return fmt.Errorf("internal error: Earlier compressed received (%d > %d)", latest.uncompressedOffset, uncompressedOffset) + } + } + i.info = append(i.info, struct { + compressedOffset int64 + uncompressedOffset int64 + }{compressedOffset: compressedOffset, uncompressedOffset: uncompressedOffset}) + return nil +} + +// Find the offset at or before the wanted (uncompressed) offset. +// If offset is 0 or positive it is the offset from the beginning of the file. +// If the uncompressed size is known, the offset must be within the file. +// If an offset outside the file is requested io.ErrUnexpectedEOF is returned. +// If the offset is negative, it is interpreted as the distance from the end of the file, +// where -1 represents the last byte. +// If offset from the end of the file is requested, but size is unknown, +// ErrUnsupported will be returned. +func (i *Index) Find(offset int64) (compressedOff, uncompressedOff int64, err error) { + if i.TotalUncompressed < 0 { + return 0, 0, ErrCorrupt + } + if offset < 0 { + offset = i.TotalUncompressed + offset + if offset < 0 { + return 0, 0, io.ErrUnexpectedEOF + } + } + if offset > i.TotalUncompressed { + return 0, 0, io.ErrUnexpectedEOF + } + if len(i.info) > 200 { + n := sort.Search(len(i.info), func(n int) bool { + return i.info[n].uncompressedOffset > offset + }) + if n == 0 { + n = 1 + } + return i.info[n-1].compressedOffset, i.info[n-1].uncompressedOffset, nil + } + for _, info := range i.info { + if info.uncompressedOffset > offset { + break + } + compressedOff = info.compressedOffset + uncompressedOff = info.uncompressedOffset + } + return compressedOff, uncompressedOff, nil +} + +// reduce to stay below maxIndexEntries +func (i *Index) reduce() { + if len(i.info) < maxIndexEntries && i.estBlockUncomp >= 1<<20 { + return + } + + // Algorithm, keep 1, remove removeN entries... + removeN := (len(i.info) + 1) / maxIndexEntries + src := i.info + j := 0 + + // Each block should be at least 1MB, but don't reduce below 1000 entries. + for i.estBlockUncomp*(int64(removeN)+1) < 1<<20 && len(i.info)/(removeN+1) > 1000 { + removeN++ + } + for idx := 0; idx < len(src); idx++ { + i.info[j] = src[idx] + j++ + idx += removeN + } + i.info = i.info[:j] + // Update maxblock estimate. + i.estBlockUncomp += i.estBlockUncomp * int64(removeN) +} + +func (i *Index) appendTo(b []byte, uncompTotal, compTotal int64) []byte { + i.reduce() + var tmp [binary.MaxVarintLen64]byte + + initSize := len(b) + // We make the start a skippable header+size. + b = append(b, ChunkTypeIndex, 0, 0, 0) + b = append(b, []byte(S2IndexHeader)...) + // Total Uncompressed size + n := binary.PutVarint(tmp[:], uncompTotal) + b = append(b, tmp[:n]...) + // Total Compressed size + n = binary.PutVarint(tmp[:], compTotal) + b = append(b, tmp[:n]...) + // Put EstBlockUncomp size + n = binary.PutVarint(tmp[:], i.estBlockUncomp) + b = append(b, tmp[:n]...) + // Put length + n = binary.PutVarint(tmp[:], int64(len(i.info))) + b = append(b, tmp[:n]...) + + // Check if we should add uncompressed offsets + var hasUncompressed byte + for idx, info := range i.info { + if idx == 0 { + if info.uncompressedOffset != 0 { + hasUncompressed = 1 + break + } + continue + } + if info.uncompressedOffset != i.info[idx-1].uncompressedOffset+i.estBlockUncomp { + hasUncompressed = 1 + break + } + } + b = append(b, hasUncompressed) + + // Add each entry + if hasUncompressed == 1 { + for idx, info := range i.info { + uOff := info.uncompressedOffset + if idx > 0 { + prev := i.info[idx-1] + uOff -= prev.uncompressedOffset + (i.estBlockUncomp) + } + n = binary.PutVarint(tmp[:], uOff) + b = append(b, tmp[:n]...) + } + } + + // Initial compressed size estimate. + cPredict := i.estBlockUncomp / 2 + + for idx, info := range i.info { + cOff := info.compressedOffset + if idx > 0 { + prev := i.info[idx-1] + cOff -= prev.compressedOffset + cPredict + // Update compressed size prediction, with half the error. + cPredict += cOff / 2 + } + n = binary.PutVarint(tmp[:], cOff) + b = append(b, tmp[:n]...) + } + + // Add Total Size. + // Stored as fixed size for easier reading. + binary.LittleEndian.PutUint32(tmp[:], uint32(len(b)-initSize+4+len(S2IndexTrailer))) + b = append(b, tmp[:4]...) + // Trailer + b = append(b, []byte(S2IndexTrailer)...) + + // Update size + chunkLen := len(b) - initSize - skippableFrameHeader + b[initSize+1] = uint8(chunkLen >> 0) + b[initSize+2] = uint8(chunkLen >> 8) + b[initSize+3] = uint8(chunkLen >> 16) + //fmt.Printf("chunklen: 0x%x Uncomp:%d, Comp:%d\n", chunkLen, uncompTotal, compTotal) + return b +} + +// Load a binary index. +// A zero value Index can be used or a previous one can be reused. +func (i *Index) Load(b []byte) ([]byte, error) { + if len(b) <= 4+len(S2IndexHeader)+len(S2IndexTrailer) { + return b, io.ErrUnexpectedEOF + } + if b[0] != ChunkTypeIndex { + return b, ErrCorrupt + } + chunkLen := int(b[1]) | int(b[2])<<8 | int(b[3])<<16 + b = b[4:] + + // Validate we have enough... + if len(b) < chunkLen { + return b, io.ErrUnexpectedEOF + } + if !bytes.Equal(b[:len(S2IndexHeader)], []byte(S2IndexHeader)) { + return b, ErrUnsupported + } + b = b[len(S2IndexHeader):] + + // Total Uncompressed + if v, n := binary.Varint(b); n <= 0 || v < 0 { + return b, ErrCorrupt + } else { + i.TotalUncompressed = v + b = b[n:] + } + + // Total Compressed + if v, n := binary.Varint(b); n <= 0 { + return b, ErrCorrupt + } else { + i.TotalCompressed = v + b = b[n:] + } + + // Read EstBlockUncomp + if v, n := binary.Varint(b); n <= 0 { + return b, ErrCorrupt + } else { + if v < 0 { + return b, ErrCorrupt + } + i.estBlockUncomp = v + b = b[n:] + } + + var entries int + if v, n := binary.Varint(b); n <= 0 { + return b, ErrCorrupt + } else { + if v < 0 || v > maxIndexEntries { + return b, ErrCorrupt + } + entries = int(v) + b = b[n:] + } + if cap(i.info) < entries { + i.allocInfos(entries) + } + i.info = i.info[:entries] + + if len(b) < 1 { + return b, io.ErrUnexpectedEOF + } + hasUncompressed := b[0] + b = b[1:] + if hasUncompressed&1 != hasUncompressed { + return b, ErrCorrupt + } + + // Add each uncompressed entry + for idx := range i.info { + var uOff int64 + if hasUncompressed != 0 { + // Load delta + if v, n := binary.Varint(b); n <= 0 { + return b, ErrCorrupt + } else { + uOff = v + b = b[n:] + } + } + + if idx > 0 { + prev := i.info[idx-1].uncompressedOffset + uOff += prev + (i.estBlockUncomp) + if uOff <= prev { + return b, ErrCorrupt + } + } + if uOff < 0 { + return b, ErrCorrupt + } + i.info[idx].uncompressedOffset = uOff + } + + // Initial compressed size estimate. + cPredict := i.estBlockUncomp / 2 + + // Add each compressed entry + for idx := range i.info { + var cOff int64 + if v, n := binary.Varint(b); n <= 0 { + return b, ErrCorrupt + } else { + cOff = v + b = b[n:] + } + + if idx > 0 { + // Update compressed size prediction, with half the error. + cPredictNew := cPredict + cOff/2 + + prev := i.info[idx-1].compressedOffset + cOff += prev + cPredict + if cOff <= prev { + return b, ErrCorrupt + } + cPredict = cPredictNew + } + if cOff < 0 { + return b, ErrCorrupt + } + i.info[idx].compressedOffset = cOff + } + if len(b) < 4+len(S2IndexTrailer) { + return b, io.ErrUnexpectedEOF + } + // Skip size... + b = b[4:] + + // Check trailer... + if !bytes.Equal(b[:len(S2IndexTrailer)], []byte(S2IndexTrailer)) { + return b, ErrCorrupt + } + return b[len(S2IndexTrailer):], nil +} + +// LoadStream will load an index from the end of the supplied stream. +// ErrUnsupported will be returned if the signature cannot be found. +// ErrCorrupt will be returned if unexpected values are found. +// io.ErrUnexpectedEOF is returned if there are too few bytes. +// IO errors are returned as-is. +func (i *Index) LoadStream(rs io.ReadSeeker) error { + // Go to end. + _, err := rs.Seek(-10, io.SeekEnd) + if err != nil { + return err + } + var tmp [10]byte + _, err = io.ReadFull(rs, tmp[:]) + if err != nil { + return err + } + // Check trailer... + if !bytes.Equal(tmp[4:4+len(S2IndexTrailer)], []byte(S2IndexTrailer)) { + return ErrUnsupported + } + sz := binary.LittleEndian.Uint32(tmp[:4]) + if sz > maxChunkSize+skippableFrameHeader { + return ErrCorrupt + } + _, err = rs.Seek(-int64(sz), io.SeekEnd) + if err != nil { + return err + } + + // Read index. + buf := make([]byte, sz) + _, err = io.ReadFull(rs, buf) + if err != nil { + return err + } + _, err = i.Load(buf) + return err +} + +// IndexStream will return an index for a stream. +// The stream structure will be checked, but +// data within blocks is not verified. +// The returned index can either be appended to the end of the stream +// or stored separately. +func IndexStream(r io.Reader) ([]byte, error) { + var i Index + var buf [maxChunkSize]byte + var readHeader bool + for { + _, err := io.ReadFull(r, buf[:4]) + if err != nil { + if err == io.EOF { + return i.appendTo(nil, i.TotalUncompressed, i.TotalCompressed), nil + } + return nil, err + } + // Start of this chunk. + startChunk := i.TotalCompressed + i.TotalCompressed += 4 + + chunkType := buf[0] + if !readHeader { + if chunkType != chunkTypeStreamIdentifier { + return nil, ErrCorrupt + } + readHeader = true + } + chunkLen := int(buf[1]) | int(buf[2])<<8 | int(buf[3])<<16 + if chunkLen < checksumSize { + return nil, ErrCorrupt + } + + i.TotalCompressed += int64(chunkLen) + _, err = io.ReadFull(r, buf[:chunkLen]) + if err != nil { + return nil, io.ErrUnexpectedEOF + } + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + // Section 4.2. Compressed data (chunk type 0x00). + // Skip checksum. + dLen, err := DecodedLen(buf[checksumSize:]) + if err != nil { + return nil, err + } + if dLen > maxBlockSize { + return nil, ErrCorrupt + } + if i.estBlockUncomp == 0 { + // Use first block for estimate... + i.estBlockUncomp = int64(dLen) + } + err = i.add(startChunk, i.TotalUncompressed) + if err != nil { + return nil, err + } + i.TotalUncompressed += int64(dLen) + continue + case chunkTypeUncompressedData: + n2 := chunkLen - checksumSize + if n2 > maxBlockSize { + return nil, ErrCorrupt + } + if i.estBlockUncomp == 0 { + // Use first block for estimate... + i.estBlockUncomp = int64(n2) + } + err = i.add(startChunk, i.TotalUncompressed) + if err != nil { + return nil, err + } + i.TotalUncompressed += int64(n2) + continue + case chunkTypeStreamIdentifier: + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(magicBody) { + return nil, ErrCorrupt + } + + if string(buf[:len(magicBody)]) != magicBody { + if string(buf[:len(magicBody)]) != magicBodySnappy { + return nil, ErrCorrupt + } + } + + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + return nil, ErrUnsupported + } + if chunkLen > maxChunkSize { + return nil, ErrUnsupported + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + } +} + +// JSON returns the index as JSON text. +func (i *Index) JSON() []byte { + x := struct { + TotalUncompressed int64 `json:"total_uncompressed"` // Total Uncompressed size if known. Will be -1 if unknown. + TotalCompressed int64 `json:"total_compressed"` // Total Compressed size if known. Will be -1 if unknown. + Offsets []struct { + CompressedOffset int64 `json:"compressed"` + UncompressedOffset int64 `json:"uncompressed"` + } `json:"offsets"` + EstBlockUncomp int64 `json:"est_block_uncompressed"` + }{ + TotalUncompressed: i.TotalUncompressed, + TotalCompressed: i.TotalCompressed, + EstBlockUncomp: i.estBlockUncomp, + } + for _, v := range i.info { + x.Offsets = append(x.Offsets, struct { + CompressedOffset int64 `json:"compressed"` + UncompressedOffset int64 `json:"uncompressed"` + }{CompressedOffset: v.compressedOffset, UncompressedOffset: v.uncompressedOffset}) + } + b, _ := json.MarshalIndent(x, "", " ") + return b +} + +// RemoveIndexHeaders will trim all headers and trailers from a given index. +// This is expected to save 20 bytes. +// These can be restored using RestoreIndexHeaders. +// This removes a layer of security, but is the most compact representation. +// Returns nil if headers contains errors. +// The returned slice references the provided slice. +func RemoveIndexHeaders(b []byte) []byte { + const save = 4 + len(S2IndexHeader) + len(S2IndexTrailer) + 4 + if len(b) <= save { + return nil + } + if b[0] != ChunkTypeIndex { + return nil + } + chunkLen := int(b[1]) | int(b[2])<<8 | int(b[3])<<16 + b = b[4:] + + // Validate we have enough... + if len(b) < chunkLen { + return nil + } + b = b[:chunkLen] + + if !bytes.Equal(b[:len(S2IndexHeader)], []byte(S2IndexHeader)) { + return nil + } + b = b[len(S2IndexHeader):] + if !bytes.HasSuffix(b, []byte(S2IndexTrailer)) { + return nil + } + b = bytes.TrimSuffix(b, []byte(S2IndexTrailer)) + + if len(b) < 4 { + return nil + } + return b[:len(b)-4] +} + +// RestoreIndexHeaders will index restore headers removed by RemoveIndexHeaders. +// No error checking is performed on the input. +// If a 0 length slice is sent, it is returned without modification. +func RestoreIndexHeaders(in []byte) []byte { + if len(in) == 0 { + return in + } + b := make([]byte, 0, 4+len(S2IndexHeader)+len(in)+len(S2IndexTrailer)+4) + b = append(b, ChunkTypeIndex, 0, 0, 0) + b = append(b, []byte(S2IndexHeader)...) + b = append(b, in...) + + var tmp [4]byte + binary.LittleEndian.PutUint32(tmp[:], uint32(len(b)+4+len(S2IndexTrailer))) + b = append(b, tmp[:4]...) + // Trailer + b = append(b, []byte(S2IndexTrailer)...) + + chunkLen := len(b) - skippableFrameHeader + b[1] = uint8(chunkLen >> 0) + b[2] = uint8(chunkLen >> 8) + b[3] = uint8(chunkLen >> 16) + return b +} diff --git a/vendor/github.com/klauspost/compress/s2/s2.go b/vendor/github.com/klauspost/compress/s2/s2.go new file mode 100644 index 0000000..dae3f73 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2/s2.go @@ -0,0 +1,143 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package s2 implements the S2 compression format. +// +// S2 is an extension of Snappy. Similar to Snappy S2 is aimed for high throughput, +// which is why it features concurrent compression for bigger payloads. +// +// Decoding is compatible with Snappy compressed content, +// but content compressed with S2 cannot be decompressed by Snappy. +// +// For more information on Snappy/S2 differences see README in: https://github.com/klauspost/compress/tree/master/s2 +// +// There are actually two S2 formats: block and stream. They are related, +// but different: trying to decompress block-compressed data as a S2 stream +// will fail, and vice versa. The block format is the Decode and Encode +// functions and the stream format is the Reader and Writer types. +// +// A "better" compression option is available. This will trade some compression +// speed +// +// The block format, the more common case, is used when the complete size (the +// number of bytes) of the original data is known upfront, at the time +// compression starts. The stream format, also known as the framing format, is +// for when that isn't always true. +// +// Blocks to not offer much data protection, so it is up to you to +// add data validation of decompressed blocks. +// +// Streams perform CRC validation of the decompressed data. +// Stream compression will also be performed on multiple CPU cores concurrently +// significantly improving throughput. +package s2 + +import ( + "bytes" + "hash/crc32" +) + +/* +Each encoded block begins with the varint-encoded length of the decoded data, +followed by a sequence of chunks. Chunks begin and end on byte boundaries. The +first byte of each chunk is broken into its 2 least and 6 most significant bits +called l and m: l ranges in [0, 4) and m ranges in [0, 64). l is the chunk tag. +Zero means a literal tag. All other values mean a copy tag. + +For literal tags: + - If m < 60, the next 1 + m bytes are literal bytes. + - Otherwise, let n be the little-endian unsigned integer denoted by the next + m - 59 bytes. The next 1 + n bytes after that are literal bytes. + +For copy tags, length bytes are copied from offset bytes ago, in the style of +Lempel-Ziv compression algorithms. In particular: + - For l == 1, the offset ranges in [0, 1<<11) and the length in [4, 12). + The length is 4 + the low 3 bits of m. The high 3 bits of m form bits 8-10 + of the offset. The next byte is bits 0-7 of the offset. + - For l == 2, the offset ranges in [0, 1<<16) and the length in [1, 65). + The length is 1 + m. The offset is the little-endian unsigned integer + denoted by the next 2 bytes. + - For l == 3, the offset ranges in [0, 1<<32) and the length in + [1, 65). The length is 1 + m. The offset is the little-endian unsigned + integer denoted by the next 4 bytes. +*/ +const ( + tagLiteral = 0x00 + tagCopy1 = 0x01 + tagCopy2 = 0x02 + tagCopy4 = 0x03 +) + +const ( + checksumSize = 4 + chunkHeaderSize = 4 + magicChunk = "\xff\x06\x00\x00" + magicBody + magicChunkSnappy = "\xff\x06\x00\x00" + magicBodySnappy + magicBodySnappy = "sNaPpY" + magicBody = "S2sTwO" + + // maxBlockSize is the maximum size of the input to encodeBlock. + // + // For the framing format (Writer type instead of Encode function), + // this is the maximum uncompressed size of a block. + maxBlockSize = 4 << 20 + + // minBlockSize is the minimum size of block setting when creating a writer. + minBlockSize = 4 << 10 + + skippableFrameHeader = 4 + maxChunkSize = 1<<24 - 1 // 16777215 + + // Default block size + defaultBlockSize = 1 << 20 + + // maxSnappyBlockSize is the maximum snappy block size. + maxSnappyBlockSize = 1 << 16 + + obufHeaderLen = checksumSize + chunkHeaderSize +) + +const ( + chunkTypeCompressedData = 0x00 + chunkTypeUncompressedData = 0x01 + ChunkTypeIndex = 0x99 + chunkTypePadding = 0xfe + chunkTypeStreamIdentifier = 0xff +) + +var crcTable = crc32.MakeTable(crc32.Castagnoli) + +// crc implements the checksum specified in section 3 of +// https://github.com/google/snappy/blob/master/framing_format.txt +func crc(b []byte) uint32 { + c := crc32.Update(0, crcTable, b) + return c>>15 | c<<17 + 0xa282ead8 +} + +// literalExtraSize returns the extra size of encoding n literals. +// n should be >= 0 and <= math.MaxUint32. +func literalExtraSize(n int64) int64 { + if n == 0 { + return 0 + } + switch { + case n < 60: + return 1 + case n < 1<<8: + return 2 + case n < 1<<16: + return 3 + case n < 1<<24: + return 4 + default: + return 5 + } +} + +type byter interface { + Bytes() []byte +} + +var _ byter = &bytes.Buffer{} diff --git a/vendor/github.com/klauspost/compress/s2sx.mod b/vendor/github.com/klauspost/compress/s2sx.mod new file mode 100644 index 0000000..2263853 --- /dev/null +++ b/vendor/github.com/klauspost/compress/s2sx.mod @@ -0,0 +1,4 @@ +module github.com/klauspost/compress + +go 1.16 + diff --git a/vendor/github.com/klauspost/compress/s2sx.sum b/vendor/github.com/klauspost/compress/s2sx.sum new file mode 100644 index 0000000..e69de29 diff --git a/vendor/github.com/klauspost/compress/snappy/.gitignore b/vendor/github.com/klauspost/compress/snappy/.gitignore new file mode 100644 index 0000000..042091d --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/.gitignore @@ -0,0 +1,16 @@ +cmd/snappytool/snappytool +testdata/bench + +# These explicitly listed benchmark data files are for an obsolete version of +# snappy_test.go. +testdata/alice29.txt +testdata/asyoulik.txt +testdata/fireworks.jpeg +testdata/geo.protodata +testdata/html +testdata/html_x_4 +testdata/kppkn.gtb +testdata/lcet10.txt +testdata/paper-100k.pdf +testdata/plrabn12.txt +testdata/urls.10K diff --git a/vendor/github.com/klauspost/compress/snappy/AUTHORS b/vendor/github.com/klauspost/compress/snappy/AUTHORS new file mode 100644 index 0000000..52ccb5a --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/AUTHORS @@ -0,0 +1,18 @@ +# This is the official list of Snappy-Go authors for copyright purposes. +# This file is distinct from the CONTRIBUTORS files. +# See the latter for an explanation. + +# Names should be added to this file as +# Name or Organization +# The email address is not required for organizations. + +# Please keep the list sorted. + +Amazon.com, Inc +Damian Gryski +Eric Buth +Google Inc. +Jan Mercl <0xjnml@gmail.com> +Klaus Post +Rodolfo Carvalho +Sebastien Binet diff --git a/vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS b/vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS new file mode 100644 index 0000000..ea6524d --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS @@ -0,0 +1,41 @@ +# This is the official list of people who can contribute +# (and typically have contributed) code to the Snappy-Go repository. +# The AUTHORS file lists the copyright holders; this file +# lists people. For example, Google employees are listed here +# but not in AUTHORS, because Google holds the copyright. +# +# The submission process automatically checks to make sure +# that people submitting code are listed in this file (by email address). +# +# Names should be added to this file only after verifying that +# the individual or the individual's organization has agreed to +# the appropriate Contributor License Agreement, found here: +# +# http://code.google.com/legal/individual-cla-v1.0.html +# http://code.google.com/legal/corporate-cla-v1.0.html +# +# The agreement for individuals can be filled out on the web. +# +# When adding J Random Contributor's name to this file, +# either J's name or J's organization's name should be +# added to the AUTHORS file, depending on whether the +# individual or corporate CLA was used. + +# Names should be added to this file like so: +# Name + +# Please keep the list sorted. + +Alex Legg +Damian Gryski +Eric Buth +Jan Mercl <0xjnml@gmail.com> +Jonathan Swinney +Kai Backman +Klaus Post +Marc-Antoine Ruel +Nigel Tao +Rob Pike +Rodolfo Carvalho +Russ Cox +Sebastien Binet diff --git a/vendor/github.com/klauspost/compress/snappy/LICENSE b/vendor/github.com/klauspost/compress/snappy/LICENSE new file mode 100644 index 0000000..6050c10 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/LICENSE @@ -0,0 +1,27 @@ +Copyright (c) 2011 The Snappy-Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/vendor/github.com/klauspost/compress/snappy/README.md b/vendor/github.com/klauspost/compress/snappy/README.md new file mode 100644 index 0000000..8271bbd --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/README.md @@ -0,0 +1,17 @@ +# snappy + +The Snappy compression format in the Go programming language. + +This is a drop-in replacement for `github.com/golang/snappy`. + +It provides a full, compatible replacement of the Snappy package by simply changing imports. + +See [Snappy Compatibility](https://github.com/klauspost/compress/tree/master/s2#snappy-compatibility) in the S2 documentation. + +"Better" compression mode is used. For buffered streams concurrent compression is used. + +For more options use the [s2 package](https://pkg.go.dev/github.com/klauspost/compress/s2). + +# usage + +Replace imports `github.com/golang/snappy` with `github.com/klauspost/compress/snappy`. diff --git a/vendor/github.com/klauspost/compress/snappy/decode.go b/vendor/github.com/klauspost/compress/snappy/decode.go new file mode 100644 index 0000000..89f1fa2 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/decode.go @@ -0,0 +1,60 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package snappy + +import ( + "io" + + "github.com/klauspost/compress/s2" +) + +var ( + // ErrCorrupt reports that the input is invalid. + ErrCorrupt = s2.ErrCorrupt + // ErrTooLarge reports that the uncompressed length is too large. + ErrTooLarge = s2.ErrTooLarge + // ErrUnsupported reports that the input isn't supported. + ErrUnsupported = s2.ErrUnsupported +) + +const ( + // maxBlockSize is the maximum size of the input to encodeBlock. It is not + // part of the wire format per se, but some parts of the encoder assume + // that an offset fits into a uint16. + // + // Also, for the framing format (Writer type instead of Encode function), + // https://github.com/google/snappy/blob/master/framing_format.txt says + // that "the uncompressed data in a chunk must be no longer than 65536 + // bytes". + maxBlockSize = 65536 +) + +// DecodedLen returns the length of the decoded block. +func DecodedLen(src []byte) (int, error) { + return s2.DecodedLen(src) +} + +// Decode returns the decoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire decoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// Decode handles the Snappy block format, not the Snappy stream format. +func Decode(dst, src []byte) ([]byte, error) { + return s2.Decode(dst, src) +} + +// NewReader returns a new Reader that decompresses from r, using the framing +// format described at +// https://github.com/google/snappy/blob/master/framing_format.txt +func NewReader(r io.Reader) *Reader { + return s2.NewReader(r, s2.ReaderMaxBlockSize(maxBlockSize)) +} + +// Reader is an io.Reader that can read Snappy-compressed bytes. +// +// Reader handles the Snappy stream format, not the Snappy block format. +type Reader = s2.Reader diff --git a/vendor/github.com/klauspost/compress/snappy/encode.go b/vendor/github.com/klauspost/compress/snappy/encode.go new file mode 100644 index 0000000..e8bd72c --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/encode.go @@ -0,0 +1,59 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package snappy + +import ( + "io" + + "github.com/klauspost/compress/s2" +) + +// Encode returns the encoded form of src. The returned slice may be a sub- +// slice of dst if dst was large enough to hold the entire encoded block. +// Otherwise, a newly allocated slice will be returned. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// Encode handles the Snappy block format, not the Snappy stream format. +func Encode(dst, src []byte) []byte { + return s2.EncodeSnappyBetter(dst, src) +} + +// MaxEncodedLen returns the maximum length of a snappy block, given its +// uncompressed length. +// +// It will return a negative value if srcLen is too large to encode. +func MaxEncodedLen(srcLen int) int { + return s2.MaxEncodedLen(srcLen) +} + +// NewWriter returns a new Writer that compresses to w. +// +// The Writer returned does not buffer writes. There is no need to Flush or +// Close such a Writer. +// +// Deprecated: the Writer returned is not suitable for many small writes, only +// for few large writes. Use NewBufferedWriter instead, which is efficient +// regardless of the frequency and shape of the writes, and remember to Close +// that Writer when done. +func NewWriter(w io.Writer) *Writer { + return s2.NewWriter(w, s2.WriterSnappyCompat(), s2.WriterBetterCompression(), s2.WriterFlushOnWrite(), s2.WriterConcurrency(1)) +} + +// NewBufferedWriter returns a new Writer that compresses to w, using the +// framing format described at +// https://github.com/google/snappy/blob/master/framing_format.txt +// +// The Writer returned buffers writes. Users must call Close to guarantee all +// data has been forwarded to the underlying io.Writer. They may also call +// Flush zero or more times before calling Close. +func NewBufferedWriter(w io.Writer) *Writer { + return s2.NewWriter(w, s2.WriterSnappyCompat(), s2.WriterBetterCompression()) +} + +// Writer is an io.Writer that can write Snappy-compressed bytes. +// +// Writer handles the Snappy stream format, not the Snappy block format. +type Writer = s2.Writer diff --git a/vendor/github.com/klauspost/compress/snappy/snappy.go b/vendor/github.com/klauspost/compress/snappy/snappy.go new file mode 100644 index 0000000..398cdc9 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/snappy.go @@ -0,0 +1,46 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package snappy implements the Snappy compression format. It aims for very +// high speeds and reasonable compression. +// +// There are actually two Snappy formats: block and stream. They are related, +// but different: trying to decompress block-compressed data as a Snappy stream +// will fail, and vice versa. The block format is the Decode and Encode +// functions and the stream format is the Reader and Writer types. +// +// The block format, the more common case, is used when the complete size (the +// number of bytes) of the original data is known upfront, at the time +// compression starts. The stream format, also known as the framing format, is +// for when that isn't always true. +// +// The canonical, C++ implementation is at https://github.com/google/snappy and +// it only implements the block format. +package snappy + +/* +Each encoded block begins with the varint-encoded length of the decoded data, +followed by a sequence of chunks. Chunks begin and end on byte boundaries. The +first byte of each chunk is broken into its 2 least and 6 most significant bits +called l and m: l ranges in [0, 4) and m ranges in [0, 64). l is the chunk tag. +Zero means a literal tag. All other values mean a copy tag. + +For literal tags: + - If m < 60, the next 1 + m bytes are literal bytes. + - Otherwise, let n be the little-endian unsigned integer denoted by the next + m - 59 bytes. The next 1 + n bytes after that are literal bytes. + +For copy tags, length bytes are copied from offset bytes ago, in the style of +Lempel-Ziv compression algorithms. In particular: + - For l == 1, the offset ranges in [0, 1<<11) and the length in [4, 12). + The length is 4 + the low 3 bits of m. The high 3 bits of m form bits 8-10 + of the offset. The next byte is bits 0-7 of the offset. + - For l == 2, the offset ranges in [0, 1<<16) and the length in [1, 65). + The length is 1 + m. The offset is the little-endian unsigned integer + denoted by the next 2 bytes. + - For l == 3, this tag is a legacy format that is no longer issued by most + encoders. Nonetheless, the offset ranges in [0, 1<<32) and the length in + [1, 65). The length is 1 + m. The offset is the little-endian unsigned + integer denoted by the next 4 bytes. +*/ diff --git a/vendor/github.com/klauspost/compress/zstd/README.md b/vendor/github.com/klauspost/compress/zstd/README.md new file mode 100644 index 0000000..beb7fa8 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/README.md @@ -0,0 +1,439 @@ +# zstd + +[Zstandard](https://facebook.github.io/zstd/) is a real-time compression algorithm, providing high compression ratios. +It offers a very wide range of compression / speed trade-off, while being backed by a very fast decoder. +A high performance compression algorithm is implemented. For now focused on speed. + +This package provides [compression](#Compressor) to and [decompression](#Decompressor) of Zstandard content. + +This package is pure Go and without use of "unsafe". + +The `zstd` package is provided as open source software using a Go standard license. + +Currently the package is heavily optimized for 64 bit processors and will be significantly slower on 32 bit processors. + +## Installation + +Install using `go get -u github.com/klauspost/compress`. The package is located in `github.com/klauspost/compress/zstd`. + +[![Go Reference](https://pkg.go.dev/badge/github.com/klauspost/compress/zstd.svg)](https://pkg.go.dev/github.com/klauspost/compress/zstd) + +## Compressor + +### Status: + +STABLE - there may always be subtle bugs, a wide variety of content has been tested and the library is actively +used by several projects. This library is being [fuzz-tested](https://github.com/klauspost/compress-fuzz) for all updates. + +There may still be specific combinations of data types/size/settings that could lead to edge cases, +so as always, testing is recommended. + +For now, a high speed (fastest) and medium-fast (default) compressor has been implemented. + +* The "Fastest" compression ratio is roughly equivalent to zstd level 1. +* The "Default" compression ratio is roughly equivalent to zstd level 3 (default). +* The "Better" compression ratio is roughly equivalent to zstd level 7. +* The "Best" compression ratio is roughly equivalent to zstd level 11. + +In terms of speed, it is typically 2x as fast as the stdlib deflate/gzip in its fastest mode. +The compression ratio compared to stdlib is around level 3, but usually 3x as fast. + + +### Usage + +An Encoder can be used for either compressing a stream via the +`io.WriteCloser` interface supported by the Encoder or as multiple independent +tasks via the `EncodeAll` function. +Smaller encodes are encouraged to use the EncodeAll function. +Use `NewWriter` to create a new instance that can be used for both. + +To create a writer with default options, do like this: + +```Go +// Compress input to output. +func Compress(in io.Reader, out io.Writer) error { + enc, err := zstd.NewWriter(out) + if err != nil { + return err + } + _, err = io.Copy(enc, in) + if err != nil { + enc.Close() + return err + } + return enc.Close() +} +``` + +Now you can encode by writing data to `enc`. The output will be finished writing when `Close()` is called. +Even if your encode fails, you should still call `Close()` to release any resources that may be held up. + +The above is fine for big encodes. However, whenever possible try to *reuse* the writer. + +To reuse the encoder, you can use the `Reset(io.Writer)` function to change to another output. +This will allow the encoder to reuse all resources and avoid wasteful allocations. + +Currently stream encoding has 'light' concurrency, meaning up to 2 goroutines can be working on part +of a stream. This is independent of the `WithEncoderConcurrency(n)`, but that is likely to change +in the future. So if you want to limit concurrency for future updates, specify the concurrency +you would like. + +If you would like stream encoding to be done without spawning async goroutines, use `WithEncoderConcurrency(1)` +which will compress input as each block is completed, blocking on writes until each has completed. + +You can specify your desired compression level using `WithEncoderLevel()` option. Currently only pre-defined +compression settings can be specified. + +#### Future Compatibility Guarantees + +This will be an evolving project. When using this package it is important to note that both the compression efficiency and speed may change. + +The goal will be to keep the default efficiency at the default zstd (level 3). +However the encoding should never be assumed to remain the same, +and you should not use hashes of compressed output for similarity checks. + +The Encoder can be assumed to produce the same output from the exact same code version. +However, the may be modes in the future that break this, +although they will not be enabled without an explicit option. + +This encoder is not designed to (and will probably never) output the exact same bitstream as the reference encoder. + +Also note, that the cgo decompressor currently does not [report all errors on invalid input](https://github.com/DataDog/zstd/issues/59), +[omits error checks](https://github.com/DataDog/zstd/issues/61), [ignores checksums](https://github.com/DataDog/zstd/issues/43) +and seems to ignore concatenated streams, even though [it is part of the spec](https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frames). + +#### Blocks + +For compressing small blocks, the returned encoder has a function called `EncodeAll(src, dst []byte) []byte`. + +`EncodeAll` will encode all input in src and append it to dst. +This function can be called concurrently. +Each call will only run on a same goroutine as the caller. + +Encoded blocks can be concatenated and the result will be the combined input stream. +Data compressed with EncodeAll can be decoded with the Decoder, using either a stream or `DecodeAll`. + +Especially when encoding blocks you should take special care to reuse the encoder. +This will effectively make it run without allocations after a warmup period. +To make it run completely without allocations, supply a destination buffer with space for all content. + +```Go +import "github.com/klauspost/compress/zstd" + +// Create a writer that caches compressors. +// For this operation type we supply a nil Reader. +var encoder, _ = zstd.NewWriter(nil) + +// Compress a buffer. +// If you have a destination buffer, the allocation in the call can also be eliminated. +func Compress(src []byte) []byte { + return encoder.EncodeAll(src, make([]byte, 0, len(src))) +} +``` + +You can control the maximum number of concurrent encodes using the `WithEncoderConcurrency(n)` +option when creating the writer. + +Using the Encoder for both a stream and individual blocks concurrently is safe. + +### Performance + +I have collected some speed examples to compare speed and compression against other compressors. + +* `file` is the input file. +* `out` is the compressor used. `zskp` is this package. `zstd` is the Datadog cgo library. `gzstd/gzkp` is gzip standard and this library. +* `level` is the compression level used. For `zskp` level 1 is "fastest", level 2 is "default"; 3 is "better", 4 is "best". +* `insize`/`outsize` is the input/output size. +* `millis` is the number of milliseconds used for compression. +* `mb/s` is megabytes (2^20 bytes) per second. + +``` +Silesia Corpus: +http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip + +This package: +file out level insize outsize millis mb/s +silesia.tar zskp 1 211947520 73821326 634 318.47 +silesia.tar zskp 2 211947520 67655404 1508 133.96 +silesia.tar zskp 3 211947520 64746933 3000 67.37 +silesia.tar zskp 4 211947520 60073508 16926 11.94 + +cgo zstd: +silesia.tar zstd 1 211947520 73605392 543 371.56 +silesia.tar zstd 3 211947520 66793289 864 233.68 +silesia.tar zstd 6 211947520 62916450 1913 105.66 +silesia.tar zstd 9 211947520 60212393 5063 39.92 + +gzip, stdlib/this package: +silesia.tar gzstd 1 211947520 80007735 1498 134.87 +silesia.tar gzkp 1 211947520 80088272 1009 200.31 + +GOB stream of binary data. Highly compressible. +https://files.klauspost.com/compress/gob-stream.7z + +file out level insize outsize millis mb/s +gob-stream zskp 1 1911399616 233948096 3230 564.34 +gob-stream zskp 2 1911399616 203997694 4997 364.73 +gob-stream zskp 3 1911399616 173526523 13435 135.68 +gob-stream zskp 4 1911399616 162195235 47559 38.33 + +gob-stream zstd 1 1911399616 249810424 2637 691.26 +gob-stream zstd 3 1911399616 208192146 3490 522.31 +gob-stream zstd 6 1911399616 193632038 6687 272.56 +gob-stream zstd 9 1911399616 177620386 16175 112.70 + +gob-stream gzstd 1 1911399616 357382013 9046 201.49 +gob-stream gzkp 1 1911399616 359136669 4885 373.08 + +The test data for the Large Text Compression Benchmark is the first +10^9 bytes of the English Wikipedia dump on Mar. 3, 2006. +http://mattmahoney.net/dc/textdata.html + +file out level insize outsize millis mb/s +enwik9 zskp 1 1000000000 343833605 3687 258.64 +enwik9 zskp 2 1000000000 317001237 7672 124.29 +enwik9 zskp 3 1000000000 291915823 15923 59.89 +enwik9 zskp 4 1000000000 261710291 77697 12.27 + +enwik9 zstd 1 1000000000 358072021 3110 306.65 +enwik9 zstd 3 1000000000 313734672 4784 199.35 +enwik9 zstd 6 1000000000 295138875 10290 92.68 +enwik9 zstd 9 1000000000 278348700 28549 33.40 + +enwik9 gzstd 1 1000000000 382578136 8608 110.78 +enwik9 gzkp 1 1000000000 382781160 5628 169.45 + +Highly compressible JSON file. +https://files.klauspost.com/compress/github-june-2days-2019.json.zst + +file out level insize outsize millis mb/s +github-june-2days-2019.json zskp 1 6273951764 697439532 9789 611.17 +github-june-2days-2019.json zskp 2 6273951764 610876538 18553 322.49 +github-june-2days-2019.json zskp 3 6273951764 517662858 44186 135.41 +github-june-2days-2019.json zskp 4 6273951764 464617114 165373 36.18 + +github-june-2days-2019.json zstd 1 6273951764 766284037 8450 708.00 +github-june-2days-2019.json zstd 3 6273951764 661889476 10927 547.57 +github-june-2days-2019.json zstd 6 6273951764 642756859 22996 260.18 +github-june-2days-2019.json zstd 9 6273951764 601974523 52413 114.16 + +github-june-2days-2019.json gzstd 1 6273951764 1164397768 26793 223.32 +github-june-2days-2019.json gzkp 1 6273951764 1120631856 17693 338.16 + +VM Image, Linux mint with a few installed applications: +https://files.klauspost.com/compress/rawstudio-mint14.7z + +file out level insize outsize millis mb/s +rawstudio-mint14.tar zskp 1 8558382592 3718400221 18206 448.29 +rawstudio-mint14.tar zskp 2 8558382592 3326118337 37074 220.15 +rawstudio-mint14.tar zskp 3 8558382592 3163842361 87306 93.49 +rawstudio-mint14.tar zskp 4 8558382592 2970480650 783862 10.41 + +rawstudio-mint14.tar zstd 1 8558382592 3609250104 17136 476.27 +rawstudio-mint14.tar zstd 3 8558382592 3341679997 29262 278.92 +rawstudio-mint14.tar zstd 6 8558382592 3235846406 77904 104.77 +rawstudio-mint14.tar zstd 9 8558382592 3160778861 140946 57.91 + +rawstudio-mint14.tar gzstd 1 8558382592 3926234992 51345 158.96 +rawstudio-mint14.tar gzkp 1 8558382592 3960117298 36722 222.26 + +CSV data: +https://files.klauspost.com/compress/nyc-taxi-data-10M.csv.zst + +file out level insize outsize millis mb/s +nyc-taxi-data-10M.csv zskp 1 3325605752 641319332 9462 335.17 +nyc-taxi-data-10M.csv zskp 2 3325605752 588976126 17570 180.50 +nyc-taxi-data-10M.csv zskp 3 3325605752 529329260 32432 97.79 +nyc-taxi-data-10M.csv zskp 4 3325605752 474949772 138025 22.98 + +nyc-taxi-data-10M.csv zstd 1 3325605752 687399637 8233 385.18 +nyc-taxi-data-10M.csv zstd 3 3325605752 598514411 10065 315.07 +nyc-taxi-data-10M.csv zstd 6 3325605752 570522953 20038 158.27 +nyc-taxi-data-10M.csv zstd 9 3325605752 517554797 64565 49.12 + +nyc-taxi-data-10M.csv gzstd 1 3325605752 928654908 21270 149.11 +nyc-taxi-data-10M.csv gzkp 1 3325605752 922273214 13929 227.68 +``` + +## Decompressor + +Staus: STABLE - there may still be subtle bugs, but a wide variety of content has been tested. + +This library is being continuously [fuzz-tested](https://github.com/klauspost/compress-fuzz), +kindly supplied by [fuzzit.dev](https://fuzzit.dev/). +The main purpose of the fuzz testing is to ensure that it is not possible to crash the decoder, +or run it past its limits with ANY input provided. + +### Usage + +The package has been designed for two main usages, big streams of data and smaller in-memory buffers. +There are two main usages of the package for these. Both of them are accessed by creating a `Decoder`. + +For streaming use a simple setup could look like this: + +```Go +import "github.com/klauspost/compress/zstd" + +func Decompress(in io.Reader, out io.Writer) error { + d, err := zstd.NewReader(in) + if err != nil { + return err + } + defer d.Close() + + // Copy content... + _, err = io.Copy(out, d) + return err +} +``` + +It is important to use the "Close" function when you no longer need the Reader to stop running goroutines, +when running with default settings. +Goroutines will exit once an error has been returned, including `io.EOF` at the end of a stream. + +Streams are decoded concurrently in 4 asynchronous stages to give the best possible throughput. +However, if you prefer synchronous decompression, use `WithDecoderConcurrency(1)` which will decompress data +as it is being requested only. + +For decoding buffers, it could look something like this: + +```Go +import "github.com/klauspost/compress/zstd" + +// Create a reader that caches decompressors. +// For this operation type we supply a nil Reader. +var decoder, _ = zstd.NewReader(nil, WithDecoderConcurrency(0)) + +// Decompress a buffer. We don't supply a destination buffer, +// so it will be allocated by the decoder. +func Decompress(src []byte) ([]byte, error) { + return decoder.DecodeAll(src, nil) +} +``` + +Both of these cases should provide the functionality needed. +The decoder can be used for *concurrent* decompression of multiple buffers. +By default 4 decompressors will be created. + +It will only allow a certain number of concurrent operations to run. +To tweak that yourself use the `WithDecoderConcurrency(n)` option when creating the decoder. +It is possible to use `WithDecoderConcurrency(0)` to create GOMAXPROCS decoders. + +### Dictionaries + +Data compressed with [dictionaries](https://github.com/facebook/zstd#the-case-for-small-data-compression) can be decompressed. + +Dictionaries are added individually to Decoders. +Dictionaries are generated by the `zstd --train` command and contains an initial state for the decoder. +To add a dictionary use the `WithDecoderDicts(dicts ...[]byte)` option with the dictionary data. +Several dictionaries can be added at once. + +The dictionary will be used automatically for the data that specifies them. +A re-used Decoder will still contain the dictionaries registered. + +When registering multiple dictionaries with the same ID, the last one will be used. + +It is possible to use dictionaries when compressing data. + +To enable a dictionary use `WithEncoderDict(dict []byte)`. Here only one dictionary will be used +and it will likely be used even if it doesn't improve compression. + +The used dictionary must be used to decompress the content. + +For any real gains, the dictionary should be built with similar data. +If an unsuitable dictionary is used the output may be slightly larger than using no dictionary. +Use the [zstd commandline tool](https://github.com/facebook/zstd/releases) to build a dictionary from sample data. +For information see [zstd dictionary information](https://github.com/facebook/zstd#the-case-for-small-data-compression). + +For now there is a fixed startup performance penalty for compressing content with dictionaries. +This will likely be improved over time. Just be aware to test performance when implementing. + +### Allocation-less operation + +The decoder has been designed to operate without allocations after a warmup. + +This means that you should *store* the decoder for best performance. +To re-use a stream decoder, use the `Reset(r io.Reader) error` to switch to another stream. +A decoder can safely be re-used even if the previous stream failed. + +To release the resources, you must call the `Close()` function on a decoder. +After this it can *no longer be reused*, but all running goroutines will be stopped. +So you *must* use this if you will no longer need the Reader. + +For decompressing smaller buffers a single decoder can be used. +When decoding buffers, you can supply a destination slice with length 0 and your expected capacity. +In this case no unneeded allocations should be made. + +### Concurrency + +The buffer decoder does everything on the same goroutine and does nothing concurrently. +It can however decode several buffers concurrently. Use `WithDecoderConcurrency(n)` to limit that. + +The stream decoder will create goroutines that: + +1) Reads input and splits the input into blocks. +2) Decompression of literals. +3) Decompression of sequences. +4) Reconstruction of output stream. + +So effectively this also means the decoder will "read ahead" and prepare data to always be available for output. + +The concurrency level will, for streams, determine how many blocks ahead the compression will start. + +Since "blocks" are quite dependent on the output of the previous block stream decoding will only have limited concurrency. + +In practice this means that concurrency is often limited to utilizing about 3 cores effectively. + +### Benchmarks + +The first two are streaming decodes and the last are smaller inputs. + +Running on AMD Ryzen 9 3950X 16-Core Processor. AMD64 assembly used. + +``` +BenchmarkDecoderSilesia-32 5 206878840 ns/op 1024.50 MB/s 49808 B/op 43 allocs/op +BenchmarkDecoderEnwik9-32 1 1271809000 ns/op 786.28 MB/s 72048 B/op 52 allocs/op + +Concurrent blocks, performance: + +BenchmarkDecoder_DecodeAllParallel/kppkn.gtb.zst-32 67356 17857 ns/op 10321.96 MB/s 22.48 pct 102 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/geo.protodata.zst-32 266656 4421 ns/op 26823.21 MB/s 11.89 pct 19 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/plrabn12.txt.zst-32 20992 56842 ns/op 8477.17 MB/s 39.90 pct 754 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/lcet10.txt.zst-32 27456 43932 ns/op 9714.01 MB/s 33.27 pct 524 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/asyoulik.txt.zst-32 78432 15047 ns/op 8319.15 MB/s 40.34 pct 66 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/alice29.txt.zst-32 65800 18436 ns/op 8249.63 MB/s 37.75 pct 88 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/html_x_4.zst-32 102993 11523 ns/op 35546.09 MB/s 3.637 pct 143 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/paper-100k.pdf.zst-32 1000000 1070 ns/op 95720.98 MB/s 80.53 pct 3 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/fireworks.jpeg.zst-32 749802 1752 ns/op 70272.35 MB/s 100.0 pct 5 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/urls.10K.zst-32 22640 52934 ns/op 13263.37 MB/s 26.25 pct 1014 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/html.zst-32 226412 5232 ns/op 19572.27 MB/s 14.49 pct 20 B/op 0 allocs/op +BenchmarkDecoder_DecodeAllParallel/comp-data.bin.zst-32 923041 1276 ns/op 3194.71 MB/s 31.26 pct 0 B/op 0 allocs/op +``` + +This reflects the performance around May 2022, but this may be out of date. + +## Zstd inside ZIP files + +It is possible to use zstandard to compress individual files inside zip archives. +While this isn't widely supported it can be useful for internal files. + +To support the compression and decompression of these files you must register a compressor and decompressor. + +It is highly recommended registering the (de)compressors on individual zip Reader/Writer and NOT +use the global registration functions. The main reason for this is that 2 registrations from +different packages will result in a panic. + +It is a good idea to only have a single compressor and decompressor, since they can be used for multiple zip +files concurrently, and using a single instance will allow reusing some resources. + +See [this example](https://pkg.go.dev/github.com/klauspost/compress/zstd#example-ZipCompressor) for +how to compress and decompress files inside zip archives. + +# Contributions + +Contributions are always welcome. +For new features/fixes, remember to add tests and for performance enhancements include benchmarks. + +For general feedback and experience reports, feel free to open an issue or write me on [Twitter](https://twitter.com/sh0dan). + +This package includes the excellent [`github.com/cespare/xxhash`](https://github.com/cespare/xxhash) package Copyright (c) 2016 Caleb Spare. diff --git a/vendor/github.com/klauspost/compress/zstd/bitreader.go b/vendor/github.com/klauspost/compress/zstd/bitreader.go new file mode 100644 index 0000000..97299d4 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bitreader.go @@ -0,0 +1,140 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "errors" + "fmt" + "io" + "math/bits" +) + +// bitReader reads a bitstream in reverse. +// The last set bit indicates the start of the stream and is used +// for aligning the input. +type bitReader struct { + in []byte + off uint // next byte to read is at in[off - 1] + value uint64 // Maybe use [16]byte, but shifting is awkward. + bitsRead uint8 +} + +// init initializes and resets the bit reader. +func (b *bitReader) init(in []byte) error { + if len(in) < 1 { + return errors.New("corrupt stream: too short") + } + b.in = in + b.off = uint(len(in)) + // The highest bit of the last byte indicates where to start + v := in[len(in)-1] + if v == 0 { + return errors.New("corrupt stream, did not find end of stream") + } + b.bitsRead = 64 + b.value = 0 + if len(in) >= 8 { + b.fillFastStart() + } else { + b.fill() + b.fill() + } + b.bitsRead += 8 - uint8(highBits(uint32(v))) + return nil +} + +// getBits will return n bits. n can be 0. +func (b *bitReader) getBits(n uint8) int { + if n == 0 /*|| b.bitsRead >= 64 */ { + return 0 + } + return int(b.get32BitsFast(n)) +} + +// get32BitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReader) get32BitsFast(n uint8) uint32 { + const regMask = 64 - 1 + v := uint32((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask)) + b.bitsRead += n + return v +} + +// fillFast() will make sure at least 32 bits are available. +// There must be at least 4 bytes available. +func (b *bitReader) fillFast() { + if b.bitsRead < 32 { + return + } + // 2 bounds checks. + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value = (b.value << 32) | uint64(low) + b.bitsRead -= 32 + b.off -= 4 +} + +// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read. +func (b *bitReader) fillFastStart() { + // Do single re-slice to avoid bounds checks. + b.value = binary.LittleEndian.Uint64(b.in[b.off-8:]) + b.bitsRead = 0 + b.off -= 8 +} + +// fill() will make sure at least 32 bits are available. +func (b *bitReader) fill() { + if b.bitsRead < 32 { + return + } + if b.off >= 4 { + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value = (b.value << 32) | uint64(low) + b.bitsRead -= 32 + b.off -= 4 + return + } + for b.off > 0 { + b.value = (b.value << 8) | uint64(b.in[b.off-1]) + b.bitsRead -= 8 + b.off-- + } +} + +// finished returns true if all bits have been read from the bit stream. +func (b *bitReader) finished() bool { + return b.off == 0 && b.bitsRead >= 64 +} + +// overread returns true if more bits have been requested than is on the stream. +func (b *bitReader) overread() bool { + return b.bitsRead > 64 +} + +// remain returns the number of bits remaining. +func (b *bitReader) remain() uint { + return b.off*8 + 64 - uint(b.bitsRead) +} + +// close the bitstream and returns an error if out-of-buffer reads occurred. +func (b *bitReader) close() error { + // Release reference. + b.in = nil + if !b.finished() { + return fmt.Errorf("%d extra bits on block, should be 0", b.remain()) + } + if b.bitsRead > 64 { + return io.ErrUnexpectedEOF + } + return nil +} + +func highBits(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} diff --git a/vendor/github.com/klauspost/compress/zstd/bitwriter.go b/vendor/github.com/klauspost/compress/zstd/bitwriter.go new file mode 100644 index 0000000..78b3c61 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bitwriter.go @@ -0,0 +1,113 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package zstd + +// bitWriter will write bits. +// First bit will be LSB of the first byte of output. +type bitWriter struct { + bitContainer uint64 + nBits uint8 + out []byte +} + +// bitMask16 is bitmasks. Has extra to avoid bounds check. +var bitMask16 = [32]uint16{ + 0, 1, 3, 7, 0xF, 0x1F, + 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, + 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF} /* up to 16 bits */ + +var bitMask32 = [32]uint32{ + 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF, + 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, + 0x1ffff, 0x3ffff, 0x7FFFF, 0xfFFFF, 0x1fFFFF, 0x3fFFFF, 0x7fFFFF, 0xffFFFF, + 0x1ffFFFF, 0x3ffFFFF, 0x7ffFFFF, 0xfffFFFF, 0x1fffFFFF, 0x3fffFFFF, 0x7fffFFFF, +} // up to 32 bits + +// addBits16NC will add up to 16 bits. +// It will not check if there is space for them, +// so the caller must ensure that it has flushed recently. +func (b *bitWriter) addBits16NC(value uint16, bits uint8) { + b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63) + b.nBits += bits +} + +// addBits32NC will add up to 31 bits. +// It will not check if there is space for them, +// so the caller must ensure that it has flushed recently. +func (b *bitWriter) addBits32NC(value uint32, bits uint8) { + b.bitContainer |= uint64(value&bitMask32[bits&31]) << (b.nBits & 63) + b.nBits += bits +} + +// addBits64NC will add up to 64 bits. +// There must be space for 32 bits. +func (b *bitWriter) addBits64NC(value uint64, bits uint8) { + if bits <= 31 { + b.addBits32Clean(uint32(value), bits) + return + } + b.addBits32Clean(uint32(value), 32) + b.flush32() + b.addBits32Clean(uint32(value>>32), bits-32) +} + +// addBits32Clean will add up to 32 bits. +// It will not check if there is space for them. +// The input must not contain more bits than specified. +func (b *bitWriter) addBits32Clean(value uint32, bits uint8) { + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated. +// It will not check if there is space for them, so the caller must ensure that it has flushed recently. +func (b *bitWriter) addBits16Clean(value uint16, bits uint8) { + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// flush32 will flush out, so there are at least 32 bits available for writing. +func (b *bitWriter) flush32() { + if b.nBits < 32 { + return + } + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24)) + b.nBits -= 32 + b.bitContainer >>= 32 +} + +// flushAlign will flush remaining full bytes and align to next byte boundary. +func (b *bitWriter) flushAlign() { + nbBytes := (b.nBits + 7) >> 3 + for i := uint8(0); i < nbBytes; i++ { + b.out = append(b.out, byte(b.bitContainer>>(i*8))) + } + b.nBits = 0 + b.bitContainer = 0 +} + +// close will write the alignment bit and write the final byte(s) +// to the output. +func (b *bitWriter) close() error { + // End mark + b.addBits16Clean(1, 1) + // flush until next byte. + b.flushAlign() + return nil +} + +// reset and continue writing by appending to out. +func (b *bitWriter) reset(out []byte) { + b.bitContainer = 0 + b.nBits = 0 + b.out = out +} diff --git a/vendor/github.com/klauspost/compress/zstd/blockdec.go b/vendor/github.com/klauspost/compress/zstd/blockdec.go new file mode 100644 index 0000000..7eed729 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/blockdec.go @@ -0,0 +1,721 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "bytes" + "encoding/binary" + "errors" + "fmt" + "io" + "io/ioutil" + "os" + "path/filepath" + "sync" + + "github.com/klauspost/compress/huff0" + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +type blockType uint8 + +//go:generate stringer -type=blockType,literalsBlockType,seqCompMode,tableIndex + +const ( + blockTypeRaw blockType = iota + blockTypeRLE + blockTypeCompressed + blockTypeReserved +) + +type literalsBlockType uint8 + +const ( + literalsBlockRaw literalsBlockType = iota + literalsBlockRLE + literalsBlockCompressed + literalsBlockTreeless +) + +const ( + // maxCompressedBlockSize is the biggest allowed compressed block size (128KB) + maxCompressedBlockSize = 128 << 10 + + compressedBlockOverAlloc = 16 + maxCompressedBlockSizeAlloc = 128<<10 + compressedBlockOverAlloc + + // Maximum possible block size (all Raw+Uncompressed). + maxBlockSize = (1 << 21) - 1 + + maxMatchLen = 131074 + maxSequences = 0x7f00 + 0xffff + + // We support slightly less than the reference decoder to be able to + // use ints on 32 bit archs. + maxOffsetBits = 30 +) + +var ( + huffDecoderPool = sync.Pool{New: func() interface{} { + return &huff0.Scratch{} + }} + + fseDecoderPool = sync.Pool{New: func() interface{} { + return &fseDecoder{} + }} +) + +type blockDec struct { + // Raw source data of the block. + data []byte + dataStorage []byte + + // Destination of the decoded data. + dst []byte + + // Buffer for literals data. + literalBuf []byte + + // Window size of the block. + WindowSize uint64 + + err error + + // Check against this crc + checkCRC []byte + + // Frame to use for singlethreaded decoding. + // Should not be used by the decoder itself since parent may be another frame. + localFrame *frameDec + + sequence []seqVals + + async struct { + newHist *history + literals []byte + seqData []byte + seqSize int // Size of uncompressed sequences + fcs uint64 + } + + // Block is RLE, this is the size. + RLESize uint32 + + Type blockType + + // Is this the last block of a frame? + Last bool + + // Use less memory + lowMem bool +} + +func (b *blockDec) String() string { + if b == nil { + return "" + } + return fmt.Sprintf("Steam Size: %d, Type: %v, Last: %t, Window: %d", len(b.data), b.Type, b.Last, b.WindowSize) +} + +func newBlockDec(lowMem bool) *blockDec { + b := blockDec{ + lowMem: lowMem, + } + return &b +} + +// reset will reset the block. +// Input must be a start of a block and will be at the end of the block when returned. +func (b *blockDec) reset(br byteBuffer, windowSize uint64) error { + b.WindowSize = windowSize + tmp, err := br.readSmall(3) + if err != nil { + println("Reading block header:", err) + return err + } + bh := uint32(tmp[0]) | (uint32(tmp[1]) << 8) | (uint32(tmp[2]) << 16) + b.Last = bh&1 != 0 + b.Type = blockType((bh >> 1) & 3) + // find size. + cSize := int(bh >> 3) + maxSize := maxCompressedBlockSizeAlloc + switch b.Type { + case blockTypeReserved: + return ErrReservedBlockType + case blockTypeRLE: + if cSize > maxCompressedBlockSize || cSize > int(b.WindowSize) { + if debugDecoder { + printf("rle block too big: csize:%d block: %+v\n", uint64(cSize), b) + } + return ErrWindowSizeExceeded + } + b.RLESize = uint32(cSize) + if b.lowMem { + maxSize = cSize + } + cSize = 1 + case blockTypeCompressed: + if debugDecoder { + println("Data size on stream:", cSize) + } + b.RLESize = 0 + maxSize = maxCompressedBlockSizeAlloc + if windowSize < maxCompressedBlockSize && b.lowMem { + maxSize = int(windowSize) + compressedBlockOverAlloc + } + if cSize > maxCompressedBlockSize || uint64(cSize) > b.WindowSize { + if debugDecoder { + printf("compressed block too big: csize:%d block: %+v\n", uint64(cSize), b) + } + return ErrCompressedSizeTooBig + } + // Empty compressed blocks must at least be 2 bytes + // for Literals_Block_Type and one for Sequences_Section_Header. + if cSize < 2 { + return ErrBlockTooSmall + } + case blockTypeRaw: + if cSize > maxCompressedBlockSize || cSize > int(b.WindowSize) { + if debugDecoder { + printf("rle block too big: csize:%d block: %+v\n", uint64(cSize), b) + } + return ErrWindowSizeExceeded + } + + b.RLESize = 0 + // We do not need a destination for raw blocks. + maxSize = -1 + default: + panic("Invalid block type") + } + + // Read block data. + if cap(b.dataStorage) < cSize { + if b.lowMem || cSize > maxCompressedBlockSize { + b.dataStorage = make([]byte, 0, cSize+compressedBlockOverAlloc) + } else { + b.dataStorage = make([]byte, 0, maxCompressedBlockSizeAlloc) + } + } + if cap(b.dst) <= maxSize { + b.dst = make([]byte, 0, maxSize+1) + } + b.data, err = br.readBig(cSize, b.dataStorage) + if err != nil { + if debugDecoder { + println("Reading block:", err, "(", cSize, ")", len(b.data)) + printf("%T", br) + } + return err + } + return nil +} + +// sendEOF will make the decoder send EOF on this frame. +func (b *blockDec) sendErr(err error) { + b.Last = true + b.Type = blockTypeReserved + b.err = err +} + +// Close will release resources. +// Closed blockDec cannot be reset. +func (b *blockDec) Close() { +} + +// decodeBuf +func (b *blockDec) decodeBuf(hist *history) error { + switch b.Type { + case blockTypeRLE: + if cap(b.dst) < int(b.RLESize) { + if b.lowMem { + b.dst = make([]byte, b.RLESize) + } else { + b.dst = make([]byte, maxBlockSize) + } + } + b.dst = b.dst[:b.RLESize] + v := b.data[0] + for i := range b.dst { + b.dst[i] = v + } + hist.appendKeep(b.dst) + return nil + case blockTypeRaw: + hist.appendKeep(b.data) + return nil + case blockTypeCompressed: + saved := b.dst + // Append directly to history + if hist.ignoreBuffer == 0 { + b.dst = hist.b + hist.b = nil + } else { + b.dst = b.dst[:0] + } + err := b.decodeCompressed(hist) + if debugDecoder { + println("Decompressed to total", len(b.dst), "bytes, hash:", xxhash.Sum64(b.dst), "error:", err) + } + if hist.ignoreBuffer == 0 { + hist.b = b.dst + b.dst = saved + } else { + hist.appendKeep(b.dst) + } + return err + case blockTypeReserved: + // Used for returning errors. + return b.err + default: + panic("Invalid block type") + } +} + +func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err error) { + // There must be at least one byte for Literals_Block_Type and one for Sequences_Section_Header + if len(in) < 2 { + return in, ErrBlockTooSmall + } + + litType := literalsBlockType(in[0] & 3) + var litRegenSize int + var litCompSize int + sizeFormat := (in[0] >> 2) & 3 + var fourStreams bool + var literals []byte + switch litType { + case literalsBlockRaw, literalsBlockRLE: + switch sizeFormat { + case 0, 2: + // Regenerated_Size uses 5 bits (0-31). Literals_Section_Header uses 1 byte. + litRegenSize = int(in[0] >> 3) + in = in[1:] + case 1: + // Regenerated_Size uses 12 bits (0-4095). Literals_Section_Header uses 2 bytes. + litRegenSize = int(in[0]>>4) + (int(in[1]) << 4) + in = in[2:] + case 3: + // Regenerated_Size uses 20 bits (0-1048575). Literals_Section_Header uses 3 bytes. + if len(in) < 3 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in)) + return in, ErrBlockTooSmall + } + litRegenSize = int(in[0]>>4) + (int(in[1]) << 4) + (int(in[2]) << 12) + in = in[3:] + } + case literalsBlockCompressed, literalsBlockTreeless: + switch sizeFormat { + case 0, 1: + // Both Regenerated_Size and Compressed_Size use 10 bits (0-1023). + if len(in) < 3 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in)) + return in, ErrBlockTooSmall + } + n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + litRegenSize = int(n & 1023) + litCompSize = int(n >> 10) + fourStreams = sizeFormat == 1 + in = in[3:] + case 2: + fourStreams = true + if len(in) < 4 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in)) + return in, ErrBlockTooSmall + } + n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20) + litRegenSize = int(n & 16383) + litCompSize = int(n >> 14) + in = in[4:] + case 3: + fourStreams = true + if len(in) < 5 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in)) + return in, ErrBlockTooSmall + } + n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20) + (uint64(in[4]) << 28) + litRegenSize = int(n & 262143) + litCompSize = int(n >> 18) + in = in[5:] + } + } + if debugDecoder { + println("literals type:", litType, "litRegenSize:", litRegenSize, "litCompSize:", litCompSize, "sizeFormat:", sizeFormat, "4X:", fourStreams) + } + if litRegenSize > int(b.WindowSize) || litRegenSize > maxCompressedBlockSize { + return in, ErrWindowSizeExceeded + } + + switch litType { + case literalsBlockRaw: + if len(in) < litRegenSize { + println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litRegenSize) + return in, ErrBlockTooSmall + } + literals = in[:litRegenSize] + in = in[litRegenSize:] + //printf("Found %d uncompressed literals\n", litRegenSize) + case literalsBlockRLE: + if len(in) < 1 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", 1) + return in, ErrBlockTooSmall + } + if cap(b.literalBuf) < litRegenSize { + if b.lowMem { + b.literalBuf = make([]byte, litRegenSize, litRegenSize+compressedBlockOverAlloc) + } else { + b.literalBuf = make([]byte, litRegenSize, maxCompressedBlockSize+compressedBlockOverAlloc) + } + } + literals = b.literalBuf[:litRegenSize] + v := in[0] + for i := range literals { + literals[i] = v + } + in = in[1:] + if debugDecoder { + printf("Found %d RLE compressed literals\n", litRegenSize) + } + case literalsBlockTreeless: + if len(in) < litCompSize { + println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize) + return in, ErrBlockTooSmall + } + // Store compressed literals, so we defer decoding until we get history. + literals = in[:litCompSize] + in = in[litCompSize:] + if debugDecoder { + printf("Found %d compressed literals\n", litCompSize) + } + huff := hist.huffTree + if huff == nil { + return in, errors.New("literal block was treeless, but no history was defined") + } + // Ensure we have space to store it. + if cap(b.literalBuf) < litRegenSize { + if b.lowMem { + b.literalBuf = make([]byte, 0, litRegenSize+compressedBlockOverAlloc) + } else { + b.literalBuf = make([]byte, 0, maxCompressedBlockSize+compressedBlockOverAlloc) + } + } + var err error + // Use our out buffer. + huff.MaxDecodedSize = litRegenSize + if fourStreams { + literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals) + } else { + literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals) + } + // Make sure we don't leak our literals buffer + if err != nil { + println("decompressing literals:", err) + return in, err + } + if len(literals) != litRegenSize { + return in, fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals)) + } + + case literalsBlockCompressed: + if len(in) < litCompSize { + println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize) + return in, ErrBlockTooSmall + } + literals = in[:litCompSize] + in = in[litCompSize:] + // Ensure we have space to store it. + if cap(b.literalBuf) < litRegenSize { + if b.lowMem { + b.literalBuf = make([]byte, 0, litRegenSize+compressedBlockOverAlloc) + } else { + b.literalBuf = make([]byte, 0, maxCompressedBlockSize+compressedBlockOverAlloc) + } + } + huff := hist.huffTree + if huff == nil || (hist.dict != nil && huff == hist.dict.litEnc) { + huff = huffDecoderPool.Get().(*huff0.Scratch) + if huff == nil { + huff = &huff0.Scratch{} + } + } + var err error + huff, literals, err = huff0.ReadTable(literals, huff) + if err != nil { + println("reading huffman table:", err) + return in, err + } + hist.huffTree = huff + huff.MaxDecodedSize = litRegenSize + // Use our out buffer. + if fourStreams { + literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals) + } else { + literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals) + } + if err != nil { + println("decoding compressed literals:", err) + return in, err + } + // Make sure we don't leak our literals buffer + if len(literals) != litRegenSize { + return in, fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals)) + } + // Re-cap to get extra size. + literals = b.literalBuf[:len(literals)] + if debugDecoder { + printf("Decompressed %d literals into %d bytes\n", litCompSize, litRegenSize) + } + } + hist.decoders.literals = literals + return in, nil +} + +// decodeCompressed will start decompressing a block. +func (b *blockDec) decodeCompressed(hist *history) error { + in := b.data + in, err := b.decodeLiterals(in, hist) + if err != nil { + return err + } + err = b.prepareSequences(in, hist) + if err != nil { + return err + } + if hist.decoders.nSeqs == 0 { + b.dst = append(b.dst, hist.decoders.literals...) + return nil + } + before := len(hist.decoders.out) + err = hist.decoders.decodeSync(hist.b[hist.ignoreBuffer:]) + if err != nil { + return err + } + if hist.decoders.maxSyncLen > 0 { + hist.decoders.maxSyncLen += uint64(before) + hist.decoders.maxSyncLen -= uint64(len(hist.decoders.out)) + } + b.dst = hist.decoders.out + hist.recentOffsets = hist.decoders.prevOffset + return nil +} + +func (b *blockDec) prepareSequences(in []byte, hist *history) (err error) { + if debugDecoder { + printf("prepareSequences: %d byte(s) input\n", len(in)) + } + // Decode Sequences + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#sequences-section + if len(in) < 1 { + return ErrBlockTooSmall + } + var nSeqs int + seqHeader := in[0] + switch { + case seqHeader < 128: + nSeqs = int(seqHeader) + in = in[1:] + case seqHeader < 255: + if len(in) < 2 { + return ErrBlockTooSmall + } + nSeqs = int(seqHeader-128)<<8 | int(in[1]) + in = in[2:] + case seqHeader == 255: + if len(in) < 3 { + return ErrBlockTooSmall + } + nSeqs = 0x7f00 + int(in[1]) + (int(in[2]) << 8) + in = in[3:] + } + if nSeqs == 0 && len(in) != 0 { + // When no sequences, there should not be any more data... + if debugDecoder { + printf("prepareSequences: 0 sequences, but %d byte(s) left on stream\n", len(in)) + } + return ErrUnexpectedBlockSize + } + + var seqs = &hist.decoders + seqs.nSeqs = nSeqs + if nSeqs > 0 { + if len(in) < 1 { + return ErrBlockTooSmall + } + br := byteReader{b: in, off: 0} + compMode := br.Uint8() + br.advance(1) + if debugDecoder { + printf("Compression modes: 0b%b", compMode) + } + for i := uint(0); i < 3; i++ { + mode := seqCompMode((compMode >> (6 - i*2)) & 3) + if debugDecoder { + println("Table", tableIndex(i), "is", mode) + } + var seq *sequenceDec + switch tableIndex(i) { + case tableLiteralLengths: + seq = &seqs.litLengths + case tableOffsets: + seq = &seqs.offsets + case tableMatchLengths: + seq = &seqs.matchLengths + default: + panic("unknown table") + } + switch mode { + case compModePredefined: + if seq.fse != nil && !seq.fse.preDefined { + fseDecoderPool.Put(seq.fse) + } + seq.fse = &fsePredef[i] + case compModeRLE: + if br.remain() < 1 { + return ErrBlockTooSmall + } + v := br.Uint8() + br.advance(1) + if seq.fse == nil || seq.fse.preDefined { + seq.fse = fseDecoderPool.Get().(*fseDecoder) + } + symb, err := decSymbolValue(v, symbolTableX[i]) + if err != nil { + printf("RLE Transform table (%v) error: %v", tableIndex(i), err) + return err + } + seq.fse.setRLE(symb) + if debugDecoder { + printf("RLE set to %+v, code: %v", symb, v) + } + case compModeFSE: + println("Reading table for", tableIndex(i)) + if seq.fse == nil || seq.fse.preDefined { + seq.fse = fseDecoderPool.Get().(*fseDecoder) + } + err := seq.fse.readNCount(&br, uint16(maxTableSymbol[i])) + if err != nil { + println("Read table error:", err) + return err + } + err = seq.fse.transform(symbolTableX[i]) + if err != nil { + println("Transform table error:", err) + return err + } + if debugDecoder { + println("Read table ok", "symbolLen:", seq.fse.symbolLen) + } + case compModeRepeat: + seq.repeat = true + } + if br.overread() { + return io.ErrUnexpectedEOF + } + } + in = br.unread() + } + if debugDecoder { + println("Literals:", len(seqs.literals), "hash:", xxhash.Sum64(seqs.literals), "and", seqs.nSeqs, "sequences.") + } + + if nSeqs == 0 { + if len(b.sequence) > 0 { + b.sequence = b.sequence[:0] + } + return nil + } + br := seqs.br + if br == nil { + br = &bitReader{} + } + if err := br.init(in); err != nil { + return err + } + + if err := seqs.initialize(br, hist, b.dst); err != nil { + println("initializing sequences:", err) + return err + } + // Extract blocks... + if false && hist.dict == nil { + fatalErr := func(err error) { + if err != nil { + panic(err) + } + } + fn := fmt.Sprintf("n-%d-lits-%d-prev-%d-%d-%d-win-%d.blk", hist.decoders.nSeqs, len(hist.decoders.literals), hist.recentOffsets[0], hist.recentOffsets[1], hist.recentOffsets[2], hist.windowSize) + var buf bytes.Buffer + fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.litLengths.fse)) + fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.matchLengths.fse)) + fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.offsets.fse)) + buf.Write(in) + ioutil.WriteFile(filepath.Join("testdata", "seqs", fn), buf.Bytes(), os.ModePerm) + } + + return nil +} + +func (b *blockDec) decodeSequences(hist *history) error { + if cap(b.sequence) < hist.decoders.nSeqs { + if b.lowMem { + b.sequence = make([]seqVals, 0, hist.decoders.nSeqs) + } else { + b.sequence = make([]seqVals, 0, 0x7F00+0xffff) + } + } + b.sequence = b.sequence[:hist.decoders.nSeqs] + if hist.decoders.nSeqs == 0 { + hist.decoders.seqSize = len(hist.decoders.literals) + return nil + } + hist.decoders.windowSize = hist.windowSize + hist.decoders.prevOffset = hist.recentOffsets + + err := hist.decoders.decode(b.sequence) + hist.recentOffsets = hist.decoders.prevOffset + return err +} + +func (b *blockDec) executeSequences(hist *history) error { + hbytes := hist.b + if len(hbytes) > hist.windowSize { + hbytes = hbytes[len(hbytes)-hist.windowSize:] + // We do not need history anymore. + if hist.dict != nil { + hist.dict.content = nil + } + } + hist.decoders.windowSize = hist.windowSize + hist.decoders.out = b.dst[:0] + err := hist.decoders.execute(b.sequence, hbytes) + if err != nil { + return err + } + return b.updateHistory(hist) +} + +func (b *blockDec) updateHistory(hist *history) error { + if len(b.data) > maxCompressedBlockSize { + return fmt.Errorf("compressed block size too large (%d)", len(b.data)) + } + // Set output and release references. + b.dst = hist.decoders.out + hist.recentOffsets = hist.decoders.prevOffset + + if b.Last { + // if last block we don't care about history. + println("Last block, no history returned") + hist.b = hist.b[:0] + return nil + } else { + hist.append(b.dst) + if debugDecoder { + println("Finished block with ", len(b.sequence), "sequences. Added", len(b.dst), "to history, now length", len(hist.b)) + } + } + hist.decoders.out, hist.decoders.literals = nil, nil + + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/blockenc.go b/vendor/github.com/klauspost/compress/zstd/blockenc.go new file mode 100644 index 0000000..12e8f6f --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/blockenc.go @@ -0,0 +1,871 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "math" + "math/bits" + + "github.com/klauspost/compress/huff0" +) + +type blockEnc struct { + size int + literals []byte + sequences []seq + coders seqCoders + litEnc *huff0.Scratch + dictLitEnc *huff0.Scratch + wr bitWriter + + extraLits int + output []byte + recentOffsets [3]uint32 + prevRecentOffsets [3]uint32 + + last bool + lowMem bool +} + +// init should be used once the block has been created. +// If called more than once, the effect is the same as calling reset. +func (b *blockEnc) init() { + if b.lowMem { + // 1K literals + if cap(b.literals) < 1<<10 { + b.literals = make([]byte, 0, 1<<10) + } + const defSeqs = 20 + if cap(b.sequences) < defSeqs { + b.sequences = make([]seq, 0, defSeqs) + } + // 1K + if cap(b.output) < 1<<10 { + b.output = make([]byte, 0, 1<<10) + } + } else { + if cap(b.literals) < maxCompressedBlockSize { + b.literals = make([]byte, 0, maxCompressedBlockSize) + } + const defSeqs = 2000 + if cap(b.sequences) < defSeqs { + b.sequences = make([]seq, 0, defSeqs) + } + if cap(b.output) < maxCompressedBlockSize { + b.output = make([]byte, 0, maxCompressedBlockSize) + } + } + + if b.coders.mlEnc == nil { + b.coders.mlEnc = &fseEncoder{} + b.coders.mlPrev = &fseEncoder{} + b.coders.ofEnc = &fseEncoder{} + b.coders.ofPrev = &fseEncoder{} + b.coders.llEnc = &fseEncoder{} + b.coders.llPrev = &fseEncoder{} + } + b.litEnc = &huff0.Scratch{WantLogLess: 4} + b.reset(nil) +} + +// initNewEncode can be used to reset offsets and encoders to the initial state. +func (b *blockEnc) initNewEncode() { + b.recentOffsets = [3]uint32{1, 4, 8} + b.litEnc.Reuse = huff0.ReusePolicyNone + b.coders.setPrev(nil, nil, nil) +} + +// reset will reset the block for a new encode, but in the same stream, +// meaning that state will be carried over, but the block content is reset. +// If a previous block is provided, the recent offsets are carried over. +func (b *blockEnc) reset(prev *blockEnc) { + b.extraLits = 0 + b.literals = b.literals[:0] + b.size = 0 + b.sequences = b.sequences[:0] + b.output = b.output[:0] + b.last = false + if prev != nil { + b.recentOffsets = prev.prevRecentOffsets + } + b.dictLitEnc = nil +} + +// reset will reset the block for a new encode, but in the same stream, +// meaning that state will be carried over, but the block content is reset. +// If a previous block is provided, the recent offsets are carried over. +func (b *blockEnc) swapEncoders(prev *blockEnc) { + b.coders.swap(&prev.coders) + b.litEnc, prev.litEnc = prev.litEnc, b.litEnc +} + +// blockHeader contains the information for a block header. +type blockHeader uint32 + +// setLast sets the 'last' indicator on a block. +func (h *blockHeader) setLast(b bool) { + if b { + *h = *h | 1 + } else { + const mask = (1 << 24) - 2 + *h = *h & mask + } +} + +// setSize will store the compressed size of a block. +func (h *blockHeader) setSize(v uint32) { + const mask = 7 + *h = (*h)&mask | blockHeader(v<<3) +} + +// setType sets the block type. +func (h *blockHeader) setType(t blockType) { + const mask = 1 | (((1 << 24) - 1) ^ 7) + *h = (*h & mask) | blockHeader(t<<1) +} + +// appendTo will append the block header to a slice. +func (h blockHeader) appendTo(b []byte) []byte { + return append(b, uint8(h), uint8(h>>8), uint8(h>>16)) +} + +// String returns a string representation of the block. +func (h blockHeader) String() string { + return fmt.Sprintf("Type: %d, Size: %d, Last:%t", (h>>1)&3, h>>3, h&1 == 1) +} + +// literalsHeader contains literals header information. +type literalsHeader uint64 + +// setType can be used to set the type of literal block. +func (h *literalsHeader) setType(t literalsBlockType) { + const mask = math.MaxUint64 - 3 + *h = (*h & mask) | literalsHeader(t) +} + +// setSize can be used to set a single size, for uncompressed and RLE content. +func (h *literalsHeader) setSize(regenLen int) { + inBits := bits.Len32(uint32(regenLen)) + // Only retain 2 bits + const mask = 3 + lh := uint64(*h & mask) + switch { + case inBits < 5: + lh |= (uint64(regenLen) << 3) | (1 << 60) + if debugEncoder { + got := int(lh>>3) & 0xff + if got != regenLen { + panic(fmt.Sprint("litRegenSize = ", regenLen, "(want) != ", got, "(got)")) + } + } + case inBits < 12: + lh |= (1 << 2) | (uint64(regenLen) << 4) | (2 << 60) + case inBits < 20: + lh |= (3 << 2) | (uint64(regenLen) << 4) | (3 << 60) + default: + panic(fmt.Errorf("internal error: block too big (%d)", regenLen)) + } + *h = literalsHeader(lh) +} + +// setSizes will set the size of a compressed literals section and the input length. +func (h *literalsHeader) setSizes(compLen, inLen int, single bool) { + compBits, inBits := bits.Len32(uint32(compLen)), bits.Len32(uint32(inLen)) + // Only retain 2 bits + const mask = 3 + lh := uint64(*h & mask) + switch { + case compBits <= 10 && inBits <= 10: + if !single { + lh |= 1 << 2 + } + lh |= (uint64(inLen) << 4) | (uint64(compLen) << (10 + 4)) | (3 << 60) + if debugEncoder { + const mmask = (1 << 24) - 1 + n := (lh >> 4) & mmask + if int(n&1023) != inLen { + panic(fmt.Sprint("regensize:", int(n&1023), "!=", inLen, inBits)) + } + if int(n>>10) != compLen { + panic(fmt.Sprint("compsize:", int(n>>10), "!=", compLen, compBits)) + } + } + case compBits <= 14 && inBits <= 14: + lh |= (2 << 2) | (uint64(inLen) << 4) | (uint64(compLen) << (14 + 4)) | (4 << 60) + if single { + panic("single stream used with more than 10 bits length.") + } + case compBits <= 18 && inBits <= 18: + lh |= (3 << 2) | (uint64(inLen) << 4) | (uint64(compLen) << (18 + 4)) | (5 << 60) + if single { + panic("single stream used with more than 10 bits length.") + } + default: + panic("internal error: block too big") + } + *h = literalsHeader(lh) +} + +// appendTo will append the literals header to a byte slice. +func (h literalsHeader) appendTo(b []byte) []byte { + size := uint8(h >> 60) + switch size { + case 1: + b = append(b, uint8(h)) + case 2: + b = append(b, uint8(h), uint8(h>>8)) + case 3: + b = append(b, uint8(h), uint8(h>>8), uint8(h>>16)) + case 4: + b = append(b, uint8(h), uint8(h>>8), uint8(h>>16), uint8(h>>24)) + case 5: + b = append(b, uint8(h), uint8(h>>8), uint8(h>>16), uint8(h>>24), uint8(h>>32)) + default: + panic(fmt.Errorf("internal error: literalsHeader has invalid size (%d)", size)) + } + return b +} + +// size returns the output size with currently set values. +func (h literalsHeader) size() int { + return int(h >> 60) +} + +func (h literalsHeader) String() string { + return fmt.Sprintf("Type: %d, SizeFormat: %d, Size: 0x%d, Bytes:%d", literalsBlockType(h&3), (h>>2)&3, h&((1<<60)-1)>>4, h>>60) +} + +// pushOffsets will push the recent offsets to the backup store. +func (b *blockEnc) pushOffsets() { + b.prevRecentOffsets = b.recentOffsets +} + +// pushOffsets will push the recent offsets to the backup store. +func (b *blockEnc) popOffsets() { + b.recentOffsets = b.prevRecentOffsets +} + +// matchOffset will adjust recent offsets and return the adjusted one, +// if it matches a previous offset. +func (b *blockEnc) matchOffset(offset, lits uint32) uint32 { + // Check if offset is one of the recent offsets. + // Adjusts the output offset accordingly. + // Gives a tiny bit of compression, typically around 1%. + if true { + if lits > 0 { + switch offset { + case b.recentOffsets[0]: + offset = 1 + case b.recentOffsets[1]: + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 2 + case b.recentOffsets[2]: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 3 + default: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset += 3 + } + } else { + switch offset { + case b.recentOffsets[1]: + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 1 + case b.recentOffsets[2]: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 2 + case b.recentOffsets[0] - 1: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 3 + default: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset += 3 + } + } + } else { + offset += 3 + } + return offset +} + +// encodeRaw can be used to set the output to a raw representation of supplied bytes. +func (b *blockEnc) encodeRaw(a []byte) { + var bh blockHeader + bh.setLast(b.last) + bh.setSize(uint32(len(a))) + bh.setType(blockTypeRaw) + b.output = bh.appendTo(b.output[:0]) + b.output = append(b.output, a...) + if debugEncoder { + println("Adding RAW block, length", len(a), "last:", b.last) + } +} + +// encodeRaw can be used to set the output to a raw representation of supplied bytes. +func (b *blockEnc) encodeRawTo(dst, src []byte) []byte { + var bh blockHeader + bh.setLast(b.last) + bh.setSize(uint32(len(src))) + bh.setType(blockTypeRaw) + dst = bh.appendTo(dst) + dst = append(dst, src...) + if debugEncoder { + println("Adding RAW block, length", len(src), "last:", b.last) + } + return dst +} + +// encodeLits can be used if the block is only litLen. +func (b *blockEnc) encodeLits(lits []byte, raw bool) error { + var bh blockHeader + bh.setLast(b.last) + bh.setSize(uint32(len(lits))) + + // Don't compress extremely small blocks + if len(lits) < 8 || (len(lits) < 32 && b.dictLitEnc == nil) || raw { + if debugEncoder { + println("Adding RAW block, length", len(lits), "last:", b.last) + } + bh.setType(blockTypeRaw) + b.output = bh.appendTo(b.output) + b.output = append(b.output, lits...) + return nil + } + + var ( + out []byte + reUsed, single bool + err error + ) + if b.dictLitEnc != nil { + b.litEnc.TransferCTable(b.dictLitEnc) + b.litEnc.Reuse = huff0.ReusePolicyAllow + b.dictLitEnc = nil + } + if len(lits) >= 1024 { + // Use 4 Streams. + out, reUsed, err = huff0.Compress4X(lits, b.litEnc) + } else if len(lits) > 32 { + // Use 1 stream + single = true + out, reUsed, err = huff0.Compress1X(lits, b.litEnc) + } else { + err = huff0.ErrIncompressible + } + + switch err { + case huff0.ErrIncompressible: + if debugEncoder { + println("Adding RAW block, length", len(lits), "last:", b.last) + } + bh.setType(blockTypeRaw) + b.output = bh.appendTo(b.output) + b.output = append(b.output, lits...) + return nil + case huff0.ErrUseRLE: + if debugEncoder { + println("Adding RLE block, length", len(lits)) + } + bh.setType(blockTypeRLE) + b.output = bh.appendTo(b.output) + b.output = append(b.output, lits[0]) + return nil + case nil: + default: + return err + } + // Compressed... + // Now, allow reuse + b.litEnc.Reuse = huff0.ReusePolicyAllow + bh.setType(blockTypeCompressed) + var lh literalsHeader + if reUsed { + if debugEncoder { + println("Reused tree, compressed to", len(out)) + } + lh.setType(literalsBlockTreeless) + } else { + if debugEncoder { + println("New tree, compressed to", len(out), "tree size:", len(b.litEnc.OutTable)) + } + lh.setType(literalsBlockCompressed) + } + // Set sizes + lh.setSizes(len(out), len(lits), single) + bh.setSize(uint32(len(out) + lh.size() + 1)) + + // Write block headers. + b.output = bh.appendTo(b.output) + b.output = lh.appendTo(b.output) + // Add compressed data. + b.output = append(b.output, out...) + // No sequences. + b.output = append(b.output, 0) + return nil +} + +// fuzzFseEncoder can be used to fuzz the FSE encoder. +func fuzzFseEncoder(data []byte) int { + if len(data) > maxSequences || len(data) < 2 { + return 0 + } + enc := fseEncoder{} + hist := enc.Histogram() + maxSym := uint8(0) + for i, v := range data { + v = v & 63 + data[i] = v + hist[v]++ + if v > maxSym { + maxSym = v + } + } + if maxSym == 0 { + // All 0 + return 0 + } + maxCount := func(a []uint32) int { + var max uint32 + for _, v := range a { + if v > max { + max = v + } + } + return int(max) + } + cnt := maxCount(hist[:maxSym]) + if cnt == len(data) { + // RLE + return 0 + } + enc.HistogramFinished(maxSym, cnt) + err := enc.normalizeCount(len(data)) + if err != nil { + return 0 + } + _, err = enc.writeCount(nil) + if err != nil { + panic(err) + } + return 1 +} + +// encode will encode the block and append the output in b.output. +// Previous offset codes must be pushed if more blocks are expected. +func (b *blockEnc) encode(org []byte, raw, rawAllLits bool) error { + if len(b.sequences) == 0 { + return b.encodeLits(b.literals, rawAllLits) + } + // We want some difference to at least account for the headers. + saved := b.size - len(b.literals) - (b.size >> 5) + if saved < 16 { + if org == nil { + return errIncompressible + } + b.popOffsets() + return b.encodeLits(org, rawAllLits) + } + + var bh blockHeader + var lh literalsHeader + bh.setLast(b.last) + bh.setType(blockTypeCompressed) + // Store offset of the block header. Needed when we know the size. + bhOffset := len(b.output) + b.output = bh.appendTo(b.output) + + var ( + out []byte + reUsed, single bool + err error + ) + if b.dictLitEnc != nil { + b.litEnc.TransferCTable(b.dictLitEnc) + b.litEnc.Reuse = huff0.ReusePolicyAllow + b.dictLitEnc = nil + } + if len(b.literals) >= 1024 && !raw { + // Use 4 Streams. + out, reUsed, err = huff0.Compress4X(b.literals, b.litEnc) + } else if len(b.literals) > 32 && !raw { + // Use 1 stream + single = true + out, reUsed, err = huff0.Compress1X(b.literals, b.litEnc) + } else { + err = huff0.ErrIncompressible + } + + switch err { + case huff0.ErrIncompressible: + lh.setType(literalsBlockRaw) + lh.setSize(len(b.literals)) + b.output = lh.appendTo(b.output) + b.output = append(b.output, b.literals...) + if debugEncoder { + println("Adding literals RAW, length", len(b.literals)) + } + case huff0.ErrUseRLE: + lh.setType(literalsBlockRLE) + lh.setSize(len(b.literals)) + b.output = lh.appendTo(b.output) + b.output = append(b.output, b.literals[0]) + if debugEncoder { + println("Adding literals RLE") + } + case nil: + // Compressed litLen... + if reUsed { + if debugEncoder { + println("reused tree") + } + lh.setType(literalsBlockTreeless) + } else { + if debugEncoder { + println("new tree, size:", len(b.litEnc.OutTable)) + } + lh.setType(literalsBlockCompressed) + if debugEncoder { + _, _, err := huff0.ReadTable(out, nil) + if err != nil { + panic(err) + } + } + } + lh.setSizes(len(out), len(b.literals), single) + if debugEncoder { + printf("Compressed %d literals to %d bytes", len(b.literals), len(out)) + println("Adding literal header:", lh) + } + b.output = lh.appendTo(b.output) + b.output = append(b.output, out...) + b.litEnc.Reuse = huff0.ReusePolicyAllow + if debugEncoder { + println("Adding literals compressed") + } + default: + if debugEncoder { + println("Adding literals ERROR:", err) + } + return err + } + // Sequence compression + + // Write the number of sequences + switch { + case len(b.sequences) < 128: + b.output = append(b.output, uint8(len(b.sequences))) + case len(b.sequences) < 0x7f00: // TODO: this could be wrong + n := len(b.sequences) + b.output = append(b.output, 128+uint8(n>>8), uint8(n)) + default: + n := len(b.sequences) - 0x7f00 + b.output = append(b.output, 255, uint8(n), uint8(n>>8)) + } + if debugEncoder { + println("Encoding", len(b.sequences), "sequences") + } + b.genCodes() + llEnc := b.coders.llEnc + ofEnc := b.coders.ofEnc + mlEnc := b.coders.mlEnc + err = llEnc.normalizeCount(len(b.sequences)) + if err != nil { + return err + } + err = ofEnc.normalizeCount(len(b.sequences)) + if err != nil { + return err + } + err = mlEnc.normalizeCount(len(b.sequences)) + if err != nil { + return err + } + + // Choose the best compression mode for each type. + // Will evaluate the new vs predefined and previous. + chooseComp := func(cur, prev, preDef *fseEncoder) (*fseEncoder, seqCompMode) { + // See if predefined/previous is better + hist := cur.count[:cur.symbolLen] + nSize := cur.approxSize(hist) + cur.maxHeaderSize() + predefSize := preDef.approxSize(hist) + prevSize := prev.approxSize(hist) + + // Add a small penalty for new encoders. + // Don't bother with extremely small (<2 byte gains). + nSize = nSize + (nSize+2*8*16)>>4 + switch { + case predefSize <= prevSize && predefSize <= nSize || forcePreDef: + if debugEncoder { + println("Using predefined", predefSize>>3, "<=", nSize>>3) + } + return preDef, compModePredefined + case prevSize <= nSize: + if debugEncoder { + println("Using previous", prevSize>>3, "<=", nSize>>3) + } + return prev, compModeRepeat + default: + if debugEncoder { + println("Using new, predef", predefSize>>3, ". previous:", prevSize>>3, ">", nSize>>3, "header max:", cur.maxHeaderSize()>>3, "bytes") + println("tl:", cur.actualTableLog, "symbolLen:", cur.symbolLen, "norm:", cur.norm[:cur.symbolLen], "hist", cur.count[:cur.symbolLen]) + } + return cur, compModeFSE + } + } + + // Write compression mode + var mode uint8 + if llEnc.useRLE { + mode |= uint8(compModeRLE) << 6 + llEnc.setRLE(b.sequences[0].llCode) + if debugEncoder { + println("llEnc.useRLE") + } + } else { + var m seqCompMode + llEnc, m = chooseComp(llEnc, b.coders.llPrev, &fsePredefEnc[tableLiteralLengths]) + mode |= uint8(m) << 6 + } + if ofEnc.useRLE { + mode |= uint8(compModeRLE) << 4 + ofEnc.setRLE(b.sequences[0].ofCode) + if debugEncoder { + println("ofEnc.useRLE") + } + } else { + var m seqCompMode + ofEnc, m = chooseComp(ofEnc, b.coders.ofPrev, &fsePredefEnc[tableOffsets]) + mode |= uint8(m) << 4 + } + + if mlEnc.useRLE { + mode |= uint8(compModeRLE) << 2 + mlEnc.setRLE(b.sequences[0].mlCode) + if debugEncoder { + println("mlEnc.useRLE, code: ", b.sequences[0].mlCode, "value", b.sequences[0].matchLen) + } + } else { + var m seqCompMode + mlEnc, m = chooseComp(mlEnc, b.coders.mlPrev, &fsePredefEnc[tableMatchLengths]) + mode |= uint8(m) << 2 + } + b.output = append(b.output, mode) + if debugEncoder { + printf("Compression modes: 0b%b", mode) + } + b.output, err = llEnc.writeCount(b.output) + if err != nil { + return err + } + start := len(b.output) + b.output, err = ofEnc.writeCount(b.output) + if err != nil { + return err + } + if false { + println("block:", b.output[start:], "tablelog", ofEnc.actualTableLog, "maxcount:", ofEnc.maxCount) + fmt.Printf("selected TableLog: %d, Symbol length: %d\n", ofEnc.actualTableLog, ofEnc.symbolLen) + for i, v := range ofEnc.norm[:ofEnc.symbolLen] { + fmt.Printf("%3d: %5d -> %4d \n", i, ofEnc.count[i], v) + } + } + b.output, err = mlEnc.writeCount(b.output) + if err != nil { + return err + } + + // Maybe in block? + wr := &b.wr + wr.reset(b.output) + + var ll, of, ml cState + + // Current sequence + seq := len(b.sequences) - 1 + s := b.sequences[seq] + llEnc.setBits(llBitsTable[:]) + mlEnc.setBits(mlBitsTable[:]) + ofEnc.setBits(nil) + + llTT, ofTT, mlTT := llEnc.ct.symbolTT[:256], ofEnc.ct.symbolTT[:256], mlEnc.ct.symbolTT[:256] + + // We have 3 bounds checks here (and in the loop). + // Since we are iterating backwards it is kinda hard to avoid. + llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode] + ll.init(wr, &llEnc.ct, llB) + of.init(wr, &ofEnc.ct, ofB) + wr.flush32() + ml.init(wr, &mlEnc.ct, mlB) + + // Each of these lookups also generates a bounds check. + wr.addBits32NC(s.litLen, llB.outBits) + wr.addBits32NC(s.matchLen, mlB.outBits) + wr.flush32() + wr.addBits32NC(s.offset, ofB.outBits) + if debugSequences { + println("Encoded seq", seq, s, "codes:", s.llCode, s.mlCode, s.ofCode, "states:", ll.state, ml.state, of.state, "bits:", llB, mlB, ofB) + } + seq-- + // Store sequences in reverse... + for seq >= 0 { + s = b.sequences[seq] + + ofB := ofTT[s.ofCode] + wr.flush32() // tablelog max is below 8 for each, so it will fill max 24 bits. + //of.encode(ofB) + nbBitsOut := (uint32(of.state) + ofB.deltaNbBits) >> 16 + dstState := int32(of.state>>(nbBitsOut&15)) + int32(ofB.deltaFindState) + wr.addBits16NC(of.state, uint8(nbBitsOut)) + of.state = of.stateTable[dstState] + + // Accumulate extra bits. + outBits := ofB.outBits & 31 + extraBits := uint64(s.offset & bitMask32[outBits]) + extraBitsN := outBits + + mlB := mlTT[s.mlCode] + //ml.encode(mlB) + nbBitsOut = (uint32(ml.state) + mlB.deltaNbBits) >> 16 + dstState = int32(ml.state>>(nbBitsOut&15)) + int32(mlB.deltaFindState) + wr.addBits16NC(ml.state, uint8(nbBitsOut)) + ml.state = ml.stateTable[dstState] + + outBits = mlB.outBits & 31 + extraBits = extraBits<> 16 + dstState = int32(ll.state>>(nbBitsOut&15)) + int32(llB.deltaFindState) + wr.addBits16NC(ll.state, uint8(nbBitsOut)) + ll.state = ll.stateTable[dstState] + + outBits = llB.outBits & 31 + extraBits = extraBits<= b.size { + // Maybe even add a bigger margin. + b.litEnc.Reuse = huff0.ReusePolicyNone + return errIncompressible + } + + // Size is output minus block header. + bh.setSize(uint32(len(b.output)-bhOffset) - 3) + if debugEncoder { + println("Rewriting block header", bh) + } + _ = bh.appendTo(b.output[bhOffset:bhOffset]) + b.coders.setPrev(llEnc, mlEnc, ofEnc) + return nil +} + +var errIncompressible = errors.New("incompressible") + +func (b *blockEnc) genCodes() { + if len(b.sequences) == 0 { + // nothing to do + return + } + if len(b.sequences) > math.MaxUint16 { + panic("can only encode up to 64K sequences") + } + // No bounds checks after here: + llH := b.coders.llEnc.Histogram() + ofH := b.coders.ofEnc.Histogram() + mlH := b.coders.mlEnc.Histogram() + for i := range llH { + llH[i] = 0 + } + for i := range ofH { + ofH[i] = 0 + } + for i := range mlH { + mlH[i] = 0 + } + + var llMax, ofMax, mlMax uint8 + for i := range b.sequences { + seq := &b.sequences[i] + v := llCode(seq.litLen) + seq.llCode = v + llH[v]++ + if v > llMax { + llMax = v + } + + v = ofCode(seq.offset) + seq.ofCode = v + ofH[v]++ + if v > ofMax { + ofMax = v + } + + v = mlCode(seq.matchLen) + seq.mlCode = v + mlH[v]++ + if v > mlMax { + mlMax = v + if debugAsserts && mlMax > maxMatchLengthSymbol { + panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d), matchlen: %d", mlMax, seq.matchLen)) + } + } + } + maxCount := func(a []uint32) int { + var max uint32 + for _, v := range a { + if v > max { + max = v + } + } + return int(max) + } + if debugAsserts && mlMax > maxMatchLengthSymbol { + panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d)", mlMax)) + } + if debugAsserts && ofMax > maxOffsetBits { + panic(fmt.Errorf("ofMax > maxOffsetBits (%d)", ofMax)) + } + if debugAsserts && llMax > maxLiteralLengthSymbol { + panic(fmt.Errorf("llMax > maxLiteralLengthSymbol (%d)", llMax)) + } + + b.coders.mlEnc.HistogramFinished(mlMax, maxCount(mlH[:mlMax+1])) + b.coders.ofEnc.HistogramFinished(ofMax, maxCount(ofH[:ofMax+1])) + b.coders.llEnc.HistogramFinished(llMax, maxCount(llH[:llMax+1])) +} diff --git a/vendor/github.com/klauspost/compress/zstd/blocktype_string.go b/vendor/github.com/klauspost/compress/zstd/blocktype_string.go new file mode 100644 index 0000000..01a01e4 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/blocktype_string.go @@ -0,0 +1,85 @@ +// Code generated by "stringer -type=blockType,literalsBlockType,seqCompMode,tableIndex"; DO NOT EDIT. + +package zstd + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[blockTypeRaw-0] + _ = x[blockTypeRLE-1] + _ = x[blockTypeCompressed-2] + _ = x[blockTypeReserved-3] +} + +const _blockType_name = "blockTypeRawblockTypeRLEblockTypeCompressedblockTypeReserved" + +var _blockType_index = [...]uint8{0, 12, 24, 43, 60} + +func (i blockType) String() string { + if i >= blockType(len(_blockType_index)-1) { + return "blockType(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _blockType_name[_blockType_index[i]:_blockType_index[i+1]] +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[literalsBlockRaw-0] + _ = x[literalsBlockRLE-1] + _ = x[literalsBlockCompressed-2] + _ = x[literalsBlockTreeless-3] +} + +const _literalsBlockType_name = "literalsBlockRawliteralsBlockRLEliteralsBlockCompressedliteralsBlockTreeless" + +var _literalsBlockType_index = [...]uint8{0, 16, 32, 55, 76} + +func (i literalsBlockType) String() string { + if i >= literalsBlockType(len(_literalsBlockType_index)-1) { + return "literalsBlockType(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _literalsBlockType_name[_literalsBlockType_index[i]:_literalsBlockType_index[i+1]] +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[compModePredefined-0] + _ = x[compModeRLE-1] + _ = x[compModeFSE-2] + _ = x[compModeRepeat-3] +} + +const _seqCompMode_name = "compModePredefinedcompModeRLEcompModeFSEcompModeRepeat" + +var _seqCompMode_index = [...]uint8{0, 18, 29, 40, 54} + +func (i seqCompMode) String() string { + if i >= seqCompMode(len(_seqCompMode_index)-1) { + return "seqCompMode(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _seqCompMode_name[_seqCompMode_index[i]:_seqCompMode_index[i+1]] +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[tableLiteralLengths-0] + _ = x[tableOffsets-1] + _ = x[tableMatchLengths-2] +} + +const _tableIndex_name = "tableLiteralLengthstableOffsetstableMatchLengths" + +var _tableIndex_index = [...]uint8{0, 19, 31, 48} + +func (i tableIndex) String() string { + if i >= tableIndex(len(_tableIndex_index)-1) { + return "tableIndex(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _tableIndex_name[_tableIndex_index[i]:_tableIndex_index[i+1]] +} diff --git a/vendor/github.com/klauspost/compress/zstd/bytebuf.go b/vendor/github.com/klauspost/compress/zstd/bytebuf.go new file mode 100644 index 0000000..2ad0207 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bytebuf.go @@ -0,0 +1,132 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "fmt" + "io" + "io/ioutil" +) + +type byteBuffer interface { + // Read up to 8 bytes. + // Returns io.ErrUnexpectedEOF if this cannot be satisfied. + readSmall(n int) ([]byte, error) + + // Read >8 bytes. + // MAY use the destination slice. + readBig(n int, dst []byte) ([]byte, error) + + // Read a single byte. + readByte() (byte, error) + + // Skip n bytes. + skipN(n int64) error +} + +// in-memory buffer +type byteBuf []byte + +func (b *byteBuf) readSmall(n int) ([]byte, error) { + if debugAsserts && n > 8 { + panic(fmt.Errorf("small read > 8 (%d). use readBig", n)) + } + bb := *b + if len(bb) < n { + return nil, io.ErrUnexpectedEOF + } + r := bb[:n] + *b = bb[n:] + return r, nil +} + +func (b *byteBuf) readBig(n int, dst []byte) ([]byte, error) { + bb := *b + if len(bb) < n { + return nil, io.ErrUnexpectedEOF + } + r := bb[:n] + *b = bb[n:] + return r, nil +} + +func (b *byteBuf) readByte() (byte, error) { + bb := *b + if len(bb) < 1 { + return 0, nil + } + r := bb[0] + *b = bb[1:] + return r, nil +} + +func (b *byteBuf) skipN(n int64) error { + bb := *b + if n < 0 { + return fmt.Errorf("negative skip (%d) requested", n) + } + if int64(len(bb)) < n { + return io.ErrUnexpectedEOF + } + *b = bb[n:] + return nil +} + +// wrapper around a reader. +type readerWrapper struct { + r io.Reader + tmp [8]byte +} + +func (r *readerWrapper) readSmall(n int) ([]byte, error) { + if debugAsserts && n > 8 { + panic(fmt.Errorf("small read > 8 (%d). use readBig", n)) + } + n2, err := io.ReadFull(r.r, r.tmp[:n]) + // We only really care about the actual bytes read. + if err != nil { + if err == io.EOF { + return nil, io.ErrUnexpectedEOF + } + if debugDecoder { + println("readSmall: got", n2, "want", n, "err", err) + } + return nil, err + } + return r.tmp[:n], nil +} + +func (r *readerWrapper) readBig(n int, dst []byte) ([]byte, error) { + if cap(dst) < n { + dst = make([]byte, n) + } + n2, err := io.ReadFull(r.r, dst[:n]) + if err == io.EOF && n > 0 { + err = io.ErrUnexpectedEOF + } + return dst[:n2], err +} + +func (r *readerWrapper) readByte() (byte, error) { + n2, err := r.r.Read(r.tmp[:1]) + if err != nil { + if err == io.EOF { + err = io.ErrUnexpectedEOF + } + return 0, err + } + if n2 != 1 { + return 0, io.ErrUnexpectedEOF + } + return r.tmp[0], nil +} + +func (r *readerWrapper) skipN(n int64) error { + n2, err := io.CopyN(ioutil.Discard, r.r, n) + if n2 != n { + err = io.ErrUnexpectedEOF + } + return err +} diff --git a/vendor/github.com/klauspost/compress/zstd/bytereader.go b/vendor/github.com/klauspost/compress/zstd/bytereader.go new file mode 100644 index 0000000..0e59a24 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bytereader.go @@ -0,0 +1,82 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +// byteReader provides a byte reader that reads +// little endian values from a byte stream. +// The input stream is manually advanced. +// The reader performs no bounds checks. +type byteReader struct { + b []byte + off int +} + +// advance the stream b n bytes. +func (b *byteReader) advance(n uint) { + b.off += int(n) +} + +// overread returns whether we have advanced too far. +func (b *byteReader) overread() bool { + return b.off > len(b.b) +} + +// Int32 returns a little endian int32 starting at current offset. +func (b byteReader) Int32() int32 { + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := int32(b2[3]) + v2 := int32(b2[2]) + v1 := int32(b2[1]) + v0 := int32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// Uint8 returns the next byte +func (b *byteReader) Uint8() uint8 { + v := b.b[b.off] + return v +} + +// Uint32 returns a little endian uint32 starting at current offset. +func (b byteReader) Uint32() uint32 { + if r := b.remain(); r < 4 { + // Very rare + v := uint32(0) + for i := 1; i <= r; i++ { + v = (v << 8) | uint32(b.b[len(b.b)-i]) + } + return v + } + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := uint32(b2[3]) + v2 := uint32(b2[2]) + v1 := uint32(b2[1]) + v0 := uint32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// Uint32NC returns a little endian uint32 starting at current offset. +// The caller must be sure if there are at least 4 bytes left. +func (b byteReader) Uint32NC() uint32 { + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := uint32(b2[3]) + v2 := uint32(b2[2]) + v1 := uint32(b2[1]) + v0 := uint32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// unread returns the unread portion of the input. +func (b byteReader) unread() []byte { + return b.b[b.off:] +} + +// remain will return the number of bytes remaining. +func (b byteReader) remain() int { + return len(b.b) - b.off +} diff --git a/vendor/github.com/klauspost/compress/zstd/decodeheader.go b/vendor/github.com/klauspost/compress/zstd/decodeheader.go new file mode 100644 index 0000000..5022e71 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/decodeheader.go @@ -0,0 +1,230 @@ +// Copyright 2020+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. + +package zstd + +import ( + "bytes" + "encoding/binary" + "errors" + "io" +) + +// HeaderMaxSize is the maximum size of a Frame and Block Header. +// If less is sent to Header.Decode it *may* still contain enough information. +const HeaderMaxSize = 14 + 3 + +// Header contains information about the first frame and block within that. +type Header struct { + // SingleSegment specifies whether the data is to be decompressed into a + // single contiguous memory segment. + // It implies that WindowSize is invalid and that FrameContentSize is valid. + SingleSegment bool + + // WindowSize is the window of data to keep while decoding. + // Will only be set if SingleSegment is false. + WindowSize uint64 + + // Dictionary ID. + // If 0, no dictionary. + DictionaryID uint32 + + // HasFCS specifies whether FrameContentSize has a valid value. + HasFCS bool + + // FrameContentSize is the expected uncompressed size of the entire frame. + FrameContentSize uint64 + + // Skippable will be true if the frame is meant to be skipped. + // This implies that FirstBlock.OK is false. + Skippable bool + + // SkippableID is the user-specific ID for the skippable frame. + // Valid values are between 0 to 15, inclusive. + SkippableID int + + // SkippableSize is the length of the user data to skip following + // the header. + SkippableSize uint32 + + // HeaderSize is the raw size of the frame header. + // + // For normal frames, it includes the size of the magic number and + // the size of the header (per section 3.1.1.1). + // It does not include the size for any data blocks (section 3.1.1.2) nor + // the size for the trailing content checksum. + // + // For skippable frames, this counts the size of the magic number + // along with the size of the size field of the payload. + // It does not include the size of the skippable payload itself. + // The total frame size is the HeaderSize plus the SkippableSize. + HeaderSize int + + // First block information. + FirstBlock struct { + // OK will be set if first block could be decoded. + OK bool + + // Is this the last block of a frame? + Last bool + + // Is the data compressed? + // If true CompressedSize will be populated. + // Unfortunately DecompressedSize cannot be determined + // without decoding the blocks. + Compressed bool + + // DecompressedSize is the expected decompressed size of the block. + // Will be 0 if it cannot be determined. + DecompressedSize int + + // CompressedSize of the data in the block. + // Does not include the block header. + // Will be equal to DecompressedSize if not Compressed. + CompressedSize int + } + + // If set there is a checksum present for the block content. + // The checksum field at the end is always 4 bytes long. + HasCheckSum bool +} + +// Decode the header from the beginning of the stream. +// This will decode the frame header and the first block header if enough bytes are provided. +// It is recommended to provide at least HeaderMaxSize bytes. +// If the frame header cannot be read an error will be returned. +// If there isn't enough input, io.ErrUnexpectedEOF is returned. +// The FirstBlock.OK will indicate if enough information was available to decode the first block header. +func (h *Header) Decode(in []byte) error { + *h = Header{} + if len(in) < 4 { + return io.ErrUnexpectedEOF + } + h.HeaderSize += 4 + b, in := in[:4], in[4:] + if !bytes.Equal(b, frameMagic) { + if !bytes.Equal(b[1:4], skippableFrameMagic) || b[0]&0xf0 != 0x50 { + return ErrMagicMismatch + } + if len(in) < 4 { + return io.ErrUnexpectedEOF + } + h.HeaderSize += 4 + h.Skippable = true + h.SkippableID = int(b[0] & 0xf) + h.SkippableSize = binary.LittleEndian.Uint32(in) + return nil + } + + // Read Window_Descriptor + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#window_descriptor + if len(in) < 1 { + return io.ErrUnexpectedEOF + } + fhd, in := in[0], in[1:] + h.HeaderSize++ + h.SingleSegment = fhd&(1<<5) != 0 + h.HasCheckSum = fhd&(1<<2) != 0 + if fhd&(1<<3) != 0 { + return errors.New("reserved bit set on frame header") + } + + if !h.SingleSegment { + if len(in) < 1 { + return io.ErrUnexpectedEOF + } + var wd byte + wd, in = in[0], in[1:] + h.HeaderSize++ + windowLog := 10 + (wd >> 3) + windowBase := uint64(1) << windowLog + windowAdd := (windowBase / 8) * uint64(wd&0x7) + h.WindowSize = windowBase + windowAdd + } + + // Read Dictionary_ID + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary_id + if size := fhd & 3; size != 0 { + if size == 3 { + size = 4 + } + if len(in) < int(size) { + return io.ErrUnexpectedEOF + } + b, in = in[:size], in[size:] + h.HeaderSize += int(size) + switch size { + case 1: + h.DictionaryID = uint32(b[0]) + case 2: + h.DictionaryID = uint32(b[0]) | (uint32(b[1]) << 8) + case 4: + h.DictionaryID = uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + } + } + + // Read Frame_Content_Size + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frame_content_size + var fcsSize int + v := fhd >> 6 + switch v { + case 0: + if h.SingleSegment { + fcsSize = 1 + } + default: + fcsSize = 1 << v + } + + if fcsSize > 0 { + h.HasFCS = true + if len(in) < fcsSize { + return io.ErrUnexpectedEOF + } + b, in = in[:fcsSize], in[fcsSize:] + h.HeaderSize += int(fcsSize) + switch fcsSize { + case 1: + h.FrameContentSize = uint64(b[0]) + case 2: + // When FCS_Field_Size is 2, the offset of 256 is added. + h.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) + 256 + case 4: + h.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) | (uint64(b[2]) << 16) | (uint64(b[3]) << 24) + case 8: + d1 := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + d2 := uint32(b[4]) | (uint32(b[5]) << 8) | (uint32(b[6]) << 16) | (uint32(b[7]) << 24) + h.FrameContentSize = uint64(d1) | (uint64(d2) << 32) + } + } + + // Frame Header done, we will not fail from now on. + if len(in) < 3 { + return nil + } + tmp := in[:3] + bh := uint32(tmp[0]) | (uint32(tmp[1]) << 8) | (uint32(tmp[2]) << 16) + h.FirstBlock.Last = bh&1 != 0 + blockType := blockType((bh >> 1) & 3) + // find size. + cSize := int(bh >> 3) + switch blockType { + case blockTypeReserved: + return nil + case blockTypeRLE: + h.FirstBlock.Compressed = true + h.FirstBlock.DecompressedSize = cSize + h.FirstBlock.CompressedSize = 1 + case blockTypeCompressed: + h.FirstBlock.Compressed = true + h.FirstBlock.CompressedSize = cSize + case blockTypeRaw: + h.FirstBlock.DecompressedSize = cSize + h.FirstBlock.CompressedSize = cSize + default: + panic("Invalid block type") + } + + h.FirstBlock.OK = true + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/decoder.go b/vendor/github.com/klauspost/compress/zstd/decoder.go new file mode 100644 index 0000000..d212f47 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/decoder.go @@ -0,0 +1,924 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "bytes" + "context" + "encoding/binary" + "io" + "sync" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +// Decoder provides decoding of zstandard streams. +// The decoder has been designed to operate without allocations after a warmup. +// This means that you should store the decoder for best performance. +// To re-use a stream decoder, use the Reset(r io.Reader) error to switch to another stream. +// A decoder can safely be re-used even if the previous stream failed. +// To release the resources, you must call the Close() function on a decoder. +type Decoder struct { + o decoderOptions + + // Unreferenced decoders, ready for use. + decoders chan *blockDec + + // Current read position used for Reader functionality. + current decoderState + + // sync stream decoding + syncStream struct { + decodedFrame uint64 + br readerWrapper + enabled bool + inFrame bool + } + + frame *frameDec + + // Custom dictionaries. + // Always uses copies. + dicts map[uint32]dict + + // streamWg is the waitgroup for all streams + streamWg sync.WaitGroup +} + +// decoderState is used for maintaining state when the decoder +// is used for streaming. +type decoderState struct { + // current block being written to stream. + decodeOutput + + // output in order to be written to stream. + output chan decodeOutput + + // cancel remaining output. + cancel context.CancelFunc + + // crc of current frame + crc *xxhash.Digest + + flushed bool +} + +var ( + // Check the interfaces we want to support. + _ = io.WriterTo(&Decoder{}) + _ = io.Reader(&Decoder{}) +) + +// NewReader creates a new decoder. +// A nil Reader can be provided in which case Reset can be used to start a decode. +// +// A Decoder can be used in two modes: +// +// 1) As a stream, or +// 2) For stateless decoding using DecodeAll. +// +// Only a single stream can be decoded concurrently, but the same decoder +// can run multiple concurrent stateless decodes. It is even possible to +// use stateless decodes while a stream is being decoded. +// +// The Reset function can be used to initiate a new stream, which is will considerably +// reduce the allocations normally caused by NewReader. +func NewReader(r io.Reader, opts ...DOption) (*Decoder, error) { + initPredefined() + var d Decoder + d.o.setDefault() + for _, o := range opts { + err := o(&d.o) + if err != nil { + return nil, err + } + } + d.current.crc = xxhash.New() + d.current.flushed = true + + if r == nil { + d.current.err = ErrDecoderNilInput + } + + // Transfer option dicts. + d.dicts = make(map[uint32]dict, len(d.o.dicts)) + for _, dc := range d.o.dicts { + d.dicts[dc.id] = dc + } + d.o.dicts = nil + + // Create decoders + d.decoders = make(chan *blockDec, d.o.concurrent) + for i := 0; i < d.o.concurrent; i++ { + dec := newBlockDec(d.o.lowMem) + dec.localFrame = newFrameDec(d.o) + d.decoders <- dec + } + + if r == nil { + return &d, nil + } + return &d, d.Reset(r) +} + +// Read bytes from the decompressed stream into p. +// Returns the number of bytes written and any error that occurred. +// When the stream is done, io.EOF will be returned. +func (d *Decoder) Read(p []byte) (int, error) { + var n int + for { + if len(d.current.b) > 0 { + filled := copy(p, d.current.b) + p = p[filled:] + d.current.b = d.current.b[filled:] + n += filled + } + if len(p) == 0 { + break + } + if len(d.current.b) == 0 { + // We have an error and no more data + if d.current.err != nil { + break + } + if !d.nextBlock(n == 0) { + return n, d.current.err + } + } + } + if len(d.current.b) > 0 { + if debugDecoder { + println("returning", n, "still bytes left:", len(d.current.b)) + } + // Only return error at end of block + return n, nil + } + if d.current.err != nil { + d.drainOutput() + } + if debugDecoder { + println("returning", n, d.current.err, len(d.decoders)) + } + return n, d.current.err +} + +// Reset will reset the decoder the supplied stream after the current has finished processing. +// Note that this functionality cannot be used after Close has been called. +// Reset can be called with a nil reader to release references to the previous reader. +// After being called with a nil reader, no other operations than Reset or DecodeAll or Close +// should be used. +func (d *Decoder) Reset(r io.Reader) error { + if d.current.err == ErrDecoderClosed { + return d.current.err + } + + d.drainOutput() + + d.syncStream.br.r = nil + if r == nil { + d.current.err = ErrDecoderNilInput + if len(d.current.b) > 0 { + d.current.b = d.current.b[:0] + } + d.current.flushed = true + return nil + } + + // If bytes buffer and < 5MB, do sync decoding anyway. + if bb, ok := r.(byter); ok && bb.Len() < 5<<20 { + bb2 := bb + if debugDecoder { + println("*bytes.Buffer detected, doing sync decode, len:", bb.Len()) + } + b := bb2.Bytes() + var dst []byte + if cap(d.current.b) > 0 { + dst = d.current.b + } + + dst, err := d.DecodeAll(b, dst[:0]) + if err == nil { + err = io.EOF + } + d.current.b = dst + d.current.err = err + d.current.flushed = true + if debugDecoder { + println("sync decode to", len(dst), "bytes, err:", err) + } + return nil + } + // Remove current block. + d.stashDecoder() + d.current.decodeOutput = decodeOutput{} + d.current.err = nil + d.current.flushed = false + d.current.d = nil + + // Ensure no-one else is still running... + d.streamWg.Wait() + if d.frame == nil { + d.frame = newFrameDec(d.o) + } + + if d.o.concurrent == 1 { + return d.startSyncDecoder(r) + } + + d.current.output = make(chan decodeOutput, d.o.concurrent) + ctx, cancel := context.WithCancel(context.Background()) + d.current.cancel = cancel + d.streamWg.Add(1) + go d.startStreamDecoder(ctx, r, d.current.output) + + return nil +} + +// drainOutput will drain the output until errEndOfStream is sent. +func (d *Decoder) drainOutput() { + if d.current.cancel != nil { + if debugDecoder { + println("cancelling current") + } + d.current.cancel() + d.current.cancel = nil + } + if d.current.d != nil { + if debugDecoder { + printf("re-adding current decoder %p, decoders: %d", d.current.d, len(d.decoders)) + } + d.decoders <- d.current.d + d.current.d = nil + d.current.b = nil + } + if d.current.output == nil || d.current.flushed { + println("current already flushed") + return + } + for v := range d.current.output { + if v.d != nil { + if debugDecoder { + printf("re-adding decoder %p", v.d) + } + d.decoders <- v.d + } + } + d.current.output = nil + d.current.flushed = true +} + +// WriteTo writes data to w until there's no more data to write or when an error occurs. +// The return value n is the number of bytes written. +// Any error encountered during the write is also returned. +func (d *Decoder) WriteTo(w io.Writer) (int64, error) { + var n int64 + for { + if len(d.current.b) > 0 { + n2, err2 := w.Write(d.current.b) + n += int64(n2) + if err2 != nil && (d.current.err == nil || d.current.err == io.EOF) { + d.current.err = err2 + } else if n2 != len(d.current.b) { + d.current.err = io.ErrShortWrite + } + } + if d.current.err != nil { + break + } + d.nextBlock(true) + } + err := d.current.err + if err != nil { + d.drainOutput() + } + if err == io.EOF { + err = nil + } + return n, err +} + +// DecodeAll allows stateless decoding of a blob of bytes. +// Output will be appended to dst, so if the destination size is known +// you can pre-allocate the destination slice to avoid allocations. +// DecodeAll can be used concurrently. +// The Decoder concurrency limits will be respected. +func (d *Decoder) DecodeAll(input, dst []byte) ([]byte, error) { + if d.decoders == nil { + return dst, ErrDecoderClosed + } + + // Grab a block decoder and frame decoder. + block := <-d.decoders + frame := block.localFrame + defer func() { + if debugDecoder { + printf("re-adding decoder: %p", block) + } + frame.rawInput = nil + frame.bBuf = nil + if frame.history.decoders.br != nil { + frame.history.decoders.br.in = nil + } + d.decoders <- block + }() + frame.bBuf = input + + for { + frame.history.reset() + err := frame.reset(&frame.bBuf) + if err != nil { + if err == io.EOF { + if debugDecoder { + println("frame reset return EOF") + } + return dst, nil + } + return dst, err + } + if frame.DictionaryID != nil { + dict, ok := d.dicts[*frame.DictionaryID] + if !ok { + return nil, ErrUnknownDictionary + } + if debugDecoder { + println("setting dict", frame.DictionaryID) + } + frame.history.setDict(&dict) + } + if frame.WindowSize > d.o.maxWindowSize { + if debugDecoder { + println("window size exceeded:", frame.WindowSize, ">", d.o.maxWindowSize) + } + return dst, ErrWindowSizeExceeded + } + if frame.FrameContentSize != fcsUnknown { + if frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) { + return dst, ErrDecoderSizeExceeded + } + if cap(dst)-len(dst) < int(frame.FrameContentSize) { + dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize)+compressedBlockOverAlloc) + copy(dst2, dst) + dst = dst2 + } + } + + if cap(dst) == 0 { + // Allocate len(input) * 2 by default if nothing is provided + // and we didn't get frame content size. + size := len(input) * 2 + // Cap to 1 MB. + if size > 1<<20 { + size = 1 << 20 + } + if uint64(size) > d.o.maxDecodedSize { + size = int(d.o.maxDecodedSize) + } + dst = make([]byte, 0, size) + } + + dst, err = frame.runDecoder(dst, block) + if err != nil { + return dst, err + } + if len(frame.bBuf) == 0 { + if debugDecoder { + println("frame dbuf empty") + } + break + } + } + return dst, nil +} + +// nextBlock returns the next block. +// If an error occurs d.err will be set. +// Optionally the function can block for new output. +// If non-blocking mode is used the returned boolean will be false +// if no data was available without blocking. +func (d *Decoder) nextBlock(blocking bool) (ok bool) { + if d.current.err != nil { + // Keep error state. + return false + } + d.current.b = d.current.b[:0] + + // SYNC: + if d.syncStream.enabled { + if !blocking { + return false + } + ok = d.nextBlockSync() + if !ok { + d.stashDecoder() + } + return ok + } + + //ASYNC: + d.stashDecoder() + if blocking { + d.current.decodeOutput, ok = <-d.current.output + } else { + select { + case d.current.decodeOutput, ok = <-d.current.output: + default: + return false + } + } + if !ok { + // This should not happen, so signal error state... + d.current.err = io.ErrUnexpectedEOF + return false + } + next := d.current.decodeOutput + if next.d != nil && next.d.async.newHist != nil { + d.current.crc.Reset() + } + if debugDecoder { + var tmp [4]byte + binary.LittleEndian.PutUint32(tmp[:], uint32(xxhash.Sum64(next.b))) + println("got", len(d.current.b), "bytes, error:", d.current.err, "data crc:", tmp) + } + + if !d.o.ignoreChecksum && len(next.b) > 0 { + n, err := d.current.crc.Write(next.b) + if err == nil { + if n != len(next.b) { + d.current.err = io.ErrShortWrite + } + } + } + if next.err == nil && next.d != nil && len(next.d.checkCRC) != 0 { + got := d.current.crc.Sum64() + var tmp [4]byte + binary.LittleEndian.PutUint32(tmp[:], uint32(got)) + if !d.o.ignoreChecksum && !bytes.Equal(tmp[:], next.d.checkCRC) { + if debugDecoder { + println("CRC Check Failed:", tmp[:], " (got) !=", next.d.checkCRC, "(on stream)") + } + d.current.err = ErrCRCMismatch + } else { + if debugDecoder { + println("CRC ok", tmp[:]) + } + } + } + + return true +} + +func (d *Decoder) nextBlockSync() (ok bool) { + if d.current.d == nil { + d.current.d = <-d.decoders + } + for len(d.current.b) == 0 { + if !d.syncStream.inFrame { + d.frame.history.reset() + d.current.err = d.frame.reset(&d.syncStream.br) + if d.current.err != nil { + return false + } + if d.frame.DictionaryID != nil { + dict, ok := d.dicts[*d.frame.DictionaryID] + if !ok { + d.current.err = ErrUnknownDictionary + return false + } else { + d.frame.history.setDict(&dict) + } + } + if d.frame.WindowSize > d.o.maxDecodedSize || d.frame.WindowSize > d.o.maxWindowSize { + d.current.err = ErrDecoderSizeExceeded + return false + } + + d.syncStream.decodedFrame = 0 + d.syncStream.inFrame = true + } + d.current.err = d.frame.next(d.current.d) + if d.current.err != nil { + return false + } + d.frame.history.ensureBlock() + if debugDecoder { + println("History trimmed:", len(d.frame.history.b), "decoded already:", d.syncStream.decodedFrame) + } + histBefore := len(d.frame.history.b) + d.current.err = d.current.d.decodeBuf(&d.frame.history) + + if d.current.err != nil { + println("error after:", d.current.err) + return false + } + d.current.b = d.frame.history.b[histBefore:] + if debugDecoder { + println("history after:", len(d.frame.history.b)) + } + + // Check frame size (before CRC) + d.syncStream.decodedFrame += uint64(len(d.current.b)) + if d.syncStream.decodedFrame > d.frame.FrameContentSize { + if debugDecoder { + printf("DecodedFrame (%d) > FrameContentSize (%d)\n", d.syncStream.decodedFrame, d.frame.FrameContentSize) + } + d.current.err = ErrFrameSizeExceeded + return false + } + + // Check FCS + if d.current.d.Last && d.frame.FrameContentSize != fcsUnknown && d.syncStream.decodedFrame != d.frame.FrameContentSize { + if debugDecoder { + printf("DecodedFrame (%d) != FrameContentSize (%d)\n", d.syncStream.decodedFrame, d.frame.FrameContentSize) + } + d.current.err = ErrFrameSizeMismatch + return false + } + + // Update/Check CRC + if d.frame.HasCheckSum { + if !d.o.ignoreChecksum { + d.frame.crc.Write(d.current.b) + } + if d.current.d.Last { + if !d.o.ignoreChecksum { + d.current.err = d.frame.checkCRC() + } else { + d.current.err = d.frame.consumeCRC() + } + if d.current.err != nil { + println("CRC error:", d.current.err) + return false + } + } + } + d.syncStream.inFrame = !d.current.d.Last + } + return true +} + +func (d *Decoder) stashDecoder() { + if d.current.d != nil { + if debugDecoder { + printf("re-adding current decoder %p", d.current.d) + } + d.decoders <- d.current.d + d.current.d = nil + } +} + +// Close will release all resources. +// It is NOT possible to reuse the decoder after this. +func (d *Decoder) Close() { + if d.current.err == ErrDecoderClosed { + return + } + d.drainOutput() + if d.current.cancel != nil { + d.current.cancel() + d.streamWg.Wait() + d.current.cancel = nil + } + if d.decoders != nil { + close(d.decoders) + for dec := range d.decoders { + dec.Close() + } + d.decoders = nil + } + if d.current.d != nil { + d.current.d.Close() + d.current.d = nil + } + d.current.err = ErrDecoderClosed +} + +// IOReadCloser returns the decoder as an io.ReadCloser for convenience. +// Any changes to the decoder will be reflected, so the returned ReadCloser +// can be reused along with the decoder. +// io.WriterTo is also supported by the returned ReadCloser. +func (d *Decoder) IOReadCloser() io.ReadCloser { + return closeWrapper{d: d} +} + +// closeWrapper wraps a function call as a closer. +type closeWrapper struct { + d *Decoder +} + +// WriteTo forwards WriteTo calls to the decoder. +func (c closeWrapper) WriteTo(w io.Writer) (n int64, err error) { + return c.d.WriteTo(w) +} + +// Read forwards read calls to the decoder. +func (c closeWrapper) Read(p []byte) (n int, err error) { + return c.d.Read(p) +} + +// Close closes the decoder. +func (c closeWrapper) Close() error { + c.d.Close() + return nil +} + +type decodeOutput struct { + d *blockDec + b []byte + err error +} + +func (d *Decoder) startSyncDecoder(r io.Reader) error { + d.frame.history.reset() + d.syncStream.br = readerWrapper{r: r} + d.syncStream.inFrame = false + d.syncStream.enabled = true + d.syncStream.decodedFrame = 0 + return nil +} + +// Create Decoder: +// ASYNC: +// Spawn 3 go routines. +// 0: Read frames and decode block literals. +// 1: Decode sequences. +// 2: Execute sequences, send to output. +func (d *Decoder) startStreamDecoder(ctx context.Context, r io.Reader, output chan decodeOutput) { + defer d.streamWg.Done() + br := readerWrapper{r: r} + + var seqDecode = make(chan *blockDec, d.o.concurrent) + var seqExecute = make(chan *blockDec, d.o.concurrent) + + // Async 1: Decode sequences... + go func() { + var hist history + var hasErr bool + + for block := range seqDecode { + if hasErr { + if block != nil { + seqExecute <- block + } + continue + } + if block.async.newHist != nil { + if debugDecoder { + println("Async 1: new history, recent:", block.async.newHist.recentOffsets) + } + hist.decoders = block.async.newHist.decoders + hist.recentOffsets = block.async.newHist.recentOffsets + hist.windowSize = block.async.newHist.windowSize + if block.async.newHist.dict != nil { + hist.setDict(block.async.newHist.dict) + } + } + if block.err != nil || block.Type != blockTypeCompressed { + hasErr = block.err != nil + seqExecute <- block + continue + } + + hist.decoders.literals = block.async.literals + block.err = block.prepareSequences(block.async.seqData, &hist) + if debugDecoder && block.err != nil { + println("prepareSequences returned:", block.err) + } + hasErr = block.err != nil + if block.err == nil { + block.err = block.decodeSequences(&hist) + if debugDecoder && block.err != nil { + println("decodeSequences returned:", block.err) + } + hasErr = block.err != nil + // block.async.sequence = hist.decoders.seq[:hist.decoders.nSeqs] + block.async.seqSize = hist.decoders.seqSize + } + seqExecute <- block + } + close(seqExecute) + }() + + var wg sync.WaitGroup + wg.Add(1) + + // Async 3: Execute sequences... + frameHistCache := d.frame.history.b + go func() { + var hist history + var decodedFrame uint64 + var fcs uint64 + var hasErr bool + for block := range seqExecute { + out := decodeOutput{err: block.err, d: block} + if block.err != nil || hasErr { + hasErr = true + output <- out + continue + } + if block.async.newHist != nil { + if debugDecoder { + println("Async 2: new history") + } + hist.windowSize = block.async.newHist.windowSize + hist.allocFrameBuffer = block.async.newHist.allocFrameBuffer + if block.async.newHist.dict != nil { + hist.setDict(block.async.newHist.dict) + } + + if cap(hist.b) < hist.allocFrameBuffer { + if cap(frameHistCache) >= hist.allocFrameBuffer { + hist.b = frameHistCache + } else { + hist.b = make([]byte, 0, hist.allocFrameBuffer) + println("Alloc history sized", hist.allocFrameBuffer) + } + } + hist.b = hist.b[:0] + fcs = block.async.fcs + decodedFrame = 0 + } + do := decodeOutput{err: block.err, d: block} + switch block.Type { + case blockTypeRLE: + if debugDecoder { + println("add rle block length:", block.RLESize) + } + + if cap(block.dst) < int(block.RLESize) { + if block.lowMem { + block.dst = make([]byte, block.RLESize) + } else { + block.dst = make([]byte, maxBlockSize) + } + } + block.dst = block.dst[:block.RLESize] + v := block.data[0] + for i := range block.dst { + block.dst[i] = v + } + hist.append(block.dst) + do.b = block.dst + case blockTypeRaw: + if debugDecoder { + println("add raw block length:", len(block.data)) + } + hist.append(block.data) + do.b = block.data + case blockTypeCompressed: + if debugDecoder { + println("execute with history length:", len(hist.b), "window:", hist.windowSize) + } + hist.decoders.seqSize = block.async.seqSize + hist.decoders.literals = block.async.literals + do.err = block.executeSequences(&hist) + hasErr = do.err != nil + if debugDecoder && hasErr { + println("executeSequences returned:", do.err) + } + do.b = block.dst + } + if !hasErr { + decodedFrame += uint64(len(do.b)) + if decodedFrame > fcs { + println("fcs exceeded", block.Last, fcs, decodedFrame) + do.err = ErrFrameSizeExceeded + hasErr = true + } else if block.Last && fcs != fcsUnknown && decodedFrame != fcs { + do.err = ErrFrameSizeMismatch + hasErr = true + } else { + if debugDecoder { + println("fcs ok", block.Last, fcs, decodedFrame) + } + } + } + output <- do + } + close(output) + frameHistCache = hist.b + wg.Done() + if debugDecoder { + println("decoder goroutines finished") + } + }() + +decodeStream: + for { + var hist history + var hasErr bool + + decodeBlock := func(block *blockDec) { + if hasErr { + if block != nil { + seqDecode <- block + } + return + } + if block.err != nil || block.Type != blockTypeCompressed { + hasErr = block.err != nil + seqDecode <- block + return + } + + remain, err := block.decodeLiterals(block.data, &hist) + block.err = err + hasErr = block.err != nil + if err == nil { + block.async.literals = hist.decoders.literals + block.async.seqData = remain + } else if debugDecoder { + println("decodeLiterals error:", err) + } + seqDecode <- block + } + frame := d.frame + if debugDecoder { + println("New frame...") + } + var historySent bool + frame.history.reset() + err := frame.reset(&br) + if debugDecoder && err != nil { + println("Frame decoder returned", err) + } + if err == nil && frame.DictionaryID != nil { + dict, ok := d.dicts[*frame.DictionaryID] + if !ok { + err = ErrUnknownDictionary + } else { + frame.history.setDict(&dict) + } + } + if err == nil && d.frame.WindowSize > d.o.maxWindowSize { + err = ErrDecoderSizeExceeded + } + if err != nil { + select { + case <-ctx.Done(): + case dec := <-d.decoders: + dec.sendErr(err) + decodeBlock(dec) + } + break decodeStream + } + + // Go through all blocks of the frame. + for { + var dec *blockDec + select { + case <-ctx.Done(): + break decodeStream + case dec = <-d.decoders: + // Once we have a decoder, we MUST return it. + } + err := frame.next(dec) + if !historySent { + h := frame.history + if debugDecoder { + println("Alloc History:", h.allocFrameBuffer) + } + hist.reset() + if h.dict != nil { + hist.setDict(h.dict) + } + dec.async.newHist = &h + dec.async.fcs = frame.FrameContentSize + historySent = true + } else { + dec.async.newHist = nil + } + if debugDecoder && err != nil { + println("next block returned error:", err) + } + dec.err = err + dec.checkCRC = nil + if dec.Last && frame.HasCheckSum && err == nil { + crc, err := frame.rawInput.readSmall(4) + if err != nil { + println("CRC missing?", err) + dec.err = err + } + var tmp [4]byte + copy(tmp[:], crc) + dec.checkCRC = tmp[:] + if debugDecoder { + println("found crc to check:", dec.checkCRC) + } + } + err = dec.err + last := dec.Last + decodeBlock(dec) + if err != nil { + break decodeStream + } + if last { + break + } + } + } + close(seqDecode) + wg.Wait() + d.frame.history.b = frameHistCache +} diff --git a/vendor/github.com/klauspost/compress/zstd/decoder_options.go b/vendor/github.com/klauspost/compress/zstd/decoder_options.go new file mode 100644 index 0000000..c70e6fa --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/decoder_options.go @@ -0,0 +1,123 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "runtime" +) + +// DOption is an option for creating a decoder. +type DOption func(*decoderOptions) error + +// options retains accumulated state of multiple options. +type decoderOptions struct { + lowMem bool + concurrent int + maxDecodedSize uint64 + maxWindowSize uint64 + dicts []dict + ignoreChecksum bool +} + +func (o *decoderOptions) setDefault() { + *o = decoderOptions{ + // use less ram: true for now, but may change. + lowMem: true, + concurrent: runtime.GOMAXPROCS(0), + maxWindowSize: MaxWindowSize, + } + if o.concurrent > 4 { + o.concurrent = 4 + } + o.maxDecodedSize = 64 << 30 +} + +// WithDecoderLowmem will set whether to use a lower amount of memory, +// but possibly have to allocate more while running. +func WithDecoderLowmem(b bool) DOption { + return func(o *decoderOptions) error { o.lowMem = b; return nil } +} + +// WithDecoderConcurrency sets the number of created decoders. +// When decoding block with DecodeAll, this will limit the number +// of possible concurrently running decodes. +// When decoding streams, this will limit the number of +// inflight blocks. +// When decoding streams and setting maximum to 1, +// no async decoding will be done. +// When a value of 0 is provided GOMAXPROCS will be used. +// By default this will be set to 4 or GOMAXPROCS, whatever is lower. +func WithDecoderConcurrency(n int) DOption { + return func(o *decoderOptions) error { + if n < 0 { + return errors.New("concurrency must be at least 1") + } + if n == 0 { + o.concurrent = runtime.GOMAXPROCS(0) + } else { + o.concurrent = n + } + return nil + } +} + +// WithDecoderMaxMemory allows to set a maximum decoded size for in-memory +// non-streaming operations or maximum window size for streaming operations. +// This can be used to control memory usage of potentially hostile content. +// Maximum is 1 << 63 bytes. Default is 64GiB. +func WithDecoderMaxMemory(n uint64) DOption { + return func(o *decoderOptions) error { + if n == 0 { + return errors.New("WithDecoderMaxMemory must be at least 1") + } + if n > 1<<63 { + return errors.New("WithDecoderMaxmemory must be less than 1 << 63") + } + o.maxDecodedSize = n + return nil + } +} + +// WithDecoderDicts allows to register one or more dictionaries for the decoder. +// If several dictionaries with the same ID is provided the last one will be used. +func WithDecoderDicts(dicts ...[]byte) DOption { + return func(o *decoderOptions) error { + for _, b := range dicts { + d, err := loadDict(b) + if err != nil { + return err + } + o.dicts = append(o.dicts, *d) + } + return nil + } +} + +// WithDecoderMaxWindow allows to set a maximum window size for decodes. +// This allows rejecting packets that will cause big memory usage. +// The Decoder will likely allocate more memory based on the WithDecoderLowmem setting. +// If WithDecoderMaxMemory is set to a lower value, that will be used. +// Default is 512MB, Maximum is ~3.75 TB as per zstandard spec. +func WithDecoderMaxWindow(size uint64) DOption { + return func(o *decoderOptions) error { + if size < MinWindowSize { + return errors.New("WithMaxWindowSize must be at least 1KB, 1024 bytes") + } + if size > (1<<41)+7*(1<<38) { + return errors.New("WithMaxWindowSize must be less than (1<<41) + 7*(1<<38) ~ 3.75TB") + } + o.maxWindowSize = size + return nil + } +} + +// IgnoreChecksum allows to forcibly ignore checksum checking. +func IgnoreChecksum(b bool) DOption { + return func(o *decoderOptions) error { + o.ignoreChecksum = b + return nil + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/dict.go b/vendor/github.com/klauspost/compress/zstd/dict.go new file mode 100644 index 0000000..a36ae83 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/dict.go @@ -0,0 +1,122 @@ +package zstd + +import ( + "bytes" + "encoding/binary" + "errors" + "fmt" + "io" + + "github.com/klauspost/compress/huff0" +) + +type dict struct { + id uint32 + + litEnc *huff0.Scratch + llDec, ofDec, mlDec sequenceDec + //llEnc, ofEnc, mlEnc []*fseEncoder + offsets [3]int + content []byte +} + +var dictMagic = [4]byte{0x37, 0xa4, 0x30, 0xec} + +// ID returns the dictionary id or 0 if d is nil. +func (d *dict) ID() uint32 { + if d == nil { + return 0 + } + return d.id +} + +// DictContentSize returns the dictionary content size or 0 if d is nil. +func (d *dict) DictContentSize() int { + if d == nil { + return 0 + } + return len(d.content) +} + +// Load a dictionary as described in +// https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format +func loadDict(b []byte) (*dict, error) { + // Check static field size. + if len(b) <= 8+(3*4) { + return nil, io.ErrUnexpectedEOF + } + d := dict{ + llDec: sequenceDec{fse: &fseDecoder{}}, + ofDec: sequenceDec{fse: &fseDecoder{}}, + mlDec: sequenceDec{fse: &fseDecoder{}}, + } + if !bytes.Equal(b[:4], dictMagic[:]) { + return nil, ErrMagicMismatch + } + d.id = binary.LittleEndian.Uint32(b[4:8]) + if d.id == 0 { + return nil, errors.New("dictionaries cannot have ID 0") + } + + // Read literal table + var err error + d.litEnc, b, err = huff0.ReadTable(b[8:], nil) + if err != nil { + return nil, err + } + d.litEnc.Reuse = huff0.ReusePolicyMust + + br := byteReader{ + b: b, + off: 0, + } + readDec := func(i tableIndex, dec *fseDecoder) error { + if err := dec.readNCount(&br, uint16(maxTableSymbol[i])); err != nil { + return err + } + if br.overread() { + return io.ErrUnexpectedEOF + } + err = dec.transform(symbolTableX[i]) + if err != nil { + println("Transform table error:", err) + return err + } + if debugDecoder || debugEncoder { + println("Read table ok", "symbolLen:", dec.symbolLen) + } + // Set decoders as predefined so they aren't reused. + dec.preDefined = true + return nil + } + + if err := readDec(tableOffsets, d.ofDec.fse); err != nil { + return nil, err + } + if err := readDec(tableMatchLengths, d.mlDec.fse); err != nil { + return nil, err + } + if err := readDec(tableLiteralLengths, d.llDec.fse); err != nil { + return nil, err + } + if br.remain() < 12 { + return nil, io.ErrUnexpectedEOF + } + + d.offsets[0] = int(br.Uint32()) + br.advance(4) + d.offsets[1] = int(br.Uint32()) + br.advance(4) + d.offsets[2] = int(br.Uint32()) + br.advance(4) + if d.offsets[0] <= 0 || d.offsets[1] <= 0 || d.offsets[2] <= 0 { + return nil, errors.New("invalid offset in dictionary") + } + d.content = make([]byte, br.remain()) + copy(d.content, br.unread()) + if d.offsets[0] > len(d.content) || d.offsets[1] > len(d.content) || d.offsets[2] > len(d.content) { + return nil, fmt.Errorf("initial offset bigger than dictionary content size %d, offsets: %v", len(d.content), d.offsets) + } + + return &d, nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_base.go b/vendor/github.com/klauspost/compress/zstd/enc_base.go new file mode 100644 index 0000000..15ae8ee --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_base.go @@ -0,0 +1,188 @@ +package zstd + +import ( + "fmt" + "math/bits" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +const ( + dictShardBits = 6 +) + +type fastBase struct { + // cur is the offset at the start of hist + cur int32 + // maximum offset. Should be at least 2x block size. + maxMatchOff int32 + hist []byte + crc *xxhash.Digest + tmp [8]byte + blk *blockEnc + lastDictID uint32 + lowMem bool +} + +// CRC returns the underlying CRC writer. +func (e *fastBase) CRC() *xxhash.Digest { + return e.crc +} + +// AppendCRC will append the CRC to the destination slice and return it. +func (e *fastBase) AppendCRC(dst []byte) []byte { + crc := e.crc.Sum(e.tmp[:0]) + dst = append(dst, crc[7], crc[6], crc[5], crc[4]) + return dst +} + +// WindowSize returns the window size of the encoder, +// or a window size small enough to contain the input size, if > 0. +func (e *fastBase) WindowSize(size int64) int32 { + if size > 0 && size < int64(e.maxMatchOff) { + b := int32(1) << uint(bits.Len(uint(size))) + // Keep minimum window. + if b < 1024 { + b = 1024 + } + return b + } + return e.maxMatchOff +} + +// Block returns the current block. +func (e *fastBase) Block() *blockEnc { + return e.blk +} + +func (e *fastBase) addBlock(src []byte) int32 { + if debugAsserts && e.cur > bufferReset { + panic(fmt.Sprintf("ecur (%d) > buffer reset (%d)", e.cur, bufferReset)) + } + // check if we have space already + if len(e.hist)+len(src) > cap(e.hist) { + if cap(e.hist) == 0 { + e.ensureHist(len(src)) + } else { + if cap(e.hist) < int(e.maxMatchOff+maxCompressedBlockSize) { + panic(fmt.Errorf("unexpected buffer cap %d, want at least %d with window %d", cap(e.hist), e.maxMatchOff+maxCompressedBlockSize, e.maxMatchOff)) + } + // Move down + offset := int32(len(e.hist)) - e.maxMatchOff + copy(e.hist[0:e.maxMatchOff], e.hist[offset:]) + e.cur += offset + e.hist = e.hist[:e.maxMatchOff] + } + } + s := int32(len(e.hist)) + e.hist = append(e.hist, src...) + return s +} + +// ensureHist will ensure that history can keep at least this many bytes. +func (e *fastBase) ensureHist(n int) { + if cap(e.hist) >= n { + return + } + l := e.maxMatchOff + if (e.lowMem && e.maxMatchOff > maxCompressedBlockSize) || e.maxMatchOff <= maxCompressedBlockSize { + l += maxCompressedBlockSize + } else { + l += e.maxMatchOff + } + // Make it at least 1MB. + if l < 1<<20 && !e.lowMem { + l = 1 << 20 + } + // Make it at least the requested size. + if l < int32(n) { + l = int32(n) + } + e.hist = make([]byte, 0, l) +} + +// useBlock will replace the block with the provided one, +// but transfer recent offsets from the previous. +func (e *fastBase) UseBlock(enc *blockEnc) { + enc.reset(e.blk) + e.blk = enc +} + +func (e *fastBase) matchlen(s, t int32, src []byte) int32 { + if debugAsserts { + if s < 0 { + err := fmt.Sprintf("s (%d) < 0", s) + panic(err) + } + if t < 0 { + err := fmt.Sprintf("s (%d) < 0", s) + panic(err) + } + if s-t > e.maxMatchOff { + err := fmt.Sprintf("s (%d) - t (%d) > maxMatchOff (%d)", s, t, e.maxMatchOff) + panic(err) + } + if len(src)-int(s) > maxCompressedBlockSize { + panic(fmt.Sprintf("len(src)-s (%d) > maxCompressedBlockSize (%d)", len(src)-int(s), maxCompressedBlockSize)) + } + } + a := src[s:] + b := src[t:] + b = b[:len(a)] + end := int32((len(a) >> 3) << 3) + for i := int32(0); i < end; i += 8 { + if diff := load6432(a, i) ^ load6432(b, i); diff != 0 { + return i + int32(bits.TrailingZeros64(diff)>>3) + } + } + + a = a[end:] + b = b[end:] + for i := range a { + if a[i] != b[i] { + return int32(i) + end + } + } + return int32(len(a)) + end +} + +// Reset the encoding table. +func (e *fastBase) resetBase(d *dict, singleBlock bool) { + if e.blk == nil { + e.blk = &blockEnc{lowMem: e.lowMem} + e.blk.init() + } else { + e.blk.reset(nil) + } + e.blk.initNewEncode() + if e.crc == nil { + e.crc = xxhash.New() + } else { + e.crc.Reset() + } + if d != nil { + low := e.lowMem + if singleBlock { + e.lowMem = true + } + e.ensureHist(d.DictContentSize() + maxCompressedBlockSize) + e.lowMem = low + } + + // We offset current position so everything will be out of reach. + // If above reset line, history will be purged. + if e.cur < bufferReset { + e.cur += e.maxMatchOff + int32(len(e.hist)) + } + e.hist = e.hist[:0] + if d != nil { + // Set offsets (currently not used) + for i, off := range d.offsets { + e.blk.recentOffsets[i] = uint32(off) + e.blk.prevRecentOffsets[i] = e.blk.recentOffsets[i] + } + // Transfer litenc. + e.blk.dictLitEnc = d.litEnc + e.hist = append(e.hist, d.content...) + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_best.go b/vendor/github.com/klauspost/compress/zstd/enc_best.go new file mode 100644 index 0000000..96028ec --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_best.go @@ -0,0 +1,558 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "bytes" + "fmt" + + "github.com/klauspost/compress" +) + +const ( + bestLongTableBits = 22 // Bits used in the long match table + bestLongTableSize = 1 << bestLongTableBits // Size of the table + bestLongLen = 8 // Bytes used for table hash + + // Note: Increasing the short table bits or making the hash shorter + // can actually lead to compression degradation since it will 'steal' more from the + // long match table and match offsets are quite big. + // This greatly depends on the type of input. + bestShortTableBits = 18 // Bits used in the short match table + bestShortTableSize = 1 << bestShortTableBits // Size of the table + bestShortLen = 4 // Bytes used for table hash + +) + +type match struct { + offset int32 + s int32 + length int32 + rep int32 + est int32 +} + +const highScore = 25000 + +// estBits will estimate output bits from predefined tables. +func (m *match) estBits(bitsPerByte int32) { + mlc := mlCode(uint32(m.length - zstdMinMatch)) + var ofc uint8 + if m.rep < 0 { + ofc = ofCode(uint32(m.s-m.offset) + 3) + } else { + ofc = ofCode(uint32(m.rep)) + } + // Cost, excluding + ofTT, mlTT := fsePredefEnc[tableOffsets].ct.symbolTT[ofc], fsePredefEnc[tableMatchLengths].ct.symbolTT[mlc] + + // Add cost of match encoding... + m.est = int32(ofTT.outBits + mlTT.outBits) + m.est += int32(ofTT.deltaNbBits>>16 + mlTT.deltaNbBits>>16) + // Subtract savings compared to literal encoding... + m.est -= (m.length * bitsPerByte) >> 10 + if m.est > 0 { + // Unlikely gain.. + m.length = 0 + m.est = highScore + } +} + +// bestFastEncoder uses 2 tables, one for short matches (5 bytes) and one for long matches. +// The long match table contains the previous entry with the same hash, +// effectively making it a "chain" of length 2. +// When we find a long match we choose between the two values and select the longest. +// When we find a short match, after checking the long, we check if we can find a long at n+1 +// and that it is longer (lazy matching). +type bestFastEncoder struct { + fastBase + table [bestShortTableSize]prevEntry + longTable [bestLongTableSize]prevEntry + dictTable []prevEntry + dictLongTable []prevEntry +} + +// Encode improves compression... +func (e *bestFastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 4 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = prevEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = prevEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + v2 := e.table[i].prev + if v < minOff { + v = 0 + v2 = 0 + } else { + v = v - e.cur + e.maxMatchOff + if v2 < minOff { + v2 = 0 + } else { + v2 = v2 - e.cur + e.maxMatchOff + } + } + e.table[i] = prevEntry{ + offset: v, + prev: v2, + } + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + v2 := e.longTable[i].prev + if v < minOff { + v = 0 + v2 = 0 + } else { + v = v - e.cur + e.maxMatchOff + if v2 < minOff { + v2 = 0 + } else { + v2 = v2 - e.cur + e.maxMatchOff + } + } + e.longTable[i] = prevEntry{ + offset: v, + prev: v2, + } + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Use this to estimate literal cost. + // Scaled by 10 bits. + bitsPerByte := int32((compress.ShannonEntropyBits(src) * 1024) / len(src)) + // Huffman can never go < 1 bit/byte + if bitsPerByte < 1024 { + bitsPerByte = 1024 + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + const kSearchStrength = 10 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + offset3 := int32(blk.recentOffsets[2]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + _ = addLiterals + + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + bestOf := func(a, b match) match { + if a.est+(a.s-b.s)*bitsPerByte>>10 < b.est+(b.s-a.s)*bitsPerByte>>10 { + return a + } + return b + } + const goodEnough = 100 + + nextHashL := hashLen(cv, bestLongTableBits, bestLongLen) + nextHashS := hashLen(cv, bestShortTableBits, bestShortLen) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + matchAt := func(offset int32, s int32, first uint32, rep int32) match { + if s-offset >= e.maxMatchOff || load3232(src, offset) != first { + return match{s: s, est: highScore} + } + if debugAsserts { + if !bytes.Equal(src[s:s+4], src[offset:offset+4]) { + panic(fmt.Sprintf("first match mismatch: %v != %v, first: %08x", src[s:s+4], src[offset:offset+4], first)) + } + } + m := match{offset: offset, s: s, length: 4 + e.matchlen(s+4, offset+4, src), rep: rep} + m.estBits(bitsPerByte) + return m + } + + best := bestOf(matchAt(candidateL.offset-e.cur, s, uint32(cv), -1), matchAt(candidateL.prev-e.cur, s, uint32(cv), -1)) + best = bestOf(best, matchAt(candidateS.offset-e.cur, s, uint32(cv), -1)) + best = bestOf(best, matchAt(candidateS.prev-e.cur, s, uint32(cv), -1)) + + if canRepeat && best.length < goodEnough { + cv32 := uint32(cv >> 8) + spp := s + 1 + best = bestOf(best, matchAt(spp-offset1, spp, cv32, 1)) + best = bestOf(best, matchAt(spp-offset2, spp, cv32, 2)) + best = bestOf(best, matchAt(spp-offset3, spp, cv32, 3)) + if best.length > 0 { + cv32 = uint32(cv >> 24) + spp += 2 + best = bestOf(best, matchAt(spp-offset1, spp, cv32, 1)) + best = bestOf(best, matchAt(spp-offset2, spp, cv32, 2)) + best = bestOf(best, matchAt(spp-offset3, spp, cv32, 3)) + } + } + // Load next and check... + e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: candidateL.offset} + e.table[nextHashS] = prevEntry{offset: s + e.cur, prev: candidateS.offset} + + // Look far ahead, unless we have a really long match already... + if best.length < goodEnough { + // No match found, move forward on input, no need to check forward... + if best.length < 4 { + s += 1 + (s-nextEmit)>>(kSearchStrength-1) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + continue + } + + s++ + candidateS = e.table[hashLen(cv>>8, bestShortTableBits, bestShortLen)] + cv = load6432(src, s) + cv2 := load6432(src, s+1) + candidateL = e.longTable[hashLen(cv, bestLongTableBits, bestLongLen)] + candidateL2 := e.longTable[hashLen(cv2, bestLongTableBits, bestLongLen)] + + // Short at s+1 + best = bestOf(best, matchAt(candidateS.offset-e.cur, s, uint32(cv), -1)) + // Long at s+1, s+2 + best = bestOf(best, matchAt(candidateL.offset-e.cur, s, uint32(cv), -1)) + best = bestOf(best, matchAt(candidateL.prev-e.cur, s, uint32(cv), -1)) + best = bestOf(best, matchAt(candidateL2.offset-e.cur, s+1, uint32(cv2), -1)) + best = bestOf(best, matchAt(candidateL2.prev-e.cur, s+1, uint32(cv2), -1)) + if false { + // Short at s+3. + // Too often worse... + best = bestOf(best, matchAt(e.table[hashLen(cv2>>8, bestShortTableBits, bestShortLen)].offset-e.cur, s+2, uint32(cv2>>8), -1)) + } + // See if we can find a better match by checking where the current best ends. + // Use that offset to see if we can find a better full match. + if sAt := best.s + best.length; sAt < sLimit { + nextHashL := hashLen(load6432(src, sAt), bestLongTableBits, bestLongLen) + candidateEnd := e.longTable[nextHashL] + if pos := candidateEnd.offset - e.cur - best.length; pos >= 0 { + bestEnd := bestOf(best, matchAt(pos, best.s, load3232(src, best.s), -1)) + if pos := candidateEnd.prev - e.cur - best.length; pos >= 0 { + bestEnd = bestOf(bestEnd, matchAt(pos, best.s, load3232(src, best.s), -1)) + } + best = bestEnd + } + } + } + + if debugAsserts { + if !bytes.Equal(src[best.s:best.s+best.length], src[best.offset:best.offset+best.length]) { + panic(fmt.Sprintf("match mismatch: %v != %v", src[best.s:best.s+best.length], src[best.offset:best.offset+best.length])) + } + } + + // We have a match, we can store the forward value + if best.rep > 0 { + s = best.s + var seq seq + seq.matchLen = uint32(best.length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := best.s + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + repIndex := best.offset + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = uint32(best.rep) + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Index match start+1 (long) -> s - 1 + index0 := s + s = best.s + best.length + + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, best.length) + + } + break encodeLoop + } + // Index skipped... + off := index0 + e.cur + for index0 < s-1 { + cv0 := load6432(src, index0) + h0 := hashLen(cv0, bestLongTableBits, bestLongLen) + h1 := hashLen(cv0, bestShortTableBits, bestShortLen) + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.table[h1] = prevEntry{offset: off, prev: e.table[h1].offset} + off++ + index0++ + } + switch best.rep { + case 2: + offset1, offset2 = offset2, offset1 + case 3: + offset1, offset2, offset3 = offset3, offset1, offset2 + } + cv = load6432(src, s) + continue + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + s = best.s + t := best.offset + offset1, offset2, offset3 = s-t, offset1, offset2 + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the n-byte match as long as possible. + l := best.length + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) -> s - 1 + index0 := s - l + 1 + // every entry + for index0 < s-1 { + cv0 := load6432(src, index0) + h0 := hashLen(cv0, bestLongTableBits, bestLongLen) + h1 := hashLen(cv0, bestShortTableBits, bestShortLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.table[h1] = prevEntry{offset: off, prev: e.table[h1].offset} + index0++ + } + + cv = load6432(src, s) + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashS := hashLen(cv, bestShortTableBits, bestShortLen) + nextHashL := hashLen(cv, bestLongTableBits, bestLongLen) + + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset} + e.table[nextHashS] = prevEntry{offset: s + e.cur, prev: e.table[nextHashS].offset} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + blk.recentOffsets[2] = uint32(offset3) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *bestFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + e.ensureHist(len(src)) + e.Encode(blk, src) +} + +// Reset will reset and set a dictionary if not nil +func (e *bestFastEncoder) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d == nil { + return + } + // Init or copy dict table + if len(e.dictTable) != len(e.table) || d.id != e.lastDictID { + if len(e.dictTable) != len(e.table) { + e.dictTable = make([]prevEntry, len(e.table)) + } + end := int32(len(d.content)) - 8 + e.maxMatchOff + for i := e.maxMatchOff; i < end; i += 4 { + const hashLog = bestShortTableBits + + cv := load6432(d.content, i-e.maxMatchOff) + nextHash := hashLen(cv, hashLog, bestShortLen) // 0 -> 4 + nextHash1 := hashLen(cv>>8, hashLog, bestShortLen) // 1 -> 5 + nextHash2 := hashLen(cv>>16, hashLog, bestShortLen) // 2 -> 6 + nextHash3 := hashLen(cv>>24, hashLog, bestShortLen) // 3 -> 7 + e.dictTable[nextHash] = prevEntry{ + prev: e.dictTable[nextHash].offset, + offset: i, + } + e.dictTable[nextHash1] = prevEntry{ + prev: e.dictTable[nextHash1].offset, + offset: i + 1, + } + e.dictTable[nextHash2] = prevEntry{ + prev: e.dictTable[nextHash2].offset, + offset: i + 2, + } + e.dictTable[nextHash3] = prevEntry{ + prev: e.dictTable[nextHash3].offset, + offset: i + 3, + } + } + e.lastDictID = d.id + } + + // Init or copy dict table + if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID { + if len(e.dictLongTable) != len(e.longTable) { + e.dictLongTable = make([]prevEntry, len(e.longTable)) + } + if len(d.content) >= 8 { + cv := load6432(d.content, 0) + h := hashLen(cv, bestLongTableBits, bestLongLen) + e.dictLongTable[h] = prevEntry{ + offset: e.maxMatchOff, + prev: e.dictLongTable[h].offset, + } + + end := int32(len(d.content)) - 8 + e.maxMatchOff + off := 8 // First to read + for i := e.maxMatchOff + 1; i < end; i++ { + cv = cv>>8 | (uint64(d.content[off]) << 56) + h := hashLen(cv, bestLongTableBits, bestLongLen) + e.dictLongTable[h] = prevEntry{ + offset: i, + prev: e.dictLongTable[h].offset, + } + off++ + } + } + e.lastDictID = d.id + } + // Reset table to initial state + copy(e.longTable[:], e.dictLongTable) + + e.cur = e.maxMatchOff + // Reset table to initial state + copy(e.table[:], e.dictTable) +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_better.go b/vendor/github.com/klauspost/compress/zstd/enc_better.go new file mode 100644 index 0000000..c769f69 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_better.go @@ -0,0 +1,1237 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import "fmt" + +const ( + betterLongTableBits = 19 // Bits used in the long match table + betterLongTableSize = 1 << betterLongTableBits // Size of the table + betterLongLen = 8 // Bytes used for table hash + + // Note: Increasing the short table bits or making the hash shorter + // can actually lead to compression degradation since it will 'steal' more from the + // long match table and match offsets are quite big. + // This greatly depends on the type of input. + betterShortTableBits = 13 // Bits used in the short match table + betterShortTableSize = 1 << betterShortTableBits // Size of the table + betterShortLen = 5 // Bytes used for table hash + + betterLongTableShardCnt = 1 << (betterLongTableBits - dictShardBits) // Number of shards in the table + betterLongTableShardSize = betterLongTableSize / betterLongTableShardCnt // Size of an individual shard + + betterShortTableShardCnt = 1 << (betterShortTableBits - dictShardBits) // Number of shards in the table + betterShortTableShardSize = betterShortTableSize / betterShortTableShardCnt // Size of an individual shard +) + +type prevEntry struct { + offset int32 + prev int32 +} + +// betterFastEncoder uses 2 tables, one for short matches (5 bytes) and one for long matches. +// The long match table contains the previous entry with the same hash, +// effectively making it a "chain" of length 2. +// When we find a long match we choose between the two values and select the longest. +// When we find a short match, after checking the long, we check if we can find a long at n+1 +// and that it is longer (lazy matching). +type betterFastEncoder struct { + fastBase + table [betterShortTableSize]tableEntry + longTable [betterLongTableSize]prevEntry +} + +type betterFastEncoderDict struct { + betterFastEncoder + dictTable []tableEntry + dictLongTable []prevEntry + shortTableShardDirty [betterShortTableShardCnt]bool + longTableShardDirty [betterLongTableShardCnt]bool + allDirty bool +} + +// Encode improves compression... +func (e *betterFastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = prevEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + v2 := e.longTable[i].prev + if v < minOff { + v = 0 + v2 = 0 + } else { + v = v - e.cur + e.maxMatchOff + if v2 < minOff { + v2 = 0 + } else { + v2 = v2 - e.cur + e.maxMatchOff + } + } + e.longTable[i] = prevEntry{ + offset: v, + prev: v2, + } + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 9 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + var matched int32 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) + nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + off := s + e.cur + e.longTable[nextHashL] = prevEntry{offset: off, prev: candidateL.offset} + e.table[nextHashS] = tableEntry{offset: off, val: uint32(cv)} + + if canRepeat { + if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Index match start+1 (long) -> s - 1 + index0 := s + repOff + s += lenght + repOff + + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + // Index skipped... + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hashLen(cv0, betterLongTableBits, betterLongLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.table[hashLen(cv1, betterShortTableBits, betterShortLen)] = tableEntry{offset: off + 1, val: uint32(cv1)} + index0 += 2 + } + cv = load6432(src, s) + continue + } + const repOff2 = 1 + + // We deviate from the reference encoder and also check offset 2. + // Still slower and not much better, so disabled. + // repIndex = s - offset2 + repOff2 + if false && repIndex >= 0 && load6432(src, repIndex) == load6432(src, s+repOff) { + // Consider history as well. + var seq seq + lenght := 8 + e.matchlen(s+8+repOff2, repIndex+8, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 2 + seq.offset = 2 + if debugSequences { + println("repeat sequence 2", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + index0 := s + repOff2 + s += lenght + repOff2 + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + + // Index skipped... + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hashLen(cv0, betterLongTableBits, betterLongLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.table[hashLen(cv1, betterShortTableBits, betterShortLen)] = tableEntry{offset: off + 1, val: uint32(cv1)} + index0 += 2 + } + cv = load6432(src, s) + // Swap offsets + offset1, offset2 = offset2, offset1 + continue + } + } + // Find the offsets of our two matches. + coffsetL := candidateL.offset - e.cur + coffsetLP := candidateL.prev - e.cur + + // Check if we have a long match. + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matched = e.matchlen(s+8, coffsetL+8, src) + 8 + t = coffsetL + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + + if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) { + // Found a long match, at least 8 bytes. + prevMatch := e.matchlen(s+8, coffsetLP+8, src) + 8 + if prevMatch > matched { + matched = prevMatch + t = coffsetLP + } + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + } + break + } + + // Check if we have a long match on prev. + if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) { + // Found a long match, at least 8 bytes. + matched = e.matchlen(s+8, coffsetLP+8, src) + 8 + t = coffsetLP + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + coffsetS := candidateS.offset - e.cur + + // Check if we have a short match. + if s-coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + matched = e.matchlen(s+4, coffsetS+4, src) + 4 + + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hashLen(cv, betterLongTableBits, betterLongLen) + candidateL = e.longTable[nextHashL] + coffsetL = candidateL.offset - e.cur + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = prevEntry{offset: s + checkAt + e.cur, prev: candidateL.offset} + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8 + if matchedNext > matched { + t = coffsetL + s += checkAt + matched = matchedNext + if debugMatches { + println("long match (after short)") + } + break + } + } + + // Check prev long... + coffsetL = candidateL.prev - e.cur + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8 + if matchedNext > matched { + t = coffsetL + s += checkAt + matched = matchedNext + if debugMatches { + println("prev long match (after short)") + } + break + } + } + t = coffsetS + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // Try to find a better match by searching for a long match at the end of the current best match + if s+matched < sLimit { + nextHashL := hashLen(load6432(src, s+matched), betterLongTableBits, betterLongLen) + cv := load3232(src, s) + candidateL := e.longTable[nextHashL] + coffsetL := candidateL.offset - e.cur - matched + if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) { + // Found a long match, at least 4 bytes. + matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4 + if matchedNext > matched { + t = coffsetL + matched = matchedNext + if debugMatches { + println("long match at end-of-match") + } + } + } + + // Check prev long... + if true { + coffsetL = candidateL.prev - e.cur - matched + if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) { + // Found a long match, at least 4 bytes. + matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4 + if matchedNext > matched { + t = coffsetL + matched = matchedNext + if debugMatches { + println("prev long match at end-of-match") + } + } + } + } + } + // A match has been found. Update recent offsets. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the n-byte match as long as possible. + l := matched + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) -> s - 1 + index0 := s - l + 1 + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hashLen(cv0, betterLongTableBits, betterLongLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.table[hashLen(cv1, betterShortTableBits, betterShortLen)] = tableEntry{offset: off + 1, val: uint32(cv1)} + index0 += 2 + } + + cv = load6432(src, s) + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) + nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) + + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset} + e.table[nextHashS] = tableEntry{offset: s + e.cur, val: uint32(cv)} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *betterFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + e.ensureHist(len(src)) + e.Encode(blk, src) +} + +// Encode improves compression... +func (e *betterFastEncoderDict) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = prevEntry{} + } + e.cur = e.maxMatchOff + e.allDirty = true + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + v2 := e.longTable[i].prev + if v < minOff { + v = 0 + v2 = 0 + } else { + v = v - e.cur + e.maxMatchOff + if v2 < minOff { + v2 = 0 + } else { + v2 = v2 - e.cur + e.maxMatchOff + } + } + e.longTable[i] = prevEntry{ + offset: v, + prev: v2, + } + } + e.allDirty = true + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 9 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + var matched int32 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) + nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + off := s + e.cur + e.longTable[nextHashL] = prevEntry{offset: off, prev: candidateL.offset} + e.markLongShardDirty(nextHashL) + e.table[nextHashS] = tableEntry{offset: off, val: uint32(cv)} + e.markShortShardDirty(nextHashS) + + if canRepeat { + if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Index match start+1 (long) -> s - 1 + index0 := s + repOff + s += lenght + repOff + + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + // Index skipped... + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hashLen(cv0, betterLongTableBits, betterLongLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.markLongShardDirty(h0) + h1 := hashLen(cv1, betterShortTableBits, betterShortLen) + e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)} + e.markShortShardDirty(h1) + index0 += 2 + } + cv = load6432(src, s) + continue + } + const repOff2 = 1 + + // We deviate from the reference encoder and also check offset 2. + // Still slower and not much better, so disabled. + // repIndex = s - offset2 + repOff2 + if false && repIndex >= 0 && load6432(src, repIndex) == load6432(src, s+repOff) { + // Consider history as well. + var seq seq + lenght := 8 + e.matchlen(s+8+repOff2, repIndex+8, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 2 + seq.offset = 2 + if debugSequences { + println("repeat sequence 2", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + index0 := s + repOff2 + s += lenght + repOff2 + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + + // Index skipped... + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hashLen(cv0, betterLongTableBits, betterLongLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.markLongShardDirty(h0) + h1 := hashLen(cv1, betterShortTableBits, betterShortLen) + e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)} + e.markShortShardDirty(h1) + index0 += 2 + } + cv = load6432(src, s) + // Swap offsets + offset1, offset2 = offset2, offset1 + continue + } + } + // Find the offsets of our two matches. + coffsetL := candidateL.offset - e.cur + coffsetLP := candidateL.prev - e.cur + + // Check if we have a long match. + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matched = e.matchlen(s+8, coffsetL+8, src) + 8 + t = coffsetL + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + + if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) { + // Found a long match, at least 8 bytes. + prevMatch := e.matchlen(s+8, coffsetLP+8, src) + 8 + if prevMatch > matched { + matched = prevMatch + t = coffsetLP + } + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + } + break + } + + // Check if we have a long match on prev. + if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) { + // Found a long match, at least 8 bytes. + matched = e.matchlen(s+8, coffsetLP+8, src) + 8 + t = coffsetLP + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + coffsetS := candidateS.offset - e.cur + + // Check if we have a short match. + if s-coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + matched = e.matchlen(s+4, coffsetS+4, src) + 4 + + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hashLen(cv, betterLongTableBits, betterLongLen) + candidateL = e.longTable[nextHashL] + coffsetL = candidateL.offset - e.cur + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = prevEntry{offset: s + checkAt + e.cur, prev: candidateL.offset} + e.markLongShardDirty(nextHashL) + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8 + if matchedNext > matched { + t = coffsetL + s += checkAt + matched = matchedNext + if debugMatches { + println("long match (after short)") + } + break + } + } + + // Check prev long... + coffsetL = candidateL.prev - e.cur + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8 + if matchedNext > matched { + t = coffsetL + s += checkAt + matched = matchedNext + if debugMatches { + println("prev long match (after short)") + } + break + } + } + t = coffsetS + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + // Try to find a better match by searching for a long match at the end of the current best match + if s+matched < sLimit { + nextHashL := hashLen(load6432(src, s+matched), betterLongTableBits, betterLongLen) + cv := load3232(src, s) + candidateL := e.longTable[nextHashL] + coffsetL := candidateL.offset - e.cur - matched + if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) { + // Found a long match, at least 4 bytes. + matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4 + if matchedNext > matched { + t = coffsetL + matched = matchedNext + if debugMatches { + println("long match at end-of-match") + } + } + } + + // Check prev long... + if true { + coffsetL = candidateL.prev - e.cur - matched + if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) { + // Found a long match, at least 4 bytes. + matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4 + if matchedNext > matched { + t = coffsetL + matched = matchedNext + if debugMatches { + println("prev long match at end-of-match") + } + } + } + } + } + // A match has been found. Update recent offsets. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the n-byte match as long as possible. + l := matched + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) -> s - 1 + index0 := s - l + 1 + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hashLen(cv0, betterLongTableBits, betterLongLen) + off := index0 + e.cur + e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset} + e.markLongShardDirty(h0) + h1 := hashLen(cv1, betterShortTableBits, betterShortLen) + e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)} + e.markShortShardDirty(h1) + index0 += 2 + } + + cv = load6432(src, s) + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) + nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) + + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset} + e.markLongShardDirty(nextHashL) + e.table[nextHashS] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.markShortShardDirty(nextHashS) + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *betterFastEncoder) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d != nil { + panic("betterFastEncoder: Reset with dict") + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *betterFastEncoderDict) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d == nil { + return + } + // Init or copy dict table + if len(e.dictTable) != len(e.table) || d.id != e.lastDictID { + if len(e.dictTable) != len(e.table) { + e.dictTable = make([]tableEntry, len(e.table)) + } + end := int32(len(d.content)) - 8 + e.maxMatchOff + for i := e.maxMatchOff; i < end; i += 4 { + const hashLog = betterShortTableBits + + cv := load6432(d.content, i-e.maxMatchOff) + nextHash := hashLen(cv, hashLog, betterShortLen) // 0 -> 4 + nextHash1 := hashLen(cv>>8, hashLog, betterShortLen) // 1 -> 5 + nextHash2 := hashLen(cv>>16, hashLog, betterShortLen) // 2 -> 6 + nextHash3 := hashLen(cv>>24, hashLog, betterShortLen) // 3 -> 7 + e.dictTable[nextHash] = tableEntry{ + val: uint32(cv), + offset: i, + } + e.dictTable[nextHash1] = tableEntry{ + val: uint32(cv >> 8), + offset: i + 1, + } + e.dictTable[nextHash2] = tableEntry{ + val: uint32(cv >> 16), + offset: i + 2, + } + e.dictTable[nextHash3] = tableEntry{ + val: uint32(cv >> 24), + offset: i + 3, + } + } + e.lastDictID = d.id + e.allDirty = true + } + + // Init or copy dict table + if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID { + if len(e.dictLongTable) != len(e.longTable) { + e.dictLongTable = make([]prevEntry, len(e.longTable)) + } + if len(d.content) >= 8 { + cv := load6432(d.content, 0) + h := hashLen(cv, betterLongTableBits, betterLongLen) + e.dictLongTable[h] = prevEntry{ + offset: e.maxMatchOff, + prev: e.dictLongTable[h].offset, + } + + end := int32(len(d.content)) - 8 + e.maxMatchOff + off := 8 // First to read + for i := e.maxMatchOff + 1; i < end; i++ { + cv = cv>>8 | (uint64(d.content[off]) << 56) + h := hashLen(cv, betterLongTableBits, betterLongLen) + e.dictLongTable[h] = prevEntry{ + offset: i, + prev: e.dictLongTable[h].offset, + } + off++ + } + } + e.lastDictID = d.id + e.allDirty = true + } + + // Reset table to initial state + { + dirtyShardCnt := 0 + if !e.allDirty { + for i := range e.shortTableShardDirty { + if e.shortTableShardDirty[i] { + dirtyShardCnt++ + } + } + } + const shardCnt = betterShortTableShardCnt + const shardSize = betterShortTableShardSize + if e.allDirty || dirtyShardCnt > shardCnt*4/6 { + copy(e.table[:], e.dictTable) + for i := range e.shortTableShardDirty { + e.shortTableShardDirty[i] = false + } + } else { + for i := range e.shortTableShardDirty { + if !e.shortTableShardDirty[i] { + continue + } + + copy(e.table[i*shardSize:(i+1)*shardSize], e.dictTable[i*shardSize:(i+1)*shardSize]) + e.shortTableShardDirty[i] = false + } + } + } + { + dirtyShardCnt := 0 + if !e.allDirty { + for i := range e.shortTableShardDirty { + if e.shortTableShardDirty[i] { + dirtyShardCnt++ + } + } + } + const shardCnt = betterLongTableShardCnt + const shardSize = betterLongTableShardSize + if e.allDirty || dirtyShardCnt > shardCnt*4/6 { + copy(e.longTable[:], e.dictLongTable) + for i := range e.longTableShardDirty { + e.longTableShardDirty[i] = false + } + } else { + for i := range e.longTableShardDirty { + if !e.longTableShardDirty[i] { + continue + } + + copy(e.longTable[i*shardSize:(i+1)*shardSize], e.dictLongTable[i*shardSize:(i+1)*shardSize]) + e.longTableShardDirty[i] = false + } + } + } + e.cur = e.maxMatchOff + e.allDirty = false +} + +func (e *betterFastEncoderDict) markLongShardDirty(entryNum uint32) { + e.longTableShardDirty[entryNum/betterLongTableShardSize] = true +} + +func (e *betterFastEncoderDict) markShortShardDirty(entryNum uint32) { + e.shortTableShardDirty[entryNum/betterShortTableShardSize] = true +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_dfast.go b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go new file mode 100644 index 0000000..7ff0c64 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go @@ -0,0 +1,1124 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import "fmt" + +const ( + dFastLongTableBits = 17 // Bits used in the long match table + dFastLongTableSize = 1 << dFastLongTableBits // Size of the table + dFastLongTableMask = dFastLongTableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks. + dFastLongLen = 8 // Bytes used for table hash + + dLongTableShardCnt = 1 << (dFastLongTableBits - dictShardBits) // Number of shards in the table + dLongTableShardSize = dFastLongTableSize / tableShardCnt // Size of an individual shard + + dFastShortTableBits = tableBits // Bits used in the short match table + dFastShortTableSize = 1 << dFastShortTableBits // Size of the table + dFastShortTableMask = dFastShortTableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks. + dFastShortLen = 5 // Bytes used for table hash + +) + +type doubleFastEncoder struct { + fastEncoder + longTable [dFastLongTableSize]tableEntry +} + +type doubleFastEncoderDict struct { + fastEncoderDict + longTable [dFastLongTableSize]tableEntry + dictLongTable []tableEntry + longTableShardDirty [dLongTableShardCnt]bool +} + +// Encode mimmics functionality in zstd_dfast.c +func (e *doubleFastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = tableEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.longTable[i].offset = v + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) + nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + + if canRepeat { + if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += lenght + repOff + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + } + // Find the offsets of our two matches. + coffsetL := s - (candidateL.offset - e.cur) + coffsetS := s - (candidateS.offset - e.cur) + + // Check if we have a long match. + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + // Check if we have a short match. + if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hashLen(cv, dFastLongTableBits, dFastLongLen) + candidateL = e.longTable[nextHashL] + coffsetL = s - (candidateL.offset - e.cur) + checkAt + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)} + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + s += checkAt + if debugMatches { + println("long match (after short)") + } + break + } + + t = candidateS.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the 4-byte match as long as possible. + l := e.matchlen(s+4, t+4, src) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) and start+2 (short) + index0 := s - l + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load6432(src, index0) + cv1 := load6432(src, index1) + te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)} + te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)} + e.longTable[hashLen(cv0, dFastLongTableBits, dFastLongLen)] = te0 + e.longTable[hashLen(cv1, dFastLongTableBits, dFastLongLen)] = te1 + cv0 >>= 8 + cv1 >>= 8 + te0.offset++ + te1.offset++ + te0.val = uint32(cv0) + te1.val = uint32(cv1) + e.table[hashLen(cv0, dFastShortTableBits, dFastShortLen)] = te0 + e.table[hashLen(cv1, dFastShortTableBits, dFastShortLen)] = te1 + + cv = load6432(src, s) + + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) + nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) + + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *doubleFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + if e.cur >= bufferReset { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = tableEntry{} + } + e.cur = e.maxMatchOff + } + + s := int32(0) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + for { + + nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) + nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + + if len(blk.sequences) > 2 { + if load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + //length := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + length := 4 + int32(matchLen(src[s+4+repOff:], src[repIndex+4:])) + + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + repOff + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + } + // Find the offsets of our two matches. + coffsetL := s - (candidateL.offset - e.cur) + coffsetS := s - (candidateS.offset - e.cur) + + // Check if we have a long match. + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d). cur: %d", s, t, e.cur)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + // Check if we have a short match. + if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hashLen(cv, dFastLongTableBits, dFastLongLen) + candidateL = e.longTable[nextHashL] + coffsetL = s - (candidateL.offset - e.cur) + checkAt + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)} + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + s += checkAt + if debugMatches { + println("long match (after short)") + } + break + } + + t = candidateS.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + // Extend the 4-byte match as long as possible. + //l := e.matchlen(s+4, t+4, src) + 4 + l := int32(matchLen(src[s+4:], src[t+4:])) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) and start+2 (short) + index0 := s - l + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load6432(src, index0) + cv1 := load6432(src, index1) + te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)} + te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)} + e.longTable[hashLen(cv0, dFastLongTableBits, dFastLongLen)] = te0 + e.longTable[hashLen(cv1, dFastLongTableBits, dFastLongLen)] = te1 + cv0 >>= 8 + cv1 >>= 8 + te0.offset++ + te1.offset++ + te0.val = uint32(cv0) + te1.val = uint32(cv1) + e.table[hashLen(cv0, dFastShortTableBits, dFastShortLen)] = te0 + e.table[hashLen(cv1, dFastShortTableBits, dFastShortLen)] = te1 + + cv = load6432(src, s) + + if len(blk.sequences) <= 2 { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashS := hashLen(cv1>>8, dFastShortTableBits, dFastShortLen) + nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) + + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + //l := 4 + e.matchlen(s+4, o2+4, src) + l := 4 + int32(matchLen(src[s+4:], src[o2+4:])) + + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } + + // We do not store history, so we must offset e.cur to avoid false matches for next user. + if e.cur < bufferReset { + e.cur += int32(len(src)) + } +} + +// Encode will encode the content, with a dictionary if initialized for it. +func (e *doubleFastEncoderDict) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = tableEntry{} + } + e.markAllShardsDirty() + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.longTable[i].offset = v + } + e.markAllShardsDirty() + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) + nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.markLongShardDirty(nextHashL) + e.table[nextHashS] = entry + e.markShardDirty(nextHashS) + + if canRepeat { + if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += lenght + repOff + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + } + // Find the offsets of our two matches. + coffsetL := s - (candidateL.offset - e.cur) + coffsetS := s - (candidateS.offset - e.cur) + + // Check if we have a long match. + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + // Check if we have a short match. + if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hashLen(cv, dFastLongTableBits, dFastLongLen) + candidateL = e.longTable[nextHashL] + coffsetL = s - (candidateL.offset - e.cur) + checkAt + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)} + e.markLongShardDirty(nextHashL) + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + s += checkAt + if debugMatches { + println("long match (after short)") + } + break + } + + t = candidateS.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the 4-byte match as long as possible. + l := e.matchlen(s+4, t+4, src) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) and start+2 (short) + index0 := s - l + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load6432(src, index0) + cv1 := load6432(src, index1) + te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)} + te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)} + longHash1 := hashLen(cv0, dFastLongTableBits, dFastLongLen) + longHash2 := hashLen(cv1, dFastLongTableBits, dFastLongLen) + e.longTable[longHash1] = te0 + e.longTable[longHash2] = te1 + e.markLongShardDirty(longHash1) + e.markLongShardDirty(longHash2) + cv0 >>= 8 + cv1 >>= 8 + te0.offset++ + te1.offset++ + te0.val = uint32(cv0) + te1.val = uint32(cv1) + hashVal1 := hashLen(cv0, dFastShortTableBits, dFastShortLen) + hashVal2 := hashLen(cv1, dFastShortTableBits, dFastShortLen) + e.table[hashVal1] = te0 + e.markShardDirty(hashVal1) + e.table[hashVal2] = te1 + e.markShardDirty(hashVal2) + + cv = load6432(src, s) + + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) + nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) + + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.markLongShardDirty(nextHashL) + e.table[nextHashS] = entry + e.markShardDirty(nextHashS) + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } + // If we encoded more than 64K mark all dirty. + if len(src) > 64<<10 { + e.markAllShardsDirty() + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *doubleFastEncoder) Reset(d *dict, singleBlock bool) { + e.fastEncoder.Reset(d, singleBlock) + if d != nil { + panic("doubleFastEncoder: Reset with dict not supported") + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *doubleFastEncoderDict) Reset(d *dict, singleBlock bool) { + allDirty := e.allDirty + e.fastEncoderDict.Reset(d, singleBlock) + if d == nil { + return + } + + // Init or copy dict table + if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID { + if len(e.dictLongTable) != len(e.longTable) { + e.dictLongTable = make([]tableEntry, len(e.longTable)) + } + if len(d.content) >= 8 { + cv := load6432(d.content, 0) + e.dictLongTable[hashLen(cv, dFastLongTableBits, dFastLongLen)] = tableEntry{ + val: uint32(cv), + offset: e.maxMatchOff, + } + end := int32(len(d.content)) - 8 + e.maxMatchOff + for i := e.maxMatchOff + 1; i < end; i++ { + cv = cv>>8 | (uint64(d.content[i-e.maxMatchOff+7]) << 56) + e.dictLongTable[hashLen(cv, dFastLongTableBits, dFastLongLen)] = tableEntry{ + val: uint32(cv), + offset: i, + } + } + } + e.lastDictID = d.id + e.allDirty = true + } + // Reset table to initial state + e.cur = e.maxMatchOff + + dirtyShardCnt := 0 + if !allDirty { + for i := range e.longTableShardDirty { + if e.longTableShardDirty[i] { + dirtyShardCnt++ + } + } + } + + if allDirty || dirtyShardCnt > dLongTableShardCnt/2 { + copy(e.longTable[:], e.dictLongTable) + for i := range e.longTableShardDirty { + e.longTableShardDirty[i] = false + } + return + } + for i := range e.longTableShardDirty { + if !e.longTableShardDirty[i] { + continue + } + + copy(e.longTable[i*dLongTableShardSize:(i+1)*dLongTableShardSize], e.dictLongTable[i*dLongTableShardSize:(i+1)*dLongTableShardSize]) + e.longTableShardDirty[i] = false + } +} + +func (e *doubleFastEncoderDict) markLongShardDirty(entryNum uint32) { + e.longTableShardDirty[entryNum/dLongTableShardSize] = true +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_fast.go b/vendor/github.com/klauspost/compress/zstd/enc_fast.go new file mode 100644 index 0000000..f51ab52 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_fast.go @@ -0,0 +1,898 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "fmt" +) + +const ( + tableBits = 15 // Bits used in the table + tableSize = 1 << tableBits // Size of the table + tableShardCnt = 1 << (tableBits - dictShardBits) // Number of shards in the table + tableShardSize = tableSize / tableShardCnt // Size of an individual shard + tableFastHashLen = 6 + tableMask = tableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks. + maxMatchLength = 131074 +) + +type tableEntry struct { + val uint32 + offset int32 +} + +type fastEncoder struct { + fastBase + table [tableSize]tableEntry +} + +type fastEncoderDict struct { + fastEncoder + dictTable []tableEntry + tableShardDirty [tableShardCnt]bool + allDirty bool +} + +// Encode mimmics functionality in zstd_fast.c +func (e *fastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + inputMargin = 8 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 2. + const stepSize = 2 + + // TEMPLATE + const hashLog = tableBits + // seems global, but would be nice to tweak. + const kSearchStrength = 6 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + // t will contain the match offset when we find one. + // When existing the search loop, we have already checked 4 bytes. + var t int32 + + // We will not use repeat offsets across blocks. + // By not using them for the first 3 matches + canRepeat := len(blk.sequences) > 2 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHash := hashLen(cv, hashLog, tableFastHashLen) + nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen) + candidate := e.table[nextHash] + candidate2 := e.table[nextHash2] + repIndex := s - offset1 + 2 + + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} + + if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) { + // Consider history as well. + var seq seq + var length int32 + length = 4 + e.matchlen(s+6, repIndex+4, src) + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + 2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + sMin := s - e.maxMatchOff + if sMin < 0 { + sMin = 0 + } + for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + 2 + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + coffset0 := s - (candidate.offset - e.cur) + coffset1 := s - (candidate2.offset - e.cur) + 1 + if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { + // found a regular match + t = candidate.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + break + } + + if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { + // found a regular match + t = candidate2.offset - e.cur + s++ + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + break + } + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + // A 4-byte match has been found. We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the 4-byte match as long as possible. + l := e.matchlen(s+4, t+4, src) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence. + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + // Don't use repeat offsets + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + + // Check offset 2 + if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) { + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + // Store this, since we have it. + nextHash := hashLen(cv, hashLog, tableFastHashLen) + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + break encodeLoop + } + // Prepare next loop. + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *fastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + const ( + inputMargin = 8 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if debugEncoder { + if len(src) > maxBlockSize { + panic("src too big") + } + } + + // Protect against e.cur wraparound. + if e.cur >= bufferReset { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = e.maxMatchOff + } + + s := int32(0) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 2. + const stepSize = 2 + + // TEMPLATE + const hashLog = tableBits + // seems global, but would be nice to tweak. + const kSearchStrength = 6 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + // t will contain the match offset when we find one. + // When existing the search loop, we have already checked 4 bytes. + var t int32 + + // We will not use repeat offsets across blocks. + // By not using them for the first 3 matches + + for { + nextHash := hashLen(cv, hashLog, tableFastHashLen) + nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen) + candidate := e.table[nextHash] + candidate2 := e.table[nextHash2] + repIndex := s - offset1 + 2 + + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} + + if len(blk.sequences) > 2 && load3232(src, repIndex) == uint32(cv>>16) { + // Consider history as well. + var seq seq + length := 4 + e.matchlen(s+6, repIndex+4, src) + + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + 2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + sMin := s - e.maxMatchOff + if sMin < 0 { + sMin = 0 + } + for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + 2 + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + coffset0 := s - (candidate.offset - e.cur) + coffset1 := s - (candidate2.offset - e.cur) + 1 + if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { + // found a regular match + t = candidate.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic(fmt.Sprintf("t (%d) < 0, candidate.offset: %d, e.cur: %d, coffset0: %d, e.maxMatchOff: %d", t, candidate.offset, e.cur, coffset0, e.maxMatchOff)) + } + break + } + + if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { + // found a regular match + t = candidate2.offset - e.cur + s++ + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + break + } + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + // A 4-byte match has been found. We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && t < 0 { + panic(fmt.Sprintf("t (%d) < 0 ", t)) + } + // Extend the 4-byte match as long as possible. + l := e.matchlen(s+4, t+4, src) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + + // Write our sequence. + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + // Don't use repeat offsets + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + + // Check offset 2 + if o2 := s - offset2; len(blk.sequences) > 2 && load3232(src, o2) == uint32(cv) { + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + // Store this, since we have it. + nextHash := hashLen(cv, hashLog, tableFastHashLen) + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + break encodeLoop + } + // Prepare next loop. + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } + // We do not store history, so we must offset e.cur to avoid false matches for next user. + if e.cur < bufferReset { + e.cur += int32(len(src)) + } +} + +// Encode will encode the content, with a dictionary if initialized for it. +func (e *fastEncoderDict) Encode(blk *blockEnc, src []byte) { + const ( + inputMargin = 8 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if e.allDirty || len(src) > 32<<10 { + e.fastEncoder.Encode(blk, src) + e.allDirty = true + return + } + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 2. + const stepSize = 2 + + // TEMPLATE + const hashLog = tableBits + // seems global, but would be nice to tweak. + const kSearchStrength = 7 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debugEncoder { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + // t will contain the match offset when we find one. + // When existing the search loop, we have already checked 4 bytes. + var t int32 + + // We will not use repeat offsets across blocks. + // By not using them for the first 3 matches + canRepeat := len(blk.sequences) > 2 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHash := hashLen(cv, hashLog, tableFastHashLen) + nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen) + candidate := e.table[nextHash] + candidate2 := e.table[nextHash2] + repIndex := s - offset1 + 2 + + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.markShardDirty(nextHash) + e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} + e.markShardDirty(nextHash2) + + if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) { + // Consider history as well. + var seq seq + var length int32 + length = 4 + e.matchlen(s+6, repIndex+4, src) + + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + 2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + sMin := s - e.maxMatchOff + if sMin < 0 { + sMin = 0 + } + for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + 2 + nextEmit = s + if s >= sLimit { + if debugEncoder { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + coffset0 := s - (candidate.offset - e.cur) + coffset1 := s - (candidate2.offset - e.cur) + 1 + if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { + // found a regular match + t = candidate.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + break + } + + if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { + // found a regular match + t = candidate2.offset - e.cur + s++ + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + break + } + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + // A 4-byte match has been found. We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the 4-byte match as long as possible. + l := e.matchlen(s+4, t+4, src) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence. + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + // Don't use repeat offsets + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + + // Check offset 2 + if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) { + // We have at least 4 byte match. + // No need to check backwards. We come straight from a match + l := 4 + e.matchlen(s+4, o2+4, src) + + // Store this, since we have it. + nextHash := hashLen(cv, hashLog, tableFastHashLen) + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.markShardDirty(nextHash) + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + break encodeLoop + } + // Prepare next loop. + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debugEncoder { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *fastEncoder) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d != nil { + panic("fastEncoder: Reset with dict") + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *fastEncoderDict) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d == nil { + return + } + + // Init or copy dict table + if len(e.dictTable) != len(e.table) || d.id != e.lastDictID { + if len(e.dictTable) != len(e.table) { + e.dictTable = make([]tableEntry, len(e.table)) + } + if true { + end := e.maxMatchOff + int32(len(d.content)) - 8 + for i := e.maxMatchOff; i < end; i += 3 { + const hashLog = tableBits + + cv := load6432(d.content, i-e.maxMatchOff) + nextHash := hashLen(cv, hashLog, tableFastHashLen) // 0 -> 5 + nextHash1 := hashLen(cv>>8, hashLog, tableFastHashLen) // 1 -> 6 + nextHash2 := hashLen(cv>>16, hashLog, tableFastHashLen) // 2 -> 7 + e.dictTable[nextHash] = tableEntry{ + val: uint32(cv), + offset: i, + } + e.dictTable[nextHash1] = tableEntry{ + val: uint32(cv >> 8), + offset: i + 1, + } + e.dictTable[nextHash2] = tableEntry{ + val: uint32(cv >> 16), + offset: i + 2, + } + } + } + e.lastDictID = d.id + e.allDirty = true + } + + e.cur = e.maxMatchOff + dirtyShardCnt := 0 + if !e.allDirty { + for i := range e.tableShardDirty { + if e.tableShardDirty[i] { + dirtyShardCnt++ + } + } + } + + const shardCnt = tableShardCnt + const shardSize = tableShardSize + if e.allDirty || dirtyShardCnt > shardCnt*4/6 { + copy(e.table[:], e.dictTable) + for i := range e.tableShardDirty { + e.tableShardDirty[i] = false + } + e.allDirty = false + return + } + for i := range e.tableShardDirty { + if !e.tableShardDirty[i] { + continue + } + + copy(e.table[i*shardSize:(i+1)*shardSize], e.dictTable[i*shardSize:(i+1)*shardSize]) + e.tableShardDirty[i] = false + } + e.allDirty = false +} + +func (e *fastEncoderDict) markAllShardsDirty() { + e.allDirty = true +} + +func (e *fastEncoderDict) markShardDirty(entryNum uint32) { + e.tableShardDirty[entryNum/tableShardSize] = true +} diff --git a/vendor/github.com/klauspost/compress/zstd/encoder.go b/vendor/github.com/klauspost/compress/zstd/encoder.go new file mode 100644 index 0000000..7aaaedb --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/encoder.go @@ -0,0 +1,641 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "crypto/rand" + "fmt" + "io" + rdebug "runtime/debug" + "sync" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +// Encoder provides encoding to Zstandard. +// An Encoder can be used for either compressing a stream via the +// io.WriteCloser interface supported by the Encoder or as multiple independent +// tasks via the EncodeAll function. +// Smaller encodes are encouraged to use the EncodeAll function. +// Use NewWriter to create a new instance. +type Encoder struct { + o encoderOptions + encoders chan encoder + state encoderState + init sync.Once +} + +type encoder interface { + Encode(blk *blockEnc, src []byte) + EncodeNoHist(blk *blockEnc, src []byte) + Block() *blockEnc + CRC() *xxhash.Digest + AppendCRC([]byte) []byte + WindowSize(size int64) int32 + UseBlock(*blockEnc) + Reset(d *dict, singleBlock bool) +} + +type encoderState struct { + w io.Writer + filling []byte + current []byte + previous []byte + encoder encoder + writing *blockEnc + err error + writeErr error + nWritten int64 + nInput int64 + frameContentSize int64 + headerWritten bool + eofWritten bool + fullFrameWritten bool + + // This waitgroup indicates an encode is running. + wg sync.WaitGroup + // This waitgroup indicates we have a block encoding/writing. + wWg sync.WaitGroup +} + +// NewWriter will create a new Zstandard encoder. +// If the encoder will be used for encoding blocks a nil writer can be used. +func NewWriter(w io.Writer, opts ...EOption) (*Encoder, error) { + initPredefined() + var e Encoder + e.o.setDefault() + for _, o := range opts { + err := o(&e.o) + if err != nil { + return nil, err + } + } + if w != nil { + e.Reset(w) + } + return &e, nil +} + +func (e *Encoder) initialize() { + if e.o.concurrent == 0 { + e.o.setDefault() + } + e.encoders = make(chan encoder, e.o.concurrent) + for i := 0; i < e.o.concurrent; i++ { + enc := e.o.encoder() + e.encoders <- enc + } +} + +// Reset will re-initialize the writer and new writes will encode to the supplied writer +// as a new, independent stream. +func (e *Encoder) Reset(w io.Writer) { + s := &e.state + s.wg.Wait() + s.wWg.Wait() + if cap(s.filling) == 0 { + s.filling = make([]byte, 0, e.o.blockSize) + } + if e.o.concurrent > 1 { + if cap(s.current) == 0 { + s.current = make([]byte, 0, e.o.blockSize) + } + if cap(s.previous) == 0 { + s.previous = make([]byte, 0, e.o.blockSize) + } + s.current = s.current[:0] + s.previous = s.previous[:0] + if s.writing == nil { + s.writing = &blockEnc{lowMem: e.o.lowMem} + s.writing.init() + } + s.writing.initNewEncode() + } + if s.encoder == nil { + s.encoder = e.o.encoder() + } + s.filling = s.filling[:0] + s.encoder.Reset(e.o.dict, false) + s.headerWritten = false + s.eofWritten = false + s.fullFrameWritten = false + s.w = w + s.err = nil + s.nWritten = 0 + s.nInput = 0 + s.writeErr = nil + s.frameContentSize = 0 +} + +// ResetContentSize will reset and set a content size for the next stream. +// If the bytes written does not match the size given an error will be returned +// when calling Close(). +// This is removed when Reset is called. +// Sizes <= 0 results in no content size set. +func (e *Encoder) ResetContentSize(w io.Writer, size int64) { + e.Reset(w) + if size >= 0 { + e.state.frameContentSize = size + } +} + +// Write data to the encoder. +// Input data will be buffered and as the buffer fills up +// content will be compressed and written to the output. +// When done writing, use Close to flush the remaining output +// and write CRC if requested. +func (e *Encoder) Write(p []byte) (n int, err error) { + s := &e.state + for len(p) > 0 { + if len(p)+len(s.filling) < e.o.blockSize { + if e.o.crc { + _, _ = s.encoder.CRC().Write(p) + } + s.filling = append(s.filling, p...) + return n + len(p), nil + } + add := p + if len(p)+len(s.filling) > e.o.blockSize { + add = add[:e.o.blockSize-len(s.filling)] + } + if e.o.crc { + _, _ = s.encoder.CRC().Write(add) + } + s.filling = append(s.filling, add...) + p = p[len(add):] + n += len(add) + if len(s.filling) < e.o.blockSize { + return n, nil + } + err := e.nextBlock(false) + if err != nil { + return n, err + } + if debugAsserts && len(s.filling) > 0 { + panic(len(s.filling)) + } + } + return n, nil +} + +// nextBlock will synchronize and start compressing input in e.state.filling. +// If an error has occurred during encoding it will be returned. +func (e *Encoder) nextBlock(final bool) error { + s := &e.state + // Wait for current block. + s.wg.Wait() + if s.err != nil { + return s.err + } + if len(s.filling) > e.o.blockSize { + return fmt.Errorf("block > maxStoreBlockSize") + } + if !s.headerWritten { + // If we have a single block encode, do a sync compression. + if final && len(s.filling) == 0 && !e.o.fullZero { + s.headerWritten = true + s.fullFrameWritten = true + s.eofWritten = true + return nil + } + if final && len(s.filling) > 0 { + s.current = e.EncodeAll(s.filling, s.current[:0]) + var n2 int + n2, s.err = s.w.Write(s.current) + if s.err != nil { + return s.err + } + s.nWritten += int64(n2) + s.nInput += int64(len(s.filling)) + s.current = s.current[:0] + s.filling = s.filling[:0] + s.headerWritten = true + s.fullFrameWritten = true + s.eofWritten = true + return nil + } + + var tmp [maxHeaderSize]byte + fh := frameHeader{ + ContentSize: uint64(s.frameContentSize), + WindowSize: uint32(s.encoder.WindowSize(s.frameContentSize)), + SingleSegment: false, + Checksum: e.o.crc, + DictID: e.o.dict.ID(), + } + + dst, err := fh.appendTo(tmp[:0]) + if err != nil { + return err + } + s.headerWritten = true + s.wWg.Wait() + var n2 int + n2, s.err = s.w.Write(dst) + if s.err != nil { + return s.err + } + s.nWritten += int64(n2) + } + if s.eofWritten { + // Ensure we only write it once. + final = false + } + + if len(s.filling) == 0 { + // Final block, but no data. + if final { + enc := s.encoder + blk := enc.Block() + blk.reset(nil) + blk.last = true + blk.encodeRaw(nil) + s.wWg.Wait() + _, s.err = s.w.Write(blk.output) + s.nWritten += int64(len(blk.output)) + s.eofWritten = true + } + return s.err + } + + // SYNC: + if e.o.concurrent == 1 { + src := s.filling + s.nInput += int64(len(s.filling)) + if debugEncoder { + println("Adding sync block,", len(src), "bytes, final:", final) + } + enc := s.encoder + blk := enc.Block() + blk.reset(nil) + enc.Encode(blk, src) + blk.last = final + if final { + s.eofWritten = true + } + + err := errIncompressible + // If we got the exact same number of literals as input, + // assume the literals cannot be compressed. + if len(src) != len(blk.literals) || len(src) != e.o.blockSize { + err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy) + } + switch err { + case errIncompressible: + if debugEncoder { + println("Storing incompressible block as raw") + } + blk.encodeRaw(src) + // In fast mode, we do not transfer offsets, so we don't have to deal with changing the. + case nil: + default: + s.err = err + return err + } + _, s.err = s.w.Write(blk.output) + s.nWritten += int64(len(blk.output)) + s.filling = s.filling[:0] + return s.err + } + + // Move blocks forward. + s.filling, s.current, s.previous = s.previous[:0], s.filling, s.current + s.nInput += int64(len(s.current)) + s.wg.Add(1) + go func(src []byte) { + if debugEncoder { + println("Adding block,", len(src), "bytes, final:", final) + } + defer func() { + if r := recover(); r != nil { + s.err = fmt.Errorf("panic while encoding: %v", r) + rdebug.PrintStack() + } + s.wg.Done() + }() + enc := s.encoder + blk := enc.Block() + enc.Encode(blk, src) + blk.last = final + if final { + s.eofWritten = true + } + // Wait for pending writes. + s.wWg.Wait() + if s.writeErr != nil { + s.err = s.writeErr + return + } + // Transfer encoders from previous write block. + blk.swapEncoders(s.writing) + // Transfer recent offsets to next. + enc.UseBlock(s.writing) + s.writing = blk + s.wWg.Add(1) + go func() { + defer func() { + if r := recover(); r != nil { + s.writeErr = fmt.Errorf("panic while encoding/writing: %v", r) + rdebug.PrintStack() + } + s.wWg.Done() + }() + err := errIncompressible + // If we got the exact same number of literals as input, + // assume the literals cannot be compressed. + if len(src) != len(blk.literals) || len(src) != e.o.blockSize { + err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy) + } + switch err { + case errIncompressible: + if debugEncoder { + println("Storing incompressible block as raw") + } + blk.encodeRaw(src) + // In fast mode, we do not transfer offsets, so we don't have to deal with changing the. + case nil: + default: + s.writeErr = err + return + } + _, s.writeErr = s.w.Write(blk.output) + s.nWritten += int64(len(blk.output)) + }() + }(s.current) + return nil +} + +// ReadFrom reads data from r until EOF or error. +// The return value n is the number of bytes read. +// Any error except io.EOF encountered during the read is also returned. +// +// The Copy function uses ReaderFrom if available. +func (e *Encoder) ReadFrom(r io.Reader) (n int64, err error) { + if debugEncoder { + println("Using ReadFrom") + } + + // Flush any current writes. + if len(e.state.filling) > 0 { + if err := e.nextBlock(false); err != nil { + return 0, err + } + } + e.state.filling = e.state.filling[:e.o.blockSize] + src := e.state.filling + for { + n2, err := r.Read(src) + if e.o.crc { + _, _ = e.state.encoder.CRC().Write(src[:n2]) + } + // src is now the unfilled part... + src = src[n2:] + n += int64(n2) + switch err { + case io.EOF: + e.state.filling = e.state.filling[:len(e.state.filling)-len(src)] + if debugEncoder { + println("ReadFrom: got EOF final block:", len(e.state.filling)) + } + return n, nil + case nil: + default: + if debugEncoder { + println("ReadFrom: got error:", err) + } + e.state.err = err + return n, err + } + if len(src) > 0 { + if debugEncoder { + println("ReadFrom: got space left in source:", len(src)) + } + continue + } + err = e.nextBlock(false) + if err != nil { + return n, err + } + e.state.filling = e.state.filling[:e.o.blockSize] + src = e.state.filling + } +} + +// Flush will send the currently written data to output +// and block until everything has been written. +// This should only be used on rare occasions where pushing the currently queued data is critical. +func (e *Encoder) Flush() error { + s := &e.state + if len(s.filling) > 0 { + err := e.nextBlock(false) + if err != nil { + return err + } + } + s.wg.Wait() + s.wWg.Wait() + if s.err != nil { + return s.err + } + return s.writeErr +} + +// Close will flush the final output and close the stream. +// The function will block until everything has been written. +// The Encoder can still be re-used after calling this. +func (e *Encoder) Close() error { + s := &e.state + if s.encoder == nil { + return nil + } + err := e.nextBlock(true) + if err != nil { + return err + } + if s.frameContentSize > 0 { + if s.nInput != s.frameContentSize { + return fmt.Errorf("frame content size %d given, but %d bytes was written", s.frameContentSize, s.nInput) + } + } + if e.state.fullFrameWritten { + return s.err + } + s.wg.Wait() + s.wWg.Wait() + + if s.err != nil { + return s.err + } + if s.writeErr != nil { + return s.writeErr + } + + // Write CRC + if e.o.crc && s.err == nil { + // heap alloc. + var tmp [4]byte + _, s.err = s.w.Write(s.encoder.AppendCRC(tmp[:0])) + s.nWritten += 4 + } + + // Add padding with content from crypto/rand.Reader + if s.err == nil && e.o.pad > 0 { + add := calcSkippableFrame(s.nWritten, int64(e.o.pad)) + frame, err := skippableFrame(s.filling[:0], add, rand.Reader) + if err != nil { + return err + } + _, s.err = s.w.Write(frame) + } + return s.err +} + +// EncodeAll will encode all input in src and append it to dst. +// This function can be called concurrently, but each call will only run on a single goroutine. +// If empty input is given, nothing is returned, unless WithZeroFrames is specified. +// Encoded blocks can be concatenated and the result will be the combined input stream. +// Data compressed with EncodeAll can be decoded with the Decoder, +// using either a stream or DecodeAll. +func (e *Encoder) EncodeAll(src, dst []byte) []byte { + if len(src) == 0 { + if e.o.fullZero { + // Add frame header. + fh := frameHeader{ + ContentSize: 0, + WindowSize: MinWindowSize, + SingleSegment: true, + // Adding a checksum would be a waste of space. + Checksum: false, + DictID: 0, + } + dst, _ = fh.appendTo(dst) + + // Write raw block as last one only. + var blk blockHeader + blk.setSize(0) + blk.setType(blockTypeRaw) + blk.setLast(true) + dst = blk.appendTo(dst) + } + return dst + } + e.init.Do(e.initialize) + enc := <-e.encoders + defer func() { + // Release encoder reference to last block. + // If a non-single block is needed the encoder will reset again. + e.encoders <- enc + }() + // Use single segments when above minimum window and below window size. + single := len(src) <= e.o.windowSize && len(src) > MinWindowSize + if e.o.single != nil { + single = *e.o.single + } + fh := frameHeader{ + ContentSize: uint64(len(src)), + WindowSize: uint32(enc.WindowSize(int64(len(src)))), + SingleSegment: single, + Checksum: e.o.crc, + DictID: e.o.dict.ID(), + } + + // If less than 1MB, allocate a buffer up front. + if len(dst) == 0 && cap(dst) == 0 && len(src) < 1<<20 && !e.o.lowMem { + dst = make([]byte, 0, len(src)) + } + dst, err := fh.appendTo(dst) + if err != nil { + panic(err) + } + + // If we can do everything in one block, prefer that. + if len(src) <= e.o.blockSize { + enc.Reset(e.o.dict, true) + // Slightly faster with no history and everything in one block. + if e.o.crc { + _, _ = enc.CRC().Write(src) + } + blk := enc.Block() + blk.last = true + if e.o.dict == nil { + enc.EncodeNoHist(blk, src) + } else { + enc.Encode(blk, src) + } + + // If we got the exact same number of literals as input, + // assume the literals cannot be compressed. + err := errIncompressible + oldout := blk.output + if len(blk.literals) != len(src) || len(src) != e.o.blockSize { + // Output directly to dst + blk.output = dst + err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy) + } + + switch err { + case errIncompressible: + if debugEncoder { + println("Storing incompressible block as raw") + } + dst = blk.encodeRawTo(dst, src) + case nil: + dst = blk.output + default: + panic(err) + } + blk.output = oldout + } else { + enc.Reset(e.o.dict, false) + blk := enc.Block() + for len(src) > 0 { + todo := src + if len(todo) > e.o.blockSize { + todo = todo[:e.o.blockSize] + } + src = src[len(todo):] + if e.o.crc { + _, _ = enc.CRC().Write(todo) + } + blk.pushOffsets() + enc.Encode(blk, todo) + if len(src) == 0 { + blk.last = true + } + err := errIncompressible + // If we got the exact same number of literals as input, + // assume the literals cannot be compressed. + if len(blk.literals) != len(todo) || len(todo) != e.o.blockSize { + err = blk.encode(todo, e.o.noEntropy, !e.o.allLitEntropy) + } + + switch err { + case errIncompressible: + if debugEncoder { + println("Storing incompressible block as raw") + } + dst = blk.encodeRawTo(dst, todo) + blk.popOffsets() + case nil: + dst = append(dst, blk.output...) + default: + panic(err) + } + blk.reset(nil) + } + } + if e.o.crc { + dst = enc.AppendCRC(dst) + } + // Add padding with content from crypto/rand.Reader + if e.o.pad > 0 { + add := calcSkippableFrame(int64(len(dst)), int64(e.o.pad)) + dst, err = skippableFrame(dst, add, rand.Reader) + if err != nil { + panic(err) + } + } + return dst +} diff --git a/vendor/github.com/klauspost/compress/zstd/encoder_options.go b/vendor/github.com/klauspost/compress/zstd/encoder_options.go new file mode 100644 index 0000000..a7c5e1a --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/encoder_options.go @@ -0,0 +1,317 @@ +package zstd + +import ( + "errors" + "fmt" + "runtime" + "strings" +) + +// EOption is an option for creating a encoder. +type EOption func(*encoderOptions) error + +// options retains accumulated state of multiple options. +type encoderOptions struct { + concurrent int + level EncoderLevel + single *bool + pad int + blockSize int + windowSize int + crc bool + fullZero bool + noEntropy bool + allLitEntropy bool + customWindow bool + customALEntropy bool + customBlockSize bool + lowMem bool + dict *dict +} + +func (o *encoderOptions) setDefault() { + *o = encoderOptions{ + concurrent: runtime.GOMAXPROCS(0), + crc: true, + single: nil, + blockSize: maxCompressedBlockSize, + windowSize: 8 << 20, + level: SpeedDefault, + allLitEntropy: true, + lowMem: false, + } +} + +// encoder returns an encoder with the selected options. +func (o encoderOptions) encoder() encoder { + switch o.level { + case SpeedFastest: + if o.dict != nil { + return &fastEncoderDict{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}} + } + return &fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}} + + case SpeedDefault: + if o.dict != nil { + return &doubleFastEncoderDict{fastEncoderDict: fastEncoderDict{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}}} + } + return &doubleFastEncoder{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}} + case SpeedBetterCompression: + if o.dict != nil { + return &betterFastEncoderDict{betterFastEncoder: betterFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}} + } + return &betterFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}} + case SpeedBestCompression: + return &bestFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}} + } + panic("unknown compression level") +} + +// WithEncoderCRC will add CRC value to output. +// Output will be 4 bytes larger. +func WithEncoderCRC(b bool) EOption { + return func(o *encoderOptions) error { o.crc = b; return nil } +} + +// WithEncoderConcurrency will set the concurrency, +// meaning the maximum number of encoders to run concurrently. +// The value supplied must be at least 1. +// For streams, setting a value of 1 will disable async compression. +// By default this will be set to GOMAXPROCS. +func WithEncoderConcurrency(n int) EOption { + return func(o *encoderOptions) error { + if n <= 0 { + return fmt.Errorf("concurrency must be at least 1") + } + o.concurrent = n + return nil + } +} + +// WithWindowSize will set the maximum allowed back-reference distance. +// The value must be a power of two between MinWindowSize and MaxWindowSize. +// A larger value will enable better compression but allocate more memory and, +// for above-default values, take considerably longer. +// The default value is determined by the compression level. +func WithWindowSize(n int) EOption { + return func(o *encoderOptions) error { + switch { + case n < MinWindowSize: + return fmt.Errorf("window size must be at least %d", MinWindowSize) + case n > MaxWindowSize: + return fmt.Errorf("window size must be at most %d", MaxWindowSize) + case (n & (n - 1)) != 0: + return errors.New("window size must be a power of 2") + } + + o.windowSize = n + o.customWindow = true + if o.blockSize > o.windowSize { + o.blockSize = o.windowSize + o.customBlockSize = true + } + return nil + } +} + +// WithEncoderPadding will add padding to all output so the size will be a multiple of n. +// This can be used to obfuscate the exact output size or make blocks of a certain size. +// The contents will be a skippable frame, so it will be invisible by the decoder. +// n must be > 0 and <= 1GB, 1<<30 bytes. +// The padded area will be filled with data from crypto/rand.Reader. +// If `EncodeAll` is used with data already in the destination, the total size will be multiple of this. +func WithEncoderPadding(n int) EOption { + return func(o *encoderOptions) error { + if n <= 0 { + return fmt.Errorf("padding must be at least 1") + } + // No need to waste our time. + if n == 1 { + o.pad = 0 + } + if n > 1<<30 { + return fmt.Errorf("padding must less than 1GB (1<<30 bytes) ") + } + o.pad = n + return nil + } +} + +// EncoderLevel predefines encoder compression levels. +// Only use the constants made available, since the actual mapping +// of these values are very likely to change and your compression could change +// unpredictably when upgrading the library. +type EncoderLevel int + +const ( + speedNotSet EncoderLevel = iota + + // SpeedFastest will choose the fastest reasonable compression. + // This is roughly equivalent to the fastest Zstandard mode. + SpeedFastest + + // SpeedDefault is the default "pretty fast" compression option. + // This is roughly equivalent to the default Zstandard mode (level 3). + SpeedDefault + + // SpeedBetterCompression will yield better compression than the default. + // Currently it is about zstd level 7-8 with ~ 2x-3x the default CPU usage. + // By using this, notice that CPU usage may go up in the future. + SpeedBetterCompression + + // SpeedBestCompression will choose the best available compression option. + // This will offer the best compression no matter the CPU cost. + SpeedBestCompression + + // speedLast should be kept as the last actual compression option. + // The is not for external usage, but is used to keep track of the valid options. + speedLast +) + +// EncoderLevelFromString will convert a string representation of an encoding level back +// to a compression level. The compare is not case sensitive. +// If the string wasn't recognized, (false, SpeedDefault) will be returned. +func EncoderLevelFromString(s string) (bool, EncoderLevel) { + for l := speedNotSet + 1; l < speedLast; l++ { + if strings.EqualFold(s, l.String()) { + return true, l + } + } + return false, SpeedDefault +} + +// EncoderLevelFromZstd will return an encoder level that closest matches the compression +// ratio of a specific zstd compression level. +// Many input values will provide the same compression level. +func EncoderLevelFromZstd(level int) EncoderLevel { + switch { + case level < 3: + return SpeedFastest + case level >= 3 && level < 6: + return SpeedDefault + case level >= 6 && level < 10: + return SpeedBetterCompression + default: + return SpeedBestCompression + } +} + +// String provides a string representation of the compression level. +func (e EncoderLevel) String() string { + switch e { + case SpeedFastest: + return "fastest" + case SpeedDefault: + return "default" + case SpeedBetterCompression: + return "better" + case SpeedBestCompression: + return "best" + default: + return "invalid" + } +} + +// WithEncoderLevel specifies a predefined compression level. +func WithEncoderLevel(l EncoderLevel) EOption { + return func(o *encoderOptions) error { + switch { + case l <= speedNotSet || l >= speedLast: + return fmt.Errorf("unknown encoder level") + } + o.level = l + if !o.customWindow { + switch o.level { + case SpeedFastest: + o.windowSize = 4 << 20 + if !o.customBlockSize { + o.blockSize = 1 << 16 + } + case SpeedDefault: + o.windowSize = 8 << 20 + case SpeedBetterCompression: + o.windowSize = 16 << 20 + case SpeedBestCompression: + o.windowSize = 32 << 20 + } + } + if !o.customALEntropy { + o.allLitEntropy = l > SpeedFastest + } + + return nil + } +} + +// WithZeroFrames will encode 0 length input as full frames. +// This can be needed for compatibility with zstandard usage, +// but is not needed for this package. +func WithZeroFrames(b bool) EOption { + return func(o *encoderOptions) error { + o.fullZero = b + return nil + } +} + +// WithAllLitEntropyCompression will apply entropy compression if no matches are found. +// Disabling this will skip incompressible data faster, but in cases with no matches but +// skewed character distribution compression is lost. +// Default value depends on the compression level selected. +func WithAllLitEntropyCompression(b bool) EOption { + return func(o *encoderOptions) error { + o.customALEntropy = true + o.allLitEntropy = b + return nil + } +} + +// WithNoEntropyCompression will always skip entropy compression of literals. +// This can be useful if content has matches, but unlikely to benefit from entropy +// compression. Usually the slight speed improvement is not worth enabling this. +func WithNoEntropyCompression(b bool) EOption { + return func(o *encoderOptions) error { + o.noEntropy = b + return nil + } +} + +// WithSingleSegment will set the "single segment" flag when EncodeAll is used. +// If this flag is set, data must be regenerated within a single continuous memory segment. +// In this case, Window_Descriptor byte is skipped, but Frame_Content_Size is necessarily present. +// As a consequence, the decoder must allocate a memory segment of size equal or larger than size of your content. +// In order to preserve the decoder from unreasonable memory requirements, +// a decoder is allowed to reject a compressed frame which requests a memory size beyond decoder's authorized range. +// For broader compatibility, decoders are recommended to support memory sizes of at least 8 MB. +// This is only a recommendation, each decoder is free to support higher or lower limits, depending on local limitations. +// If this is not specified, block encodes will automatically choose this based on the input size and the window size. +// This setting has no effect on streamed encodes. +func WithSingleSegment(b bool) EOption { + return func(o *encoderOptions) error { + o.single = &b + return nil + } +} + +// WithLowerEncoderMem will trade in some memory cases trade less memory usage for +// slower encoding speed. +// This will not change the window size which is the primary function for reducing +// memory usage. See WithWindowSize. +func WithLowerEncoderMem(b bool) EOption { + return func(o *encoderOptions) error { + o.lowMem = b + return nil + } +} + +// WithEncoderDict allows to register a dictionary that will be used for the encode. +// The encoder *may* choose to use no dictionary instead for certain payloads. +func WithEncoderDict(dict []byte) EOption { + return func(o *encoderOptions) error { + d, err := loadDict(dict) + if err != nil { + return err + } + o.dict = d + return nil + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/framedec.go b/vendor/github.com/klauspost/compress/zstd/framedec.go new file mode 100644 index 0000000..9568a4b --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/framedec.go @@ -0,0 +1,419 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "bytes" + "encoding/hex" + "errors" + "io" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +type frameDec struct { + o decoderOptions + crc *xxhash.Digest + + WindowSize uint64 + + // Frame history passed between blocks + history history + + rawInput byteBuffer + + // Byte buffer that can be reused for small input blocks. + bBuf byteBuf + + FrameContentSize uint64 + + DictionaryID *uint32 + HasCheckSum bool + SingleSegment bool +} + +const ( + // MinWindowSize is the minimum Window Size, which is 1 KB. + MinWindowSize = 1 << 10 + + // MaxWindowSize is the maximum encoder window size + // and the default decoder maximum window size. + MaxWindowSize = 1 << 29 +) + +var ( + frameMagic = []byte{0x28, 0xb5, 0x2f, 0xfd} + skippableFrameMagic = []byte{0x2a, 0x4d, 0x18} +) + +func newFrameDec(o decoderOptions) *frameDec { + if o.maxWindowSize > o.maxDecodedSize { + o.maxWindowSize = o.maxDecodedSize + } + d := frameDec{ + o: o, + } + return &d +} + +// reset will read the frame header and prepare for block decoding. +// If nothing can be read from the input, io.EOF will be returned. +// Any other error indicated that the stream contained data, but +// there was a problem. +func (d *frameDec) reset(br byteBuffer) error { + d.HasCheckSum = false + d.WindowSize = 0 + var signature [4]byte + for { + var err error + // Check if we can read more... + b, err := br.readSmall(1) + switch err { + case io.EOF, io.ErrUnexpectedEOF: + return io.EOF + default: + return err + case nil: + signature[0] = b[0] + } + // Read the rest, don't allow io.ErrUnexpectedEOF + b, err = br.readSmall(3) + switch err { + case io.EOF: + return io.EOF + default: + return err + case nil: + copy(signature[1:], b) + } + + if !bytes.Equal(signature[1:4], skippableFrameMagic) || signature[0]&0xf0 != 0x50 { + if debugDecoder { + println("Not skippable", hex.EncodeToString(signature[:]), hex.EncodeToString(skippableFrameMagic)) + } + // Break if not skippable frame. + break + } + // Read size to skip + b, err = br.readSmall(4) + if err != nil { + if debugDecoder { + println("Reading Frame Size", err) + } + return err + } + n := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + println("Skipping frame with", n, "bytes.") + err = br.skipN(int64(n)) + if err != nil { + if debugDecoder { + println("Reading discarded frame", err) + } + return err + } + } + if !bytes.Equal(signature[:], frameMagic) { + if debugDecoder { + println("Got magic numbers: ", signature, "want:", frameMagic) + } + return ErrMagicMismatch + } + + // Read Frame_Header_Descriptor + fhd, err := br.readByte() + if err != nil { + if debugDecoder { + println("Reading Frame_Header_Descriptor", err) + } + return err + } + d.SingleSegment = fhd&(1<<5) != 0 + + if fhd&(1<<3) != 0 { + return errors.New("reserved bit set on frame header") + } + + // Read Window_Descriptor + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#window_descriptor + d.WindowSize = 0 + if !d.SingleSegment { + wd, err := br.readByte() + if err != nil { + if debugDecoder { + println("Reading Window_Descriptor", err) + } + return err + } + printf("raw: %x, mantissa: %d, exponent: %d\n", wd, wd&7, wd>>3) + windowLog := 10 + (wd >> 3) + windowBase := uint64(1) << windowLog + windowAdd := (windowBase / 8) * uint64(wd&0x7) + d.WindowSize = windowBase + windowAdd + } + + // Read Dictionary_ID + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary_id + d.DictionaryID = nil + if size := fhd & 3; size != 0 { + if size == 3 { + size = 4 + } + + b, err := br.readSmall(int(size)) + if err != nil { + println("Reading Dictionary_ID", err) + return err + } + var id uint32 + switch size { + case 1: + id = uint32(b[0]) + case 2: + id = uint32(b[0]) | (uint32(b[1]) << 8) + case 4: + id = uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + } + if debugDecoder { + println("Dict size", size, "ID:", id) + } + if id > 0 { + // ID 0 means "sorry, no dictionary anyway". + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary-format + d.DictionaryID = &id + } + } + + // Read Frame_Content_Size + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frame_content_size + var fcsSize int + v := fhd >> 6 + switch v { + case 0: + if d.SingleSegment { + fcsSize = 1 + } + default: + fcsSize = 1 << v + } + d.FrameContentSize = fcsUnknown + if fcsSize > 0 { + b, err := br.readSmall(fcsSize) + if err != nil { + println("Reading Frame content", err) + return err + } + switch fcsSize { + case 1: + d.FrameContentSize = uint64(b[0]) + case 2: + // When FCS_Field_Size is 2, the offset of 256 is added. + d.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) + 256 + case 4: + d.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) | (uint64(b[2]) << 16) | (uint64(b[3]) << 24) + case 8: + d1 := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + d2 := uint32(b[4]) | (uint32(b[5]) << 8) | (uint32(b[6]) << 16) | (uint32(b[7]) << 24) + d.FrameContentSize = uint64(d1) | (uint64(d2) << 32) + } + if debugDecoder { + println("Read FCS:", d.FrameContentSize) + } + } + + // Move this to shared. + d.HasCheckSum = fhd&(1<<2) != 0 + if d.HasCheckSum { + if d.crc == nil { + d.crc = xxhash.New() + } + d.crc.Reset() + } + + if d.WindowSize > d.o.maxWindowSize { + if debugDecoder { + printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize) + } + return ErrWindowSizeExceeded + } + + if d.WindowSize == 0 && d.SingleSegment { + // We may not need window in this case. + d.WindowSize = d.FrameContentSize + if d.WindowSize < MinWindowSize { + d.WindowSize = MinWindowSize + } + if d.WindowSize > d.o.maxDecodedSize { + if debugDecoder { + printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize) + } + return ErrDecoderSizeExceeded + } + } + + // The minimum Window_Size is 1 KB. + if d.WindowSize < MinWindowSize { + if debugDecoder { + println("got window size: ", d.WindowSize) + } + return ErrWindowSizeTooSmall + } + d.history.windowSize = int(d.WindowSize) + if !d.o.lowMem || d.history.windowSize < maxBlockSize { + // Alloc 2x window size if not low-mem, or very small window size. + d.history.allocFrameBuffer = d.history.windowSize * 2 + } else { + // Alloc with one additional block + d.history.allocFrameBuffer = d.history.windowSize + maxBlockSize + } + + if debugDecoder { + println("Frame: Dict:", d.DictionaryID, "FrameContentSize:", d.FrameContentSize, "singleseg:", d.SingleSegment, "window:", d.WindowSize, "crc:", d.HasCheckSum) + } + + // history contains input - maybe we do something + d.rawInput = br + return nil +} + +// next will start decoding the next block from stream. +func (d *frameDec) next(block *blockDec) error { + if debugDecoder { + println("decoding new block") + } + err := block.reset(d.rawInput, d.WindowSize) + if err != nil { + println("block error:", err) + // Signal the frame decoder we have a problem. + block.sendErr(err) + return err + } + return nil +} + +// checkCRC will check the checksum if the frame has one. +// Will return ErrCRCMismatch if crc check failed, otherwise nil. +func (d *frameDec) checkCRC() error { + if !d.HasCheckSum { + return nil + } + + // We can overwrite upper tmp now + want, err := d.rawInput.readSmall(4) + if err != nil { + println("CRC missing?", err) + return err + } + + if d.o.ignoreChecksum { + return nil + } + + var tmp [4]byte + got := d.crc.Sum64() + // Flip to match file order. + tmp[0] = byte(got >> 0) + tmp[1] = byte(got >> 8) + tmp[2] = byte(got >> 16) + tmp[3] = byte(got >> 24) + + if !bytes.Equal(tmp[:], want) { + if debugDecoder { + println("CRC Check Failed:", tmp[:], "!=", want) + } + return ErrCRCMismatch + } + if debugDecoder { + println("CRC ok", tmp[:]) + } + return nil +} + +// consumeCRC reads the checksum data if the frame has one. +func (d *frameDec) consumeCRC() error { + if d.HasCheckSum { + _, err := d.rawInput.readSmall(4) + if err != nil { + println("CRC missing?", err) + return err + } + } + + return nil +} + +// runDecoder will create a sync decoder that will decode a block of data. +func (d *frameDec) runDecoder(dst []byte, dec *blockDec) ([]byte, error) { + saved := d.history.b + + // We use the history for output to avoid copying it. + d.history.b = dst + d.history.ignoreBuffer = len(dst) + // Store input length, so we only check new data. + crcStart := len(dst) + d.history.decoders.maxSyncLen = 0 + if d.FrameContentSize != fcsUnknown { + d.history.decoders.maxSyncLen = d.FrameContentSize + uint64(len(dst)) + if d.history.decoders.maxSyncLen > d.o.maxDecodedSize { + return dst, ErrDecoderSizeExceeded + } + if uint64(cap(dst)) < d.history.decoders.maxSyncLen { + // Alloc for output + dst2 := make([]byte, len(dst), d.history.decoders.maxSyncLen+compressedBlockOverAlloc) + copy(dst2, dst) + dst = dst2 + } + } + var err error + for { + err = dec.reset(d.rawInput, d.WindowSize) + if err != nil { + break + } + if debugDecoder { + println("next block:", dec) + } + err = dec.decodeBuf(&d.history) + if err != nil { + break + } + if uint64(len(d.history.b)) > d.o.maxDecodedSize { + err = ErrDecoderSizeExceeded + break + } + if uint64(len(d.history.b)-crcStart) > d.FrameContentSize { + println("runDecoder: FrameContentSize exceeded", uint64(len(d.history.b)-crcStart), ">", d.FrameContentSize) + err = ErrFrameSizeExceeded + break + } + if dec.Last { + break + } + if debugDecoder { + println("runDecoder: FrameContentSize", uint64(len(d.history.b)-crcStart), "<=", d.FrameContentSize) + } + } + dst = d.history.b + if err == nil { + if d.FrameContentSize != fcsUnknown && uint64(len(d.history.b)-crcStart) != d.FrameContentSize { + err = ErrFrameSizeMismatch + } else if d.HasCheckSum { + if d.o.ignoreChecksum { + err = d.consumeCRC() + } else { + var n int + n, err = d.crc.Write(dst[crcStart:]) + if err == nil { + if n != len(dst)-crcStart { + err = io.ErrShortWrite + } else { + err = d.checkCRC() + } + } + } + } + } + d.history.b = saved + return dst, err +} diff --git a/vendor/github.com/klauspost/compress/zstd/frameenc.go b/vendor/github.com/klauspost/compress/zstd/frameenc.go new file mode 100644 index 0000000..4ef7f5a --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/frameenc.go @@ -0,0 +1,137 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "fmt" + "io" + "math" + "math/bits" +) + +type frameHeader struct { + ContentSize uint64 + WindowSize uint32 + SingleSegment bool + Checksum bool + DictID uint32 +} + +const maxHeaderSize = 14 + +func (f frameHeader) appendTo(dst []byte) ([]byte, error) { + dst = append(dst, frameMagic...) + var fhd uint8 + if f.Checksum { + fhd |= 1 << 2 + } + if f.SingleSegment { + fhd |= 1 << 5 + } + + var dictIDContent []byte + if f.DictID > 0 { + var tmp [4]byte + if f.DictID < 256 { + fhd |= 1 + tmp[0] = uint8(f.DictID) + dictIDContent = tmp[:1] + } else if f.DictID < 1<<16 { + fhd |= 2 + binary.LittleEndian.PutUint16(tmp[:2], uint16(f.DictID)) + dictIDContent = tmp[:2] + } else { + fhd |= 3 + binary.LittleEndian.PutUint32(tmp[:4], f.DictID) + dictIDContent = tmp[:4] + } + } + var fcs uint8 + if f.ContentSize >= 256 { + fcs++ + } + if f.ContentSize >= 65536+256 { + fcs++ + } + if f.ContentSize >= 0xffffffff { + fcs++ + } + + fhd |= fcs << 6 + + dst = append(dst, fhd) + if !f.SingleSegment { + const winLogMin = 10 + windowLog := (bits.Len32(f.WindowSize-1) - winLogMin) << 3 + dst = append(dst, uint8(windowLog)) + } + if f.DictID > 0 { + dst = append(dst, dictIDContent...) + } + switch fcs { + case 0: + if f.SingleSegment { + dst = append(dst, uint8(f.ContentSize)) + } + // Unless SingleSegment is set, framessizes < 256 are nto stored. + case 1: + f.ContentSize -= 256 + dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8)) + case 2: + dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8), uint8(f.ContentSize>>16), uint8(f.ContentSize>>24)) + case 3: + dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8), uint8(f.ContentSize>>16), uint8(f.ContentSize>>24), + uint8(f.ContentSize>>32), uint8(f.ContentSize>>40), uint8(f.ContentSize>>48), uint8(f.ContentSize>>56)) + default: + panic("invalid fcs") + } + return dst, nil +} + +const skippableFrameHeader = 4 + 4 + +// calcSkippableFrame will return a total size to be added for written +// to be divisible by multiple. +// The value will always be > skippableFrameHeader. +// The function will panic if written < 0 or wantMultiple <= 0. +func calcSkippableFrame(written, wantMultiple int64) int { + if wantMultiple <= 0 { + panic("wantMultiple <= 0") + } + if written < 0 { + panic("written < 0") + } + leftOver := written % wantMultiple + if leftOver == 0 { + return 0 + } + toAdd := wantMultiple - leftOver + for toAdd < skippableFrameHeader { + toAdd += wantMultiple + } + return int(toAdd) +} + +// skippableFrame will add a skippable frame with a total size of bytes. +// total should be >= skippableFrameHeader and < math.MaxUint32. +func skippableFrame(dst []byte, total int, r io.Reader) ([]byte, error) { + if total == 0 { + return dst, nil + } + if total < skippableFrameHeader { + return dst, fmt.Errorf("requested skippable frame (%d) < 8", total) + } + if int64(total) > math.MaxUint32 { + return dst, fmt.Errorf("requested skippable frame (%d) > max uint32", total) + } + dst = append(dst, 0x50, 0x2a, 0x4d, 0x18) + f := uint32(total - skippableFrameHeader) + dst = append(dst, uint8(f), uint8(f>>8), uint8(f>>16), uint8(f>>24)) + start := len(dst) + dst = append(dst, make([]byte, f)...) + _, err := io.ReadFull(r, dst[start:]) + return dst, err +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go new file mode 100644 index 0000000..2f8860a --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go @@ -0,0 +1,307 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "errors" + "fmt" + "io" +) + +const ( + tablelogAbsoluteMax = 9 +) + +const ( + /*!MEMORY_USAGE : + * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.) + * Increasing memory usage improves compression ratio + * Reduced memory usage can improve speed, due to cache effect + * Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */ + maxMemoryUsage = tablelogAbsoluteMax + 2 + + maxTableLog = maxMemoryUsage - 2 + maxTablesize = 1 << maxTableLog + maxTableMask = (1 << maxTableLog) - 1 + minTablelog = 5 + maxSymbolValue = 255 +) + +// fseDecoder provides temporary storage for compression and decompression. +type fseDecoder struct { + dt [maxTablesize]decSymbol // Decompression table. + symbolLen uint16 // Length of active part of the symbol table. + actualTableLog uint8 // Selected tablelog. + maxBits uint8 // Maximum number of additional bits + + // used for table creation to avoid allocations. + stateTable [256]uint16 + norm [maxSymbolValue + 1]int16 + preDefined bool +} + +// tableStep returns the next table index. +func tableStep(tableSize uint32) uint32 { + return (tableSize >> 1) + (tableSize >> 3) + 3 +} + +// readNCount will read the symbol distribution so decoding tables can be constructed. +func (s *fseDecoder) readNCount(b *byteReader, maxSymbol uint16) error { + var ( + charnum uint16 + previous0 bool + ) + if b.remain() < 4 { + return errors.New("input too small") + } + bitStream := b.Uint32NC() + nbBits := uint((bitStream & 0xF) + minTablelog) // extract tableLog + if nbBits > tablelogAbsoluteMax { + println("Invalid tablelog:", nbBits) + return errors.New("tableLog too large") + } + bitStream >>= 4 + bitCount := uint(4) + + s.actualTableLog = uint8(nbBits) + remaining := int32((1 << nbBits) + 1) + threshold := int32(1 << nbBits) + gotTotal := int32(0) + nbBits++ + + for remaining > 1 && charnum <= maxSymbol { + if previous0 { + //println("prev0") + n0 := charnum + for (bitStream & 0xFFFF) == 0xFFFF { + //println("24 x 0") + n0 += 24 + if r := b.remain(); r > 5 { + b.advance(2) + // The check above should make sure we can read 32 bits + bitStream = b.Uint32NC() >> bitCount + } else { + // end of bit stream + bitStream >>= 16 + bitCount += 16 + } + } + //printf("bitstream: %d, 0b%b", bitStream&3, bitStream) + for (bitStream & 3) == 3 { + n0 += 3 + bitStream >>= 2 + bitCount += 2 + } + n0 += uint16(bitStream & 3) + bitCount += 2 + + if n0 > maxSymbolValue { + return errors.New("maxSymbolValue too small") + } + //println("inserting ", n0-charnum, "zeroes from idx", charnum, "ending before", n0) + for charnum < n0 { + s.norm[uint8(charnum)] = 0 + charnum++ + } + + if r := b.remain(); r >= 7 || r-int(bitCount>>3) >= 4 { + b.advance(bitCount >> 3) + bitCount &= 7 + // The check above should make sure we can read 32 bits + bitStream = b.Uint32NC() >> bitCount + } else { + bitStream >>= 2 + } + } + + max := (2*threshold - 1) - remaining + var count int32 + + if int32(bitStream)&(threshold-1) < max { + count = int32(bitStream) & (threshold - 1) + if debugAsserts && nbBits < 1 { + panic("nbBits underflow") + } + bitCount += nbBits - 1 + } else { + count = int32(bitStream) & (2*threshold - 1) + if count >= threshold { + count -= max + } + bitCount += nbBits + } + + // extra accuracy + count-- + if count < 0 { + // -1 means +1 + remaining += count + gotTotal -= count + } else { + remaining -= count + gotTotal += count + } + s.norm[charnum&0xff] = int16(count) + charnum++ + previous0 = count == 0 + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + + if r := b.remain(); r >= 7 || r-int(bitCount>>3) >= 4 { + b.advance(bitCount >> 3) + bitCount &= 7 + // The check above should make sure we can read 32 bits + bitStream = b.Uint32NC() >> (bitCount & 31) + } else { + bitCount -= (uint)(8 * (len(b.b) - 4 - b.off)) + b.off = len(b.b) - 4 + bitStream = b.Uint32() >> (bitCount & 31) + } + } + s.symbolLen = charnum + if s.symbolLen <= 1 { + return fmt.Errorf("symbolLen (%d) too small", s.symbolLen) + } + if s.symbolLen > maxSymbolValue+1 { + return fmt.Errorf("symbolLen (%d) too big", s.symbolLen) + } + if remaining != 1 { + return fmt.Errorf("corruption detected (remaining %d != 1)", remaining) + } + if bitCount > 32 { + return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount) + } + if gotTotal != 1<> 3) + return s.buildDtable() +} + +func (s *fseDecoder) mustReadFrom(r io.Reader) { + fatalErr := func(err error) { + if err != nil { + panic(err) + } + } + // dt [maxTablesize]decSymbol // Decompression table. + // symbolLen uint16 // Length of active part of the symbol table. + // actualTableLog uint8 // Selected tablelog. + // maxBits uint8 // Maximum number of additional bits + // // used for table creation to avoid allocations. + // stateTable [256]uint16 + // norm [maxSymbolValue + 1]int16 + // preDefined bool + fatalErr(binary.Read(r, binary.LittleEndian, &s.dt)) + fatalErr(binary.Read(r, binary.LittleEndian, &s.symbolLen)) + fatalErr(binary.Read(r, binary.LittleEndian, &s.actualTableLog)) + fatalErr(binary.Read(r, binary.LittleEndian, &s.maxBits)) + fatalErr(binary.Read(r, binary.LittleEndian, &s.stateTable)) + fatalErr(binary.Read(r, binary.LittleEndian, &s.norm)) + fatalErr(binary.Read(r, binary.LittleEndian, &s.preDefined)) +} + +// decSymbol contains information about a state entry, +// Including the state offset base, the output symbol and +// the number of bits to read for the low part of the destination state. +// Using a composite uint64 is faster than a struct with separate members. +type decSymbol uint64 + +func newDecSymbol(nbits, addBits uint8, newState uint16, baseline uint32) decSymbol { + return decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32) +} + +func (d decSymbol) nbBits() uint8 { + return uint8(d) +} + +func (d decSymbol) addBits() uint8 { + return uint8(d >> 8) +} + +func (d decSymbol) newState() uint16 { + return uint16(d >> 16) +} + +func (d decSymbol) baselineInt() int { + return int(d >> 32) +} + +func (d *decSymbol) setNBits(nBits uint8) { + const mask = 0xffffffffffffff00 + *d = (*d & mask) | decSymbol(nBits) +} + +func (d *decSymbol) setAddBits(addBits uint8) { + const mask = 0xffffffffffff00ff + *d = (*d & mask) | (decSymbol(addBits) << 8) +} + +func (d *decSymbol) setNewState(state uint16) { + const mask = 0xffffffff0000ffff + *d = (*d & mask) | decSymbol(state)<<16 +} + +func (d *decSymbol) setExt(addBits uint8, baseline uint32) { + const mask = 0xffff00ff + *d = (*d & mask) | (decSymbol(addBits) << 8) | (decSymbol(baseline) << 32) +} + +// decSymbolValue returns the transformed decSymbol for the given symbol. +func decSymbolValue(symb uint8, t []baseOffset) (decSymbol, error) { + if int(symb) >= len(t) { + return 0, fmt.Errorf("rle symbol %d >= max %d", symb, len(t)) + } + lu := t[symb] + return newDecSymbol(0, lu.addBits, 0, lu.baseLine), nil +} + +// setRLE will set the decoder til RLE mode. +func (s *fseDecoder) setRLE(symbol decSymbol) { + s.actualTableLog = 0 + s.maxBits = symbol.addBits() + s.dt[0] = symbol +} + +// transform will transform the decoder table into a table usable for +// decoding without having to apply the transformation while decoding. +// The state will contain the base value and the number of bits to read. +func (s *fseDecoder) transform(t []baseOffset) error { + tableSize := uint16(1 << s.actualTableLog) + s.maxBits = 0 + for i, v := range s.dt[:tableSize] { + add := v.addBits() + if int(add) >= len(t) { + return fmt.Errorf("invalid decoding table entry %d, symbol %d >= max (%d)", i, v.addBits(), len(t)) + } + lu := t[add] + if lu.addBits > s.maxBits { + s.maxBits = lu.addBits + } + v.setExt(lu.addBits, lu.baseLine) + s.dt[i] = v + } + return nil +} + +type fseState struct { + dt []decSymbol + state decSymbol +} + +// Initialize and decodeAsync first state and symbol. +func (s *fseState) init(br *bitReader, tableLog uint8, dt []decSymbol) { + s.dt = dt + br.fill() + s.state = dt[br.getBits(tableLog)] +} + +// final returns the current state symbol without decoding the next. +func (s decSymbol) final() (int, uint8) { + return s.baselineInt(), s.addBits() +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go new file mode 100644 index 0000000..c881d28 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go @@ -0,0 +1,64 @@ +//go:build amd64 && !appengine && !noasm && gc +// +build amd64,!appengine,!noasm,gc + +package zstd + +import ( + "fmt" +) + +type buildDtableAsmContext struct { + // inputs + stateTable *uint16 + norm *int16 + dt *uint64 + + // outputs --- set by the procedure in the case of error; + // for interpretation please see the error handling part below + errParam1 uint64 + errParam2 uint64 +} + +// buildDtable_asm is an x86 assembly implementation of fseDecoder.buildDtable. +// Function returns non-zero exit code on error. +// go:noescape +func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int + +// please keep in sync with _generate/gen_fse.go +const ( + errorCorruptedNormalizedCounter = 1 + errorNewStateTooBig = 2 + errorNewStateNoBits = 3 +) + +// buildDtable will build the decoding table. +func (s *fseDecoder) buildDtable() error { + ctx := buildDtableAsmContext{ + stateTable: &s.stateTable[0], + norm: &s.norm[0], + dt: (*uint64)(&s.dt[0]), + } + code := buildDtable_asm(s, &ctx) + + if code != 0 { + switch code { + case errorCorruptedNormalizedCounter: + position := ctx.errParam1 + return fmt.Errorf("corrupted input (position=%d, expected 0)", position) + + case errorNewStateTooBig: + newState := decSymbol(ctx.errParam1) + size := ctx.errParam2 + return fmt.Errorf("newState (%d) outside table size (%d)", newState, size) + + case errorNewStateNoBits: + newState := decSymbol(ctx.errParam1) + oldState := decSymbol(ctx.errParam2) + return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, oldState) + + default: + return fmt.Errorf("buildDtable_asm returned unhandled nonzero code = %d", code) + } + } + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s new file mode 100644 index 0000000..da32b44 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s @@ -0,0 +1,127 @@ +// Code generated by command: go run gen_fse.go -out ../fse_decoder_amd64.s -pkg=zstd. DO NOT EDIT. + +//go:build !appengine && !noasm && gc && !noasm +// +build !appengine,!noasm,gc,!noasm + +// func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int +TEXT ·buildDtable_asm(SB), $0-24 + MOVQ ctx+8(FP), CX + MOVQ s+0(FP), DI + + // Load values + MOVBQZX 4098(DI), DX + XORQ AX, AX + BTSQ DX, AX + MOVQ (CX), BX + MOVQ 16(CX), SI + LEAQ -1(AX), R8 + MOVQ 8(CX), CX + MOVWQZX 4096(DI), DI + + // End load values + // Init, lay down lowprob symbols + XORQ R9, R9 + JMP init_main_loop_condition + +init_main_loop: + MOVWQSX (CX)(R9*2), R10 + CMPW R10, $-1 + JNE do_not_update_high_threshold + MOVB R9, 1(SI)(R8*8) + DECQ R8 + MOVQ $0x0000000000000001, R10 + +do_not_update_high_threshold: + MOVW R10, (BX)(R9*2) + INCQ R9 + +init_main_loop_condition: + CMPQ R9, DI + JL init_main_loop + + // Spread symbols + // Calculate table step + MOVQ AX, R9 + SHRQ $0x01, R9 + MOVQ AX, R10 + SHRQ $0x03, R10 + LEAQ 3(R9)(R10*1), R9 + + // Fill add bits values + LEAQ -1(AX), R10 + XORQ R11, R11 + XORQ R12, R12 + JMP spread_main_loop_condition + +spread_main_loop: + XORQ R13, R13 + MOVWQSX (CX)(R12*2), R14 + JMP spread_inner_loop_condition + +spread_inner_loop: + MOVB R12, 1(SI)(R11*8) + +adjust_position: + ADDQ R9, R11 + ANDQ R10, R11 + CMPQ R11, R8 + JG adjust_position + INCQ R13 + +spread_inner_loop_condition: + CMPQ R13, R14 + JL spread_inner_loop + INCQ R12 + +spread_main_loop_condition: + CMPQ R12, DI + JL spread_main_loop + TESTQ R11, R11 + JZ spread_check_ok + MOVQ ctx+8(FP), AX + MOVQ R11, 24(AX) + MOVQ $+1, ret+16(FP) + RET + +spread_check_ok: + // Build Decoding table + XORQ DI, DI + +build_table_main_table: + MOVBQZX 1(SI)(DI*8), CX + MOVWQZX (BX)(CX*2), R8 + LEAQ 1(R8), R9 + MOVW R9, (BX)(CX*2) + MOVQ R8, R9 + BSRQ R9, R9 + MOVQ DX, CX + SUBQ R9, CX + SHLQ CL, R8 + SUBQ AX, R8 + MOVB CL, (SI)(DI*8) + MOVW R8, 2(SI)(DI*8) + CMPQ R8, AX + JLE build_table_check1_ok + MOVQ ctx+8(FP), CX + MOVQ R8, 24(CX) + MOVQ AX, 32(CX) + MOVQ $+2, ret+16(FP) + RET + +build_table_check1_ok: + TESTB CL, CL + JNZ build_table_check2_ok + CMPW R8, DI + JNE build_table_check2_ok + MOVQ ctx+8(FP), AX + MOVQ R8, 24(AX) + MOVQ DI, 32(AX) + MOVQ $+3, ret+16(FP) + RET + +build_table_check2_ok: + INCQ DI + CMPQ DI, AX + JL build_table_main_table + MOVQ $+0, ret+16(FP) + RET diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go new file mode 100644 index 0000000..332e51f --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go @@ -0,0 +1,72 @@ +//go:build !amd64 || appengine || !gc || noasm +// +build !amd64 appengine !gc noasm + +package zstd + +import ( + "errors" + "fmt" +) + +// buildDtable will build the decoding table. +func (s *fseDecoder) buildDtable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + symbolNext := s.stateTable[:256] + + // Init, lay down lowprob symbols + { + for i, v := range s.norm[:s.symbolLen] { + if v == -1 { + s.dt[highThreshold].setAddBits(uint8(i)) + highThreshold-- + symbolNext[i] = 1 + } else { + symbolNext[i] = uint16(v) + } + } + } + + // Spread symbols + { + tableMask := tableSize - 1 + step := tableStep(tableSize) + position := uint32(0) + for ss, v := range s.norm[:s.symbolLen] { + for i := 0; i < int(v); i++ { + s.dt[position].setAddBits(uint8(ss)) + position = (position + step) & tableMask + for position > highThreshold { + // lowprob area + position = (position + step) & tableMask + } + } + } + if position != 0 { + // position must reach all cells once, otherwise normalizedCounter is incorrect + return errors.New("corrupted input (position != 0)") + } + } + + // Build Decoding table + { + tableSize := uint16(1 << s.actualTableLog) + for u, v := range s.dt[:tableSize] { + symbol := v.addBits() + nextState := symbolNext[symbol] + symbolNext[symbol] = nextState + 1 + nBits := s.actualTableLog - byte(highBits(uint32(nextState))) + s.dt[u&maxTableMask].setNBits(nBits) + newState := (nextState << nBits) - tableSize + if newState > tableSize { + return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize) + } + if newState == uint16(u) && nBits == 0 { + // Seems weird that this is possible with nbits > 0. + return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u) + } + s.dt[u&maxTableMask].setNewState(newState) + } + } + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_encoder.go b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go new file mode 100644 index 0000000..ab26326 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go @@ -0,0 +1,701 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "math" +) + +const ( + // For encoding we only support up to + maxEncTableLog = 8 + maxEncTablesize = 1 << maxTableLog + maxEncTableMask = (1 << maxTableLog) - 1 + minEncTablelog = 5 + maxEncSymbolValue = maxMatchLengthSymbol +) + +// Scratch provides temporary storage for compression and decompression. +type fseEncoder struct { + symbolLen uint16 // Length of active part of the symbol table. + actualTableLog uint8 // Selected tablelog. + ct cTable // Compression tables. + maxCount int // count of the most probable symbol + zeroBits bool // no bits has prob > 50%. + clearCount bool // clear count + useRLE bool // This encoder is for RLE + preDefined bool // This encoder is predefined. + reUsed bool // Set to know when the encoder has been reused. + rleVal uint8 // RLE Symbol + maxBits uint8 // Maximum output bits after transform. + + // TODO: Technically zstd should be fine with 64 bytes. + count [256]uint32 + norm [256]int16 +} + +// cTable contains tables used for compression. +type cTable struct { + tableSymbol []byte + stateTable []uint16 + symbolTT []symbolTransform +} + +// symbolTransform contains the state transform for a symbol. +type symbolTransform struct { + deltaNbBits uint32 + deltaFindState int16 + outBits uint8 +} + +// String prints values as a human readable string. +func (s symbolTransform) String() string { + return fmt.Sprintf("{deltabits: %08x, findstate:%d outbits:%d}", s.deltaNbBits, s.deltaFindState, s.outBits) +} + +// Histogram allows to populate the histogram and skip that step in the compression, +// It otherwise allows to inspect the histogram when compression is done. +// To indicate that you have populated the histogram call HistogramFinished +// with the value of the highest populated symbol, as well as the number of entries +// in the most populated entry. These are accepted at face value. +func (s *fseEncoder) Histogram() *[256]uint32 { + return &s.count +} + +// HistogramFinished can be called to indicate that the histogram has been populated. +// maxSymbol is the index of the highest set symbol of the next data segment. +// maxCount is the number of entries in the most populated entry. +// These are accepted at face value. +func (s *fseEncoder) HistogramFinished(maxSymbol uint8, maxCount int) { + s.maxCount = maxCount + s.symbolLen = uint16(maxSymbol) + 1 + s.clearCount = maxCount != 0 +} + +// allocCtable will allocate tables needed for compression. +// If existing tables a re big enough, they are simply re-used. +func (s *fseEncoder) allocCtable() { + tableSize := 1 << s.actualTableLog + // get tableSymbol that is big enough. + if cap(s.ct.tableSymbol) < tableSize { + s.ct.tableSymbol = make([]byte, tableSize) + } + s.ct.tableSymbol = s.ct.tableSymbol[:tableSize] + + ctSize := tableSize + if cap(s.ct.stateTable) < ctSize { + s.ct.stateTable = make([]uint16, ctSize) + } + s.ct.stateTable = s.ct.stateTable[:ctSize] + + if cap(s.ct.symbolTT) < 256 { + s.ct.symbolTT = make([]symbolTransform, 256) + } + s.ct.symbolTT = s.ct.symbolTT[:256] +} + +// buildCTable will populate the compression table so it is ready to be used. +func (s *fseEncoder) buildCTable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + var cumul [256]int16 + + s.allocCtable() + tableSymbol := s.ct.tableSymbol[:tableSize] + // symbol start positions + { + cumul[0] = 0 + for ui, v := range s.norm[:s.symbolLen-1] { + u := byte(ui) // one less than reference + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = u + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + } + // Encode last symbol separately to avoid overflowing u + u := int(s.symbolLen - 1) + v := s.norm[s.symbolLen-1] + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = byte(u) + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + if uint32(cumul[s.symbolLen]) != tableSize { + return fmt.Errorf("internal error: expected cumul[s.symbolLen] (%d) == tableSize (%d)", cumul[s.symbolLen], tableSize) + } + cumul[s.symbolLen] = int16(tableSize) + 1 + } + // Spread symbols + s.zeroBits = false + { + step := tableStep(tableSize) + tableMask := tableSize - 1 + var position uint32 + // if any symbol > largeLimit, we may have 0 bits output. + largeLimit := int16(1 << (s.actualTableLog - 1)) + for ui, v := range s.norm[:s.symbolLen] { + symbol := byte(ui) + if v > largeLimit { + s.zeroBits = true + } + for nbOccurrences := int16(0); nbOccurrences < v; nbOccurrences++ { + tableSymbol[position] = symbol + position = (position + step) & tableMask + for position > highThreshold { + position = (position + step) & tableMask + } /* Low proba area */ + } + } + + // Check if we have gone through all positions + if position != 0 { + return errors.New("position!=0") + } + } + + // Build table + table := s.ct.stateTable + { + tsi := int(tableSize) + for u, v := range tableSymbol { + // TableU16 : sorted by symbol order; gives next state value + table[cumul[v]] = uint16(tsi + u) + cumul[v]++ + } + } + + // Build Symbol Transformation Table + { + total := int16(0) + symbolTT := s.ct.symbolTT[:s.symbolLen] + tableLog := s.actualTableLog + tl := (uint32(tableLog) << 16) - (1 << tableLog) + for i, v := range s.norm[:s.symbolLen] { + switch v { + case 0: + case -1, 1: + symbolTT[i].deltaNbBits = tl + symbolTT[i].deltaFindState = total - 1 + total++ + default: + maxBitsOut := uint32(tableLog) - highBit(uint32(v-1)) + minStatePlus := uint32(v) << maxBitsOut + symbolTT[i].deltaNbBits = (maxBitsOut << 16) - minStatePlus + symbolTT[i].deltaFindState = total - v + total += v + } + } + if total != int16(tableSize) { + return fmt.Errorf("total mismatch %d (got) != %d (want)", total, tableSize) + } + } + return nil +} + +var rtbTable = [...]uint32{0, 473195, 504333, 520860, 550000, 700000, 750000, 830000} + +func (s *fseEncoder) setRLE(val byte) { + s.allocCtable() + s.actualTableLog = 0 + s.ct.stateTable = s.ct.stateTable[:1] + s.ct.symbolTT[val] = symbolTransform{ + deltaFindState: 0, + deltaNbBits: 0, + } + if debugEncoder { + println("setRLE: val", val, "symbolTT", s.ct.symbolTT[val]) + } + s.rleVal = val + s.useRLE = true +} + +// setBits will set output bits for the transform. +// if nil is provided, the number of bits is equal to the index. +func (s *fseEncoder) setBits(transform []byte) { + if s.reUsed || s.preDefined { + return + } + if s.useRLE { + if transform == nil { + s.ct.symbolTT[s.rleVal].outBits = s.rleVal + s.maxBits = s.rleVal + return + } + s.maxBits = transform[s.rleVal] + s.ct.symbolTT[s.rleVal].outBits = s.maxBits + return + } + if transform == nil { + for i := range s.ct.symbolTT[:s.symbolLen] { + s.ct.symbolTT[i].outBits = uint8(i) + } + s.maxBits = uint8(s.symbolLen - 1) + return + } + s.maxBits = 0 + for i, v := range transform[:s.symbolLen] { + s.ct.symbolTT[i].outBits = v + if v > s.maxBits { + // We could assume bits always going up, but we play safe. + s.maxBits = v + } + } +} + +// normalizeCount will normalize the count of the symbols so +// the total is equal to the table size. +// If successful, compression tables will also be made ready. +func (s *fseEncoder) normalizeCount(length int) error { + if s.reUsed { + return nil + } + s.optimalTableLog(length) + var ( + tableLog = s.actualTableLog + scale = 62 - uint64(tableLog) + step = (1 << 62) / uint64(length) + vStep = uint64(1) << (scale - 20) + stillToDistribute = int16(1 << tableLog) + largest int + largestP int16 + lowThreshold = (uint32)(length >> tableLog) + ) + if s.maxCount == length { + s.useRLE = true + return nil + } + s.useRLE = false + for i, cnt := range s.count[:s.symbolLen] { + // already handled + // if (count[s] == s.length) return 0; /* rle special case */ + + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + stillToDistribute-- + } else { + proba := (int16)((uint64(cnt) * step) >> scale) + if proba < 8 { + restToBeat := vStep * uint64(rtbTable[proba]) + v := uint64(cnt)*step - (uint64(proba) << scale) + if v > restToBeat { + proba++ + } + } + if proba > largestP { + largestP = proba + largest = i + } + s.norm[i] = proba + stillToDistribute -= proba + } + } + + if -stillToDistribute >= (s.norm[largest] >> 1) { + // corner case, need another normalization method + err := s.normalizeCount2(length) + if err != nil { + return err + } + if debugAsserts { + err = s.validateNorm() + if err != nil { + return err + } + } + return s.buildCTable() + } + s.norm[largest] += stillToDistribute + if debugAsserts { + err := s.validateNorm() + if err != nil { + return err + } + } + return s.buildCTable() +} + +// Secondary normalization method. +// To be used when primary method fails. +func (s *fseEncoder) normalizeCount2(length int) error { + const notYetAssigned = -2 + var ( + distributed uint32 + total = uint32(length) + tableLog = s.actualTableLog + lowThreshold = total >> tableLog + lowOne = (total * 3) >> (tableLog + 1) + ) + for i, cnt := range s.count[:s.symbolLen] { + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + distributed++ + total -= cnt + continue + } + if cnt <= lowOne { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + s.norm[i] = notYetAssigned + } + toDistribute := (1 << tableLog) - distributed + + if (total / toDistribute) > lowOne { + // risk of rounding to zero + lowOne = (total * 3) / (toDistribute * 2) + for i, cnt := range s.count[:s.symbolLen] { + if (s.norm[i] == notYetAssigned) && (cnt <= lowOne) { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + } + toDistribute = (1 << tableLog) - distributed + } + if distributed == uint32(s.symbolLen)+1 { + // all values are pretty poor; + // probably incompressible data (should have already been detected); + // find max, then give all remaining points to max + var maxV int + var maxC uint32 + for i, cnt := range s.count[:s.symbolLen] { + if cnt > maxC { + maxV = i + maxC = cnt + } + } + s.norm[maxV] += int16(toDistribute) + return nil + } + + if total == 0 { + // all of the symbols were low enough for the lowOne or lowThreshold + for i := uint32(0); toDistribute > 0; i = (i + 1) % (uint32(s.symbolLen)) { + if s.norm[i] > 0 { + toDistribute-- + s.norm[i]++ + } + } + return nil + } + + var ( + vStepLog = 62 - uint64(tableLog) + mid = uint64((1 << (vStepLog - 1)) - 1) + rStep = (((1 << vStepLog) * uint64(toDistribute)) + mid) / uint64(total) // scale on remaining + tmpTotal = mid + ) + for i, cnt := range s.count[:s.symbolLen] { + if s.norm[i] == notYetAssigned { + var ( + end = tmpTotal + uint64(cnt)*rStep + sStart = uint32(tmpTotal >> vStepLog) + sEnd = uint32(end >> vStepLog) + weight = sEnd - sStart + ) + if weight < 1 { + return errors.New("weight < 1") + } + s.norm[i] = int16(weight) + tmpTotal = end + } + } + return nil +} + +// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog +func (s *fseEncoder) optimalTableLog(length int) { + tableLog := uint8(maxEncTableLog) + minBitsSrc := highBit(uint32(length)) + 1 + minBitsSymbols := highBit(uint32(s.symbolLen-1)) + 2 + minBits := uint8(minBitsSymbols) + if minBitsSrc < minBitsSymbols { + minBits = uint8(minBitsSrc) + } + + maxBitsSrc := uint8(highBit(uint32(length-1))) - 2 + if maxBitsSrc < tableLog { + // Accuracy can be reduced + tableLog = maxBitsSrc + } + if minBits > tableLog { + tableLog = minBits + } + // Need a minimum to safely represent all symbol values + if tableLog < minEncTablelog { + tableLog = minEncTablelog + } + if tableLog > maxEncTableLog { + tableLog = maxEncTableLog + } + s.actualTableLog = tableLog +} + +// validateNorm validates the normalized histogram table. +func (s *fseEncoder) validateNorm() (err error) { + var total int + for _, v := range s.norm[:s.symbolLen] { + if v >= 0 { + total += int(v) + } else { + total -= int(v) + } + } + defer func() { + if err == nil { + return + } + fmt.Printf("selected TableLog: %d, Symbol length: %d\n", s.actualTableLog, s.symbolLen) + for i, v := range s.norm[:s.symbolLen] { + fmt.Printf("%3d: %5d -> %4d \n", i, s.count[i], v) + } + }() + if total != (1 << s.actualTableLog) { + return fmt.Errorf("warning: Total == %d != %d", total, 1<> 3) + 3 + 2 + + // Write Table Size + bitStream = uint32(tableLog - minEncTablelog) + bitCount = uint(4) + remaining = int16(tableSize + 1) /* +1 for extra accuracy */ + threshold = int16(tableSize) + nbBits = uint(tableLog + 1) + outP = len(out) + ) + if cap(out) < outP+maxHeaderSize { + out = append(out, make([]byte, maxHeaderSize*3)...) + out = out[:len(out)-maxHeaderSize*3] + } + out = out[:outP+maxHeaderSize] + + // stops at 1 + for remaining > 1 { + if previous0 { + start := charnum + for s.norm[charnum] == 0 { + charnum++ + } + for charnum >= start+24 { + start += 24 + bitStream += uint32(0xFFFF) << bitCount + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + } + for charnum >= start+3 { + start += 3 + bitStream += 3 << bitCount + bitCount += 2 + } + bitStream += uint32(charnum-start) << bitCount + bitCount += 2 + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + count := s.norm[charnum] + charnum++ + max := (2*threshold - 1) - remaining + if count < 0 { + remaining += count + } else { + remaining -= count + } + count++ // +1 for extra accuracy + if count >= threshold { + count += max // [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ + } + bitStream += uint32(count) << bitCount + bitCount += nbBits + if count < max { + bitCount-- + } + + previous0 = count == 1 + if remaining < 1 { + return nil, errors.New("internal error: remaining < 1") + } + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + if outP+2 > len(out) { + return nil, fmt.Errorf("internal error: %d > %d, maxheader: %d, sl: %d, tl: %d, normcount: %v", outP+2, len(out), maxHeaderSize, s.symbolLen, int(tableLog), s.norm[:s.symbolLen]) + } + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += int((bitCount + 7) / 8) + + if charnum > s.symbolLen { + return nil, errors.New("internal error: charnum > s.symbolLen") + } + return out[:outP], nil +} + +// Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits) +// note 1 : assume symbolValue is valid (<= maxSymbolValue) +// note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits * +func (s *fseEncoder) bitCost(symbolValue uint8, accuracyLog uint32) uint32 { + minNbBits := s.ct.symbolTT[symbolValue].deltaNbBits >> 16 + threshold := (minNbBits + 1) << 16 + if debugAsserts { + if !(s.actualTableLog < 16) { + panic("!s.actualTableLog < 16") + } + // ensure enough room for renormalization double shift + if !(uint8(accuracyLog) < 31-s.actualTableLog) { + panic("!uint8(accuracyLog) < 31-s.actualTableLog") + } + } + tableSize := uint32(1) << s.actualTableLog + deltaFromThreshold := threshold - (s.ct.symbolTT[symbolValue].deltaNbBits + tableSize) + // linear interpolation (very approximate) + normalizedDeltaFromThreshold := (deltaFromThreshold << accuracyLog) >> s.actualTableLog + bitMultiplier := uint32(1) << accuracyLog + if debugAsserts { + if s.ct.symbolTT[symbolValue].deltaNbBits+tableSize > threshold { + panic("s.ct.symbolTT[symbolValue].deltaNbBits+tableSize > threshold") + } + if normalizedDeltaFromThreshold > bitMultiplier { + panic("normalizedDeltaFromThreshold > bitMultiplier") + } + } + return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold +} + +// Returns the cost in bits of encoding the distribution in count using ctable. +// Histogram should only be up to the last non-zero symbol. +// Returns an -1 if ctable cannot represent all the symbols in count. +func (s *fseEncoder) approxSize(hist []uint32) uint32 { + if int(s.symbolLen) < len(hist) { + // More symbols than we have. + return math.MaxUint32 + } + if s.useRLE { + // We will never reuse RLE encoders. + return math.MaxUint32 + } + const kAccuracyLog = 8 + badCost := (uint32(s.actualTableLog) + 1) << kAccuracyLog + var cost uint32 + for i, v := range hist { + if v == 0 { + continue + } + if s.norm[i] == 0 { + return math.MaxUint32 + } + bitCost := s.bitCost(uint8(i), kAccuracyLog) + if bitCost > badCost { + return math.MaxUint32 + } + cost += v * bitCost + } + return cost >> kAccuracyLog +} + +// maxHeaderSize returns the maximum header size in bits. +// This is not exact size, but we want a penalty for new tables anyway. +func (s *fseEncoder) maxHeaderSize() uint32 { + if s.preDefined { + return 0 + } + if s.useRLE { + return 8 + } + return (((uint32(s.symbolLen) * uint32(s.actualTableLog)) >> 3) + 3) * 8 +} + +// cState contains the compression state of a stream. +type cState struct { + bw *bitWriter + stateTable []uint16 + state uint16 +} + +// init will initialize the compression state to the first symbol of the stream. +func (c *cState) init(bw *bitWriter, ct *cTable, first symbolTransform) { + c.bw = bw + c.stateTable = ct.stateTable + if len(c.stateTable) == 1 { + // RLE + c.stateTable[0] = uint16(0) + c.state = 0 + return + } + nbBitsOut := (first.deltaNbBits + (1 << 15)) >> 16 + im := int32((nbBitsOut << 16) - first.deltaNbBits) + lu := (im >> nbBitsOut) + int32(first.deltaFindState) + c.state = c.stateTable[lu] +} + +// flush will write the tablelog to the output and flush the remaining full bytes. +func (c *cState) flush(tableLog uint8) { + c.bw.flush32() + c.bw.addBits16NC(c.state, tableLog) +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_predefined.go b/vendor/github.com/klauspost/compress/zstd/fse_predefined.go new file mode 100644 index 0000000..474cb77 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_predefined.go @@ -0,0 +1,158 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "fmt" + "math" + "sync" +) + +var ( + // fsePredef are the predefined fse tables as defined here: + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions + // These values are already transformed. + fsePredef [3]fseDecoder + + // fsePredefEnc are the predefined encoder based on fse tables as defined here: + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions + // These values are already transformed. + fsePredefEnc [3]fseEncoder + + // symbolTableX contain the transformations needed for each type as defined in + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets + symbolTableX [3][]baseOffset + + // maxTableSymbol is the biggest supported symbol for each table type + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets + maxTableSymbol = [3]uint8{tableLiteralLengths: maxLiteralLengthSymbol, tableOffsets: maxOffsetLengthSymbol, tableMatchLengths: maxMatchLengthSymbol} + + // bitTables is the bits table for each table. + bitTables = [3][]byte{tableLiteralLengths: llBitsTable[:], tableOffsets: nil, tableMatchLengths: mlBitsTable[:]} +) + +type tableIndex uint8 + +const ( + // indexes for fsePredef and symbolTableX + tableLiteralLengths tableIndex = 0 + tableOffsets tableIndex = 1 + tableMatchLengths tableIndex = 2 + + maxLiteralLengthSymbol = 35 + maxOffsetLengthSymbol = 30 + maxMatchLengthSymbol = 52 +) + +// baseOffset is used for calculating transformations. +type baseOffset struct { + baseLine uint32 + addBits uint8 +} + +// fillBase will precalculate base offsets with the given bit distributions. +func fillBase(dst []baseOffset, base uint32, bits ...uint8) { + if len(bits) != len(dst) { + panic(fmt.Sprintf("len(dst) (%d) != len(bits) (%d)", len(dst), len(bits))) + } + for i, bit := range bits { + if base > math.MaxInt32 { + panic("invalid decoding table, base overflows int32") + } + + dst[i] = baseOffset{ + baseLine: base, + addBits: bit, + } + base += 1 << bit + } +} + +var predef sync.Once + +func initPredefined() { + predef.Do(func() { + // Literals length codes + tmp := make([]baseOffset, 36) + for i := range tmp[:16] { + tmp[i] = baseOffset{ + baseLine: uint32(i), + addBits: 0, + } + } + fillBase(tmp[16:], 16, 1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16) + symbolTableX[tableLiteralLengths] = tmp + + // Match length codes + tmp = make([]baseOffset, 53) + for i := range tmp[:32] { + tmp[i] = baseOffset{ + // The transformation adds the 3 length. + baseLine: uint32(i) + 3, + addBits: 0, + } + } + fillBase(tmp[32:], 35, 1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16) + symbolTableX[tableMatchLengths] = tmp + + // Offset codes + tmp = make([]baseOffset, maxOffsetBits+1) + tmp[1] = baseOffset{ + baseLine: 1, + addBits: 1, + } + fillBase(tmp[2:], 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30) + symbolTableX[tableOffsets] = tmp + + // Fill predefined tables and transform them. + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions + for i := range fsePredef[:] { + f := &fsePredef[i] + switch tableIndex(i) { + case tableLiteralLengths: + // https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L243 + f.actualTableLog = 6 + copy(f.norm[:], []int16{4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1, + -1, -1, -1, -1}) + f.symbolLen = 36 + case tableOffsets: + // https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L281 + f.actualTableLog = 5 + copy(f.norm[:], []int16{ + 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, -1, -1, -1}) + f.symbolLen = 29 + case tableMatchLengths: + //https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L304 + f.actualTableLog = 6 + copy(f.norm[:], []int16{ + 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, + -1, -1, -1, -1, -1}) + f.symbolLen = 53 + } + if err := f.buildDtable(); err != nil { + panic(fmt.Errorf("building table %v: %v", tableIndex(i), err)) + } + if err := f.transform(symbolTableX[i]); err != nil { + panic(fmt.Errorf("building table %v: %v", tableIndex(i), err)) + } + f.preDefined = true + + // Create encoder as well + enc := &fsePredefEnc[i] + copy(enc.norm[:], f.norm[:]) + enc.symbolLen = f.symbolLen + enc.actualTableLog = f.actualTableLog + if err := enc.buildCTable(); err != nil { + panic(fmt.Errorf("building encoding table %v: %v", tableIndex(i), err)) + } + enc.setBits(bitTables[i]) + enc.preDefined = true + } + }) +} diff --git a/vendor/github.com/klauspost/compress/zstd/hash.go b/vendor/github.com/klauspost/compress/zstd/hash.go new file mode 100644 index 0000000..5d73c21 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/hash.go @@ -0,0 +1,35 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +const ( + prime3bytes = 506832829 + prime4bytes = 2654435761 + prime5bytes = 889523592379 + prime6bytes = 227718039650203 + prime7bytes = 58295818150454627 + prime8bytes = 0xcf1bbcdcb7a56463 +) + +// hashLen returns a hash of the lowest mls bytes of with length output bits. +// mls must be >=3 and <=8. Any other value will return hash for 4 bytes. +// length should always be < 32. +// Preferably length and mls should be a constant for inlining. +func hashLen(u uint64, length, mls uint8) uint32 { + switch mls { + case 3: + return (uint32(u<<8) * prime3bytes) >> (32 - length) + case 5: + return uint32(((u << (64 - 40)) * prime5bytes) >> (64 - length)) + case 6: + return uint32(((u << (64 - 48)) * prime6bytes) >> (64 - length)) + case 7: + return uint32(((u << (64 - 56)) * prime7bytes) >> (64 - length)) + case 8: + return uint32((u * prime8bytes) >> (64 - length)) + default: + return (uint32(u) * prime4bytes) >> (32 - length) + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/history.go b/vendor/github.com/klauspost/compress/zstd/history.go new file mode 100644 index 0000000..28b4015 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/history.go @@ -0,0 +1,119 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "github.com/klauspost/compress/huff0" +) + +// history contains the information transferred between blocks. +type history struct { + // Literal decompression + huffTree *huff0.Scratch + + // Sequence decompression + decoders sequenceDecs + recentOffsets [3]int + + // History buffer... + b []byte + + // ignoreBuffer is meant to ignore a number of bytes + // when checking for matches in history + ignoreBuffer int + + windowSize int + allocFrameBuffer int // needed? + error bool + dict *dict +} + +// reset will reset the history to initial state of a frame. +// The history must already have been initialized to the desired size. +func (h *history) reset() { + h.b = h.b[:0] + h.ignoreBuffer = 0 + h.error = false + h.recentOffsets = [3]int{1, 4, 8} + if f := h.decoders.litLengths.fse; f != nil && !f.preDefined { + fseDecoderPool.Put(f) + } + if f := h.decoders.offsets.fse; f != nil && !f.preDefined { + fseDecoderPool.Put(f) + } + if f := h.decoders.matchLengths.fse; f != nil && !f.preDefined { + fseDecoderPool.Put(f) + } + h.decoders = sequenceDecs{br: h.decoders.br} + if h.huffTree != nil { + if h.dict == nil || h.dict.litEnc != h.huffTree { + huffDecoderPool.Put(h.huffTree) + } + } + h.huffTree = nil + h.dict = nil + //printf("history created: %+v (l: %d, c: %d)", *h, len(h.b), cap(h.b)) +} + +func (h *history) setDict(dict *dict) { + if dict == nil { + return + } + h.dict = dict + h.decoders.litLengths = dict.llDec + h.decoders.offsets = dict.ofDec + h.decoders.matchLengths = dict.mlDec + h.decoders.dict = dict.content + h.recentOffsets = dict.offsets + h.huffTree = dict.litEnc +} + +// append bytes to history. +// This function will make sure there is space for it, +// if the buffer has been allocated with enough extra space. +func (h *history) append(b []byte) { + if len(b) >= h.windowSize { + // Discard all history by simply overwriting + h.b = h.b[:h.windowSize] + copy(h.b, b[len(b)-h.windowSize:]) + return + } + + // If there is space, append it. + if len(b) < cap(h.b)-len(h.b) { + h.b = append(h.b, b...) + return + } + + // Move data down so we only have window size left. + // We know we have less than window size in b at this point. + discard := len(b) + len(h.b) - h.windowSize + copy(h.b, h.b[discard:]) + h.b = h.b[:h.windowSize] + copy(h.b[h.windowSize-len(b):], b) +} + +// ensureBlock will ensure there is space for at least one block... +func (h *history) ensureBlock() { + if cap(h.b) < h.allocFrameBuffer { + h.b = make([]byte, 0, h.allocFrameBuffer) + return + } + + avail := cap(h.b) - len(h.b) + if avail >= h.windowSize || avail > maxCompressedBlockSize { + return + } + // Move data down so we only have window size left. + // We know we have less than window size in b at this point. + discard := len(h.b) - h.windowSize + copy(h.b, h.b[discard:]) + h.b = h.b[:h.windowSize] +} + +// append bytes to history without ever discarding anything. +func (h *history) appendKeep(b []byte) { + h.b = append(h.b, b...) +} diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt new file mode 100644 index 0000000..24b5306 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt @@ -0,0 +1,22 @@ +Copyright (c) 2016 Caleb Spare + +MIT License + +Permission is hereby granted, free of charge, to any person obtaining +a copy of this software and associated documentation files (the +"Software"), to deal in the Software without restriction, including +without limitation the rights to use, copy, modify, merge, publish, +distribute, sublicense, and/or sell copies of the Software, and to +permit persons to whom the Software is furnished to do so, subject to +the following conditions: + +The above copyright notice and this permission notice shall be +included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND +NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE +LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION +OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md new file mode 100644 index 0000000..69aa3bb --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md @@ -0,0 +1,58 @@ +# xxhash + +VENDORED: Go to [github.com/cespare/xxhash](https://github.com/cespare/xxhash) for original package. + + +[![GoDoc](https://godoc.org/github.com/cespare/xxhash?status.svg)](https://godoc.org/github.com/cespare/xxhash) +[![Build Status](https://travis-ci.org/cespare/xxhash.svg?branch=master)](https://travis-ci.org/cespare/xxhash) + +xxhash is a Go implementation of the 64-bit +[xxHash](http://cyan4973.github.io/xxHash/) algorithm, XXH64. This is a +high-quality hashing algorithm that is much faster than anything in the Go +standard library. + +This package provides a straightforward API: + +``` +func Sum64(b []byte) uint64 +func Sum64String(s string) uint64 +type Digest struct{ ... } + func New() *Digest +``` + +The `Digest` type implements hash.Hash64. Its key methods are: + +``` +func (*Digest) Write([]byte) (int, error) +func (*Digest) WriteString(string) (int, error) +func (*Digest) Sum64() uint64 +``` + +This implementation provides a fast pure-Go implementation and an even faster +assembly implementation for amd64. + +## Benchmarks + +Here are some quick benchmarks comparing the pure-Go and assembly +implementations of Sum64. + +| input size | purego | asm | +| --- | --- | --- | +| 5 B | 979.66 MB/s | 1291.17 MB/s | +| 100 B | 7475.26 MB/s | 7973.40 MB/s | +| 4 KB | 17573.46 MB/s | 17602.65 MB/s | +| 10 MB | 17131.46 MB/s | 17142.16 MB/s | + +These numbers were generated on Ubuntu 18.04 with an Intel i7-8700K CPU using +the following commands under Go 1.11.2: + +``` +$ go test -tags purego -benchtime 10s -bench '/xxhash,direct,bytes' +$ go test -benchtime 10s -bench '/xxhash,direct,bytes' +``` + +## Projects using this package + +- [InfluxDB](https://github.com/influxdata/influxdb) +- [Prometheus](https://github.com/prometheus/prometheus) +- [FreeCache](https://github.com/coocood/freecache) diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go new file mode 100644 index 0000000..2c112a0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go @@ -0,0 +1,237 @@ +// Package xxhash implements the 64-bit variant of xxHash (XXH64) as described +// at http://cyan4973.github.io/xxHash/. +// THIS IS VENDORED: Go to github.com/cespare/xxhash for original package. + +package xxhash + +import ( + "encoding/binary" + "errors" + "math/bits" +) + +const ( + prime1 uint64 = 11400714785074694791 + prime2 uint64 = 14029467366897019727 + prime3 uint64 = 1609587929392839161 + prime4 uint64 = 9650029242287828579 + prime5 uint64 = 2870177450012600261 +) + +// NOTE(caleb): I'm using both consts and vars of the primes. Using consts where +// possible in the Go code is worth a small (but measurable) performance boost +// by avoiding some MOVQs. Vars are needed for the asm and also are useful for +// convenience in the Go code in a few places where we need to intentionally +// avoid constant arithmetic (e.g., v1 := prime1 + prime2 fails because the +// result overflows a uint64). +var ( + prime1v = prime1 + prime2v = prime2 + prime3v = prime3 + prime4v = prime4 + prime5v = prime5 +) + +// Digest implements hash.Hash64. +type Digest struct { + v1 uint64 + v2 uint64 + v3 uint64 + v4 uint64 + total uint64 + mem [32]byte + n int // how much of mem is used +} + +// New creates a new Digest that computes the 64-bit xxHash algorithm. +func New() *Digest { + var d Digest + d.Reset() + return &d +} + +// Reset clears the Digest's state so that it can be reused. +func (d *Digest) Reset() { + d.v1 = prime1v + prime2 + d.v2 = prime2 + d.v3 = 0 + d.v4 = -prime1v + d.total = 0 + d.n = 0 +} + +// Size always returns 8 bytes. +func (d *Digest) Size() int { return 8 } + +// BlockSize always returns 32 bytes. +func (d *Digest) BlockSize() int { return 32 } + +// Write adds more data to d. It always returns len(b), nil. +func (d *Digest) Write(b []byte) (n int, err error) { + n = len(b) + d.total += uint64(n) + + if d.n+n < 32 { + // This new data doesn't even fill the current block. + copy(d.mem[d.n:], b) + d.n += n + return + } + + if d.n > 0 { + // Finish off the partial block. + copy(d.mem[d.n:], b) + d.v1 = round(d.v1, u64(d.mem[0:8])) + d.v2 = round(d.v2, u64(d.mem[8:16])) + d.v3 = round(d.v3, u64(d.mem[16:24])) + d.v4 = round(d.v4, u64(d.mem[24:32])) + b = b[32-d.n:] + d.n = 0 + } + + if len(b) >= 32 { + // One or more full blocks left. + nw := writeBlocks(d, b) + b = b[nw:] + } + + // Store any remaining partial block. + copy(d.mem[:], b) + d.n = len(b) + + return +} + +// Sum appends the current hash to b and returns the resulting slice. +func (d *Digest) Sum(b []byte) []byte { + s := d.Sum64() + return append( + b, + byte(s>>56), + byte(s>>48), + byte(s>>40), + byte(s>>32), + byte(s>>24), + byte(s>>16), + byte(s>>8), + byte(s), + ) +} + +// Sum64 returns the current hash. +func (d *Digest) Sum64() uint64 { + var h uint64 + + if d.total >= 32 { + v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4 + h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + h = mergeRound(h, v1) + h = mergeRound(h, v2) + h = mergeRound(h, v3) + h = mergeRound(h, v4) + } else { + h = d.v3 + prime5 + } + + h += d.total + + i, end := 0, d.n + for ; i+8 <= end; i += 8 { + k1 := round(0, u64(d.mem[i:i+8])) + h ^= k1 + h = rol27(h)*prime1 + prime4 + } + if i+4 <= end { + h ^= uint64(u32(d.mem[i:i+4])) * prime1 + h = rol23(h)*prime2 + prime3 + i += 4 + } + for i < end { + h ^= uint64(d.mem[i]) * prime5 + h = rol11(h) * prime1 + i++ + } + + h ^= h >> 33 + h *= prime2 + h ^= h >> 29 + h *= prime3 + h ^= h >> 32 + + return h +} + +const ( + magic = "xxh\x06" + marshaledSize = len(magic) + 8*5 + 32 +) + +// MarshalBinary implements the encoding.BinaryMarshaler interface. +func (d *Digest) MarshalBinary() ([]byte, error) { + b := make([]byte, 0, marshaledSize) + b = append(b, magic...) + b = appendUint64(b, d.v1) + b = appendUint64(b, d.v2) + b = appendUint64(b, d.v3) + b = appendUint64(b, d.v4) + b = appendUint64(b, d.total) + b = append(b, d.mem[:d.n]...) + b = b[:len(b)+len(d.mem)-d.n] + return b, nil +} + +// UnmarshalBinary implements the encoding.BinaryUnmarshaler interface. +func (d *Digest) UnmarshalBinary(b []byte) error { + if len(b) < len(magic) || string(b[:len(magic)]) != magic { + return errors.New("xxhash: invalid hash state identifier") + } + if len(b) != marshaledSize { + return errors.New("xxhash: invalid hash state size") + } + b = b[len(magic):] + b, d.v1 = consumeUint64(b) + b, d.v2 = consumeUint64(b) + b, d.v3 = consumeUint64(b) + b, d.v4 = consumeUint64(b) + b, d.total = consumeUint64(b) + copy(d.mem[:], b) + d.n = int(d.total % uint64(len(d.mem))) + return nil +} + +func appendUint64(b []byte, x uint64) []byte { + var a [8]byte + binary.LittleEndian.PutUint64(a[:], x) + return append(b, a[:]...) +} + +func consumeUint64(b []byte) ([]byte, uint64) { + x := u64(b) + return b[8:], x +} + +func u64(b []byte) uint64 { return binary.LittleEndian.Uint64(b) } +func u32(b []byte) uint32 { return binary.LittleEndian.Uint32(b) } + +func round(acc, input uint64) uint64 { + acc += input * prime2 + acc = rol31(acc) + acc *= prime1 + return acc +} + +func mergeRound(acc, val uint64) uint64 { + val = round(0, val) + acc ^= val + acc = acc*prime1 + prime4 + return acc +} + +func rol1(x uint64) uint64 { return bits.RotateLeft64(x, 1) } +func rol7(x uint64) uint64 { return bits.RotateLeft64(x, 7) } +func rol11(x uint64) uint64 { return bits.RotateLeft64(x, 11) } +func rol12(x uint64) uint64 { return bits.RotateLeft64(x, 12) } +func rol18(x uint64) uint64 { return bits.RotateLeft64(x, 18) } +func rol23(x uint64) uint64 { return bits.RotateLeft64(x, 23) } +func rol27(x uint64) uint64 { return bits.RotateLeft64(x, 27) } +func rol31(x uint64) uint64 { return bits.RotateLeft64(x, 31) } diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s new file mode 100644 index 0000000..cea1785 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s @@ -0,0 +1,216 @@ +// +build !appengine +// +build gc +// +build !purego +// +build !noasm + +#include "textflag.h" + +// Register allocation: +// AX h +// SI pointer to advance through b +// DX n +// BX loop end +// R8 v1, k1 +// R9 v2 +// R10 v3 +// R11 v4 +// R12 tmp +// R13 prime1v +// R14 prime2v +// DI prime4v + +// round reads from and advances the buffer pointer in SI. +// It assumes that R13 has prime1v and R14 has prime2v. +#define round(r) \ + MOVQ (SI), R12 \ + ADDQ $8, SI \ + IMULQ R14, R12 \ + ADDQ R12, r \ + ROLQ $31, r \ + IMULQ R13, r + +// mergeRound applies a merge round on the two registers acc and val. +// It assumes that R13 has prime1v, R14 has prime2v, and DI has prime4v. +#define mergeRound(acc, val) \ + IMULQ R14, val \ + ROLQ $31, val \ + IMULQ R13, val \ + XORQ val, acc \ + IMULQ R13, acc \ + ADDQ DI, acc + +// func Sum64(b []byte) uint64 +TEXT ·Sum64(SB), NOSPLIT, $0-32 + // Load fixed primes. + MOVQ ·prime1v(SB), R13 + MOVQ ·prime2v(SB), R14 + MOVQ ·prime4v(SB), DI + + // Load slice. + MOVQ b_base+0(FP), SI + MOVQ b_len+8(FP), DX + LEAQ (SI)(DX*1), BX + + // The first loop limit will be len(b)-32. + SUBQ $32, BX + + // Check whether we have at least one block. + CMPQ DX, $32 + JLT noBlocks + + // Set up initial state (v1, v2, v3, v4). + MOVQ R13, R8 + ADDQ R14, R8 + MOVQ R14, R9 + XORQ R10, R10 + XORQ R11, R11 + SUBQ R13, R11 + + // Loop until SI > BX. +blockLoop: + round(R8) + round(R9) + round(R10) + round(R11) + + CMPQ SI, BX + JLE blockLoop + + MOVQ R8, AX + ROLQ $1, AX + MOVQ R9, R12 + ROLQ $7, R12 + ADDQ R12, AX + MOVQ R10, R12 + ROLQ $12, R12 + ADDQ R12, AX + MOVQ R11, R12 + ROLQ $18, R12 + ADDQ R12, AX + + mergeRound(AX, R8) + mergeRound(AX, R9) + mergeRound(AX, R10) + mergeRound(AX, R11) + + JMP afterBlocks + +noBlocks: + MOVQ ·prime5v(SB), AX + +afterBlocks: + ADDQ DX, AX + + // Right now BX has len(b)-32, and we want to loop until SI > len(b)-8. + ADDQ $24, BX + + CMPQ SI, BX + JG fourByte + +wordLoop: + // Calculate k1. + MOVQ (SI), R8 + ADDQ $8, SI + IMULQ R14, R8 + ROLQ $31, R8 + IMULQ R13, R8 + + XORQ R8, AX + ROLQ $27, AX + IMULQ R13, AX + ADDQ DI, AX + + CMPQ SI, BX + JLE wordLoop + +fourByte: + ADDQ $4, BX + CMPQ SI, BX + JG singles + + MOVL (SI), R8 + ADDQ $4, SI + IMULQ R13, R8 + XORQ R8, AX + + ROLQ $23, AX + IMULQ R14, AX + ADDQ ·prime3v(SB), AX + +singles: + ADDQ $4, BX + CMPQ SI, BX + JGE finalize + +singlesLoop: + MOVBQZX (SI), R12 + ADDQ $1, SI + IMULQ ·prime5v(SB), R12 + XORQ R12, AX + + ROLQ $11, AX + IMULQ R13, AX + + CMPQ SI, BX + JL singlesLoop + +finalize: + MOVQ AX, R12 + SHRQ $33, R12 + XORQ R12, AX + IMULQ R14, AX + MOVQ AX, R12 + SHRQ $29, R12 + XORQ R12, AX + IMULQ ·prime3v(SB), AX + MOVQ AX, R12 + SHRQ $32, R12 + XORQ R12, AX + + MOVQ AX, ret+24(FP) + RET + +// writeBlocks uses the same registers as above except that it uses AX to store +// the d pointer. + +// func writeBlocks(d *Digest, b []byte) int +TEXT ·writeBlocks(SB), NOSPLIT, $0-40 + // Load fixed primes needed for round. + MOVQ ·prime1v(SB), R13 + MOVQ ·prime2v(SB), R14 + + // Load slice. + MOVQ b_base+8(FP), SI + MOVQ b_len+16(FP), DX + LEAQ (SI)(DX*1), BX + SUBQ $32, BX + + // Load vN from d. + MOVQ d+0(FP), AX + MOVQ 0(AX), R8 // v1 + MOVQ 8(AX), R9 // v2 + MOVQ 16(AX), R10 // v3 + MOVQ 24(AX), R11 // v4 + + // We don't need to check the loop condition here; this function is + // always called with at least one block of data to process. +blockLoop: + round(R8) + round(R9) + round(R10) + round(R11) + + CMPQ SI, BX + JLE blockLoop + + // Copy vN back to d. + MOVQ R8, 0(AX) + MOVQ R9, 8(AX) + MOVQ R10, 16(AX) + MOVQ R11, 24(AX) + + // The number of bytes written is SI minus the old base pointer. + SUBQ b_base+8(FP), SI + MOVQ SI, ret+32(FP) + + RET diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_arm64.s b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_arm64.s new file mode 100644 index 0000000..4d64a17 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_arm64.s @@ -0,0 +1,186 @@ +// +build gc,!purego,!noasm + +#include "textflag.h" + +// Register allocation. +#define digest R1 +#define h R2 // Return value. +#define p R3 // Input pointer. +#define len R4 +#define nblocks R5 // len / 32. +#define prime1 R7 +#define prime2 R8 +#define prime3 R9 +#define prime4 R10 +#define prime5 R11 +#define v1 R12 +#define v2 R13 +#define v3 R14 +#define v4 R15 +#define x1 R20 +#define x2 R21 +#define x3 R22 +#define x4 R23 + +#define round(acc, x) \ + MADD prime2, acc, x, acc \ + ROR $64-31, acc \ + MUL prime1, acc \ + +// x = round(0, x). +#define round0(x) \ + MUL prime2, x \ + ROR $64-31, x \ + MUL prime1, x \ + +#define mergeRound(x) \ + round0(x) \ + EOR x, h \ + MADD h, prime4, prime1, h \ + +// Update v[1-4] with 32-byte blocks. Assumes len >= 32. +#define blocksLoop() \ + LSR $5, len, nblocks \ + PCALIGN $16 \ + loop: \ + LDP.P 32(p), (x1, x2) \ + round(v1, x1) \ + LDP -16(p), (x3, x4) \ + round(v2, x2) \ + SUB $1, nblocks \ + round(v3, x3) \ + round(v4, x4) \ + CBNZ nblocks, loop \ + +// The primes are repeated here to ensure that they're stored +// in a contiguous array, so we can load them with LDP. +DATA primes<> +0(SB)/8, $11400714785074694791 +DATA primes<> +8(SB)/8, $14029467366897019727 +DATA primes<>+16(SB)/8, $1609587929392839161 +DATA primes<>+24(SB)/8, $9650029242287828579 +DATA primes<>+32(SB)/8, $2870177450012600261 +GLOBL primes<>(SB), NOPTR+RODATA, $40 + +// func Sum64(b []byte) uint64 +TEXT ·Sum64(SB), NOFRAME+NOSPLIT, $0-32 + LDP b_base+0(FP), (p, len) + + LDP primes<> +0(SB), (prime1, prime2) + LDP primes<>+16(SB), (prime3, prime4) + MOVD primes<>+32(SB), prime5 + + CMP $32, len + CSEL LO, prime5, ZR, h // if len < 32 { h = prime5 } else { h = 0 } + BLO afterLoop + + ADD prime1, prime2, v1 + MOVD prime2, v2 + MOVD $0, v3 + NEG prime1, v4 + + blocksLoop() + + ROR $64-1, v1, x1 + ROR $64-7, v2, x2 + ADD x1, x2 + ROR $64-12, v3, x3 + ROR $64-18, v4, x4 + ADD x3, x4 + ADD x2, x4, h + + mergeRound(v1) + mergeRound(v2) + mergeRound(v3) + mergeRound(v4) + +afterLoop: + ADD len, h + + TBZ $4, len, try8 + LDP.P 16(p), (x1, x2) + + round0(x1) + ROR $64-27, h + EOR x1 @> 64-27, h, h + MADD h, prime4, prime1, h + + round0(x2) + ROR $64-27, h + EOR x2 @> 64-27, h + MADD h, prime4, prime1, h + +try8: + TBZ $3, len, try4 + MOVD.P 8(p), x1 + + round0(x1) + ROR $64-27, h + EOR x1 @> 64-27, h + MADD h, prime4, prime1, h + +try4: + TBZ $2, len, try2 + MOVWU.P 4(p), x2 + + MUL prime1, x2 + ROR $64-23, h + EOR x2 @> 64-23, h + MADD h, prime3, prime2, h + +try2: + TBZ $1, len, try1 + MOVHU.P 2(p), x3 + AND $255, x3, x1 + LSR $8, x3, x2 + + MUL prime5, x1 + ROR $64-11, h + EOR x1 @> 64-11, h + MUL prime1, h + + MUL prime5, x2 + ROR $64-11, h + EOR x2 @> 64-11, h + MUL prime1, h + +try1: + TBZ $0, len, end + MOVBU (p), x4 + + MUL prime5, x4 + ROR $64-11, h + EOR x4 @> 64-11, h + MUL prime1, h + +end: + EOR h >> 33, h + MUL prime2, h + EOR h >> 29, h + MUL prime3, h + EOR h >> 32, h + + MOVD h, ret+24(FP) + RET + +// func writeBlocks(d *Digest, b []byte) int +// +// Assumes len(b) >= 32. +TEXT ·writeBlocks(SB), NOFRAME+NOSPLIT, $0-40 + LDP primes<>(SB), (prime1, prime2) + + // Load state. Assume v[1-4] are stored contiguously. + MOVD d+0(FP), digest + LDP 0(digest), (v1, v2) + LDP 16(digest), (v3, v4) + + LDP b_base+8(FP), (p, len) + + blocksLoop() + + // Store updated state. + STP (v1, v2), 0(digest) + STP (v3, v4), 16(digest) + + BIC $31, len + MOVD len, ret+32(FP) + RET diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_asm.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_asm.go new file mode 100644 index 0000000..1a1fac9 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_asm.go @@ -0,0 +1,16 @@ +//go:build (amd64 || arm64) && !appengine && gc && !purego && !noasm +// +build amd64 arm64 +// +build !appengine +// +build gc +// +build !purego +// +build !noasm + +package xxhash + +// Sum64 computes the 64-bit xxHash digest of b. +// +//go:noescape +func Sum64(b []byte) uint64 + +//go:noescape +func writeBlocks(d *Digest, b []byte) int diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go new file mode 100644 index 0000000..209cb4a --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go @@ -0,0 +1,77 @@ +//go:build (!amd64 && !arm64) || appengine || !gc || purego || noasm +// +build !amd64,!arm64 appengine !gc purego noasm + +package xxhash + +// Sum64 computes the 64-bit xxHash digest of b. +func Sum64(b []byte) uint64 { + // A simpler version would be + // d := New() + // d.Write(b) + // return d.Sum64() + // but this is faster, particularly for small inputs. + + n := len(b) + var h uint64 + + if n >= 32 { + v1 := prime1v + prime2 + v2 := prime2 + v3 := uint64(0) + v4 := -prime1v + for len(b) >= 32 { + v1 = round(v1, u64(b[0:8:len(b)])) + v2 = round(v2, u64(b[8:16:len(b)])) + v3 = round(v3, u64(b[16:24:len(b)])) + v4 = round(v4, u64(b[24:32:len(b)])) + b = b[32:len(b):len(b)] + } + h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + h = mergeRound(h, v1) + h = mergeRound(h, v2) + h = mergeRound(h, v3) + h = mergeRound(h, v4) + } else { + h = prime5 + } + + h += uint64(n) + + i, end := 0, len(b) + for ; i+8 <= end; i += 8 { + k1 := round(0, u64(b[i:i+8:len(b)])) + h ^= k1 + h = rol27(h)*prime1 + prime4 + } + if i+4 <= end { + h ^= uint64(u32(b[i:i+4:len(b)])) * prime1 + h = rol23(h)*prime2 + prime3 + i += 4 + } + for ; i < end; i++ { + h ^= uint64(b[i]) * prime5 + h = rol11(h) * prime1 + } + + h ^= h >> 33 + h *= prime2 + h ^= h >> 29 + h *= prime3 + h ^= h >> 32 + + return h +} + +func writeBlocks(d *Digest, b []byte) int { + v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4 + n := len(b) + for len(b) >= 32 { + v1 = round(v1, u64(b[0:8:len(b)])) + v2 = round(v2, u64(b[8:16:len(b)])) + v3 = round(v3, u64(b[16:24:len(b)])) + v4 = round(v4, u64(b[24:32:len(b)])) + b = b[32:len(b):len(b)] + } + d.v1, d.v2, d.v3, d.v4 = v1, v2, v3, v4 + return n - len(b) +} diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go new file mode 100644 index 0000000..6f3b0cb --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go @@ -0,0 +1,11 @@ +package xxhash + +// Sum64String computes the 64-bit xxHash digest of s. +func Sum64String(s string) uint64 { + return Sum64([]byte(s)) +} + +// WriteString adds more data to d. It always returns len(s), nil. +func (d *Digest) WriteString(s string) (n int, err error) { + return d.Write([]byte(s)) +} diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec.go b/vendor/github.com/klauspost/compress/zstd/seqdec.go new file mode 100644 index 0000000..df04472 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqdec.go @@ -0,0 +1,491 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "io" +) + +type seq struct { + litLen uint32 + matchLen uint32 + offset uint32 + + // Codes are stored here for the encoder + // so they only have to be looked up once. + llCode, mlCode, ofCode uint8 +} + +type seqVals struct { + ll, ml, mo int +} + +func (s seq) String() string { + if s.offset <= 3 { + if s.offset == 0 { + return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset: INVALID (0)") + } + return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset, " (repeat)") + } + return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset-3, " (new)") +} + +type seqCompMode uint8 + +const ( + compModePredefined seqCompMode = iota + compModeRLE + compModeFSE + compModeRepeat +) + +type sequenceDec struct { + // decoder keeps track of the current state and updates it from the bitstream. + fse *fseDecoder + state fseState + repeat bool +} + +// init the state of the decoder with input from stream. +func (s *sequenceDec) init(br *bitReader) error { + if s.fse == nil { + return errors.New("sequence decoder not defined") + } + s.state.init(br, s.fse.actualTableLog, s.fse.dt[:1< cap(s.out) { + addBytes := s.seqSize + len(s.out) + s.out = append(s.out, make([]byte, addBytes)...) + s.out = s.out[:len(s.out)-addBytes] + } + + if debugDecoder { + printf("Execute %d seqs with hist %d, dict %d, literals: %d into %d bytes\n", len(seqs), len(hist), len(s.dict), len(s.literals), s.seqSize) + } + + var t = len(s.out) + out := s.out[:t+s.seqSize] + + for _, seq := range seqs { + // Add literals + copy(out[t:], s.literals[:seq.ll]) + t += seq.ll + s.literals = s.literals[seq.ll:] + + // Copy from dictionary... + if seq.mo > t+len(hist) || seq.mo > s.windowSize { + if len(s.dict) == 0 { + return fmt.Errorf("match offset (%d) bigger than current history (%d)", seq.mo, t+len(hist)) + } + + // we may be in dictionary. + dictO := len(s.dict) - (seq.mo - (t + len(hist))) + if dictO < 0 || dictO >= len(s.dict) { + return fmt.Errorf("match offset (%d) bigger than current history+dict (%d)", seq.mo, t+len(hist)+len(s.dict)) + } + end := dictO + seq.ml + if end > len(s.dict) { + n := len(s.dict) - dictO + copy(out[t:], s.dict[dictO:]) + t += n + seq.ml -= n + } else { + copy(out[t:], s.dict[dictO:end]) + t += end - dictO + continue + } + } + + // Copy from history. + if v := seq.mo - t; v > 0 { + // v is the start position in history from end. + start := len(hist) - v + if seq.ml > v { + // Some goes into current block. + // Copy remainder of history + copy(out[t:], hist[start:]) + t += v + seq.ml -= v + } else { + copy(out[t:], hist[start:start+seq.ml]) + t += seq.ml + continue + } + } + // We must be in current buffer now + if seq.ml > 0 { + start := t - seq.mo + if seq.ml <= t-start { + // No overlap + copy(out[t:], out[start:start+seq.ml]) + t += seq.ml + continue + } else { + // Overlapping copy + // Extend destination slice and copy one byte at the time. + src := out[start : start+seq.ml] + dst := out[t:] + dst = dst[:len(src)] + t += len(src) + // Destination is the space we just added. + for i := range src { + dst[i] = src[i] + } + } + } + } + + // Add final literals + copy(out[t:], s.literals) + if debugDecoder { + t += len(s.literals) + if t != len(out) { + panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize)) + } + } + s.out = out + + return nil +} + +// decode sequences from the stream with the provided history. +func (s *sequenceDecs) decodeSync(hist []byte) error { + supported, err := s.decodeSyncSimple(hist) + if supported { + return err + } + + br := s.br + seqs := s.nSeqs + startSize := len(s.out) + // Grab full sizes tables, to avoid bounds checks. + llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize] + llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state + out := s.out + maxBlockSize := maxCompressedBlockSize + if s.windowSize < maxBlockSize { + maxBlockSize = s.windowSize + } + + for i := seqs - 1; i >= 0; i-- { + if br.overread() { + printf("reading sequence %d, exceeded available data\n", seqs-i) + return io.ErrUnexpectedEOF + } + var ll, mo, ml int + if br.off > 4+((maxOffsetBits+16+16)>>3) { + // inlined function: + // ll, mo, ml = s.nextFast(br, llState, mlState, ofState) + + // Final will not read from stream. + var llB, mlB, moB uint8 + ll, llB = llState.final() + ml, mlB = mlState.final() + mo, moB = ofState.final() + + // extra bits are stored in reverse order. + br.fillFast() + mo += br.getBits(moB) + if s.maxBits > 32 { + br.fillFast() + } + ml += br.getBits(mlB) + ll += br.getBits(llB) + + if moB > 1 { + s.prevOffset[2] = s.prevOffset[1] + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = mo + } else { + // mo = s.adjustOffset(mo, ll, moB) + // Inlined for rather big speedup + if ll == 0 { + // There is an exception though, when current sequence's literals_length = 0. + // In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2, + // an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte. + mo++ + } + + if mo == 0 { + mo = s.prevOffset[0] + } else { + var temp int + if mo == 3 { + temp = s.prevOffset[0] - 1 + } else { + temp = s.prevOffset[mo] + } + + if temp == 0 { + // 0 is not valid; input is corrupted; force offset to 1 + println("WARNING: temp was 0") + temp = 1 + } + + if mo != 1 { + s.prevOffset[2] = s.prevOffset[1] + } + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = temp + mo = temp + } + } + br.fillFast() + } else { + ll, mo, ml = s.next(br, llState, mlState, ofState) + br.fill() + } + + if debugSequences { + println("Seq", seqs-i-1, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml) + } + + if ll > len(s.literals) { + return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, len(s.literals)) + } + size := ll + ml + len(out) + if size-startSize > maxBlockSize { + return fmt.Errorf("output (%d) bigger than max block size (%d)", size-startSize, maxBlockSize) + } + if size > cap(out) { + // Not enough size, which can happen under high volume block streaming conditions + // but could be if destination slice is too small for sync operations. + // over-allocating here can create a large amount of GC pressure so we try to keep + // it as contained as possible + used := len(out) - startSize + addBytes := 256 + ll + ml + used>>2 + // Clamp to max block size. + if used+addBytes > maxBlockSize { + addBytes = maxBlockSize - used + } + out = append(out, make([]byte, addBytes)...) + out = out[:len(out)-addBytes] + } + if ml > maxMatchLen { + return fmt.Errorf("match len (%d) bigger than max allowed length", ml) + } + + // Add literals + out = append(out, s.literals[:ll]...) + s.literals = s.literals[ll:] + + if mo == 0 && ml > 0 { + return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml) + } + + if mo > len(out)+len(hist) || mo > s.windowSize { + if len(s.dict) == 0 { + return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist)-startSize) + } + + // we may be in dictionary. + dictO := len(s.dict) - (mo - (len(out) + len(hist))) + if dictO < 0 || dictO >= len(s.dict) { + return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist)-startSize) + } + end := dictO + ml + if end > len(s.dict) { + out = append(out, s.dict[dictO:]...) + ml -= len(s.dict) - dictO + } else { + out = append(out, s.dict[dictO:end]...) + mo = 0 + ml = 0 + } + } + + // Copy from history. + // TODO: Blocks without history could be made to ignore this completely. + if v := mo - len(out); v > 0 { + // v is the start position in history from end. + start := len(hist) - v + if ml > v { + // Some goes into current block. + // Copy remainder of history + out = append(out, hist[start:]...) + ml -= v + } else { + out = append(out, hist[start:start+ml]...) + ml = 0 + } + } + // We must be in current buffer now + if ml > 0 { + start := len(out) - mo + if ml <= len(out)-start { + // No overlap + out = append(out, out[start:start+ml]...) + } else { + // Overlapping copy + // Extend destination slice and copy one byte at the time. + out = out[:len(out)+ml] + src := out[start : start+ml] + // Destination is the space we just added. + dst := out[len(out)-ml:] + dst = dst[:len(src)] + for i := range src { + dst[i] = src[i] + } + } + } + if i == 0 { + // This is the last sequence, so we shouldn't update state. + break + } + + // Manually inlined, ~ 5-20% faster + // Update all 3 states at once. Approx 20% faster. + nBits := llState.nbBits() + mlState.nbBits() + ofState.nbBits() + if nBits == 0 { + llState = llTable[llState.newState()&maxTableMask] + mlState = mlTable[mlState.newState()&maxTableMask] + ofState = ofTable[ofState.newState()&maxTableMask] + } else { + bits := br.get32BitsFast(nBits) + + lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31)) + llState = llTable[(llState.newState()+lowBits)&maxTableMask] + + lowBits = uint16(bits >> (ofState.nbBits() & 31)) + lowBits &= bitMask[mlState.nbBits()&15] + mlState = mlTable[(mlState.newState()+lowBits)&maxTableMask] + + lowBits = uint16(bits) & bitMask[ofState.nbBits()&15] + ofState = ofTable[(ofState.newState()+lowBits)&maxTableMask] + } + } + + // Check if space for literals + if size := len(s.literals) + len(s.out) - startSize; size > maxBlockSize { + return fmt.Errorf("output (%d) bigger than max block size (%d)", size, maxBlockSize) + } + + // Add final literals + s.out = append(out, s.literals...) + return br.close() +} + +var bitMask [16]uint16 + +func init() { + for i := range bitMask[:] { + bitMask[i] = uint16((1 << uint(i)) - 1) + } +} + +func (s *sequenceDecs) next(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) { + // Final will not read from stream. + ll, llB := llState.final() + ml, mlB := mlState.final() + mo, moB := ofState.final() + + // extra bits are stored in reverse order. + br.fill() + if s.maxBits <= 32 { + mo += br.getBits(moB) + ml += br.getBits(mlB) + ll += br.getBits(llB) + } else { + mo += br.getBits(moB) + br.fill() + // matchlength+literal length, max 32 bits + ml += br.getBits(mlB) + ll += br.getBits(llB) + + } + mo = s.adjustOffset(mo, ll, moB) + return +} + +func (s *sequenceDecs) adjustOffset(offset, litLen int, offsetB uint8) int { + if offsetB > 1 { + s.prevOffset[2] = s.prevOffset[1] + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = offset + return offset + } + + if litLen == 0 { + // There is an exception though, when current sequence's literals_length = 0. + // In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2, + // an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte. + offset++ + } + + if offset == 0 { + return s.prevOffset[0] + } + var temp int + if offset == 3 { + temp = s.prevOffset[0] - 1 + } else { + temp = s.prevOffset[offset] + } + + if temp == 0 { + // 0 is not valid; input is corrupted; force offset to 1 + println("temp was 0") + temp = 1 + } + + if offset != 1 { + s.prevOffset[2] = s.prevOffset[1] + } + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = temp + return temp +} diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go new file mode 100644 index 0000000..7598c10 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go @@ -0,0 +1,368 @@ +//go:build amd64 && !appengine && !noasm && gc +// +build amd64,!appengine,!noasm,gc + +package zstd + +import ( + "fmt" + + "github.com/klauspost/compress/internal/cpuinfo" +) + +type decodeSyncAsmContext struct { + llTable []decSymbol + mlTable []decSymbol + ofTable []decSymbol + llState uint64 + mlState uint64 + ofState uint64 + iteration int + litRemain int + out []byte + outPosition int + literals []byte + litPosition int + history []byte + windowSize int + ll int // set on error (not for all errors, please refer to _generate/gen.go) + ml int // set on error (not for all errors, please refer to _generate/gen.go) + mo int // set on error (not for all errors, please refer to _generate/gen.go) +} + +// sequenceDecs_decodeSync_amd64 implements the main loop of sequenceDecs.decodeSync in x86 asm. +// +// Please refer to seqdec_generic.go for the reference implementation. +//go:noescape +func sequenceDecs_decodeSync_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int + +// sequenceDecs_decodeSync_bmi2 implements the main loop of sequenceDecs.decodeSync in x86 asm with BMI2 extensions. +//go:noescape +func sequenceDecs_decodeSync_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int + +// sequenceDecs_decodeSync_safe_amd64 does the same as above, but does not write more than output buffer. +//go:noescape +func sequenceDecs_decodeSync_safe_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int + +// sequenceDecs_decodeSync_safe_bmi2 does the same as above, but does not write more than output buffer. +//go:noescape +func sequenceDecs_decodeSync_safe_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int + +// decode sequences from the stream with the provided history but without a dictionary. +func (s *sequenceDecs) decodeSyncSimple(hist []byte) (bool, error) { + if len(s.dict) > 0 { + return false, nil + } + if s.maxSyncLen == 0 && cap(s.out)-len(s.out) < maxCompressedBlockSize { + return false, nil + } + + // FIXME: Using unsafe memory copies leads to rare, random crashes + // with fuzz testing. It is therefore disabled for now. + const useSafe = true + /* + useSafe := false + if s.maxSyncLen == 0 && cap(s.out)-len(s.out) < maxCompressedBlockSizeAlloc { + useSafe = true + } + if s.maxSyncLen > 0 && cap(s.out)-len(s.out)-compressedBlockOverAlloc < int(s.maxSyncLen) { + useSafe = true + } + if cap(s.literals) < len(s.literals)+compressedBlockOverAlloc { + useSafe = true + } + */ + + br := s.br + + maxBlockSize := maxCompressedBlockSize + if s.windowSize < maxBlockSize { + maxBlockSize = s.windowSize + } + + ctx := decodeSyncAsmContext{ + llTable: s.litLengths.fse.dt[:maxTablesize], + mlTable: s.matchLengths.fse.dt[:maxTablesize], + ofTable: s.offsets.fse.dt[:maxTablesize], + llState: uint64(s.litLengths.state.state), + mlState: uint64(s.matchLengths.state.state), + ofState: uint64(s.offsets.state.state), + iteration: s.nSeqs - 1, + litRemain: len(s.literals), + out: s.out, + outPosition: len(s.out), + literals: s.literals, + windowSize: s.windowSize, + history: hist, + } + + s.seqSize = 0 + startSize := len(s.out) + + var errCode int + if cpuinfo.HasBMI2() { + if useSafe { + errCode = sequenceDecs_decodeSync_safe_bmi2(s, br, &ctx) + } else { + errCode = sequenceDecs_decodeSync_bmi2(s, br, &ctx) + } + } else { + if useSafe { + errCode = sequenceDecs_decodeSync_safe_amd64(s, br, &ctx) + } else { + errCode = sequenceDecs_decodeSync_amd64(s, br, &ctx) + } + } + switch errCode { + case noError: + break + + case errorMatchLenOfsMismatch: + return true, fmt.Errorf("zero matchoff and matchlen (%d) > 0", ctx.ml) + + case errorMatchLenTooBig: + return true, fmt.Errorf("match len (%d) bigger than max allowed length", ctx.ml) + + case errorMatchOffTooBig: + return true, fmt.Errorf("match offset (%d) bigger than current history (%d)", + ctx.mo, ctx.outPosition+len(hist)-startSize) + + case errorNotEnoughLiterals: + return true, fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", + ctx.ll, ctx.litRemain+ctx.ll) + + case errorNotEnoughSpace: + size := ctx.outPosition + ctx.ll + ctx.ml + if debugDecoder { + println("msl:", s.maxSyncLen, "cap", cap(s.out), "bef:", startSize, "sz:", size-startSize, "mbs:", maxBlockSize, "outsz:", cap(s.out)-startSize) + } + return true, fmt.Errorf("output (%d) bigger than max block size (%d)", size-startSize, maxBlockSize) + + default: + return true, fmt.Errorf("sequenceDecs_decode returned erronous code %d", errCode) + } + + s.seqSize += ctx.litRemain + if s.seqSize > maxBlockSize { + return true, fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize) + } + err := br.close() + if err != nil { + printf("Closing sequences: %v, %+v\n", err, *br) + return true, err + } + + s.literals = s.literals[ctx.litPosition:] + t := ctx.outPosition + s.out = s.out[:t] + + // Add final literals + s.out = append(s.out, s.literals...) + if debugDecoder { + t += len(s.literals) + if t != len(s.out) { + panic(fmt.Errorf("length mismatch, want %d, got %d", len(s.out), t)) + } + } + + return true, nil +} + +// -------------------------------------------------------------------------------- + +type decodeAsmContext struct { + llTable []decSymbol + mlTable []decSymbol + ofTable []decSymbol + llState uint64 + mlState uint64 + ofState uint64 + iteration int + seqs []seqVals + litRemain int +} + +const noError = 0 + +// error reported when mo == 0 && ml > 0 +const errorMatchLenOfsMismatch = 1 + +// error reported when ml > maxMatchLen +const errorMatchLenTooBig = 2 + +// error reported when mo > available history or mo > s.windowSize +const errorMatchOffTooBig = 3 + +// error reported when the sum of literal lengths exeeceds the literal buffer size +const errorNotEnoughLiterals = 4 + +// error reported when capacity of `out` is too small +const errorNotEnoughSpace = 5 + +// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm. +// +// Please refer to seqdec_generic.go for the reference implementation. +//go:noescape +func sequenceDecs_decode_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int + +// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm. +// +// Please refer to seqdec_generic.go for the reference implementation. +//go:noescape +func sequenceDecs_decode_56_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int + +// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm with BMI2 extensions. +//go:noescape +func sequenceDecs_decode_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int + +// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm with BMI2 extensions. +//go:noescape +func sequenceDecs_decode_56_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int + +// decode sequences from the stream without the provided history. +func (s *sequenceDecs) decode(seqs []seqVals) error { + br := s.br + + maxBlockSize := maxCompressedBlockSize + if s.windowSize < maxBlockSize { + maxBlockSize = s.windowSize + } + + ctx := decodeAsmContext{ + llTable: s.litLengths.fse.dt[:maxTablesize], + mlTable: s.matchLengths.fse.dt[:maxTablesize], + ofTable: s.offsets.fse.dt[:maxTablesize], + llState: uint64(s.litLengths.state.state), + mlState: uint64(s.matchLengths.state.state), + ofState: uint64(s.offsets.state.state), + seqs: seqs, + iteration: len(seqs) - 1, + litRemain: len(s.literals), + } + + s.seqSize = 0 + lte56bits := s.maxBits+s.offsets.fse.actualTableLog+s.matchLengths.fse.actualTableLog+s.litLengths.fse.actualTableLog <= 56 + var errCode int + if cpuinfo.HasBMI2() { + if lte56bits { + errCode = sequenceDecs_decode_56_bmi2(s, br, &ctx) + } else { + errCode = sequenceDecs_decode_bmi2(s, br, &ctx) + } + } else { + if lte56bits { + errCode = sequenceDecs_decode_56_amd64(s, br, &ctx) + } else { + errCode = sequenceDecs_decode_amd64(s, br, &ctx) + } + } + if errCode != 0 { + i := len(seqs) - ctx.iteration - 1 + switch errCode { + case errorMatchLenOfsMismatch: + ml := ctx.seqs[i].ml + return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml) + + case errorMatchLenTooBig: + ml := ctx.seqs[i].ml + return fmt.Errorf("match len (%d) bigger than max allowed length", ml) + + case errorNotEnoughLiterals: + ll := ctx.seqs[i].ll + return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, ctx.litRemain+ll) + } + + return fmt.Errorf("sequenceDecs_decode_amd64 returned erronous code %d", errCode) + } + + if ctx.litRemain < 0 { + return fmt.Errorf("literal count is too big: total available %d, total requested %d", + len(s.literals), len(s.literals)-ctx.litRemain) + } + + s.seqSize += ctx.litRemain + if s.seqSize > maxBlockSize { + return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize) + } + err := br.close() + if err != nil { + printf("Closing sequences: %v, %+v\n", err, *br) + } + return err +} + +// -------------------------------------------------------------------------------- + +type executeAsmContext struct { + seqs []seqVals + seqIndex int + out []byte + history []byte + literals []byte + outPosition int + litPosition int + windowSize int +} + +// sequenceDecs_executeSimple_amd64 implements the main loop of sequenceDecs.executeSimple in x86 asm. +// +// Returns false if a match offset is too big. +// +// Please refer to seqdec_generic.go for the reference implementation. +//go:noescape +func sequenceDecs_executeSimple_amd64(ctx *executeAsmContext) bool + +// Same as above, but with safe memcopies +//go:noescape +func sequenceDecs_executeSimple_safe_amd64(ctx *executeAsmContext) bool + +// executeSimple handles cases when dictionary is not used. +func (s *sequenceDecs) executeSimple(seqs []seqVals, hist []byte) error { + // Ensure we have enough output size... + if len(s.out)+s.seqSize+compressedBlockOverAlloc > cap(s.out) { + addBytes := s.seqSize + len(s.out) + compressedBlockOverAlloc + s.out = append(s.out, make([]byte, addBytes)...) + s.out = s.out[:len(s.out)-addBytes] + } + + if debugDecoder { + printf("Execute %d seqs with literals: %d into %d bytes\n", len(seqs), len(s.literals), s.seqSize) + } + + var t = len(s.out) + out := s.out[:t+s.seqSize] + + ctx := executeAsmContext{ + seqs: seqs, + seqIndex: 0, + out: out, + history: hist, + outPosition: t, + litPosition: 0, + literals: s.literals, + windowSize: s.windowSize, + } + var ok bool + if cap(s.literals) < len(s.literals)+compressedBlockOverAlloc { + ok = sequenceDecs_executeSimple_safe_amd64(&ctx) + } else { + ok = sequenceDecs_executeSimple_amd64(&ctx) + } + if !ok { + return fmt.Errorf("match offset (%d) bigger than current history (%d)", + seqs[ctx.seqIndex].mo, ctx.outPosition+len(hist)) + } + s.literals = s.literals[ctx.litPosition:] + t = ctx.outPosition + + // Add final literals + copy(out[t:], s.literals) + if debugDecoder { + t += len(s.literals) + if t != len(out) { + panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize)) + } + } + s.out = out + + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s new file mode 100644 index 0000000..27e7677 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s @@ -0,0 +1,4100 @@ +// Code generated by command: go run gen.go -out ../seqdec_amd64.s -pkg=zstd. DO NOT EDIT. + +//go:build !appengine && !noasm && gc && !noasm +// +build !appengine,!noasm,gc,!noasm + +// func sequenceDecs_decode_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int +// Requires: CMOV +TEXT ·sequenceDecs_decode_amd64(SB), $8-32 + MOVQ br+8(FP), AX + MOVQ 32(AX), DX + MOVBQZX 40(AX), BX + MOVQ 24(AX), SI + MOVQ (AX), AX + ADDQ SI, AX + MOVQ AX, (SP) + MOVQ ctx+16(FP), AX + MOVQ 72(AX), DI + MOVQ 80(AX), R8 + MOVQ 88(AX), R9 + MOVQ 104(AX), R10 + MOVQ s+0(FP), AX + MOVQ 144(AX), R11 + MOVQ 152(AX), R12 + MOVQ 160(AX), R13 + +sequenceDecs_decode_amd64_main_loop: + MOVQ (SP), R14 + + // Fill bitreader to have enough for the offset and match length. + CMPQ SI, $0x08 + JL sequenceDecs_decode_amd64_fill_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R14 + MOVQ (R14), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decode_amd64_fill_end + +sequenceDecs_decode_amd64_fill_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decode_amd64_fill_end + CMPQ BX, $0x07 + JLE sequenceDecs_decode_amd64_fill_end + SHLQ $0x08, DX + SUBQ $0x01, R14 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R14), AX + ORQ AX, DX + JMP sequenceDecs_decode_amd64_fill_byte_by_byte + +sequenceDecs_decode_amd64_fill_end: + // Update offset + MOVQ R9, AX + MOVQ BX, CX + MOVQ DX, R15 + SHLQ CL, R15 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decode_amd64_of_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decode_amd64_of_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decode_amd64_of_update_zero + NEGQ CX + SHRQ CL, R15 + ADDQ R15, AX + +sequenceDecs_decode_amd64_of_update_zero: + MOVQ AX, 16(R10) + + // Update match length + MOVQ R8, AX + MOVQ BX, CX + MOVQ DX, R15 + SHLQ CL, R15 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decode_amd64_ml_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decode_amd64_ml_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decode_amd64_ml_update_zero + NEGQ CX + SHRQ CL, R15 + ADDQ R15, AX + +sequenceDecs_decode_amd64_ml_update_zero: + MOVQ AX, 8(R10) + + // Fill bitreader to have enough for the remaining + CMPQ SI, $0x08 + JL sequenceDecs_decode_amd64_fill_2_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R14 + MOVQ (R14), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decode_amd64_fill_2_end + +sequenceDecs_decode_amd64_fill_2_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decode_amd64_fill_2_end + CMPQ BX, $0x07 + JLE sequenceDecs_decode_amd64_fill_2_end + SHLQ $0x08, DX + SUBQ $0x01, R14 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R14), AX + ORQ AX, DX + JMP sequenceDecs_decode_amd64_fill_2_byte_by_byte + +sequenceDecs_decode_amd64_fill_2_end: + // Update literal length + MOVQ DI, AX + MOVQ BX, CX + MOVQ DX, R15 + SHLQ CL, R15 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decode_amd64_ll_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decode_amd64_ll_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decode_amd64_ll_update_zero + NEGQ CX + SHRQ CL, R15 + ADDQ R15, AX + +sequenceDecs_decode_amd64_ll_update_zero: + MOVQ AX, (R10) + + // Fill bitreader for state updates + MOVQ R14, (SP) + MOVQ R9, AX + SHRQ $0x08, AX + MOVBQZX AL, AX + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decode_amd64_skip_update + + // Update Literal Length State + MOVBQZX DI, R14 + SHRQ $0x10, DI + MOVWQZX DI, DI + LEAQ (BX)(R14*1), CX + MOVQ DX, R15 + MOVQ CX, BX + ROLQ CL, R15 + MOVL $0x00000001, BP + MOVB R14, CL + SHLL CL, BP + DECL BP + ANDQ BP, R15 + ADDQ R15, DI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(DI*8), DI + + // Update Match Length State + MOVBQZX R8, R14 + SHRQ $0x10, R8 + MOVWQZX R8, R8 + LEAQ (BX)(R14*1), CX + MOVQ DX, R15 + MOVQ CX, BX + ROLQ CL, R15 + MOVL $0x00000001, BP + MOVB R14, CL + SHLL CL, BP + DECL BP + ANDQ BP, R15 + ADDQ R15, R8 + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Offset State + MOVBQZX R9, R14 + SHRQ $0x10, R9 + MOVWQZX R9, R9 + LEAQ (BX)(R14*1), CX + MOVQ DX, R15 + MOVQ CX, BX + ROLQ CL, R15 + MOVL $0x00000001, BP + MOVB R14, CL + SHLL CL, BP + DECL BP + ANDQ BP, R15 + ADDQ R15, R9 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R9*8), R9 + +sequenceDecs_decode_amd64_skip_update: + // Adjust offset + MOVQ 16(R10), CX + CMPQ AX, $0x01 + JBE sequenceDecs_decode_amd64_adjust_offsetB_1_or_0 + MOVQ R12, R13 + MOVQ R11, R12 + MOVQ CX, R11 + JMP sequenceDecs_decode_amd64_after_adjust + +sequenceDecs_decode_amd64_adjust_offsetB_1_or_0: + CMPQ (R10), $0x00000000 + JNE sequenceDecs_decode_amd64_adjust_offset_maybezero + INCQ CX + JMP sequenceDecs_decode_amd64_adjust_offset_nonzero + +sequenceDecs_decode_amd64_adjust_offset_maybezero: + TESTQ CX, CX + JNZ sequenceDecs_decode_amd64_adjust_offset_nonzero + MOVQ R11, CX + JMP sequenceDecs_decode_amd64_after_adjust + +sequenceDecs_decode_amd64_adjust_offset_nonzero: + CMPQ CX, $0x01 + JB sequenceDecs_decode_amd64_adjust_zero + JEQ sequenceDecs_decode_amd64_adjust_one + CMPQ CX, $0x02 + JA sequenceDecs_decode_amd64_adjust_three + JMP sequenceDecs_decode_amd64_adjust_two + +sequenceDecs_decode_amd64_adjust_zero: + MOVQ R11, AX + JMP sequenceDecs_decode_amd64_adjust_test_temp_valid + +sequenceDecs_decode_amd64_adjust_one: + MOVQ R12, AX + JMP sequenceDecs_decode_amd64_adjust_test_temp_valid + +sequenceDecs_decode_amd64_adjust_two: + MOVQ R13, AX + JMP sequenceDecs_decode_amd64_adjust_test_temp_valid + +sequenceDecs_decode_amd64_adjust_three: + LEAQ -1(R11), AX + +sequenceDecs_decode_amd64_adjust_test_temp_valid: + TESTQ AX, AX + JNZ sequenceDecs_decode_amd64_adjust_temp_valid + MOVQ $0x00000001, AX + +sequenceDecs_decode_amd64_adjust_temp_valid: + CMPQ CX, $0x01 + CMOVQNE R12, R13 + MOVQ R11, R12 + MOVQ AX, R11 + MOVQ AX, CX + +sequenceDecs_decode_amd64_after_adjust: + MOVQ CX, 16(R10) + + // Check values + MOVQ 8(R10), AX + MOVQ (R10), R14 + LEAQ (AX)(R14*1), R15 + MOVQ s+0(FP), BP + ADDQ R15, 256(BP) + MOVQ ctx+16(FP), R15 + SUBQ R14, 128(R15) + JS error_not_enough_literals + CMPQ AX, $0x00020002 + JA sequenceDecs_decode_amd64_error_match_len_too_big + TESTQ CX, CX + JNZ sequenceDecs_decode_amd64_match_len_ofs_ok + TESTQ AX, AX + JNZ sequenceDecs_decode_amd64_error_match_len_ofs_mismatch + +sequenceDecs_decode_amd64_match_len_ofs_ok: + ADDQ $0x18, R10 + MOVQ ctx+16(FP), AX + DECQ 96(AX) + JNS sequenceDecs_decode_amd64_main_loop + MOVQ s+0(FP), AX + MOVQ R11, 144(AX) + MOVQ R12, 152(AX) + MOVQ R13, 160(AX) + MOVQ br+8(FP), AX + MOVQ DX, 32(AX) + MOVB BL, 40(AX) + MOVQ SI, 24(AX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decode_amd64_error_match_len_ofs_mismatch: + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decode_amd64_error_match_len_too_big: + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_decode_56_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int +// Requires: CMOV +TEXT ·sequenceDecs_decode_56_amd64(SB), $8-32 + MOVQ br+8(FP), AX + MOVQ 32(AX), DX + MOVBQZX 40(AX), BX + MOVQ 24(AX), SI + MOVQ (AX), AX + ADDQ SI, AX + MOVQ AX, (SP) + MOVQ ctx+16(FP), AX + MOVQ 72(AX), DI + MOVQ 80(AX), R8 + MOVQ 88(AX), R9 + MOVQ 104(AX), R10 + MOVQ s+0(FP), AX + MOVQ 144(AX), R11 + MOVQ 152(AX), R12 + MOVQ 160(AX), R13 + +sequenceDecs_decode_56_amd64_main_loop: + MOVQ (SP), R14 + + // Fill bitreader to have enough for the offset and match length. + CMPQ SI, $0x08 + JL sequenceDecs_decode_56_amd64_fill_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R14 + MOVQ (R14), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decode_56_amd64_fill_end + +sequenceDecs_decode_56_amd64_fill_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decode_56_amd64_fill_end + CMPQ BX, $0x07 + JLE sequenceDecs_decode_56_amd64_fill_end + SHLQ $0x08, DX + SUBQ $0x01, R14 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R14), AX + ORQ AX, DX + JMP sequenceDecs_decode_56_amd64_fill_byte_by_byte + +sequenceDecs_decode_56_amd64_fill_end: + // Update offset + MOVQ R9, AX + MOVQ BX, CX + MOVQ DX, R15 + SHLQ CL, R15 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decode_56_amd64_of_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decode_56_amd64_of_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decode_56_amd64_of_update_zero + NEGQ CX + SHRQ CL, R15 + ADDQ R15, AX + +sequenceDecs_decode_56_amd64_of_update_zero: + MOVQ AX, 16(R10) + + // Update match length + MOVQ R8, AX + MOVQ BX, CX + MOVQ DX, R15 + SHLQ CL, R15 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decode_56_amd64_ml_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decode_56_amd64_ml_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decode_56_amd64_ml_update_zero + NEGQ CX + SHRQ CL, R15 + ADDQ R15, AX + +sequenceDecs_decode_56_amd64_ml_update_zero: + MOVQ AX, 8(R10) + + // Update literal length + MOVQ DI, AX + MOVQ BX, CX + MOVQ DX, R15 + SHLQ CL, R15 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decode_56_amd64_ll_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decode_56_amd64_ll_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decode_56_amd64_ll_update_zero + NEGQ CX + SHRQ CL, R15 + ADDQ R15, AX + +sequenceDecs_decode_56_amd64_ll_update_zero: + MOVQ AX, (R10) + + // Fill bitreader for state updates + MOVQ R14, (SP) + MOVQ R9, AX + SHRQ $0x08, AX + MOVBQZX AL, AX + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decode_56_amd64_skip_update + + // Update Literal Length State + MOVBQZX DI, R14 + SHRQ $0x10, DI + MOVWQZX DI, DI + LEAQ (BX)(R14*1), CX + MOVQ DX, R15 + MOVQ CX, BX + ROLQ CL, R15 + MOVL $0x00000001, BP + MOVB R14, CL + SHLL CL, BP + DECL BP + ANDQ BP, R15 + ADDQ R15, DI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(DI*8), DI + + // Update Match Length State + MOVBQZX R8, R14 + SHRQ $0x10, R8 + MOVWQZX R8, R8 + LEAQ (BX)(R14*1), CX + MOVQ DX, R15 + MOVQ CX, BX + ROLQ CL, R15 + MOVL $0x00000001, BP + MOVB R14, CL + SHLL CL, BP + DECL BP + ANDQ BP, R15 + ADDQ R15, R8 + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Offset State + MOVBQZX R9, R14 + SHRQ $0x10, R9 + MOVWQZX R9, R9 + LEAQ (BX)(R14*1), CX + MOVQ DX, R15 + MOVQ CX, BX + ROLQ CL, R15 + MOVL $0x00000001, BP + MOVB R14, CL + SHLL CL, BP + DECL BP + ANDQ BP, R15 + ADDQ R15, R9 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R9*8), R9 + +sequenceDecs_decode_56_amd64_skip_update: + // Adjust offset + MOVQ 16(R10), CX + CMPQ AX, $0x01 + JBE sequenceDecs_decode_56_amd64_adjust_offsetB_1_or_0 + MOVQ R12, R13 + MOVQ R11, R12 + MOVQ CX, R11 + JMP sequenceDecs_decode_56_amd64_after_adjust + +sequenceDecs_decode_56_amd64_adjust_offsetB_1_or_0: + CMPQ (R10), $0x00000000 + JNE sequenceDecs_decode_56_amd64_adjust_offset_maybezero + INCQ CX + JMP sequenceDecs_decode_56_amd64_adjust_offset_nonzero + +sequenceDecs_decode_56_amd64_adjust_offset_maybezero: + TESTQ CX, CX + JNZ sequenceDecs_decode_56_amd64_adjust_offset_nonzero + MOVQ R11, CX + JMP sequenceDecs_decode_56_amd64_after_adjust + +sequenceDecs_decode_56_amd64_adjust_offset_nonzero: + CMPQ CX, $0x01 + JB sequenceDecs_decode_56_amd64_adjust_zero + JEQ sequenceDecs_decode_56_amd64_adjust_one + CMPQ CX, $0x02 + JA sequenceDecs_decode_56_amd64_adjust_three + JMP sequenceDecs_decode_56_amd64_adjust_two + +sequenceDecs_decode_56_amd64_adjust_zero: + MOVQ R11, AX + JMP sequenceDecs_decode_56_amd64_adjust_test_temp_valid + +sequenceDecs_decode_56_amd64_adjust_one: + MOVQ R12, AX + JMP sequenceDecs_decode_56_amd64_adjust_test_temp_valid + +sequenceDecs_decode_56_amd64_adjust_two: + MOVQ R13, AX + JMP sequenceDecs_decode_56_amd64_adjust_test_temp_valid + +sequenceDecs_decode_56_amd64_adjust_three: + LEAQ -1(R11), AX + +sequenceDecs_decode_56_amd64_adjust_test_temp_valid: + TESTQ AX, AX + JNZ sequenceDecs_decode_56_amd64_adjust_temp_valid + MOVQ $0x00000001, AX + +sequenceDecs_decode_56_amd64_adjust_temp_valid: + CMPQ CX, $0x01 + CMOVQNE R12, R13 + MOVQ R11, R12 + MOVQ AX, R11 + MOVQ AX, CX + +sequenceDecs_decode_56_amd64_after_adjust: + MOVQ CX, 16(R10) + + // Check values + MOVQ 8(R10), AX + MOVQ (R10), R14 + LEAQ (AX)(R14*1), R15 + MOVQ s+0(FP), BP + ADDQ R15, 256(BP) + MOVQ ctx+16(FP), R15 + SUBQ R14, 128(R15) + JS error_not_enough_literals + CMPQ AX, $0x00020002 + JA sequenceDecs_decode_56_amd64_error_match_len_too_big + TESTQ CX, CX + JNZ sequenceDecs_decode_56_amd64_match_len_ofs_ok + TESTQ AX, AX + JNZ sequenceDecs_decode_56_amd64_error_match_len_ofs_mismatch + +sequenceDecs_decode_56_amd64_match_len_ofs_ok: + ADDQ $0x18, R10 + MOVQ ctx+16(FP), AX + DECQ 96(AX) + JNS sequenceDecs_decode_56_amd64_main_loop + MOVQ s+0(FP), AX + MOVQ R11, 144(AX) + MOVQ R12, 152(AX) + MOVQ R13, 160(AX) + MOVQ br+8(FP), AX + MOVQ DX, 32(AX) + MOVB BL, 40(AX) + MOVQ SI, 24(AX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decode_56_amd64_error_match_len_ofs_mismatch: + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decode_56_amd64_error_match_len_too_big: + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_decode_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int +// Requires: BMI, BMI2, CMOV +TEXT ·sequenceDecs_decode_bmi2(SB), $8-32 + MOVQ br+8(FP), CX + MOVQ 32(CX), AX + MOVBQZX 40(CX), DX + MOVQ 24(CX), BX + MOVQ (CX), CX + ADDQ BX, CX + MOVQ CX, (SP) + MOVQ ctx+16(FP), CX + MOVQ 72(CX), SI + MOVQ 80(CX), DI + MOVQ 88(CX), R8 + MOVQ 104(CX), R9 + MOVQ s+0(FP), CX + MOVQ 144(CX), R10 + MOVQ 152(CX), R11 + MOVQ 160(CX), R12 + +sequenceDecs_decode_bmi2_main_loop: + MOVQ (SP), R13 + + // Fill bitreader to have enough for the offset and match length. + CMPQ BX, $0x08 + JL sequenceDecs_decode_bmi2_fill_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R13 + MOVQ (R13), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decode_bmi2_fill_end + +sequenceDecs_decode_bmi2_fill_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decode_bmi2_fill_end + CMPQ DX, $0x07 + JLE sequenceDecs_decode_bmi2_fill_end + SHLQ $0x08, AX + SUBQ $0x01, R13 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R13), CX + ORQ CX, AX + JMP sequenceDecs_decode_bmi2_fill_byte_by_byte + +sequenceDecs_decode_bmi2_fill_end: + // Update offset + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R14 + MOVQ AX, R15 + LEAQ (DX)(R14*1), CX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + MOVQ CX, DX + MOVQ R8, CX + SHRQ $0x20, CX + ADDQ R15, CX + MOVQ CX, 16(R9) + + // Update match length + MOVQ $0x00000808, CX + BEXTRQ CX, DI, R14 + MOVQ AX, R15 + LEAQ (DX)(R14*1), CX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + MOVQ CX, DX + MOVQ DI, CX + SHRQ $0x20, CX + ADDQ R15, CX + MOVQ CX, 8(R9) + + // Fill bitreader to have enough for the remaining + CMPQ BX, $0x08 + JL sequenceDecs_decode_bmi2_fill_2_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R13 + MOVQ (R13), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decode_bmi2_fill_2_end + +sequenceDecs_decode_bmi2_fill_2_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decode_bmi2_fill_2_end + CMPQ DX, $0x07 + JLE sequenceDecs_decode_bmi2_fill_2_end + SHLQ $0x08, AX + SUBQ $0x01, R13 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R13), CX + ORQ CX, AX + JMP sequenceDecs_decode_bmi2_fill_2_byte_by_byte + +sequenceDecs_decode_bmi2_fill_2_end: + // Update literal length + MOVQ $0x00000808, CX + BEXTRQ CX, SI, R14 + MOVQ AX, R15 + LEAQ (DX)(R14*1), CX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + MOVQ CX, DX + MOVQ SI, CX + SHRQ $0x20, CX + ADDQ R15, CX + MOVQ CX, (R9) + + // Fill bitreader for state updates + MOVQ R13, (SP) + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R13 + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decode_bmi2_skip_update + LEAQ (SI)(DI*1), R14 + ADDQ R8, R14 + MOVBQZX R14, R14 + LEAQ (DX)(R14*1), CX + MOVQ AX, R15 + MOVQ CX, DX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + + // Update Offset State + BZHIQ R8, R15, CX + SHRXQ R8, R15, R15 + MOVQ $0x00001010, R14 + BEXTRQ R14, R8, R8 + ADDQ CX, R8 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Match Length State + BZHIQ DI, R15, CX + SHRXQ DI, R15, R15 + MOVQ $0x00001010, R14 + BEXTRQ R14, DI, DI + ADDQ CX, DI + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(DI*8), DI + + // Update Literal Length State + BZHIQ SI, R15, CX + MOVQ $0x00001010, R14 + BEXTRQ R14, SI, SI + ADDQ CX, SI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(SI*8), SI + +sequenceDecs_decode_bmi2_skip_update: + // Adjust offset + MOVQ 16(R9), CX + CMPQ R13, $0x01 + JBE sequenceDecs_decode_bmi2_adjust_offsetB_1_or_0 + MOVQ R11, R12 + MOVQ R10, R11 + MOVQ CX, R10 + JMP sequenceDecs_decode_bmi2_after_adjust + +sequenceDecs_decode_bmi2_adjust_offsetB_1_or_0: + CMPQ (R9), $0x00000000 + JNE sequenceDecs_decode_bmi2_adjust_offset_maybezero + INCQ CX + JMP sequenceDecs_decode_bmi2_adjust_offset_nonzero + +sequenceDecs_decode_bmi2_adjust_offset_maybezero: + TESTQ CX, CX + JNZ sequenceDecs_decode_bmi2_adjust_offset_nonzero + MOVQ R10, CX + JMP sequenceDecs_decode_bmi2_after_adjust + +sequenceDecs_decode_bmi2_adjust_offset_nonzero: + CMPQ CX, $0x01 + JB sequenceDecs_decode_bmi2_adjust_zero + JEQ sequenceDecs_decode_bmi2_adjust_one + CMPQ CX, $0x02 + JA sequenceDecs_decode_bmi2_adjust_three + JMP sequenceDecs_decode_bmi2_adjust_two + +sequenceDecs_decode_bmi2_adjust_zero: + MOVQ R10, R13 + JMP sequenceDecs_decode_bmi2_adjust_test_temp_valid + +sequenceDecs_decode_bmi2_adjust_one: + MOVQ R11, R13 + JMP sequenceDecs_decode_bmi2_adjust_test_temp_valid + +sequenceDecs_decode_bmi2_adjust_two: + MOVQ R12, R13 + JMP sequenceDecs_decode_bmi2_adjust_test_temp_valid + +sequenceDecs_decode_bmi2_adjust_three: + LEAQ -1(R10), R13 + +sequenceDecs_decode_bmi2_adjust_test_temp_valid: + TESTQ R13, R13 + JNZ sequenceDecs_decode_bmi2_adjust_temp_valid + MOVQ $0x00000001, R13 + +sequenceDecs_decode_bmi2_adjust_temp_valid: + CMPQ CX, $0x01 + CMOVQNE R11, R12 + MOVQ R10, R11 + MOVQ R13, R10 + MOVQ R13, CX + +sequenceDecs_decode_bmi2_after_adjust: + MOVQ CX, 16(R9) + + // Check values + MOVQ 8(R9), R13 + MOVQ (R9), R14 + LEAQ (R13)(R14*1), R15 + MOVQ s+0(FP), BP + ADDQ R15, 256(BP) + MOVQ ctx+16(FP), R15 + SUBQ R14, 128(R15) + JS error_not_enough_literals + CMPQ R13, $0x00020002 + JA sequenceDecs_decode_bmi2_error_match_len_too_big + TESTQ CX, CX + JNZ sequenceDecs_decode_bmi2_match_len_ofs_ok + TESTQ R13, R13 + JNZ sequenceDecs_decode_bmi2_error_match_len_ofs_mismatch + +sequenceDecs_decode_bmi2_match_len_ofs_ok: + ADDQ $0x18, R9 + MOVQ ctx+16(FP), CX + DECQ 96(CX) + JNS sequenceDecs_decode_bmi2_main_loop + MOVQ s+0(FP), CX + MOVQ R10, 144(CX) + MOVQ R11, 152(CX) + MOVQ R12, 160(CX) + MOVQ br+8(FP), CX + MOVQ AX, 32(CX) + MOVB DL, 40(CX) + MOVQ BX, 24(CX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decode_bmi2_error_match_len_ofs_mismatch: + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decode_bmi2_error_match_len_too_big: + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_decode_56_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int +// Requires: BMI, BMI2, CMOV +TEXT ·sequenceDecs_decode_56_bmi2(SB), $8-32 + MOVQ br+8(FP), CX + MOVQ 32(CX), AX + MOVBQZX 40(CX), DX + MOVQ 24(CX), BX + MOVQ (CX), CX + ADDQ BX, CX + MOVQ CX, (SP) + MOVQ ctx+16(FP), CX + MOVQ 72(CX), SI + MOVQ 80(CX), DI + MOVQ 88(CX), R8 + MOVQ 104(CX), R9 + MOVQ s+0(FP), CX + MOVQ 144(CX), R10 + MOVQ 152(CX), R11 + MOVQ 160(CX), R12 + +sequenceDecs_decode_56_bmi2_main_loop: + MOVQ (SP), R13 + + // Fill bitreader to have enough for the offset and match length. + CMPQ BX, $0x08 + JL sequenceDecs_decode_56_bmi2_fill_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R13 + MOVQ (R13), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decode_56_bmi2_fill_end + +sequenceDecs_decode_56_bmi2_fill_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decode_56_bmi2_fill_end + CMPQ DX, $0x07 + JLE sequenceDecs_decode_56_bmi2_fill_end + SHLQ $0x08, AX + SUBQ $0x01, R13 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R13), CX + ORQ CX, AX + JMP sequenceDecs_decode_56_bmi2_fill_byte_by_byte + +sequenceDecs_decode_56_bmi2_fill_end: + // Update offset + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R14 + MOVQ AX, R15 + LEAQ (DX)(R14*1), CX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + MOVQ CX, DX + MOVQ R8, CX + SHRQ $0x20, CX + ADDQ R15, CX + MOVQ CX, 16(R9) + + // Update match length + MOVQ $0x00000808, CX + BEXTRQ CX, DI, R14 + MOVQ AX, R15 + LEAQ (DX)(R14*1), CX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + MOVQ CX, DX + MOVQ DI, CX + SHRQ $0x20, CX + ADDQ R15, CX + MOVQ CX, 8(R9) + + // Update literal length + MOVQ $0x00000808, CX + BEXTRQ CX, SI, R14 + MOVQ AX, R15 + LEAQ (DX)(R14*1), CX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + MOVQ CX, DX + MOVQ SI, CX + SHRQ $0x20, CX + ADDQ R15, CX + MOVQ CX, (R9) + + // Fill bitreader for state updates + MOVQ R13, (SP) + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R13 + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decode_56_bmi2_skip_update + LEAQ (SI)(DI*1), R14 + ADDQ R8, R14 + MOVBQZX R14, R14 + LEAQ (DX)(R14*1), CX + MOVQ AX, R15 + MOVQ CX, DX + ROLQ CL, R15 + BZHIQ R14, R15, R15 + + // Update Offset State + BZHIQ R8, R15, CX + SHRXQ R8, R15, R15 + MOVQ $0x00001010, R14 + BEXTRQ R14, R8, R8 + ADDQ CX, R8 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Match Length State + BZHIQ DI, R15, CX + SHRXQ DI, R15, R15 + MOVQ $0x00001010, R14 + BEXTRQ R14, DI, DI + ADDQ CX, DI + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(DI*8), DI + + // Update Literal Length State + BZHIQ SI, R15, CX + MOVQ $0x00001010, R14 + BEXTRQ R14, SI, SI + ADDQ CX, SI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(SI*8), SI + +sequenceDecs_decode_56_bmi2_skip_update: + // Adjust offset + MOVQ 16(R9), CX + CMPQ R13, $0x01 + JBE sequenceDecs_decode_56_bmi2_adjust_offsetB_1_or_0 + MOVQ R11, R12 + MOVQ R10, R11 + MOVQ CX, R10 + JMP sequenceDecs_decode_56_bmi2_after_adjust + +sequenceDecs_decode_56_bmi2_adjust_offsetB_1_or_0: + CMPQ (R9), $0x00000000 + JNE sequenceDecs_decode_56_bmi2_adjust_offset_maybezero + INCQ CX + JMP sequenceDecs_decode_56_bmi2_adjust_offset_nonzero + +sequenceDecs_decode_56_bmi2_adjust_offset_maybezero: + TESTQ CX, CX + JNZ sequenceDecs_decode_56_bmi2_adjust_offset_nonzero + MOVQ R10, CX + JMP sequenceDecs_decode_56_bmi2_after_adjust + +sequenceDecs_decode_56_bmi2_adjust_offset_nonzero: + CMPQ CX, $0x01 + JB sequenceDecs_decode_56_bmi2_adjust_zero + JEQ sequenceDecs_decode_56_bmi2_adjust_one + CMPQ CX, $0x02 + JA sequenceDecs_decode_56_bmi2_adjust_three + JMP sequenceDecs_decode_56_bmi2_adjust_two + +sequenceDecs_decode_56_bmi2_adjust_zero: + MOVQ R10, R13 + JMP sequenceDecs_decode_56_bmi2_adjust_test_temp_valid + +sequenceDecs_decode_56_bmi2_adjust_one: + MOVQ R11, R13 + JMP sequenceDecs_decode_56_bmi2_adjust_test_temp_valid + +sequenceDecs_decode_56_bmi2_adjust_two: + MOVQ R12, R13 + JMP sequenceDecs_decode_56_bmi2_adjust_test_temp_valid + +sequenceDecs_decode_56_bmi2_adjust_three: + LEAQ -1(R10), R13 + +sequenceDecs_decode_56_bmi2_adjust_test_temp_valid: + TESTQ R13, R13 + JNZ sequenceDecs_decode_56_bmi2_adjust_temp_valid + MOVQ $0x00000001, R13 + +sequenceDecs_decode_56_bmi2_adjust_temp_valid: + CMPQ CX, $0x01 + CMOVQNE R11, R12 + MOVQ R10, R11 + MOVQ R13, R10 + MOVQ R13, CX + +sequenceDecs_decode_56_bmi2_after_adjust: + MOVQ CX, 16(R9) + + // Check values + MOVQ 8(R9), R13 + MOVQ (R9), R14 + LEAQ (R13)(R14*1), R15 + MOVQ s+0(FP), BP + ADDQ R15, 256(BP) + MOVQ ctx+16(FP), R15 + SUBQ R14, 128(R15) + JS error_not_enough_literals + CMPQ R13, $0x00020002 + JA sequenceDecs_decode_56_bmi2_error_match_len_too_big + TESTQ CX, CX + JNZ sequenceDecs_decode_56_bmi2_match_len_ofs_ok + TESTQ R13, R13 + JNZ sequenceDecs_decode_56_bmi2_error_match_len_ofs_mismatch + +sequenceDecs_decode_56_bmi2_match_len_ofs_ok: + ADDQ $0x18, R9 + MOVQ ctx+16(FP), CX + DECQ 96(CX) + JNS sequenceDecs_decode_56_bmi2_main_loop + MOVQ s+0(FP), CX + MOVQ R10, 144(CX) + MOVQ R11, 152(CX) + MOVQ R12, 160(CX) + MOVQ br+8(FP), CX + MOVQ AX, 32(CX) + MOVB DL, 40(CX) + MOVQ BX, 24(CX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decode_56_bmi2_error_match_len_ofs_mismatch: + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decode_56_bmi2_error_match_len_too_big: + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_executeSimple_amd64(ctx *executeAsmContext) bool +// Requires: SSE +TEXT ·sequenceDecs_executeSimple_amd64(SB), $8-9 + MOVQ ctx+0(FP), R10 + MOVQ 8(R10), CX + TESTQ CX, CX + JZ empty_seqs + MOVQ (R10), AX + MOVQ 24(R10), DX + MOVQ 32(R10), BX + MOVQ 80(R10), SI + MOVQ 104(R10), DI + MOVQ 120(R10), R8 + MOVQ 56(R10), R9 + MOVQ 64(R10), R10 + ADDQ R10, R9 + + // seqsBase += 24 * seqIndex + LEAQ (DX)(DX*2), R11 + SHLQ $0x03, R11 + ADDQ R11, AX + + // outBase += outPosition + ADDQ DI, BX + +main_loop: + MOVQ (AX), R11 + MOVQ 16(AX), R12 + MOVQ 8(AX), R13 + + // Copy literals + TESTQ R11, R11 + JZ check_offset + XORQ R14, R14 + +copy_1: + MOVUPS (SI)(R14*1), X0 + MOVUPS X0, (BX)(R14*1) + ADDQ $0x10, R14 + CMPQ R14, R11 + JB copy_1 + ADDQ R11, SI + ADDQ R11, BX + ADDQ R11, DI + + // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) +check_offset: + LEAQ (DI)(R10*1), R11 + CMPQ R12, R11 + JG error_match_off_too_big + CMPQ R12, R8 + JG error_match_off_too_big + + // Copy match from history + MOVQ R12, R11 + SUBQ DI, R11 + JLS copy_match + MOVQ R9, R14 + SUBQ R11, R14 + CMPQ R13, R11 + JG copy_all_from_history + MOVQ R13, R11 + SUBQ $0x10, R11 + JB copy_4_small + +copy_4_loop: + MOVUPS (R14), X0 + MOVUPS X0, (BX) + ADDQ $0x10, R14 + ADDQ $0x10, BX + SUBQ $0x10, R11 + JAE copy_4_loop + LEAQ 16(R14)(R11*1), R14 + LEAQ 16(BX)(R11*1), BX + MOVUPS -16(R14), X0 + MOVUPS X0, -16(BX) + JMP copy_4_end + +copy_4_small: + CMPQ R13, $0x03 + JE copy_4_move_3 + CMPQ R13, $0x08 + JB copy_4_move_4through7 + JMP copy_4_move_8through16 + +copy_4_move_3: + MOVW (R14), R11 + MOVB 2(R14), R12 + MOVW R11, (BX) + MOVB R12, 2(BX) + ADDQ R13, R14 + ADDQ R13, BX + JMP copy_4_end + +copy_4_move_4through7: + MOVL (R14), R11 + MOVL -4(R14)(R13*1), R12 + MOVL R11, (BX) + MOVL R12, -4(BX)(R13*1) + ADDQ R13, R14 + ADDQ R13, BX + JMP copy_4_end + +copy_4_move_8through16: + MOVQ (R14), R11 + MOVQ -8(R14)(R13*1), R12 + MOVQ R11, (BX) + MOVQ R12, -8(BX)(R13*1) + ADDQ R13, R14 + ADDQ R13, BX + +copy_4_end: + ADDQ R13, DI + ADDQ $0x18, AX + INCQ DX + CMPQ DX, CX + JB main_loop + JMP loop_finished + +copy_all_from_history: + MOVQ R11, R15 + SUBQ $0x10, R15 + JB copy_5_small + +copy_5_loop: + MOVUPS (R14), X0 + MOVUPS X0, (BX) + ADDQ $0x10, R14 + ADDQ $0x10, BX + SUBQ $0x10, R15 + JAE copy_5_loop + LEAQ 16(R14)(R15*1), R14 + LEAQ 16(BX)(R15*1), BX + MOVUPS -16(R14), X0 + MOVUPS X0, -16(BX) + JMP copy_5_end + +copy_5_small: + CMPQ R11, $0x03 + JE copy_5_move_3 + JB copy_5_move_1or2 + CMPQ R11, $0x08 + JB copy_5_move_4through7 + JMP copy_5_move_8through16 + +copy_5_move_1or2: + MOVB (R14), R15 + MOVB -1(R14)(R11*1), BP + MOVB R15, (BX) + MOVB BP, -1(BX)(R11*1) + ADDQ R11, R14 + ADDQ R11, BX + JMP copy_5_end + +copy_5_move_3: + MOVW (R14), R15 + MOVB 2(R14), BP + MOVW R15, (BX) + MOVB BP, 2(BX) + ADDQ R11, R14 + ADDQ R11, BX + JMP copy_5_end + +copy_5_move_4through7: + MOVL (R14), R15 + MOVL -4(R14)(R11*1), BP + MOVL R15, (BX) + MOVL BP, -4(BX)(R11*1) + ADDQ R11, R14 + ADDQ R11, BX + JMP copy_5_end + +copy_5_move_8through16: + MOVQ (R14), R15 + MOVQ -8(R14)(R11*1), BP + MOVQ R15, (BX) + MOVQ BP, -8(BX)(R11*1) + ADDQ R11, R14 + ADDQ R11, BX + +copy_5_end: + ADDQ R11, DI + SUBQ R11, R13 + + // Copy match from the current buffer +copy_match: + MOVQ BX, R11 + SUBQ R12, R11 + + // ml <= mo + CMPQ R13, R12 + JA copy_overlapping_match + + // Copy non-overlapping match + ADDQ R13, DI + MOVQ BX, R12 + ADDQ R13, BX + +copy_2: + MOVUPS (R11), X0 + MOVUPS X0, (R12) + ADDQ $0x10, R11 + ADDQ $0x10, R12 + SUBQ $0x10, R13 + JHI copy_2 + JMP handle_loop + + // Copy overlapping match +copy_overlapping_match: + ADDQ R13, DI + +copy_slow_3: + MOVB (R11), R12 + MOVB R12, (BX) + INCQ R11 + INCQ BX + DECQ R13 + JNZ copy_slow_3 + +handle_loop: + ADDQ $0x18, AX + INCQ DX + CMPQ DX, CX + JB main_loop + +loop_finished: + // Return value + MOVB $0x01, ret+8(FP) + + // Update the context + MOVQ ctx+0(FP), AX + MOVQ DX, 24(AX) + MOVQ DI, 104(AX) + MOVQ 80(AX), CX + SUBQ CX, SI + MOVQ SI, 112(AX) + RET + +error_match_off_too_big: + // Return value + MOVB $0x00, ret+8(FP) + + // Update the context + MOVQ ctx+0(FP), AX + MOVQ DX, 24(AX) + MOVQ DI, 104(AX) + MOVQ 80(AX), CX + SUBQ CX, SI + MOVQ SI, 112(AX) + RET + +empty_seqs: + // Return value + MOVB $0x01, ret+8(FP) + RET + +// func sequenceDecs_executeSimple_safe_amd64(ctx *executeAsmContext) bool +// Requires: SSE +TEXT ·sequenceDecs_executeSimple_safe_amd64(SB), $8-9 + MOVQ ctx+0(FP), R10 + MOVQ 8(R10), CX + TESTQ CX, CX + JZ empty_seqs + MOVQ (R10), AX + MOVQ 24(R10), DX + MOVQ 32(R10), BX + MOVQ 80(R10), SI + MOVQ 104(R10), DI + MOVQ 120(R10), R8 + MOVQ 56(R10), R9 + MOVQ 64(R10), R10 + ADDQ R10, R9 + + // seqsBase += 24 * seqIndex + LEAQ (DX)(DX*2), R11 + SHLQ $0x03, R11 + ADDQ R11, AX + + // outBase += outPosition + ADDQ DI, BX + +main_loop: + MOVQ (AX), R11 + MOVQ 16(AX), R12 + MOVQ 8(AX), R13 + + // Copy literals + TESTQ R11, R11 + JZ check_offset + MOVQ R11, R14 + SUBQ $0x10, R14 + JB copy_1_small + +copy_1_loop: + MOVUPS (SI), X0 + MOVUPS X0, (BX) + ADDQ $0x10, SI + ADDQ $0x10, BX + SUBQ $0x10, R14 + JAE copy_1_loop + LEAQ 16(SI)(R14*1), SI + LEAQ 16(BX)(R14*1), BX + MOVUPS -16(SI), X0 + MOVUPS X0, -16(BX) + JMP copy_1_end + +copy_1_small: + CMPQ R11, $0x03 + JE copy_1_move_3 + JB copy_1_move_1or2 + CMPQ R11, $0x08 + JB copy_1_move_4through7 + JMP copy_1_move_8through16 + +copy_1_move_1or2: + MOVB (SI), R14 + MOVB -1(SI)(R11*1), R15 + MOVB R14, (BX) + MOVB R15, -1(BX)(R11*1) + ADDQ R11, SI + ADDQ R11, BX + JMP copy_1_end + +copy_1_move_3: + MOVW (SI), R14 + MOVB 2(SI), R15 + MOVW R14, (BX) + MOVB R15, 2(BX) + ADDQ R11, SI + ADDQ R11, BX + JMP copy_1_end + +copy_1_move_4through7: + MOVL (SI), R14 + MOVL -4(SI)(R11*1), R15 + MOVL R14, (BX) + MOVL R15, -4(BX)(R11*1) + ADDQ R11, SI + ADDQ R11, BX + JMP copy_1_end + +copy_1_move_8through16: + MOVQ (SI), R14 + MOVQ -8(SI)(R11*1), R15 + MOVQ R14, (BX) + MOVQ R15, -8(BX)(R11*1) + ADDQ R11, SI + ADDQ R11, BX + +copy_1_end: + ADDQ R11, DI + + // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) +check_offset: + LEAQ (DI)(R10*1), R11 + CMPQ R12, R11 + JG error_match_off_too_big + CMPQ R12, R8 + JG error_match_off_too_big + + // Copy match from history + MOVQ R12, R11 + SUBQ DI, R11 + JLS copy_match + MOVQ R9, R14 + SUBQ R11, R14 + CMPQ R13, R11 + JG copy_all_from_history + MOVQ R13, R11 + SUBQ $0x10, R11 + JB copy_4_small + +copy_4_loop: + MOVUPS (R14), X0 + MOVUPS X0, (BX) + ADDQ $0x10, R14 + ADDQ $0x10, BX + SUBQ $0x10, R11 + JAE copy_4_loop + LEAQ 16(R14)(R11*1), R14 + LEAQ 16(BX)(R11*1), BX + MOVUPS -16(R14), X0 + MOVUPS X0, -16(BX) + JMP copy_4_end + +copy_4_small: + CMPQ R13, $0x03 + JE copy_4_move_3 + CMPQ R13, $0x08 + JB copy_4_move_4through7 + JMP copy_4_move_8through16 + +copy_4_move_3: + MOVW (R14), R11 + MOVB 2(R14), R12 + MOVW R11, (BX) + MOVB R12, 2(BX) + ADDQ R13, R14 + ADDQ R13, BX + JMP copy_4_end + +copy_4_move_4through7: + MOVL (R14), R11 + MOVL -4(R14)(R13*1), R12 + MOVL R11, (BX) + MOVL R12, -4(BX)(R13*1) + ADDQ R13, R14 + ADDQ R13, BX + JMP copy_4_end + +copy_4_move_8through16: + MOVQ (R14), R11 + MOVQ -8(R14)(R13*1), R12 + MOVQ R11, (BX) + MOVQ R12, -8(BX)(R13*1) + ADDQ R13, R14 + ADDQ R13, BX + +copy_4_end: + ADDQ R13, DI + ADDQ $0x18, AX + INCQ DX + CMPQ DX, CX + JB main_loop + JMP loop_finished + +copy_all_from_history: + MOVQ R11, R15 + SUBQ $0x10, R15 + JB copy_5_small + +copy_5_loop: + MOVUPS (R14), X0 + MOVUPS X0, (BX) + ADDQ $0x10, R14 + ADDQ $0x10, BX + SUBQ $0x10, R15 + JAE copy_5_loop + LEAQ 16(R14)(R15*1), R14 + LEAQ 16(BX)(R15*1), BX + MOVUPS -16(R14), X0 + MOVUPS X0, -16(BX) + JMP copy_5_end + +copy_5_small: + CMPQ R11, $0x03 + JE copy_5_move_3 + JB copy_5_move_1or2 + CMPQ R11, $0x08 + JB copy_5_move_4through7 + JMP copy_5_move_8through16 + +copy_5_move_1or2: + MOVB (R14), R15 + MOVB -1(R14)(R11*1), BP + MOVB R15, (BX) + MOVB BP, -1(BX)(R11*1) + ADDQ R11, R14 + ADDQ R11, BX + JMP copy_5_end + +copy_5_move_3: + MOVW (R14), R15 + MOVB 2(R14), BP + MOVW R15, (BX) + MOVB BP, 2(BX) + ADDQ R11, R14 + ADDQ R11, BX + JMP copy_5_end + +copy_5_move_4through7: + MOVL (R14), R15 + MOVL -4(R14)(R11*1), BP + MOVL R15, (BX) + MOVL BP, -4(BX)(R11*1) + ADDQ R11, R14 + ADDQ R11, BX + JMP copy_5_end + +copy_5_move_8through16: + MOVQ (R14), R15 + MOVQ -8(R14)(R11*1), BP + MOVQ R15, (BX) + MOVQ BP, -8(BX)(R11*1) + ADDQ R11, R14 + ADDQ R11, BX + +copy_5_end: + ADDQ R11, DI + SUBQ R11, R13 + + // Copy match from the current buffer +copy_match: + MOVQ BX, R11 + SUBQ R12, R11 + + // ml <= mo + CMPQ R13, R12 + JA copy_overlapping_match + + // Copy non-overlapping match + ADDQ R13, DI + MOVQ R13, R12 + SUBQ $0x10, R12 + JB copy_2_small + +copy_2_loop: + MOVUPS (R11), X0 + MOVUPS X0, (BX) + ADDQ $0x10, R11 + ADDQ $0x10, BX + SUBQ $0x10, R12 + JAE copy_2_loop + LEAQ 16(R11)(R12*1), R11 + LEAQ 16(BX)(R12*1), BX + MOVUPS -16(R11), X0 + MOVUPS X0, -16(BX) + JMP copy_2_end + +copy_2_small: + CMPQ R13, $0x03 + JE copy_2_move_3 + JB copy_2_move_1or2 + CMPQ R13, $0x08 + JB copy_2_move_4through7 + JMP copy_2_move_8through16 + +copy_2_move_1or2: + MOVB (R11), R12 + MOVB -1(R11)(R13*1), R14 + MOVB R12, (BX) + MOVB R14, -1(BX)(R13*1) + ADDQ R13, R11 + ADDQ R13, BX + JMP copy_2_end + +copy_2_move_3: + MOVW (R11), R12 + MOVB 2(R11), R14 + MOVW R12, (BX) + MOVB R14, 2(BX) + ADDQ R13, R11 + ADDQ R13, BX + JMP copy_2_end + +copy_2_move_4through7: + MOVL (R11), R12 + MOVL -4(R11)(R13*1), R14 + MOVL R12, (BX) + MOVL R14, -4(BX)(R13*1) + ADDQ R13, R11 + ADDQ R13, BX + JMP copy_2_end + +copy_2_move_8through16: + MOVQ (R11), R12 + MOVQ -8(R11)(R13*1), R14 + MOVQ R12, (BX) + MOVQ R14, -8(BX)(R13*1) + ADDQ R13, R11 + ADDQ R13, BX + +copy_2_end: + JMP handle_loop + + // Copy overlapping match +copy_overlapping_match: + ADDQ R13, DI + +copy_slow_3: + MOVB (R11), R12 + MOVB R12, (BX) + INCQ R11 + INCQ BX + DECQ R13 + JNZ copy_slow_3 + +handle_loop: + ADDQ $0x18, AX + INCQ DX + CMPQ DX, CX + JB main_loop + +loop_finished: + // Return value + MOVB $0x01, ret+8(FP) + + // Update the context + MOVQ ctx+0(FP), AX + MOVQ DX, 24(AX) + MOVQ DI, 104(AX) + MOVQ 80(AX), CX + SUBQ CX, SI + MOVQ SI, 112(AX) + RET + +error_match_off_too_big: + // Return value + MOVB $0x00, ret+8(FP) + + // Update the context + MOVQ ctx+0(FP), AX + MOVQ DX, 24(AX) + MOVQ DI, 104(AX) + MOVQ 80(AX), CX + SUBQ CX, SI + MOVQ SI, 112(AX) + RET + +empty_seqs: + // Return value + MOVB $0x01, ret+8(FP) + RET + +// func sequenceDecs_decodeSync_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int +// Requires: CMOV, SSE +TEXT ·sequenceDecs_decodeSync_amd64(SB), $64-32 + MOVQ br+8(FP), AX + MOVQ 32(AX), DX + MOVBQZX 40(AX), BX + MOVQ 24(AX), SI + MOVQ (AX), AX + ADDQ SI, AX + MOVQ AX, (SP) + MOVQ ctx+16(FP), AX + MOVQ 72(AX), DI + MOVQ 80(AX), R8 + MOVQ 88(AX), R9 + XORQ CX, CX + MOVQ CX, 8(SP) + MOVQ CX, 16(SP) + MOVQ CX, 24(SP) + MOVQ 112(AX), R10 + MOVQ 128(AX), CX + MOVQ CX, 32(SP) + MOVQ 144(AX), R11 + MOVQ 136(AX), R12 + MOVQ 200(AX), CX + MOVQ CX, 56(SP) + MOVQ 176(AX), CX + MOVQ CX, 48(SP) + MOVQ 184(AX), AX + MOVQ AX, 40(SP) + MOVQ 40(SP), AX + ADDQ AX, 48(SP) + + // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) + ADDQ R10, 32(SP) + + // outBase += outPosition + ADDQ R12, R10 + +sequenceDecs_decodeSync_amd64_main_loop: + MOVQ (SP), R13 + + // Fill bitreader to have enough for the offset and match length. + CMPQ SI, $0x08 + JL sequenceDecs_decodeSync_amd64_fill_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R13 + MOVQ (R13), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decodeSync_amd64_fill_end + +sequenceDecs_decodeSync_amd64_fill_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decodeSync_amd64_fill_end + CMPQ BX, $0x07 + JLE sequenceDecs_decodeSync_amd64_fill_end + SHLQ $0x08, DX + SUBQ $0x01, R13 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R13), AX + ORQ AX, DX + JMP sequenceDecs_decodeSync_amd64_fill_byte_by_byte + +sequenceDecs_decodeSync_amd64_fill_end: + // Update offset + MOVQ R9, AX + MOVQ BX, CX + MOVQ DX, R14 + SHLQ CL, R14 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decodeSync_amd64_of_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decodeSync_amd64_of_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decodeSync_amd64_of_update_zero + NEGQ CX + SHRQ CL, R14 + ADDQ R14, AX + +sequenceDecs_decodeSync_amd64_of_update_zero: + MOVQ AX, 8(SP) + + // Update match length + MOVQ R8, AX + MOVQ BX, CX + MOVQ DX, R14 + SHLQ CL, R14 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decodeSync_amd64_ml_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decodeSync_amd64_ml_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decodeSync_amd64_ml_update_zero + NEGQ CX + SHRQ CL, R14 + ADDQ R14, AX + +sequenceDecs_decodeSync_amd64_ml_update_zero: + MOVQ AX, 16(SP) + + // Fill bitreader to have enough for the remaining + CMPQ SI, $0x08 + JL sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R13 + MOVQ (R13), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decodeSync_amd64_fill_2_end + +sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decodeSync_amd64_fill_2_end + CMPQ BX, $0x07 + JLE sequenceDecs_decodeSync_amd64_fill_2_end + SHLQ $0x08, DX + SUBQ $0x01, R13 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R13), AX + ORQ AX, DX + JMP sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte + +sequenceDecs_decodeSync_amd64_fill_2_end: + // Update literal length + MOVQ DI, AX + MOVQ BX, CX + MOVQ DX, R14 + SHLQ CL, R14 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decodeSync_amd64_ll_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decodeSync_amd64_ll_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decodeSync_amd64_ll_update_zero + NEGQ CX + SHRQ CL, R14 + ADDQ R14, AX + +sequenceDecs_decodeSync_amd64_ll_update_zero: + MOVQ AX, 24(SP) + + // Fill bitreader for state updates + MOVQ R13, (SP) + MOVQ R9, AX + SHRQ $0x08, AX + MOVBQZX AL, AX + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decodeSync_amd64_skip_update + + // Update Literal Length State + MOVBQZX DI, R13 + SHRQ $0x10, DI + MOVWQZX DI, DI + LEAQ (BX)(R13*1), CX + MOVQ DX, R14 + MOVQ CX, BX + ROLQ CL, R14 + MOVL $0x00000001, R15 + MOVB R13, CL + SHLL CL, R15 + DECL R15 + ANDQ R15, R14 + ADDQ R14, DI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(DI*8), DI + + // Update Match Length State + MOVBQZX R8, R13 + SHRQ $0x10, R8 + MOVWQZX R8, R8 + LEAQ (BX)(R13*1), CX + MOVQ DX, R14 + MOVQ CX, BX + ROLQ CL, R14 + MOVL $0x00000001, R15 + MOVB R13, CL + SHLL CL, R15 + DECL R15 + ANDQ R15, R14 + ADDQ R14, R8 + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Offset State + MOVBQZX R9, R13 + SHRQ $0x10, R9 + MOVWQZX R9, R9 + LEAQ (BX)(R13*1), CX + MOVQ DX, R14 + MOVQ CX, BX + ROLQ CL, R14 + MOVL $0x00000001, R15 + MOVB R13, CL + SHLL CL, R15 + DECL R15 + ANDQ R15, R14 + ADDQ R14, R9 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R9*8), R9 + +sequenceDecs_decodeSync_amd64_skip_update: + // Adjust offset + MOVQ s+0(FP), CX + MOVQ 8(SP), R13 + CMPQ AX, $0x01 + JBE sequenceDecs_decodeSync_amd64_adjust_offsetB_1_or_0 + MOVUPS 144(CX), X0 + MOVQ R13, 144(CX) + MOVUPS X0, 152(CX) + JMP sequenceDecs_decodeSync_amd64_after_adjust + +sequenceDecs_decodeSync_amd64_adjust_offsetB_1_or_0: + CMPQ 24(SP), $0x00000000 + JNE sequenceDecs_decodeSync_amd64_adjust_offset_maybezero + INCQ R13 + JMP sequenceDecs_decodeSync_amd64_adjust_offset_nonzero + +sequenceDecs_decodeSync_amd64_adjust_offset_maybezero: + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_amd64_adjust_offset_nonzero + MOVQ 144(CX), R13 + JMP sequenceDecs_decodeSync_amd64_after_adjust + +sequenceDecs_decodeSync_amd64_adjust_offset_nonzero: + MOVQ R13, AX + XORQ R14, R14 + MOVQ $-1, R15 + CMPQ R13, $0x03 + CMOVQEQ R14, AX + CMOVQEQ R15, R14 + ADDQ 144(CX)(AX*8), R14 + JNZ sequenceDecs_decodeSync_amd64_adjust_temp_valid + MOVQ $0x00000001, R14 + +sequenceDecs_decodeSync_amd64_adjust_temp_valid: + CMPQ R13, $0x01 + JZ sequenceDecs_decodeSync_amd64_adjust_skip + MOVQ 152(CX), AX + MOVQ AX, 160(CX) + +sequenceDecs_decodeSync_amd64_adjust_skip: + MOVQ 144(CX), AX + MOVQ AX, 152(CX) + MOVQ R14, 144(CX) + MOVQ R14, R13 + +sequenceDecs_decodeSync_amd64_after_adjust: + MOVQ R13, 8(SP) + + // Check values + MOVQ 16(SP), AX + MOVQ 24(SP), CX + LEAQ (AX)(CX*1), R14 + MOVQ s+0(FP), R15 + ADDQ R14, 256(R15) + MOVQ ctx+16(FP), R14 + SUBQ CX, 104(R14) + JS error_not_enough_literals + CMPQ AX, $0x00020002 + JA sequenceDecs_decodeSync_amd64_error_match_len_too_big + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_amd64_match_len_ofs_ok + TESTQ AX, AX + JNZ sequenceDecs_decodeSync_amd64_error_match_len_ofs_mismatch + +sequenceDecs_decodeSync_amd64_match_len_ofs_ok: + MOVQ 24(SP), AX + MOVQ 8(SP), CX + MOVQ 16(SP), R13 + + // Check if we have enough space in s.out + LEAQ (AX)(R13*1), R14 + ADDQ R10, R14 + CMPQ R14, 32(SP) + JA error_not_enough_space + + // Copy literals + TESTQ AX, AX + JZ check_offset + XORQ R14, R14 + +copy_1: + MOVUPS (R11)(R14*1), X0 + MOVUPS X0, (R10)(R14*1) + ADDQ $0x10, R14 + CMPQ R14, AX + JB copy_1 + ADDQ AX, R11 + ADDQ AX, R10 + ADDQ AX, R12 + + // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) +check_offset: + MOVQ R12, AX + ADDQ 40(SP), AX + CMPQ CX, AX + JG error_match_off_too_big + CMPQ CX, 56(SP) + JG error_match_off_too_big + + // Copy match from history + MOVQ CX, AX + SUBQ R12, AX + JLS copy_match + MOVQ 48(SP), R14 + SUBQ AX, R14 + CMPQ R13, AX + JG copy_all_from_history + MOVQ R13, AX + SUBQ $0x10, AX + JB copy_4_small + +copy_4_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R10) + ADDQ $0x10, R14 + ADDQ $0x10, R10 + SUBQ $0x10, AX + JAE copy_4_loop + LEAQ 16(R14)(AX*1), R14 + LEAQ 16(R10)(AX*1), R10 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R10) + JMP copy_4_end + +copy_4_small: + CMPQ R13, $0x03 + JE copy_4_move_3 + CMPQ R13, $0x08 + JB copy_4_move_4through7 + JMP copy_4_move_8through16 + +copy_4_move_3: + MOVW (R14), AX + MOVB 2(R14), CL + MOVW AX, (R10) + MOVB CL, 2(R10) + ADDQ R13, R14 + ADDQ R13, R10 + JMP copy_4_end + +copy_4_move_4through7: + MOVL (R14), AX + MOVL -4(R14)(R13*1), CX + MOVL AX, (R10) + MOVL CX, -4(R10)(R13*1) + ADDQ R13, R14 + ADDQ R13, R10 + JMP copy_4_end + +copy_4_move_8through16: + MOVQ (R14), AX + MOVQ -8(R14)(R13*1), CX + MOVQ AX, (R10) + MOVQ CX, -8(R10)(R13*1) + ADDQ R13, R14 + ADDQ R13, R10 + +copy_4_end: + ADDQ R13, R12 + JMP handle_loop + JMP loop_finished + +copy_all_from_history: + MOVQ AX, R15 + SUBQ $0x10, R15 + JB copy_5_small + +copy_5_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R10) + ADDQ $0x10, R14 + ADDQ $0x10, R10 + SUBQ $0x10, R15 + JAE copy_5_loop + LEAQ 16(R14)(R15*1), R14 + LEAQ 16(R10)(R15*1), R10 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R10) + JMP copy_5_end + +copy_5_small: + CMPQ AX, $0x03 + JE copy_5_move_3 + JB copy_5_move_1or2 + CMPQ AX, $0x08 + JB copy_5_move_4through7 + JMP copy_5_move_8through16 + +copy_5_move_1or2: + MOVB (R14), R15 + MOVB -1(R14)(AX*1), BP + MOVB R15, (R10) + MOVB BP, -1(R10)(AX*1) + ADDQ AX, R14 + ADDQ AX, R10 + JMP copy_5_end + +copy_5_move_3: + MOVW (R14), R15 + MOVB 2(R14), BP + MOVW R15, (R10) + MOVB BP, 2(R10) + ADDQ AX, R14 + ADDQ AX, R10 + JMP copy_5_end + +copy_5_move_4through7: + MOVL (R14), R15 + MOVL -4(R14)(AX*1), BP + MOVL R15, (R10) + MOVL BP, -4(R10)(AX*1) + ADDQ AX, R14 + ADDQ AX, R10 + JMP copy_5_end + +copy_5_move_8through16: + MOVQ (R14), R15 + MOVQ -8(R14)(AX*1), BP + MOVQ R15, (R10) + MOVQ BP, -8(R10)(AX*1) + ADDQ AX, R14 + ADDQ AX, R10 + +copy_5_end: + ADDQ AX, R12 + SUBQ AX, R13 + + // Copy match from the current buffer +copy_match: + MOVQ R10, AX + SUBQ CX, AX + + // ml <= mo + CMPQ R13, CX + JA copy_overlapping_match + + // Copy non-overlapping match + ADDQ R13, R12 + MOVQ R10, CX + ADDQ R13, R10 + +copy_2: + MOVUPS (AX), X0 + MOVUPS X0, (CX) + ADDQ $0x10, AX + ADDQ $0x10, CX + SUBQ $0x10, R13 + JHI copy_2 + JMP handle_loop + + // Copy overlapping match +copy_overlapping_match: + ADDQ R13, R12 + +copy_slow_3: + MOVB (AX), CL + MOVB CL, (R10) + INCQ AX + INCQ R10 + DECQ R13 + JNZ copy_slow_3 + +handle_loop: + MOVQ ctx+16(FP), AX + DECQ 96(AX) + JNS sequenceDecs_decodeSync_amd64_main_loop + +loop_finished: + MOVQ br+8(FP), AX + MOVQ DX, 32(AX) + MOVB BL, 40(AX) + MOVQ SI, 24(AX) + + // Update the context + MOVQ ctx+16(FP), AX + MOVQ R12, 136(AX) + MOVQ 144(AX), CX + SUBQ CX, R11 + MOVQ R11, 168(AX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decodeSync_amd64_error_match_len_ofs_mismatch: + MOVQ 16(SP), AX + MOVQ ctx+16(FP), CX + MOVQ AX, 216(CX) + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decodeSync_amd64_error_match_len_too_big: + MOVQ ctx+16(FP), AX + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error +error_match_off_too_big: + MOVQ ctx+16(FP), AX + MOVQ 8(SP), CX + MOVQ CX, 224(AX) + MOVQ R12, 136(AX) + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error +error_not_enough_space: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ R12, 136(AX) + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_decodeSync_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int +// Requires: BMI, BMI2, CMOV, SSE +TEXT ·sequenceDecs_decodeSync_bmi2(SB), $64-32 + MOVQ br+8(FP), CX + MOVQ 32(CX), AX + MOVBQZX 40(CX), DX + MOVQ 24(CX), BX + MOVQ (CX), CX + ADDQ BX, CX + MOVQ CX, (SP) + MOVQ ctx+16(FP), CX + MOVQ 72(CX), SI + MOVQ 80(CX), DI + MOVQ 88(CX), R8 + XORQ R9, R9 + MOVQ R9, 8(SP) + MOVQ R9, 16(SP) + MOVQ R9, 24(SP) + MOVQ 112(CX), R9 + MOVQ 128(CX), R10 + MOVQ R10, 32(SP) + MOVQ 144(CX), R10 + MOVQ 136(CX), R11 + MOVQ 200(CX), R12 + MOVQ R12, 56(SP) + MOVQ 176(CX), R12 + MOVQ R12, 48(SP) + MOVQ 184(CX), CX + MOVQ CX, 40(SP) + MOVQ 40(SP), CX + ADDQ CX, 48(SP) + + // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) + ADDQ R9, 32(SP) + + // outBase += outPosition + ADDQ R11, R9 + +sequenceDecs_decodeSync_bmi2_main_loop: + MOVQ (SP), R12 + + // Fill bitreader to have enough for the offset and match length. + CMPQ BX, $0x08 + JL sequenceDecs_decodeSync_bmi2_fill_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R12 + MOVQ (R12), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decodeSync_bmi2_fill_end + +sequenceDecs_decodeSync_bmi2_fill_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decodeSync_bmi2_fill_end + CMPQ DX, $0x07 + JLE sequenceDecs_decodeSync_bmi2_fill_end + SHLQ $0x08, AX + SUBQ $0x01, R12 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R12), CX + ORQ CX, AX + JMP sequenceDecs_decodeSync_bmi2_fill_byte_by_byte + +sequenceDecs_decodeSync_bmi2_fill_end: + // Update offset + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R13 + MOVQ AX, R14 + LEAQ (DX)(R13*1), CX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + MOVQ CX, DX + MOVQ R8, CX + SHRQ $0x20, CX + ADDQ R14, CX + MOVQ CX, 8(SP) + + // Update match length + MOVQ $0x00000808, CX + BEXTRQ CX, DI, R13 + MOVQ AX, R14 + LEAQ (DX)(R13*1), CX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + MOVQ CX, DX + MOVQ DI, CX + SHRQ $0x20, CX + ADDQ R14, CX + MOVQ CX, 16(SP) + + // Fill bitreader to have enough for the remaining + CMPQ BX, $0x08 + JL sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R12 + MOVQ (R12), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decodeSync_bmi2_fill_2_end + +sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decodeSync_bmi2_fill_2_end + CMPQ DX, $0x07 + JLE sequenceDecs_decodeSync_bmi2_fill_2_end + SHLQ $0x08, AX + SUBQ $0x01, R12 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R12), CX + ORQ CX, AX + JMP sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte + +sequenceDecs_decodeSync_bmi2_fill_2_end: + // Update literal length + MOVQ $0x00000808, CX + BEXTRQ CX, SI, R13 + MOVQ AX, R14 + LEAQ (DX)(R13*1), CX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + MOVQ CX, DX + MOVQ SI, CX + SHRQ $0x20, CX + ADDQ R14, CX + MOVQ CX, 24(SP) + + // Fill bitreader for state updates + MOVQ R12, (SP) + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R12 + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decodeSync_bmi2_skip_update + LEAQ (SI)(DI*1), R13 + ADDQ R8, R13 + MOVBQZX R13, R13 + LEAQ (DX)(R13*1), CX + MOVQ AX, R14 + MOVQ CX, DX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + + // Update Offset State + BZHIQ R8, R14, CX + SHRXQ R8, R14, R14 + MOVQ $0x00001010, R13 + BEXTRQ R13, R8, R8 + ADDQ CX, R8 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Match Length State + BZHIQ DI, R14, CX + SHRXQ DI, R14, R14 + MOVQ $0x00001010, R13 + BEXTRQ R13, DI, DI + ADDQ CX, DI + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(DI*8), DI + + // Update Literal Length State + BZHIQ SI, R14, CX + MOVQ $0x00001010, R13 + BEXTRQ R13, SI, SI + ADDQ CX, SI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(SI*8), SI + +sequenceDecs_decodeSync_bmi2_skip_update: + // Adjust offset + MOVQ s+0(FP), CX + MOVQ 8(SP), R13 + CMPQ R12, $0x01 + JBE sequenceDecs_decodeSync_bmi2_adjust_offsetB_1_or_0 + MOVUPS 144(CX), X0 + MOVQ R13, 144(CX) + MOVUPS X0, 152(CX) + JMP sequenceDecs_decodeSync_bmi2_after_adjust + +sequenceDecs_decodeSync_bmi2_adjust_offsetB_1_or_0: + CMPQ 24(SP), $0x00000000 + JNE sequenceDecs_decodeSync_bmi2_adjust_offset_maybezero + INCQ R13 + JMP sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero + +sequenceDecs_decodeSync_bmi2_adjust_offset_maybezero: + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero + MOVQ 144(CX), R13 + JMP sequenceDecs_decodeSync_bmi2_after_adjust + +sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero: + MOVQ R13, R12 + XORQ R14, R14 + MOVQ $-1, R15 + CMPQ R13, $0x03 + CMOVQEQ R14, R12 + CMOVQEQ R15, R14 + ADDQ 144(CX)(R12*8), R14 + JNZ sequenceDecs_decodeSync_bmi2_adjust_temp_valid + MOVQ $0x00000001, R14 + +sequenceDecs_decodeSync_bmi2_adjust_temp_valid: + CMPQ R13, $0x01 + JZ sequenceDecs_decodeSync_bmi2_adjust_skip + MOVQ 152(CX), R12 + MOVQ R12, 160(CX) + +sequenceDecs_decodeSync_bmi2_adjust_skip: + MOVQ 144(CX), R12 + MOVQ R12, 152(CX) + MOVQ R14, 144(CX) + MOVQ R14, R13 + +sequenceDecs_decodeSync_bmi2_after_adjust: + MOVQ R13, 8(SP) + + // Check values + MOVQ 16(SP), CX + MOVQ 24(SP), R12 + LEAQ (CX)(R12*1), R14 + MOVQ s+0(FP), R15 + ADDQ R14, 256(R15) + MOVQ ctx+16(FP), R14 + SUBQ R12, 104(R14) + JS error_not_enough_literals + CMPQ CX, $0x00020002 + JA sequenceDecs_decodeSync_bmi2_error_match_len_too_big + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_bmi2_match_len_ofs_ok + TESTQ CX, CX + JNZ sequenceDecs_decodeSync_bmi2_error_match_len_ofs_mismatch + +sequenceDecs_decodeSync_bmi2_match_len_ofs_ok: + MOVQ 24(SP), CX + MOVQ 8(SP), R12 + MOVQ 16(SP), R13 + + // Check if we have enough space in s.out + LEAQ (CX)(R13*1), R14 + ADDQ R9, R14 + CMPQ R14, 32(SP) + JA error_not_enough_space + + // Copy literals + TESTQ CX, CX + JZ check_offset + XORQ R14, R14 + +copy_1: + MOVUPS (R10)(R14*1), X0 + MOVUPS X0, (R9)(R14*1) + ADDQ $0x10, R14 + CMPQ R14, CX + JB copy_1 + ADDQ CX, R10 + ADDQ CX, R9 + ADDQ CX, R11 + + // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) +check_offset: + MOVQ R11, CX + ADDQ 40(SP), CX + CMPQ R12, CX + JG error_match_off_too_big + CMPQ R12, 56(SP) + JG error_match_off_too_big + + // Copy match from history + MOVQ R12, CX + SUBQ R11, CX + JLS copy_match + MOVQ 48(SP), R14 + SUBQ CX, R14 + CMPQ R13, CX + JG copy_all_from_history + MOVQ R13, CX + SUBQ $0x10, CX + JB copy_4_small + +copy_4_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R9) + ADDQ $0x10, R14 + ADDQ $0x10, R9 + SUBQ $0x10, CX + JAE copy_4_loop + LEAQ 16(R14)(CX*1), R14 + LEAQ 16(R9)(CX*1), R9 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R9) + JMP copy_4_end + +copy_4_small: + CMPQ R13, $0x03 + JE copy_4_move_3 + CMPQ R13, $0x08 + JB copy_4_move_4through7 + JMP copy_4_move_8through16 + +copy_4_move_3: + MOVW (R14), CX + MOVB 2(R14), R12 + MOVW CX, (R9) + MOVB R12, 2(R9) + ADDQ R13, R14 + ADDQ R13, R9 + JMP copy_4_end + +copy_4_move_4through7: + MOVL (R14), CX + MOVL -4(R14)(R13*1), R12 + MOVL CX, (R9) + MOVL R12, -4(R9)(R13*1) + ADDQ R13, R14 + ADDQ R13, R9 + JMP copy_4_end + +copy_4_move_8through16: + MOVQ (R14), CX + MOVQ -8(R14)(R13*1), R12 + MOVQ CX, (R9) + MOVQ R12, -8(R9)(R13*1) + ADDQ R13, R14 + ADDQ R13, R9 + +copy_4_end: + ADDQ R13, R11 + JMP handle_loop + JMP loop_finished + +copy_all_from_history: + MOVQ CX, R15 + SUBQ $0x10, R15 + JB copy_5_small + +copy_5_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R9) + ADDQ $0x10, R14 + ADDQ $0x10, R9 + SUBQ $0x10, R15 + JAE copy_5_loop + LEAQ 16(R14)(R15*1), R14 + LEAQ 16(R9)(R15*1), R9 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R9) + JMP copy_5_end + +copy_5_small: + CMPQ CX, $0x03 + JE copy_5_move_3 + JB copy_5_move_1or2 + CMPQ CX, $0x08 + JB copy_5_move_4through7 + JMP copy_5_move_8through16 + +copy_5_move_1or2: + MOVB (R14), R15 + MOVB -1(R14)(CX*1), BP + MOVB R15, (R9) + MOVB BP, -1(R9)(CX*1) + ADDQ CX, R14 + ADDQ CX, R9 + JMP copy_5_end + +copy_5_move_3: + MOVW (R14), R15 + MOVB 2(R14), BP + MOVW R15, (R9) + MOVB BP, 2(R9) + ADDQ CX, R14 + ADDQ CX, R9 + JMP copy_5_end + +copy_5_move_4through7: + MOVL (R14), R15 + MOVL -4(R14)(CX*1), BP + MOVL R15, (R9) + MOVL BP, -4(R9)(CX*1) + ADDQ CX, R14 + ADDQ CX, R9 + JMP copy_5_end + +copy_5_move_8through16: + MOVQ (R14), R15 + MOVQ -8(R14)(CX*1), BP + MOVQ R15, (R9) + MOVQ BP, -8(R9)(CX*1) + ADDQ CX, R14 + ADDQ CX, R9 + +copy_5_end: + ADDQ CX, R11 + SUBQ CX, R13 + + // Copy match from the current buffer +copy_match: + MOVQ R9, CX + SUBQ R12, CX + + // ml <= mo + CMPQ R13, R12 + JA copy_overlapping_match + + // Copy non-overlapping match + ADDQ R13, R11 + MOVQ R9, R12 + ADDQ R13, R9 + +copy_2: + MOVUPS (CX), X0 + MOVUPS X0, (R12) + ADDQ $0x10, CX + ADDQ $0x10, R12 + SUBQ $0x10, R13 + JHI copy_2 + JMP handle_loop + + // Copy overlapping match +copy_overlapping_match: + ADDQ R13, R11 + +copy_slow_3: + MOVB (CX), R12 + MOVB R12, (R9) + INCQ CX + INCQ R9 + DECQ R13 + JNZ copy_slow_3 + +handle_loop: + MOVQ ctx+16(FP), CX + DECQ 96(CX) + JNS sequenceDecs_decodeSync_bmi2_main_loop + +loop_finished: + MOVQ br+8(FP), CX + MOVQ AX, 32(CX) + MOVB DL, 40(CX) + MOVQ BX, 24(CX) + + // Update the context + MOVQ ctx+16(FP), AX + MOVQ R11, 136(AX) + MOVQ 144(AX), CX + SUBQ CX, R10 + MOVQ R10, 168(AX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decodeSync_bmi2_error_match_len_ofs_mismatch: + MOVQ 16(SP), AX + MOVQ ctx+16(FP), CX + MOVQ AX, 216(CX) + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decodeSync_bmi2_error_match_len_too_big: + MOVQ ctx+16(FP), AX + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error +error_match_off_too_big: + MOVQ ctx+16(FP), AX + MOVQ 8(SP), CX + MOVQ CX, 224(AX) + MOVQ R11, 136(AX) + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error +error_not_enough_space: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ R11, 136(AX) + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_decodeSync_safe_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int +// Requires: CMOV, SSE +TEXT ·sequenceDecs_decodeSync_safe_amd64(SB), $64-32 + MOVQ br+8(FP), AX + MOVQ 32(AX), DX + MOVBQZX 40(AX), BX + MOVQ 24(AX), SI + MOVQ (AX), AX + ADDQ SI, AX + MOVQ AX, (SP) + MOVQ ctx+16(FP), AX + MOVQ 72(AX), DI + MOVQ 80(AX), R8 + MOVQ 88(AX), R9 + XORQ CX, CX + MOVQ CX, 8(SP) + MOVQ CX, 16(SP) + MOVQ CX, 24(SP) + MOVQ 112(AX), R10 + MOVQ 128(AX), CX + MOVQ CX, 32(SP) + MOVQ 144(AX), R11 + MOVQ 136(AX), R12 + MOVQ 200(AX), CX + MOVQ CX, 56(SP) + MOVQ 176(AX), CX + MOVQ CX, 48(SP) + MOVQ 184(AX), AX + MOVQ AX, 40(SP) + MOVQ 40(SP), AX + ADDQ AX, 48(SP) + + // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) + ADDQ R10, 32(SP) + + // outBase += outPosition + ADDQ R12, R10 + +sequenceDecs_decodeSync_safe_amd64_main_loop: + MOVQ (SP), R13 + + // Fill bitreader to have enough for the offset and match length. + CMPQ SI, $0x08 + JL sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R13 + MOVQ (R13), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decodeSync_safe_amd64_fill_end + +sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decodeSync_safe_amd64_fill_end + CMPQ BX, $0x07 + JLE sequenceDecs_decodeSync_safe_amd64_fill_end + SHLQ $0x08, DX + SUBQ $0x01, R13 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R13), AX + ORQ AX, DX + JMP sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte + +sequenceDecs_decodeSync_safe_amd64_fill_end: + // Update offset + MOVQ R9, AX + MOVQ BX, CX + MOVQ DX, R14 + SHLQ CL, R14 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decodeSync_safe_amd64_of_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decodeSync_safe_amd64_of_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decodeSync_safe_amd64_of_update_zero + NEGQ CX + SHRQ CL, R14 + ADDQ R14, AX + +sequenceDecs_decodeSync_safe_amd64_of_update_zero: + MOVQ AX, 8(SP) + + // Update match length + MOVQ R8, AX + MOVQ BX, CX + MOVQ DX, R14 + SHLQ CL, R14 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decodeSync_safe_amd64_ml_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decodeSync_safe_amd64_ml_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decodeSync_safe_amd64_ml_update_zero + NEGQ CX + SHRQ CL, R14 + ADDQ R14, AX + +sequenceDecs_decodeSync_safe_amd64_ml_update_zero: + MOVQ AX, 16(SP) + + // Fill bitreader to have enough for the remaining + CMPQ SI, $0x08 + JL sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte + MOVQ BX, AX + SHRQ $0x03, AX + SUBQ AX, R13 + MOVQ (R13), DX + SUBQ AX, SI + ANDQ $0x07, BX + JMP sequenceDecs_decodeSync_safe_amd64_fill_2_end + +sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte: + CMPQ SI, $0x00 + JLE sequenceDecs_decodeSync_safe_amd64_fill_2_end + CMPQ BX, $0x07 + JLE sequenceDecs_decodeSync_safe_amd64_fill_2_end + SHLQ $0x08, DX + SUBQ $0x01, R13 + SUBQ $0x01, SI + SUBQ $0x08, BX + MOVBQZX (R13), AX + ORQ AX, DX + JMP sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte + +sequenceDecs_decodeSync_safe_amd64_fill_2_end: + // Update literal length + MOVQ DI, AX + MOVQ BX, CX + MOVQ DX, R14 + SHLQ CL, R14 + MOVB AH, CL + SHRQ $0x20, AX + TESTQ CX, CX + JZ sequenceDecs_decodeSync_safe_amd64_ll_update_zero + ADDQ CX, BX + CMPQ BX, $0x40 + JA sequenceDecs_decodeSync_safe_amd64_ll_update_zero + CMPQ CX, $0x40 + JAE sequenceDecs_decodeSync_safe_amd64_ll_update_zero + NEGQ CX + SHRQ CL, R14 + ADDQ R14, AX + +sequenceDecs_decodeSync_safe_amd64_ll_update_zero: + MOVQ AX, 24(SP) + + // Fill bitreader for state updates + MOVQ R13, (SP) + MOVQ R9, AX + SHRQ $0x08, AX + MOVBQZX AL, AX + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decodeSync_safe_amd64_skip_update + + // Update Literal Length State + MOVBQZX DI, R13 + SHRQ $0x10, DI + MOVWQZX DI, DI + LEAQ (BX)(R13*1), CX + MOVQ DX, R14 + MOVQ CX, BX + ROLQ CL, R14 + MOVL $0x00000001, R15 + MOVB R13, CL + SHLL CL, R15 + DECL R15 + ANDQ R15, R14 + ADDQ R14, DI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(DI*8), DI + + // Update Match Length State + MOVBQZX R8, R13 + SHRQ $0x10, R8 + MOVWQZX R8, R8 + LEAQ (BX)(R13*1), CX + MOVQ DX, R14 + MOVQ CX, BX + ROLQ CL, R14 + MOVL $0x00000001, R15 + MOVB R13, CL + SHLL CL, R15 + DECL R15 + ANDQ R15, R14 + ADDQ R14, R8 + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Offset State + MOVBQZX R9, R13 + SHRQ $0x10, R9 + MOVWQZX R9, R9 + LEAQ (BX)(R13*1), CX + MOVQ DX, R14 + MOVQ CX, BX + ROLQ CL, R14 + MOVL $0x00000001, R15 + MOVB R13, CL + SHLL CL, R15 + DECL R15 + ANDQ R15, R14 + ADDQ R14, R9 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R9*8), R9 + +sequenceDecs_decodeSync_safe_amd64_skip_update: + // Adjust offset + MOVQ s+0(FP), CX + MOVQ 8(SP), R13 + CMPQ AX, $0x01 + JBE sequenceDecs_decodeSync_safe_amd64_adjust_offsetB_1_or_0 + MOVUPS 144(CX), X0 + MOVQ R13, 144(CX) + MOVUPS X0, 152(CX) + JMP sequenceDecs_decodeSync_safe_amd64_after_adjust + +sequenceDecs_decodeSync_safe_amd64_adjust_offsetB_1_or_0: + CMPQ 24(SP), $0x00000000 + JNE sequenceDecs_decodeSync_safe_amd64_adjust_offset_maybezero + INCQ R13 + JMP sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero + +sequenceDecs_decodeSync_safe_amd64_adjust_offset_maybezero: + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero + MOVQ 144(CX), R13 + JMP sequenceDecs_decodeSync_safe_amd64_after_adjust + +sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero: + MOVQ R13, AX + XORQ R14, R14 + MOVQ $-1, R15 + CMPQ R13, $0x03 + CMOVQEQ R14, AX + CMOVQEQ R15, R14 + ADDQ 144(CX)(AX*8), R14 + JNZ sequenceDecs_decodeSync_safe_amd64_adjust_temp_valid + MOVQ $0x00000001, R14 + +sequenceDecs_decodeSync_safe_amd64_adjust_temp_valid: + CMPQ R13, $0x01 + JZ sequenceDecs_decodeSync_safe_amd64_adjust_skip + MOVQ 152(CX), AX + MOVQ AX, 160(CX) + +sequenceDecs_decodeSync_safe_amd64_adjust_skip: + MOVQ 144(CX), AX + MOVQ AX, 152(CX) + MOVQ R14, 144(CX) + MOVQ R14, R13 + +sequenceDecs_decodeSync_safe_amd64_after_adjust: + MOVQ R13, 8(SP) + + // Check values + MOVQ 16(SP), AX + MOVQ 24(SP), CX + LEAQ (AX)(CX*1), R14 + MOVQ s+0(FP), R15 + ADDQ R14, 256(R15) + MOVQ ctx+16(FP), R14 + SUBQ CX, 104(R14) + JS error_not_enough_literals + CMPQ AX, $0x00020002 + JA sequenceDecs_decodeSync_safe_amd64_error_match_len_too_big + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_safe_amd64_match_len_ofs_ok + TESTQ AX, AX + JNZ sequenceDecs_decodeSync_safe_amd64_error_match_len_ofs_mismatch + +sequenceDecs_decodeSync_safe_amd64_match_len_ofs_ok: + MOVQ 24(SP), AX + MOVQ 8(SP), CX + MOVQ 16(SP), R13 + + // Check if we have enough space in s.out + LEAQ (AX)(R13*1), R14 + ADDQ R10, R14 + CMPQ R14, 32(SP) + JA error_not_enough_space + + // Copy literals + TESTQ AX, AX + JZ check_offset + MOVQ AX, R14 + SUBQ $0x10, R14 + JB copy_1_small + +copy_1_loop: + MOVUPS (R11), X0 + MOVUPS X0, (R10) + ADDQ $0x10, R11 + ADDQ $0x10, R10 + SUBQ $0x10, R14 + JAE copy_1_loop + LEAQ 16(R11)(R14*1), R11 + LEAQ 16(R10)(R14*1), R10 + MOVUPS -16(R11), X0 + MOVUPS X0, -16(R10) + JMP copy_1_end + +copy_1_small: + CMPQ AX, $0x03 + JE copy_1_move_3 + JB copy_1_move_1or2 + CMPQ AX, $0x08 + JB copy_1_move_4through7 + JMP copy_1_move_8through16 + +copy_1_move_1or2: + MOVB (R11), R14 + MOVB -1(R11)(AX*1), R15 + MOVB R14, (R10) + MOVB R15, -1(R10)(AX*1) + ADDQ AX, R11 + ADDQ AX, R10 + JMP copy_1_end + +copy_1_move_3: + MOVW (R11), R14 + MOVB 2(R11), R15 + MOVW R14, (R10) + MOVB R15, 2(R10) + ADDQ AX, R11 + ADDQ AX, R10 + JMP copy_1_end + +copy_1_move_4through7: + MOVL (R11), R14 + MOVL -4(R11)(AX*1), R15 + MOVL R14, (R10) + MOVL R15, -4(R10)(AX*1) + ADDQ AX, R11 + ADDQ AX, R10 + JMP copy_1_end + +copy_1_move_8through16: + MOVQ (R11), R14 + MOVQ -8(R11)(AX*1), R15 + MOVQ R14, (R10) + MOVQ R15, -8(R10)(AX*1) + ADDQ AX, R11 + ADDQ AX, R10 + +copy_1_end: + ADDQ AX, R12 + + // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) +check_offset: + MOVQ R12, AX + ADDQ 40(SP), AX + CMPQ CX, AX + JG error_match_off_too_big + CMPQ CX, 56(SP) + JG error_match_off_too_big + + // Copy match from history + MOVQ CX, AX + SUBQ R12, AX + JLS copy_match + MOVQ 48(SP), R14 + SUBQ AX, R14 + CMPQ R13, AX + JG copy_all_from_history + MOVQ R13, AX + SUBQ $0x10, AX + JB copy_4_small + +copy_4_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R10) + ADDQ $0x10, R14 + ADDQ $0x10, R10 + SUBQ $0x10, AX + JAE copy_4_loop + LEAQ 16(R14)(AX*1), R14 + LEAQ 16(R10)(AX*1), R10 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R10) + JMP copy_4_end + +copy_4_small: + CMPQ R13, $0x03 + JE copy_4_move_3 + CMPQ R13, $0x08 + JB copy_4_move_4through7 + JMP copy_4_move_8through16 + +copy_4_move_3: + MOVW (R14), AX + MOVB 2(R14), CL + MOVW AX, (R10) + MOVB CL, 2(R10) + ADDQ R13, R14 + ADDQ R13, R10 + JMP copy_4_end + +copy_4_move_4through7: + MOVL (R14), AX + MOVL -4(R14)(R13*1), CX + MOVL AX, (R10) + MOVL CX, -4(R10)(R13*1) + ADDQ R13, R14 + ADDQ R13, R10 + JMP copy_4_end + +copy_4_move_8through16: + MOVQ (R14), AX + MOVQ -8(R14)(R13*1), CX + MOVQ AX, (R10) + MOVQ CX, -8(R10)(R13*1) + ADDQ R13, R14 + ADDQ R13, R10 + +copy_4_end: + ADDQ R13, R12 + JMP handle_loop + JMP loop_finished + +copy_all_from_history: + MOVQ AX, R15 + SUBQ $0x10, R15 + JB copy_5_small + +copy_5_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R10) + ADDQ $0x10, R14 + ADDQ $0x10, R10 + SUBQ $0x10, R15 + JAE copy_5_loop + LEAQ 16(R14)(R15*1), R14 + LEAQ 16(R10)(R15*1), R10 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R10) + JMP copy_5_end + +copy_5_small: + CMPQ AX, $0x03 + JE copy_5_move_3 + JB copy_5_move_1or2 + CMPQ AX, $0x08 + JB copy_5_move_4through7 + JMP copy_5_move_8through16 + +copy_5_move_1or2: + MOVB (R14), R15 + MOVB -1(R14)(AX*1), BP + MOVB R15, (R10) + MOVB BP, -1(R10)(AX*1) + ADDQ AX, R14 + ADDQ AX, R10 + JMP copy_5_end + +copy_5_move_3: + MOVW (R14), R15 + MOVB 2(R14), BP + MOVW R15, (R10) + MOVB BP, 2(R10) + ADDQ AX, R14 + ADDQ AX, R10 + JMP copy_5_end + +copy_5_move_4through7: + MOVL (R14), R15 + MOVL -4(R14)(AX*1), BP + MOVL R15, (R10) + MOVL BP, -4(R10)(AX*1) + ADDQ AX, R14 + ADDQ AX, R10 + JMP copy_5_end + +copy_5_move_8through16: + MOVQ (R14), R15 + MOVQ -8(R14)(AX*1), BP + MOVQ R15, (R10) + MOVQ BP, -8(R10)(AX*1) + ADDQ AX, R14 + ADDQ AX, R10 + +copy_5_end: + ADDQ AX, R12 + SUBQ AX, R13 + + // Copy match from the current buffer +copy_match: + MOVQ R10, AX + SUBQ CX, AX + + // ml <= mo + CMPQ R13, CX + JA copy_overlapping_match + + // Copy non-overlapping match + ADDQ R13, R12 + MOVQ R13, CX + SUBQ $0x10, CX + JB copy_2_small + +copy_2_loop: + MOVUPS (AX), X0 + MOVUPS X0, (R10) + ADDQ $0x10, AX + ADDQ $0x10, R10 + SUBQ $0x10, CX + JAE copy_2_loop + LEAQ 16(AX)(CX*1), AX + LEAQ 16(R10)(CX*1), R10 + MOVUPS -16(AX), X0 + MOVUPS X0, -16(R10) + JMP copy_2_end + +copy_2_small: + CMPQ R13, $0x03 + JE copy_2_move_3 + JB copy_2_move_1or2 + CMPQ R13, $0x08 + JB copy_2_move_4through7 + JMP copy_2_move_8through16 + +copy_2_move_1or2: + MOVB (AX), CL + MOVB -1(AX)(R13*1), R14 + MOVB CL, (R10) + MOVB R14, -1(R10)(R13*1) + ADDQ R13, AX + ADDQ R13, R10 + JMP copy_2_end + +copy_2_move_3: + MOVW (AX), CX + MOVB 2(AX), R14 + MOVW CX, (R10) + MOVB R14, 2(R10) + ADDQ R13, AX + ADDQ R13, R10 + JMP copy_2_end + +copy_2_move_4through7: + MOVL (AX), CX + MOVL -4(AX)(R13*1), R14 + MOVL CX, (R10) + MOVL R14, -4(R10)(R13*1) + ADDQ R13, AX + ADDQ R13, R10 + JMP copy_2_end + +copy_2_move_8through16: + MOVQ (AX), CX + MOVQ -8(AX)(R13*1), R14 + MOVQ CX, (R10) + MOVQ R14, -8(R10)(R13*1) + ADDQ R13, AX + ADDQ R13, R10 + +copy_2_end: + JMP handle_loop + + // Copy overlapping match +copy_overlapping_match: + ADDQ R13, R12 + +copy_slow_3: + MOVB (AX), CL + MOVB CL, (R10) + INCQ AX + INCQ R10 + DECQ R13 + JNZ copy_slow_3 + +handle_loop: + MOVQ ctx+16(FP), AX + DECQ 96(AX) + JNS sequenceDecs_decodeSync_safe_amd64_main_loop + +loop_finished: + MOVQ br+8(FP), AX + MOVQ DX, 32(AX) + MOVB BL, 40(AX) + MOVQ SI, 24(AX) + + // Update the context + MOVQ ctx+16(FP), AX + MOVQ R12, 136(AX) + MOVQ 144(AX), CX + SUBQ CX, R11 + MOVQ R11, 168(AX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decodeSync_safe_amd64_error_match_len_ofs_mismatch: + MOVQ 16(SP), AX + MOVQ ctx+16(FP), CX + MOVQ AX, 216(CX) + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decodeSync_safe_amd64_error_match_len_too_big: + MOVQ ctx+16(FP), AX + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error +error_match_off_too_big: + MOVQ ctx+16(FP), AX + MOVQ 8(SP), CX + MOVQ CX, 224(AX) + MOVQ R12, 136(AX) + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error +error_not_enough_space: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ R12, 136(AX) + MOVQ $0x00000005, ret+24(FP) + RET + +// func sequenceDecs_decodeSync_safe_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int +// Requires: BMI, BMI2, CMOV, SSE +TEXT ·sequenceDecs_decodeSync_safe_bmi2(SB), $64-32 + MOVQ br+8(FP), CX + MOVQ 32(CX), AX + MOVBQZX 40(CX), DX + MOVQ 24(CX), BX + MOVQ (CX), CX + ADDQ BX, CX + MOVQ CX, (SP) + MOVQ ctx+16(FP), CX + MOVQ 72(CX), SI + MOVQ 80(CX), DI + MOVQ 88(CX), R8 + XORQ R9, R9 + MOVQ R9, 8(SP) + MOVQ R9, 16(SP) + MOVQ R9, 24(SP) + MOVQ 112(CX), R9 + MOVQ 128(CX), R10 + MOVQ R10, 32(SP) + MOVQ 144(CX), R10 + MOVQ 136(CX), R11 + MOVQ 200(CX), R12 + MOVQ R12, 56(SP) + MOVQ 176(CX), R12 + MOVQ R12, 48(SP) + MOVQ 184(CX), CX + MOVQ CX, 40(SP) + MOVQ 40(SP), CX + ADDQ CX, 48(SP) + + // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) + ADDQ R9, 32(SP) + + // outBase += outPosition + ADDQ R11, R9 + +sequenceDecs_decodeSync_safe_bmi2_main_loop: + MOVQ (SP), R12 + + // Fill bitreader to have enough for the offset and match length. + CMPQ BX, $0x08 + JL sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R12 + MOVQ (R12), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decodeSync_safe_bmi2_fill_end + +sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decodeSync_safe_bmi2_fill_end + CMPQ DX, $0x07 + JLE sequenceDecs_decodeSync_safe_bmi2_fill_end + SHLQ $0x08, AX + SUBQ $0x01, R12 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R12), CX + ORQ CX, AX + JMP sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte + +sequenceDecs_decodeSync_safe_bmi2_fill_end: + // Update offset + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R13 + MOVQ AX, R14 + LEAQ (DX)(R13*1), CX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + MOVQ CX, DX + MOVQ R8, CX + SHRQ $0x20, CX + ADDQ R14, CX + MOVQ CX, 8(SP) + + // Update match length + MOVQ $0x00000808, CX + BEXTRQ CX, DI, R13 + MOVQ AX, R14 + LEAQ (DX)(R13*1), CX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + MOVQ CX, DX + MOVQ DI, CX + SHRQ $0x20, CX + ADDQ R14, CX + MOVQ CX, 16(SP) + + // Fill bitreader to have enough for the remaining + CMPQ BX, $0x08 + JL sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte + MOVQ DX, CX + SHRQ $0x03, CX + SUBQ CX, R12 + MOVQ (R12), AX + SUBQ CX, BX + ANDQ $0x07, DX + JMP sequenceDecs_decodeSync_safe_bmi2_fill_2_end + +sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte: + CMPQ BX, $0x00 + JLE sequenceDecs_decodeSync_safe_bmi2_fill_2_end + CMPQ DX, $0x07 + JLE sequenceDecs_decodeSync_safe_bmi2_fill_2_end + SHLQ $0x08, AX + SUBQ $0x01, R12 + SUBQ $0x01, BX + SUBQ $0x08, DX + MOVBQZX (R12), CX + ORQ CX, AX + JMP sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte + +sequenceDecs_decodeSync_safe_bmi2_fill_2_end: + // Update literal length + MOVQ $0x00000808, CX + BEXTRQ CX, SI, R13 + MOVQ AX, R14 + LEAQ (DX)(R13*1), CX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + MOVQ CX, DX + MOVQ SI, CX + SHRQ $0x20, CX + ADDQ R14, CX + MOVQ CX, 24(SP) + + // Fill bitreader for state updates + MOVQ R12, (SP) + MOVQ $0x00000808, CX + BEXTRQ CX, R8, R12 + MOVQ ctx+16(FP), CX + CMPQ 96(CX), $0x00 + JZ sequenceDecs_decodeSync_safe_bmi2_skip_update + LEAQ (SI)(DI*1), R13 + ADDQ R8, R13 + MOVBQZX R13, R13 + LEAQ (DX)(R13*1), CX + MOVQ AX, R14 + MOVQ CX, DX + ROLQ CL, R14 + BZHIQ R13, R14, R14 + + // Update Offset State + BZHIQ R8, R14, CX + SHRXQ R8, R14, R14 + MOVQ $0x00001010, R13 + BEXTRQ R13, R8, R8 + ADDQ CX, R8 + + // Load ctx.ofTable + MOVQ ctx+16(FP), CX + MOVQ 48(CX), CX + MOVQ (CX)(R8*8), R8 + + // Update Match Length State + BZHIQ DI, R14, CX + SHRXQ DI, R14, R14 + MOVQ $0x00001010, R13 + BEXTRQ R13, DI, DI + ADDQ CX, DI + + // Load ctx.mlTable + MOVQ ctx+16(FP), CX + MOVQ 24(CX), CX + MOVQ (CX)(DI*8), DI + + // Update Literal Length State + BZHIQ SI, R14, CX + MOVQ $0x00001010, R13 + BEXTRQ R13, SI, SI + ADDQ CX, SI + + // Load ctx.llTable + MOVQ ctx+16(FP), CX + MOVQ (CX), CX + MOVQ (CX)(SI*8), SI + +sequenceDecs_decodeSync_safe_bmi2_skip_update: + // Adjust offset + MOVQ s+0(FP), CX + MOVQ 8(SP), R13 + CMPQ R12, $0x01 + JBE sequenceDecs_decodeSync_safe_bmi2_adjust_offsetB_1_or_0 + MOVUPS 144(CX), X0 + MOVQ R13, 144(CX) + MOVUPS X0, 152(CX) + JMP sequenceDecs_decodeSync_safe_bmi2_after_adjust + +sequenceDecs_decodeSync_safe_bmi2_adjust_offsetB_1_or_0: + CMPQ 24(SP), $0x00000000 + JNE sequenceDecs_decodeSync_safe_bmi2_adjust_offset_maybezero + INCQ R13 + JMP sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero + +sequenceDecs_decodeSync_safe_bmi2_adjust_offset_maybezero: + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero + MOVQ 144(CX), R13 + JMP sequenceDecs_decodeSync_safe_bmi2_after_adjust + +sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero: + MOVQ R13, R12 + XORQ R14, R14 + MOVQ $-1, R15 + CMPQ R13, $0x03 + CMOVQEQ R14, R12 + CMOVQEQ R15, R14 + ADDQ 144(CX)(R12*8), R14 + JNZ sequenceDecs_decodeSync_safe_bmi2_adjust_temp_valid + MOVQ $0x00000001, R14 + +sequenceDecs_decodeSync_safe_bmi2_adjust_temp_valid: + CMPQ R13, $0x01 + JZ sequenceDecs_decodeSync_safe_bmi2_adjust_skip + MOVQ 152(CX), R12 + MOVQ R12, 160(CX) + +sequenceDecs_decodeSync_safe_bmi2_adjust_skip: + MOVQ 144(CX), R12 + MOVQ R12, 152(CX) + MOVQ R14, 144(CX) + MOVQ R14, R13 + +sequenceDecs_decodeSync_safe_bmi2_after_adjust: + MOVQ R13, 8(SP) + + // Check values + MOVQ 16(SP), CX + MOVQ 24(SP), R12 + LEAQ (CX)(R12*1), R14 + MOVQ s+0(FP), R15 + ADDQ R14, 256(R15) + MOVQ ctx+16(FP), R14 + SUBQ R12, 104(R14) + JS error_not_enough_literals + CMPQ CX, $0x00020002 + JA sequenceDecs_decodeSync_safe_bmi2_error_match_len_too_big + TESTQ R13, R13 + JNZ sequenceDecs_decodeSync_safe_bmi2_match_len_ofs_ok + TESTQ CX, CX + JNZ sequenceDecs_decodeSync_safe_bmi2_error_match_len_ofs_mismatch + +sequenceDecs_decodeSync_safe_bmi2_match_len_ofs_ok: + MOVQ 24(SP), CX + MOVQ 8(SP), R12 + MOVQ 16(SP), R13 + + // Check if we have enough space in s.out + LEAQ (CX)(R13*1), R14 + ADDQ R9, R14 + CMPQ R14, 32(SP) + JA error_not_enough_space + + // Copy literals + TESTQ CX, CX + JZ check_offset + MOVQ CX, R14 + SUBQ $0x10, R14 + JB copy_1_small + +copy_1_loop: + MOVUPS (R10), X0 + MOVUPS X0, (R9) + ADDQ $0x10, R10 + ADDQ $0x10, R9 + SUBQ $0x10, R14 + JAE copy_1_loop + LEAQ 16(R10)(R14*1), R10 + LEAQ 16(R9)(R14*1), R9 + MOVUPS -16(R10), X0 + MOVUPS X0, -16(R9) + JMP copy_1_end + +copy_1_small: + CMPQ CX, $0x03 + JE copy_1_move_3 + JB copy_1_move_1or2 + CMPQ CX, $0x08 + JB copy_1_move_4through7 + JMP copy_1_move_8through16 + +copy_1_move_1or2: + MOVB (R10), R14 + MOVB -1(R10)(CX*1), R15 + MOVB R14, (R9) + MOVB R15, -1(R9)(CX*1) + ADDQ CX, R10 + ADDQ CX, R9 + JMP copy_1_end + +copy_1_move_3: + MOVW (R10), R14 + MOVB 2(R10), R15 + MOVW R14, (R9) + MOVB R15, 2(R9) + ADDQ CX, R10 + ADDQ CX, R9 + JMP copy_1_end + +copy_1_move_4through7: + MOVL (R10), R14 + MOVL -4(R10)(CX*1), R15 + MOVL R14, (R9) + MOVL R15, -4(R9)(CX*1) + ADDQ CX, R10 + ADDQ CX, R9 + JMP copy_1_end + +copy_1_move_8through16: + MOVQ (R10), R14 + MOVQ -8(R10)(CX*1), R15 + MOVQ R14, (R9) + MOVQ R15, -8(R9)(CX*1) + ADDQ CX, R10 + ADDQ CX, R9 + +copy_1_end: + ADDQ CX, R11 + + // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) +check_offset: + MOVQ R11, CX + ADDQ 40(SP), CX + CMPQ R12, CX + JG error_match_off_too_big + CMPQ R12, 56(SP) + JG error_match_off_too_big + + // Copy match from history + MOVQ R12, CX + SUBQ R11, CX + JLS copy_match + MOVQ 48(SP), R14 + SUBQ CX, R14 + CMPQ R13, CX + JG copy_all_from_history + MOVQ R13, CX + SUBQ $0x10, CX + JB copy_4_small + +copy_4_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R9) + ADDQ $0x10, R14 + ADDQ $0x10, R9 + SUBQ $0x10, CX + JAE copy_4_loop + LEAQ 16(R14)(CX*1), R14 + LEAQ 16(R9)(CX*1), R9 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R9) + JMP copy_4_end + +copy_4_small: + CMPQ R13, $0x03 + JE copy_4_move_3 + CMPQ R13, $0x08 + JB copy_4_move_4through7 + JMP copy_4_move_8through16 + +copy_4_move_3: + MOVW (R14), CX + MOVB 2(R14), R12 + MOVW CX, (R9) + MOVB R12, 2(R9) + ADDQ R13, R14 + ADDQ R13, R9 + JMP copy_4_end + +copy_4_move_4through7: + MOVL (R14), CX + MOVL -4(R14)(R13*1), R12 + MOVL CX, (R9) + MOVL R12, -4(R9)(R13*1) + ADDQ R13, R14 + ADDQ R13, R9 + JMP copy_4_end + +copy_4_move_8through16: + MOVQ (R14), CX + MOVQ -8(R14)(R13*1), R12 + MOVQ CX, (R9) + MOVQ R12, -8(R9)(R13*1) + ADDQ R13, R14 + ADDQ R13, R9 + +copy_4_end: + ADDQ R13, R11 + JMP handle_loop + JMP loop_finished + +copy_all_from_history: + MOVQ CX, R15 + SUBQ $0x10, R15 + JB copy_5_small + +copy_5_loop: + MOVUPS (R14), X0 + MOVUPS X0, (R9) + ADDQ $0x10, R14 + ADDQ $0x10, R9 + SUBQ $0x10, R15 + JAE copy_5_loop + LEAQ 16(R14)(R15*1), R14 + LEAQ 16(R9)(R15*1), R9 + MOVUPS -16(R14), X0 + MOVUPS X0, -16(R9) + JMP copy_5_end + +copy_5_small: + CMPQ CX, $0x03 + JE copy_5_move_3 + JB copy_5_move_1or2 + CMPQ CX, $0x08 + JB copy_5_move_4through7 + JMP copy_5_move_8through16 + +copy_5_move_1or2: + MOVB (R14), R15 + MOVB -1(R14)(CX*1), BP + MOVB R15, (R9) + MOVB BP, -1(R9)(CX*1) + ADDQ CX, R14 + ADDQ CX, R9 + JMP copy_5_end + +copy_5_move_3: + MOVW (R14), R15 + MOVB 2(R14), BP + MOVW R15, (R9) + MOVB BP, 2(R9) + ADDQ CX, R14 + ADDQ CX, R9 + JMP copy_5_end + +copy_5_move_4through7: + MOVL (R14), R15 + MOVL -4(R14)(CX*1), BP + MOVL R15, (R9) + MOVL BP, -4(R9)(CX*1) + ADDQ CX, R14 + ADDQ CX, R9 + JMP copy_5_end + +copy_5_move_8through16: + MOVQ (R14), R15 + MOVQ -8(R14)(CX*1), BP + MOVQ R15, (R9) + MOVQ BP, -8(R9)(CX*1) + ADDQ CX, R14 + ADDQ CX, R9 + +copy_5_end: + ADDQ CX, R11 + SUBQ CX, R13 + + // Copy match from the current buffer +copy_match: + MOVQ R9, CX + SUBQ R12, CX + + // ml <= mo + CMPQ R13, R12 + JA copy_overlapping_match + + // Copy non-overlapping match + ADDQ R13, R11 + MOVQ R13, R12 + SUBQ $0x10, R12 + JB copy_2_small + +copy_2_loop: + MOVUPS (CX), X0 + MOVUPS X0, (R9) + ADDQ $0x10, CX + ADDQ $0x10, R9 + SUBQ $0x10, R12 + JAE copy_2_loop + LEAQ 16(CX)(R12*1), CX + LEAQ 16(R9)(R12*1), R9 + MOVUPS -16(CX), X0 + MOVUPS X0, -16(R9) + JMP copy_2_end + +copy_2_small: + CMPQ R13, $0x03 + JE copy_2_move_3 + JB copy_2_move_1or2 + CMPQ R13, $0x08 + JB copy_2_move_4through7 + JMP copy_2_move_8through16 + +copy_2_move_1or2: + MOVB (CX), R12 + MOVB -1(CX)(R13*1), R14 + MOVB R12, (R9) + MOVB R14, -1(R9)(R13*1) + ADDQ R13, CX + ADDQ R13, R9 + JMP copy_2_end + +copy_2_move_3: + MOVW (CX), R12 + MOVB 2(CX), R14 + MOVW R12, (R9) + MOVB R14, 2(R9) + ADDQ R13, CX + ADDQ R13, R9 + JMP copy_2_end + +copy_2_move_4through7: + MOVL (CX), R12 + MOVL -4(CX)(R13*1), R14 + MOVL R12, (R9) + MOVL R14, -4(R9)(R13*1) + ADDQ R13, CX + ADDQ R13, R9 + JMP copy_2_end + +copy_2_move_8through16: + MOVQ (CX), R12 + MOVQ -8(CX)(R13*1), R14 + MOVQ R12, (R9) + MOVQ R14, -8(R9)(R13*1) + ADDQ R13, CX + ADDQ R13, R9 + +copy_2_end: + JMP handle_loop + + // Copy overlapping match +copy_overlapping_match: + ADDQ R13, R11 + +copy_slow_3: + MOVB (CX), R12 + MOVB R12, (R9) + INCQ CX + INCQ R9 + DECQ R13 + JNZ copy_slow_3 + +handle_loop: + MOVQ ctx+16(FP), CX + DECQ 96(CX) + JNS sequenceDecs_decodeSync_safe_bmi2_main_loop + +loop_finished: + MOVQ br+8(FP), CX + MOVQ AX, 32(CX) + MOVB DL, 40(CX) + MOVQ BX, 24(CX) + + // Update the context + MOVQ ctx+16(FP), AX + MOVQ R11, 136(AX) + MOVQ 144(AX), CX + SUBQ CX, R10 + MOVQ R10, 168(AX) + + // Return success + MOVQ $0x00000000, ret+24(FP) + RET + + // Return with match length error +sequenceDecs_decodeSync_safe_bmi2_error_match_len_ofs_mismatch: + MOVQ 16(SP), AX + MOVQ ctx+16(FP), CX + MOVQ AX, 216(CX) + MOVQ $0x00000001, ret+24(FP) + RET + + // Return with match too long error +sequenceDecs_decodeSync_safe_bmi2_error_match_len_too_big: + MOVQ ctx+16(FP), AX + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ $0x00000002, ret+24(FP) + RET + + // Return with match offset too long error +error_match_off_too_big: + MOVQ ctx+16(FP), AX + MOVQ 8(SP), CX + MOVQ CX, 224(AX) + MOVQ R11, 136(AX) + MOVQ $0x00000003, ret+24(FP) + RET + + // Return with not enough literals error +error_not_enough_literals: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ $0x00000004, ret+24(FP) + RET + + // Return with not enough output space error +error_not_enough_space: + MOVQ ctx+16(FP), AX + MOVQ 24(SP), CX + MOVQ CX, 208(AX) + MOVQ 16(SP), CX + MOVQ CX, 216(AX) + MOVQ R11, 136(AX) + MOVQ $0x00000005, ret+24(FP) + RET diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec_generic.go b/vendor/github.com/klauspost/compress/zstd/seqdec_generic.go new file mode 100644 index 0000000..c3452bc --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqdec_generic.go @@ -0,0 +1,237 @@ +//go:build !amd64 || appengine || !gc || noasm +// +build !amd64 appengine !gc noasm + +package zstd + +import ( + "fmt" + "io" +) + +// decode sequences from the stream with the provided history but without dictionary. +func (s *sequenceDecs) decodeSyncSimple(hist []byte) (bool, error) { + return false, nil +} + +// decode sequences from the stream without the provided history. +func (s *sequenceDecs) decode(seqs []seqVals) error { + br := s.br + + // Grab full sizes tables, to avoid bounds checks. + llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize] + llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state + s.seqSize = 0 + litRemain := len(s.literals) + + maxBlockSize := maxCompressedBlockSize + if s.windowSize < maxBlockSize { + maxBlockSize = s.windowSize + } + for i := range seqs { + var ll, mo, ml int + if br.off > 4+((maxOffsetBits+16+16)>>3) { + // inlined function: + // ll, mo, ml = s.nextFast(br, llState, mlState, ofState) + + // Final will not read from stream. + var llB, mlB, moB uint8 + ll, llB = llState.final() + ml, mlB = mlState.final() + mo, moB = ofState.final() + + // extra bits are stored in reverse order. + br.fillFast() + mo += br.getBits(moB) + if s.maxBits > 32 { + br.fillFast() + } + ml += br.getBits(mlB) + ll += br.getBits(llB) + + if moB > 1 { + s.prevOffset[2] = s.prevOffset[1] + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = mo + } else { + // mo = s.adjustOffset(mo, ll, moB) + // Inlined for rather big speedup + if ll == 0 { + // There is an exception though, when current sequence's literals_length = 0. + // In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2, + // an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte. + mo++ + } + + if mo == 0 { + mo = s.prevOffset[0] + } else { + var temp int + if mo == 3 { + temp = s.prevOffset[0] - 1 + } else { + temp = s.prevOffset[mo] + } + + if temp == 0 { + // 0 is not valid; input is corrupted; force offset to 1 + println("WARNING: temp was 0") + temp = 1 + } + + if mo != 1 { + s.prevOffset[2] = s.prevOffset[1] + } + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = temp + mo = temp + } + } + br.fillFast() + } else { + if br.overread() { + if debugDecoder { + printf("reading sequence %d, exceeded available data\n", i) + } + return io.ErrUnexpectedEOF + } + ll, mo, ml = s.next(br, llState, mlState, ofState) + br.fill() + } + + if debugSequences { + println("Seq", i, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml) + } + // Evaluate. + // We might be doing this async, so do it early. + if mo == 0 && ml > 0 { + return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml) + } + if ml > maxMatchLen { + return fmt.Errorf("match len (%d) bigger than max allowed length", ml) + } + s.seqSize += ll + ml + if s.seqSize > maxBlockSize { + return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize) + } + litRemain -= ll + if litRemain < 0 { + return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, litRemain+ll) + } + seqs[i] = seqVals{ + ll: ll, + ml: ml, + mo: mo, + } + if i == len(seqs)-1 { + // This is the last sequence, so we shouldn't update state. + break + } + + // Manually inlined, ~ 5-20% faster + // Update all 3 states at once. Approx 20% faster. + nBits := llState.nbBits() + mlState.nbBits() + ofState.nbBits() + if nBits == 0 { + llState = llTable[llState.newState()&maxTableMask] + mlState = mlTable[mlState.newState()&maxTableMask] + ofState = ofTable[ofState.newState()&maxTableMask] + } else { + bits := br.get32BitsFast(nBits) + lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31)) + llState = llTable[(llState.newState()+lowBits)&maxTableMask] + + lowBits = uint16(bits >> (ofState.nbBits() & 31)) + lowBits &= bitMask[mlState.nbBits()&15] + mlState = mlTable[(mlState.newState()+lowBits)&maxTableMask] + + lowBits = uint16(bits) & bitMask[ofState.nbBits()&15] + ofState = ofTable[(ofState.newState()+lowBits)&maxTableMask] + } + } + s.seqSize += litRemain + if s.seqSize > maxBlockSize { + return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize) + } + err := br.close() + if err != nil { + printf("Closing sequences: %v, %+v\n", err, *br) + } + return err +} + +// executeSimple handles cases when a dictionary is not used. +func (s *sequenceDecs) executeSimple(seqs []seqVals, hist []byte) error { + // Ensure we have enough output size... + if len(s.out)+s.seqSize > cap(s.out) { + addBytes := s.seqSize + len(s.out) + s.out = append(s.out, make([]byte, addBytes)...) + s.out = s.out[:len(s.out)-addBytes] + } + + if debugDecoder { + printf("Execute %d seqs with literals: %d into %d bytes\n", len(seqs), len(s.literals), s.seqSize) + } + + var t = len(s.out) + out := s.out[:t+s.seqSize] + + for _, seq := range seqs { + // Add literals + copy(out[t:], s.literals[:seq.ll]) + t += seq.ll + s.literals = s.literals[seq.ll:] + + // Malformed input + if seq.mo > t+len(hist) || seq.mo > s.windowSize { + return fmt.Errorf("match offset (%d) bigger than current history (%d)", seq.mo, t+len(hist)) + } + + // Copy from history. + if v := seq.mo - t; v > 0 { + // v is the start position in history from end. + start := len(hist) - v + if seq.ml > v { + // Some goes into the current block. + // Copy remainder of history + copy(out[t:], hist[start:]) + t += v + seq.ml -= v + } else { + copy(out[t:], hist[start:start+seq.ml]) + t += seq.ml + continue + } + } + + // We must be in the current buffer now + if seq.ml > 0 { + start := t - seq.mo + if seq.ml <= t-start { + // No overlap + copy(out[t:], out[start:start+seq.ml]) + t += seq.ml + } else { + // Overlapping copy + // Extend destination slice and copy one byte at the time. + src := out[start : start+seq.ml] + dst := out[t:] + dst = dst[:len(src)] + t += len(src) + // Destination is the space we just added. + for i := range src { + dst[i] = src[i] + } + } + } + } + // Add final literals + copy(out[t:], s.literals) + if debugDecoder { + t += len(s.literals) + if t != len(out) { + panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize)) + } + } + s.out = out + + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/seqenc.go b/vendor/github.com/klauspost/compress/zstd/seqenc.go new file mode 100644 index 0000000..8014174 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqenc.go @@ -0,0 +1,114 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import "math/bits" + +type seqCoders struct { + llEnc, ofEnc, mlEnc *fseEncoder + llPrev, ofPrev, mlPrev *fseEncoder +} + +// swap coders with another (block). +func (s *seqCoders) swap(other *seqCoders) { + *s, *other = *other, *s +} + +// setPrev will update the previous encoders to the actually used ones +// and make sure a fresh one is in the main slot. +func (s *seqCoders) setPrev(ll, ml, of *fseEncoder) { + compareSwap := func(used *fseEncoder, current, prev **fseEncoder) { + // We used the new one, more current to history and reuse the previous history + if *current == used { + *prev, *current = *current, *prev + c := *current + p := *prev + c.reUsed = false + p.reUsed = true + return + } + if used == *prev { + return + } + // Ensure we cannot reuse by accident + prevEnc := *prev + prevEnc.symbolLen = 0 + } + compareSwap(ll, &s.llEnc, &s.llPrev) + compareSwap(ml, &s.mlEnc, &s.mlPrev) + compareSwap(of, &s.ofEnc, &s.ofPrev) +} + +func highBit(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} + +var llCodeTable = [64]byte{0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + 16, 16, 17, 17, 18, 18, 19, 19, + 20, 20, 20, 20, 21, 21, 21, 21, + 22, 22, 22, 22, 22, 22, 22, 22, + 23, 23, 23, 23, 23, 23, 23, 23, + 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24} + +// Up to 6 bits +const maxLLCode = 35 + +// llBitsTable translates from ll code to number of bits. +var llBitsTable = [maxLLCode + 1]byte{ + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 1, 1, 1, 1, 2, 2, 3, 3, + 4, 6, 7, 8, 9, 10, 11, 12, + 13, 14, 15, 16} + +// llCode returns the code that represents the literal length requested. +func llCode(litLength uint32) uint8 { + const llDeltaCode = 19 + if litLength <= 63 { + // Compiler insists on bounds check (Go 1.12) + return llCodeTable[litLength&63] + } + return uint8(highBit(litLength)) + llDeltaCode +} + +var mlCodeTable = [128]byte{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, + 32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37, + 38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39, + 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, + 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, + 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, + 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42} + +// Up to 6 bits +const maxMLCode = 52 + +// mlBitsTable translates from ml code to number of bits. +var mlBitsTable = [maxMLCode + 1]byte{ + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 1, 1, 1, 1, 2, 2, 3, 3, + 4, 4, 5, 7, 8, 9, 10, 11, + 12, 13, 14, 15, 16} + +// note : mlBase = matchLength - MINMATCH; +// because it's the format it's stored in seqStore->sequences +func mlCode(mlBase uint32) uint8 { + const mlDeltaCode = 36 + if mlBase <= 127 { + // Compiler insists on bounds check (Go 1.12) + return mlCodeTable[mlBase&127] + } + return uint8(highBit(mlBase)) + mlDeltaCode +} + +func ofCode(offset uint32) uint8 { + // A valid offset will always be > 0. + return uint8(bits.Len32(offset) - 1) +} diff --git a/vendor/github.com/klauspost/compress/zstd/snappy.go b/vendor/github.com/klauspost/compress/zstd/snappy.go new file mode 100644 index 0000000..9e1baad --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/snappy.go @@ -0,0 +1,435 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "errors" + "hash/crc32" + "io" + + "github.com/klauspost/compress/huff0" + snappy "github.com/klauspost/compress/internal/snapref" +) + +const ( + snappyTagLiteral = 0x00 + snappyTagCopy1 = 0x01 + snappyTagCopy2 = 0x02 + snappyTagCopy4 = 0x03 +) + +const ( + snappyChecksumSize = 4 + snappyMagicBody = "sNaPpY" + + // snappyMaxBlockSize is the maximum size of the input to encodeBlock. It is not + // part of the wire format per se, but some parts of the encoder assume + // that an offset fits into a uint16. + // + // Also, for the framing format (Writer type instead of Encode function), + // https://github.com/google/snappy/blob/master/framing_format.txt says + // that "the uncompressed data in a chunk must be no longer than 65536 + // bytes". + snappyMaxBlockSize = 65536 + + // snappyMaxEncodedLenOfMaxBlockSize equals MaxEncodedLen(snappyMaxBlockSize), but is + // hard coded to be a const instead of a variable, so that obufLen can also + // be a const. Their equivalence is confirmed by + // TestMaxEncodedLenOfMaxBlockSize. + snappyMaxEncodedLenOfMaxBlockSize = 76490 +) + +const ( + chunkTypeCompressedData = 0x00 + chunkTypeUncompressedData = 0x01 + chunkTypePadding = 0xfe + chunkTypeStreamIdentifier = 0xff +) + +var ( + // ErrSnappyCorrupt reports that the input is invalid. + ErrSnappyCorrupt = errors.New("snappy: corrupt input") + // ErrSnappyTooLarge reports that the uncompressed length is too large. + ErrSnappyTooLarge = errors.New("snappy: decoded block is too large") + // ErrSnappyUnsupported reports that the input isn't supported. + ErrSnappyUnsupported = errors.New("snappy: unsupported input") + + errUnsupportedLiteralLength = errors.New("snappy: unsupported literal length") +) + +// SnappyConverter can read SnappyConverter-compressed streams and convert them to zstd. +// Conversion is done by converting the stream directly from Snappy without intermediate +// full decoding. +// Therefore the compression ratio is much less than what can be done by a full decompression +// and compression, and a faulty Snappy stream may lead to a faulty Zstandard stream without +// any errors being generated. +// No CRC value is being generated and not all CRC values of the Snappy stream are checked. +// However, it provides really fast recompression of Snappy streams. +// The converter can be reused to avoid allocations, even after errors. +type SnappyConverter struct { + r io.Reader + err error + buf []byte + block *blockEnc +} + +// Convert the Snappy stream supplied in 'in' and write the zStandard stream to 'w'. +// If any error is detected on the Snappy stream it is returned. +// The number of bytes written is returned. +func (r *SnappyConverter) Convert(in io.Reader, w io.Writer) (int64, error) { + initPredefined() + r.err = nil + r.r = in + if r.block == nil { + r.block = &blockEnc{} + r.block.init() + } + r.block.initNewEncode() + if len(r.buf) != snappyMaxEncodedLenOfMaxBlockSize+snappyChecksumSize { + r.buf = make([]byte, snappyMaxEncodedLenOfMaxBlockSize+snappyChecksumSize) + } + r.block.litEnc.Reuse = huff0.ReusePolicyNone + var written int64 + var readHeader bool + { + var header []byte + var n int + header, r.err = frameHeader{WindowSize: snappyMaxBlockSize}.appendTo(r.buf[:0]) + + n, r.err = w.Write(header) + if r.err != nil { + return written, r.err + } + written += int64(n) + } + + for { + if !r.readFull(r.buf[:4], true) { + // Add empty last block + r.block.reset(nil) + r.block.last = true + err := r.block.encodeLits(r.block.literals, false) + if err != nil { + return written, err + } + n, err := w.Write(r.block.output) + if err != nil { + return written, err + } + written += int64(n) + + return written, r.err + } + chunkType := r.buf[0] + if !readHeader { + if chunkType != chunkTypeStreamIdentifier { + println("chunkType != chunkTypeStreamIdentifier", chunkType) + r.err = ErrSnappyCorrupt + return written, r.err + } + readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + if chunkLen > len(r.buf) { + println("chunkLen > len(r.buf)", chunkType) + r.err = ErrSnappyUnsupported + return written, r.err + } + + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + // Section 4.2. Compressed data (chunk type 0x00). + if chunkLen < snappyChecksumSize { + println("chunkLen < snappyChecksumSize", chunkLen, snappyChecksumSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return written, r.err + } + //checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[snappyChecksumSize:] + + n, hdr, err := snappyDecodedLen(buf) + if err != nil { + r.err = err + return written, r.err + } + buf = buf[hdr:] + if n > snappyMaxBlockSize { + println("n > snappyMaxBlockSize", n, snappyMaxBlockSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + r.block.reset(nil) + r.block.pushOffsets() + if err := decodeSnappy(r.block, buf); err != nil { + r.err = err + return written, r.err + } + if r.block.size+r.block.extraLits != n { + printf("invalid size, want %d, got %d\n", n, r.block.size+r.block.extraLits) + r.err = ErrSnappyCorrupt + return written, r.err + } + err = r.block.encode(nil, false, false) + switch err { + case errIncompressible: + r.block.popOffsets() + r.block.reset(nil) + r.block.literals, err = snappy.Decode(r.block.literals[:n], r.buf[snappyChecksumSize:chunkLen]) + if err != nil { + return written, err + } + err = r.block.encodeLits(r.block.literals, false) + if err != nil { + return written, err + } + case nil: + default: + return written, err + } + + n, r.err = w.Write(r.block.output) + if r.err != nil { + return written, err + } + written += int64(n) + continue + case chunkTypeUncompressedData: + if debugEncoder { + println("Uncompressed, chunklen", chunkLen) + } + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < snappyChecksumSize { + println("chunkLen < snappyChecksumSize", chunkLen, snappyChecksumSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + r.block.reset(nil) + buf := r.buf[:snappyChecksumSize] + if !r.readFull(buf, false) { + return written, r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + // Read directly into r.decoded instead of via r.buf. + n := chunkLen - snappyChecksumSize + if n > snappyMaxBlockSize { + println("n > snappyMaxBlockSize", n, snappyMaxBlockSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + r.block.literals = r.block.literals[:n] + if !r.readFull(r.block.literals, false) { + return written, r.err + } + if snappyCRC(r.block.literals) != checksum { + println("literals crc mismatch") + r.err = ErrSnappyCorrupt + return written, r.err + } + err := r.block.encodeLits(r.block.literals, false) + if err != nil { + return written, err + } + n, r.err = w.Write(r.block.output) + if r.err != nil { + return written, err + } + written += int64(n) + continue + + case chunkTypeStreamIdentifier: + if debugEncoder { + println("stream id", chunkLen, len(snappyMagicBody)) + } + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(snappyMagicBody) { + println("chunkLen != len(snappyMagicBody)", chunkLen, len(snappyMagicBody)) + r.err = ErrSnappyCorrupt + return written, r.err + } + if !r.readFull(r.buf[:len(snappyMagicBody)], false) { + return written, r.err + } + for i := 0; i < len(snappyMagicBody); i++ { + if r.buf[i] != snappyMagicBody[i] { + println("r.buf[i] != snappyMagicBody[i]", r.buf[i], snappyMagicBody[i], i) + r.err = ErrSnappyCorrupt + return written, r.err + } + } + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + println("chunkType <= 0x7f") + r.err = ErrSnappyUnsupported + return written, r.err + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + if !r.readFull(r.buf[:chunkLen], false) { + return written, r.err + } + } +} + +// decodeSnappy writes the decoding of src to dst. It assumes that the varint-encoded +// length of the decompressed bytes has already been read. +func decodeSnappy(blk *blockEnc, src []byte) error { + //decodeRef(make([]byte, snappyMaxBlockSize), src) + var s, length int + lits := blk.extraLits + var offset uint32 + for s < len(src) { + switch src[s] & 0x03 { + case snappyTagLiteral: + x := uint32(src[s] >> 2) + switch { + case x < 60: + s++ + case x == 60: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + x = uint32(src[s-1]) + case x == 61: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + x = uint32(src[s-2]) | uint32(src[s-1])<<8 + case x == 62: + s += 4 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + case x == 63: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + if x > snappyMaxBlockSize { + println("x > snappyMaxBlockSize", x, snappyMaxBlockSize) + return ErrSnappyCorrupt + } + length = int(x) + 1 + if length <= 0 { + println("length <= 0 ", length) + + return errUnsupportedLiteralLength + } + //if length > snappyMaxBlockSize-d || uint32(length) > len(src)-s { + // return ErrSnappyCorrupt + //} + + blk.literals = append(blk.literals, src[s:s+length]...) + //println(length, "litLen") + lits += length + s += length + continue + + case snappyTagCopy1: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, len(src)) + return ErrSnappyCorrupt + } + length = 4 + int(src[s-2])>>2&0x7 + offset = uint32(src[s-2])&0xe0<<3 | uint32(src[s-1]) + + case snappyTagCopy2: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, len(src)) + return ErrSnappyCorrupt + } + length = 1 + int(src[s-3])>>2 + offset = uint32(src[s-2]) | uint32(src[s-1])<<8 + + case snappyTagCopy4: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, len(src)) + return ErrSnappyCorrupt + } + length = 1 + int(src[s-5])>>2 + offset = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + + if offset <= 0 || blk.size+lits < int(offset) /*|| length > len(blk)-d */ { + println("offset <= 0 || blk.size+lits < int(offset)", offset, blk.size+lits, int(offset), blk.size, lits) + + return ErrSnappyCorrupt + } + + // Check if offset is one of the recent offsets. + // Adjusts the output offset accordingly. + // Gives a tiny bit of compression, typically around 1%. + if false { + offset = blk.matchOffset(offset, uint32(lits)) + } else { + offset += 3 + } + + blk.sequences = append(blk.sequences, seq{ + litLen: uint32(lits), + offset: offset, + matchLen: uint32(length) - zstdMinMatch, + }) + blk.size += length + lits + lits = 0 + } + blk.extraLits = lits + return nil +} + +func (r *SnappyConverter) readFull(p []byte, allowEOF bool) (ok bool) { + if _, r.err = io.ReadFull(r.r, p); r.err != nil { + if r.err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) { + r.err = ErrSnappyCorrupt + } + return false + } + return true +} + +var crcTable = crc32.MakeTable(crc32.Castagnoli) + +// crc implements the checksum specified in section 3 of +// https://github.com/google/snappy/blob/master/framing_format.txt +func snappyCRC(b []byte) uint32 { + c := crc32.Update(0, crcTable, b) + return c>>15 | c<<17 + 0xa282ead8 +} + +// snappyDecodedLen returns the length of the decoded block and the number of bytes +// that the length header occupied. +func snappyDecodedLen(src []byte) (blockLen, headerLen int, err error) { + v, n := binary.Uvarint(src) + if n <= 0 || v > 0xffffffff { + return 0, 0, ErrSnappyCorrupt + } + + const wordSize = 32 << (^uint(0) >> 32 & 1) + if wordSize == 32 && v > 0x7fffffff { + return 0, 0, ErrSnappyTooLarge + } + return int(v), n, nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/zip.go b/vendor/github.com/klauspost/compress/zstd/zip.go new file mode 100644 index 0000000..29c15c8 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/zip.go @@ -0,0 +1,141 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. + +package zstd + +import ( + "errors" + "io" + "sync" +) + +// ZipMethodWinZip is the method for Zstandard compressed data inside Zip files for WinZip. +// See https://www.winzip.com/win/en/comp_info.html +const ZipMethodWinZip = 93 + +// ZipMethodPKWare is the original method number used by PKWARE to indicate Zstandard compression. +// Deprecated: This has been deprecated by PKWARE, use ZipMethodWinZip instead for compression. +// See https://pkware.cachefly.net/webdocs/APPNOTE/APPNOTE-6.3.9.TXT +const ZipMethodPKWare = 20 + +// zipReaderPool is the default reader pool. +var zipReaderPool = sync.Pool{New: func() interface{} { + z, err := NewReader(nil, WithDecoderLowmem(true), WithDecoderMaxWindow(128<<20), WithDecoderConcurrency(1)) + if err != nil { + panic(err) + } + return z +}} + +// newZipReader creates a pooled zip decompressor. +func newZipReader(opts ...DOption) func(r io.Reader) io.ReadCloser { + pool := &zipReaderPool + if len(opts) > 0 { + opts = append([]DOption{WithDecoderLowmem(true), WithDecoderMaxWindow(128 << 20)}, opts...) + // Force concurrency 1 + opts = append(opts, WithDecoderConcurrency(1)) + // Create our own pool + pool = &sync.Pool{} + } + return func(r io.Reader) io.ReadCloser { + dec, ok := pool.Get().(*Decoder) + if ok { + dec.Reset(r) + } else { + d, err := NewReader(r, opts...) + if err != nil { + panic(err) + } + dec = d + } + return &pooledZipReader{dec: dec, pool: pool} + } +} + +type pooledZipReader struct { + mu sync.Mutex // guards Close and Read + pool *sync.Pool + dec *Decoder +} + +func (r *pooledZipReader) Read(p []byte) (n int, err error) { + r.mu.Lock() + defer r.mu.Unlock() + if r.dec == nil { + return 0, errors.New("read after close or EOF") + } + dec, err := r.dec.Read(p) + if err == io.EOF { + r.dec.Reset(nil) + r.pool.Put(r.dec) + r.dec = nil + } + return dec, err +} + +func (r *pooledZipReader) Close() error { + r.mu.Lock() + defer r.mu.Unlock() + var err error + if r.dec != nil { + err = r.dec.Reset(nil) + r.pool.Put(r.dec) + r.dec = nil + } + return err +} + +type pooledZipWriter struct { + mu sync.Mutex // guards Close and Read + enc *Encoder + pool *sync.Pool +} + +func (w *pooledZipWriter) Write(p []byte) (n int, err error) { + w.mu.Lock() + defer w.mu.Unlock() + if w.enc == nil { + return 0, errors.New("Write after Close") + } + return w.enc.Write(p) +} + +func (w *pooledZipWriter) Close() error { + w.mu.Lock() + defer w.mu.Unlock() + var err error + if w.enc != nil { + err = w.enc.Close() + w.pool.Put(w.enc) + w.enc = nil + } + return err +} + +// ZipCompressor returns a compressor that can be registered with zip libraries. +// The provided encoder options will be used on all encodes. +func ZipCompressor(opts ...EOption) func(w io.Writer) (io.WriteCloser, error) { + var pool sync.Pool + return func(w io.Writer) (io.WriteCloser, error) { + enc, ok := pool.Get().(*Encoder) + if ok { + enc.Reset(w) + } else { + var err error + enc, err = NewWriter(w, opts...) + if err != nil { + return nil, err + } + } + return &pooledZipWriter{enc: enc, pool: &pool}, nil + } +} + +// ZipDecompressor returns a decompressor that can be registered with zip libraries. +// See ZipCompressor for example. +// Options can be specified. WithDecoderConcurrency(1) is forced, +// and by default a 128MB maximum decompression window is specified. +// The window size can be overridden if required. +func ZipDecompressor(opts ...DOption) func(r io.Reader) io.ReadCloser { + return newZipReader(opts...) +} diff --git a/vendor/github.com/klauspost/compress/zstd/zstd.go b/vendor/github.com/klauspost/compress/zstd/zstd.go new file mode 100644 index 0000000..3eb3f1c --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/zstd.go @@ -0,0 +1,152 @@ +// Package zstd provides decompression of zstandard files. +// +// For advanced usage and examples, go to the README: https://github.com/klauspost/compress/tree/master/zstd#zstd +package zstd + +import ( + "bytes" + "encoding/binary" + "errors" + "log" + "math" + "math/bits" +) + +// enable debug printing +const debug = false + +// enable encoding debug printing +const debugEncoder = debug + +// enable decoding debug printing +const debugDecoder = debug + +// Enable extra assertions. +const debugAsserts = debug || false + +// print sequence details +const debugSequences = false + +// print detailed matching information +const debugMatches = false + +// force encoder to use predefined tables. +const forcePreDef = false + +// zstdMinMatch is the minimum zstd match length. +const zstdMinMatch = 3 + +// Reset the buffer offset when reaching this. +const bufferReset = math.MaxInt32 - MaxWindowSize + +// fcsUnknown is used for unknown frame content size. +const fcsUnknown = math.MaxUint64 + +var ( + // ErrReservedBlockType is returned when a reserved block type is found. + // Typically this indicates wrong or corrupted input. + ErrReservedBlockType = errors.New("invalid input: reserved block type encountered") + + // ErrCompressedSizeTooBig is returned when a block is bigger than allowed. + // Typically this indicates wrong or corrupted input. + ErrCompressedSizeTooBig = errors.New("invalid input: compressed size too big") + + // ErrBlockTooSmall is returned when a block is too small to be decoded. + // Typically returned on invalid input. + ErrBlockTooSmall = errors.New("block too small") + + // ErrUnexpectedBlockSize is returned when a block has unexpected size. + // Typically returned on invalid input. + ErrUnexpectedBlockSize = errors.New("unexpected block size") + + // ErrMagicMismatch is returned when a "magic" number isn't what is expected. + // Typically this indicates wrong or corrupted input. + ErrMagicMismatch = errors.New("invalid input: magic number mismatch") + + // ErrWindowSizeExceeded is returned when a reference exceeds the valid window size. + // Typically this indicates wrong or corrupted input. + ErrWindowSizeExceeded = errors.New("window size exceeded") + + // ErrWindowSizeTooSmall is returned when no window size is specified. + // Typically this indicates wrong or corrupted input. + ErrWindowSizeTooSmall = errors.New("invalid input: window size was too small") + + // ErrDecoderSizeExceeded is returned if decompressed size exceeds the configured limit. + ErrDecoderSizeExceeded = errors.New("decompressed size exceeds configured limit") + + // ErrUnknownDictionary is returned if the dictionary ID is unknown. + // For the time being dictionaries are not supported. + ErrUnknownDictionary = errors.New("unknown dictionary") + + // ErrFrameSizeExceeded is returned if the stated frame size is exceeded. + // This is only returned if SingleSegment is specified on the frame. + ErrFrameSizeExceeded = errors.New("frame size exceeded") + + // ErrFrameSizeMismatch is returned if the stated frame size does not match the expected size. + // This is only returned if SingleSegment is specified on the frame. + ErrFrameSizeMismatch = errors.New("frame size does not match size on stream") + + // ErrCRCMismatch is returned if CRC mismatches. + ErrCRCMismatch = errors.New("CRC check failed") + + // ErrDecoderClosed will be returned if the Decoder was used after + // Close has been called. + ErrDecoderClosed = errors.New("decoder used after Close") + + // ErrDecoderNilInput is returned when a nil Reader was provided + // and an operation other than Reset/DecodeAll/Close was attempted. + ErrDecoderNilInput = errors.New("nil input provided as reader") +) + +func println(a ...interface{}) { + if debug || debugDecoder || debugEncoder { + log.Println(a...) + } +} + +func printf(format string, a ...interface{}) { + if debug || debugDecoder || debugEncoder { + log.Printf(format, a...) + } +} + +// matchLen returns the maximum length. +// a must be the shortest of the two. +// The function also returns whether all bytes matched. +func matchLen(a, b []byte) int { + b = b[:len(a)] + for i := 0; i < len(a)-7; i += 8 { + if diff := load64(a, i) ^ load64(b, i); diff != 0 { + return i + (bits.TrailingZeros64(diff) >> 3) + } + } + + checked := (len(a) >> 3) << 3 + a = a[checked:] + b = b[checked:] + for i := range a { + if a[i] != b[i] { + return i + checked + } + } + return len(a) + checked +} + +func load3232(b []byte, i int32) uint32 { + return binary.LittleEndian.Uint32(b[i:]) +} + +func load6432(b []byte, i int32) uint64 { + return binary.LittleEndian.Uint64(b[i:]) +} + +func load64(b []byte, i int) uint64 { + return binary.LittleEndian.Uint64(b[i:]) +} + +type byter interface { + Bytes() []byte + Len() int +} + +var _ byter = &bytes.Buffer{} diff --git a/vendor/github.com/pierrec/lz4/v4/.gitignore b/vendor/github.com/pierrec/lz4/v4/.gitignore new file mode 100644 index 0000000..5d7e88d --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/.gitignore @@ -0,0 +1,36 @@ +# Created by https://www.gitignore.io/api/macos + +### macOS ### +*.DS_Store +.AppleDouble +.LSOverride + +# Icon must end with two \r +Icon + + +# Thumbnails +._* + +# Files that might appear in the root of a volume +.DocumentRevisions-V100 +.fseventsd +.Spotlight-V100 +.TemporaryItems +.Trashes +.VolumeIcon.icns +.com.apple.timemachine.donotpresent + +# Directories potentially created on remote AFP share +.AppleDB +.AppleDesktop +Network Trash Folder +Temporary Items +.apdisk + +# End of https://www.gitignore.io/api/macos + +cmd/*/*exe +.idea + +fuzz/*.zip diff --git a/vendor/github.com/pierrec/lz4/v4/LICENSE b/vendor/github.com/pierrec/lz4/v4/LICENSE new file mode 100644 index 0000000..bd899d8 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/LICENSE @@ -0,0 +1,28 @@ +Copyright (c) 2015, Pierre Curto +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +* Redistributions of source code must retain the above copyright notice, this + list of conditions and the following disclaimer. + +* Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. + +* Neither the name of xxHash nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + diff --git a/vendor/github.com/pierrec/lz4/v4/README.md b/vendor/github.com/pierrec/lz4/v4/README.md new file mode 100644 index 0000000..4629c9d --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/README.md @@ -0,0 +1,92 @@ +# lz4 : LZ4 compression in pure Go + +[![Go Reference](https://pkg.go.dev/badge/github.com/pierrec/lz4/v4.svg)](https://pkg.go.dev/github.com/pierrec/lz4/v4) +[![CI](https://github.com/pierrec/lz4/workflows/ci/badge.svg)](https://github.com/pierrec/lz4/actions) +[![Go Report Card](https://goreportcard.com/badge/github.com/pierrec/lz4)](https://goreportcard.com/report/github.com/pierrec/lz4) +[![GitHub tag (latest SemVer)](https://img.shields.io/github/tag/pierrec/lz4.svg?style=social)](https://github.com/pierrec/lz4/tags) + +## Overview + +This package provides a streaming interface to [LZ4 data streams](http://fastcompression.blogspot.fr/2013/04/lz4-streaming-format-final.html) as well as low level compress and uncompress functions for LZ4 data blocks. +The implementation is based on the reference C [one](https://github.com/lz4/lz4). + +## Install + +Assuming you have the go toolchain installed: + +``` +go get github.com/pierrec/lz4/v4 +``` + +There is a command line interface tool to compress and decompress LZ4 files. + +``` +go install github.com/pierrec/lz4/v4/cmd/lz4c +``` + +Usage + +``` +Usage of lz4c: + -version + print the program version + +Subcommands: +Compress the given files or from stdin to stdout. +compress [arguments] [ ...] + -bc + enable block checksum + -l int + compression level (0=fastest) + -sc + disable stream checksum + -size string + block max size [64K,256K,1M,4M] (default "4M") + +Uncompress the given files or from stdin to stdout. +uncompress [arguments] [ ...] + +``` + + +## Example + +``` +// Compress and uncompress an input string. +s := "hello world" +r := strings.NewReader(s) + +// The pipe will uncompress the data from the writer. +pr, pw := io.Pipe() +zw := lz4.NewWriter(pw) +zr := lz4.NewReader(pr) + +go func() { + // Compress the input string. + _, _ = io.Copy(zw, r) + _ = zw.Close() // Make sure the writer is closed + _ = pw.Close() // Terminate the pipe +}() + +_, _ = io.Copy(os.Stdout, zr) + +// Output: +// hello world +``` + +## Contributing + +Contributions are very welcome for bug fixing, performance improvements...! + +- Open an issue with a proper description +- Send a pull request with appropriate test case(s) + +## Contributors + +Thanks to all [contributors](https://github.com/pierrec/lz4/graphs/contributors) so far! + +Special thanks to [@Zariel](https://github.com/Zariel) for his asm implementation of the decoder. + +Special thanks to [@greatroar](https://github.com/greatroar) for his work on the asm implementations of the decoder for amd64 and arm64. + +Special thanks to [@klauspost](https://github.com/klauspost) for his work on optimizing the code. diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go new file mode 100644 index 0000000..9054998 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go @@ -0,0 +1,482 @@ +package lz4block + +import ( + "encoding/binary" + "math/bits" + "sync" + + "github.com/pierrec/lz4/v4/internal/lz4errors" +) + +const ( + // The following constants are used to setup the compression algorithm. + minMatch = 4 // the minimum size of the match sequence size (4 bytes) + winSizeLog = 16 // LZ4 64Kb window size limit + winSize = 1 << winSizeLog + winMask = winSize - 1 // 64Kb window of previous data for dependent blocks + + // hashLog determines the size of the hash table used to quickly find a previous match position. + // Its value influences the compression speed and memory usage, the lower the faster, + // but at the expense of the compression ratio. + // 16 seems to be the best compromise for fast compression. + hashLog = 16 + htSize = 1 << hashLog + + mfLimit = 10 + minMatch // The last match cannot start within the last 14 bytes. +) + +func recoverBlock(e *error) { + if r := recover(); r != nil && *e == nil { + *e = lz4errors.ErrInvalidSourceShortBuffer + } +} + +// blockHash hashes the lower five bytes of x into a value < htSize. +func blockHash(x uint64) uint32 { + const prime6bytes = 227718039650203 + x &= 1<<40 - 1 + return uint32((x * prime6bytes) >> (64 - hashLog)) +} + +func CompressBlockBound(n int) int { + return n + n/255 + 16 +} + +func UncompressBlock(src, dst, dict []byte) (int, error) { + if len(src) == 0 { + return 0, nil + } + if di := decodeBlock(dst, src, dict); di >= 0 { + return di, nil + } + return 0, lz4errors.ErrInvalidSourceShortBuffer +} + +type Compressor struct { + // Offsets are at most 64kiB, so we can store only the lower 16 bits of + // match positions: effectively, an offset from some 64kiB block boundary. + // + // When we retrieve such an offset, we interpret it as relative to the last + // block boundary si &^ 0xffff, or the one before, (si &^ 0xffff) - 0x10000, + // depending on which of these is inside the current window. If a table + // entry was generated more than 64kiB back in the input, we find out by + // inspecting the input stream. + table [htSize]uint16 + + // Bitmap indicating which positions in the table are in use. + // This allows us to quickly reset the table for reuse, + // without having to zero everything. + inUse [htSize / 32]uint32 +} + +// Get returns the position of a presumptive match for the hash h. +// The match may be a false positive due to a hash collision or an old entry. +// If si < winSize, the return value may be negative. +func (c *Compressor) get(h uint32, si int) int { + h &= htSize - 1 + i := 0 + if c.inUse[h/32]&(1<<(h%32)) != 0 { + i = int(c.table[h]) + } + i += si &^ winMask + if i >= si { + // Try previous 64kiB block (negative when in first block). + i -= winSize + } + return i +} + +func (c *Compressor) put(h uint32, si int) { + h &= htSize - 1 + c.table[h] = uint16(si) + c.inUse[h/32] |= 1 << (h % 32) +} + +func (c *Compressor) reset() { c.inUse = [htSize / 32]uint32{} } + +var compressorPool = sync.Pool{New: func() interface{} { return new(Compressor) }} + +func CompressBlock(src, dst []byte) (int, error) { + c := compressorPool.Get().(*Compressor) + n, err := c.CompressBlock(src, dst) + compressorPool.Put(c) + return n, err +} + +func (c *Compressor) CompressBlock(src, dst []byte) (int, error) { + // Zero out reused table to avoid non-deterministic output (issue #65). + c.reset() + + // Return 0, nil only if the destination buffer size is < CompressBlockBound. + isNotCompressible := len(dst) < CompressBlockBound(len(src)) + + // adaptSkipLog sets how quickly the compressor begins skipping blocks when data is incompressible. + // This significantly speeds up incompressible data and usually has very small impact on compression. + // bytes to skip = 1 + (bytes since last match >> adaptSkipLog) + const adaptSkipLog = 7 + + // si: Current position of the search. + // anchor: Position of the current literals. + var si, di, anchor int + sn := len(src) - mfLimit + if sn <= 0 { + goto lastLiterals + } + + // Fast scan strategy: the hash table only stores the last five-byte sequences. + for si < sn { + // Hash the next five bytes (sequence)... + match := binary.LittleEndian.Uint64(src[si:]) + h := blockHash(match) + h2 := blockHash(match >> 8) + + // We check a match at s, s+1 and s+2 and pick the first one we get. + // Checking 3 only requires us to load the source one. + ref := c.get(h, si) + ref2 := c.get(h2, si+1) + c.put(h, si) + c.put(h2, si+1) + + offset := si - ref + + if offset <= 0 || offset >= winSize || uint32(match) != binary.LittleEndian.Uint32(src[ref:]) { + // No match. Start calculating another hash. + // The processor can usually do this out-of-order. + h = blockHash(match >> 16) + ref3 := c.get(h, si+2) + + // Check the second match at si+1 + si += 1 + offset = si - ref2 + + if offset <= 0 || offset >= winSize || uint32(match>>8) != binary.LittleEndian.Uint32(src[ref2:]) { + // No match. Check the third match at si+2 + si += 1 + offset = si - ref3 + c.put(h, si) + + if offset <= 0 || offset >= winSize || uint32(match>>16) != binary.LittleEndian.Uint32(src[ref3:]) { + // Skip one extra byte (at si+3) before we check 3 matches again. + si += 2 + (si-anchor)>>adaptSkipLog + continue + } + } + } + + // Match found. + lLen := si - anchor // Literal length. + // We already matched 4 bytes. + mLen := 4 + + // Extend backwards if we can, reducing literals. + tOff := si - offset - 1 + for lLen > 0 && tOff >= 0 && src[si-1] == src[tOff] { + si-- + tOff-- + lLen-- + mLen++ + } + + // Add the match length, so we continue search at the end. + // Use mLen to store the offset base. + si, mLen = si+mLen, si+minMatch + + // Find the longest match by looking by batches of 8 bytes. + for si+8 <= sn { + x := binary.LittleEndian.Uint64(src[si:]) ^ binary.LittleEndian.Uint64(src[si-offset:]) + if x == 0 { + si += 8 + } else { + // Stop is first non-zero byte. + si += bits.TrailingZeros64(x) >> 3 + break + } + } + + mLen = si - mLen + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + if mLen < 0xF { + dst[di] = byte(mLen) + } else { + dst[di] = 0xF + } + + // Encode literals length. + if lLen < 0xF { + dst[di] |= byte(lLen << 4) + } else { + dst[di] |= 0xF0 + di++ + l := lLen - 0xF + for ; l >= 0xFF && di < len(dst); l -= 0xFF { + dst[di] = 0xFF + di++ + } + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + dst[di] = byte(l) + } + di++ + + // Literals. + if di+lLen > len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + copy(dst[di:di+lLen], src[anchor:anchor+lLen]) + di += lLen + 2 + anchor = si + + // Encode offset. + if di > len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + dst[di-2], dst[di-1] = byte(offset), byte(offset>>8) + + // Encode match length part 2. + if mLen >= 0xF { + for mLen -= 0xF; mLen >= 0xFF && di < len(dst); mLen -= 0xFF { + dst[di] = 0xFF + di++ + } + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + dst[di] = byte(mLen) + di++ + } + // Check if we can load next values. + if si >= sn { + break + } + // Hash match end-2 + h = blockHash(binary.LittleEndian.Uint64(src[si-2:])) + c.put(h, si-2) + } + +lastLiterals: + if isNotCompressible && anchor == 0 { + // Incompressible. + return 0, nil + } + + // Last literals. + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + lLen := len(src) - anchor + if lLen < 0xF { + dst[di] = byte(lLen << 4) + } else { + dst[di] = 0xF0 + di++ + for lLen -= 0xF; lLen >= 0xFF && di < len(dst); lLen -= 0xFF { + dst[di] = 0xFF + di++ + } + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + dst[di] = byte(lLen) + } + di++ + + // Write the last literals. + if isNotCompressible && di >= anchor { + // Incompressible. + return 0, nil + } + if di+len(src)-anchor > len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } + di += copy(dst[di:di+len(src)-anchor], src[anchor:]) + return di, nil +} + +// blockHash hashes 4 bytes into a value < winSize. +func blockHashHC(x uint32) uint32 { + const hasher uint32 = 2654435761 // Knuth multiplicative hash. + return x * hasher >> (32 - winSizeLog) +} + +type CompressorHC struct { + // hashTable: stores the last position found for a given hash + // chainTable: stores previous positions for a given hash + hashTable, chainTable [htSize]int + needsReset bool +} + +var compressorHCPool = sync.Pool{New: func() interface{} { return new(CompressorHC) }} + +func CompressBlockHC(src, dst []byte, depth CompressionLevel) (int, error) { + c := compressorHCPool.Get().(*CompressorHC) + n, err := c.CompressBlock(src, dst, depth) + compressorHCPool.Put(c) + return n, err +} + +func (c *CompressorHC) CompressBlock(src, dst []byte, depth CompressionLevel) (_ int, err error) { + if c.needsReset { + // Zero out reused table to avoid non-deterministic output (issue #65). + c.hashTable = [htSize]int{} + c.chainTable = [htSize]int{} + } + c.needsReset = true // Only false on first call. + + defer recoverBlock(&err) + + // Return 0, nil only if the destination buffer size is < CompressBlockBound. + isNotCompressible := len(dst) < CompressBlockBound(len(src)) + + // adaptSkipLog sets how quickly the compressor begins skipping blocks when data is incompressible. + // This significantly speeds up incompressible data and usually has very small impact on compression. + // bytes to skip = 1 + (bytes since last match >> adaptSkipLog) + const adaptSkipLog = 7 + + var si, di, anchor int + sn := len(src) - mfLimit + if sn <= 0 { + goto lastLiterals + } + + if depth == 0 { + depth = winSize + } + + for si < sn { + // Hash the next 4 bytes (sequence). + match := binary.LittleEndian.Uint32(src[si:]) + h := blockHashHC(match) + + // Follow the chain until out of window and give the longest match. + mLen := 0 + offset := 0 + for next, try := c.hashTable[h], depth; try > 0 && next > 0 && si-next < winSize; next, try = c.chainTable[next&winMask], try-1 { + // The first (mLen==0) or next byte (mLen>=minMatch) at current match length + // must match to improve on the match length. + if src[next+mLen] != src[si+mLen] { + continue + } + ml := 0 + // Compare the current position with a previous with the same hash. + for ml < sn-si { + x := binary.LittleEndian.Uint64(src[next+ml:]) ^ binary.LittleEndian.Uint64(src[si+ml:]) + if x == 0 { + ml += 8 + } else { + // Stop is first non-zero byte. + ml += bits.TrailingZeros64(x) >> 3 + break + } + } + if ml < minMatch || ml <= mLen { + // Match too small (>adaptSkipLog + continue + } + + // Match found. + // Update hash/chain tables with overlapping bytes: + // si already hashed, add everything from si+1 up to the match length. + winStart := si + 1 + if ws := si + mLen - winSize; ws > winStart { + winStart = ws + } + for si, ml := winStart, si+mLen; si < ml; { + match >>= 8 + match |= uint32(src[si+3]) << 24 + h := blockHashHC(match) + c.chainTable[si&winMask] = c.hashTable[h] + c.hashTable[h] = si + si++ + } + + lLen := si - anchor + si += mLen + mLen -= minMatch // Match length does not include minMatch. + + if mLen < 0xF { + dst[di] = byte(mLen) + } else { + dst[di] = 0xF + } + + // Encode literals length. + if lLen < 0xF { + dst[di] |= byte(lLen << 4) + } else { + dst[di] |= 0xF0 + di++ + l := lLen - 0xF + for ; l >= 0xFF; l -= 0xFF { + dst[di] = 0xFF + di++ + } + dst[di] = byte(l) + } + di++ + + // Literals. + copy(dst[di:di+lLen], src[anchor:anchor+lLen]) + di += lLen + anchor = si + + // Encode offset. + di += 2 + dst[di-2], dst[di-1] = byte(offset), byte(offset>>8) + + // Encode match length part 2. + if mLen >= 0xF { + for mLen -= 0xF; mLen >= 0xFF; mLen -= 0xFF { + dst[di] = 0xFF + di++ + } + dst[di] = byte(mLen) + di++ + } + } + + if isNotCompressible && anchor == 0 { + // Incompressible. + return 0, nil + } + + // Last literals. +lastLiterals: + lLen := len(src) - anchor + if lLen < 0xF { + dst[di] = byte(lLen << 4) + } else { + dst[di] = 0xF0 + di++ + lLen -= 0xF + for ; lLen >= 0xFF; lLen -= 0xFF { + dst[di] = 0xFF + di++ + } + dst[di] = byte(lLen) + } + di++ + + // Write the last literals. + if isNotCompressible && di >= anchor { + // Incompressible. + return 0, nil + } + di += copy(dst[di:di+len(src)-anchor], src[anchor:]) + return di, nil +} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go new file mode 100644 index 0000000..a1bfa99 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go @@ -0,0 +1,90 @@ +// Package lz4block provides LZ4 BlockSize types and pools of buffers. +package lz4block + +import "sync" + +const ( + Block64Kb uint32 = 1 << (16 + iota*2) + Block256Kb + Block1Mb + Block4Mb +) + +// In legacy mode all blocks are compressed regardless +// of the compressed size: use the bound size. +var Block8Mb = uint32(CompressBlockBound(8 << 20)) + +var ( + BlockPool64K = sync.Pool{New: func() interface{} { return make([]byte, Block64Kb) }} + BlockPool256K = sync.Pool{New: func() interface{} { return make([]byte, Block256Kb) }} + BlockPool1M = sync.Pool{New: func() interface{} { return make([]byte, Block1Mb) }} + BlockPool4M = sync.Pool{New: func() interface{} { return make([]byte, Block4Mb) }} + BlockPool8M = sync.Pool{New: func() interface{} { return make([]byte, Block8Mb) }} +) + +func Index(b uint32) BlockSizeIndex { + switch b { + case Block64Kb: + return 4 + case Block256Kb: + return 5 + case Block1Mb: + return 6 + case Block4Mb: + return 7 + case Block8Mb: // only valid in legacy mode + return 3 + } + return 0 +} + +func IsValid(b uint32) bool { + return Index(b) > 0 +} + +type BlockSizeIndex uint8 + +func (b BlockSizeIndex) IsValid() bool { + switch b { + case 4, 5, 6, 7: + return true + } + return false +} + +func (b BlockSizeIndex) Get() []byte { + var buf interface{} + switch b { + case 4: + buf = BlockPool64K.Get() + case 5: + buf = BlockPool256K.Get() + case 6: + buf = BlockPool1M.Get() + case 7: + buf = BlockPool4M.Get() + case 3: + buf = BlockPool8M.Get() + } + return buf.([]byte) +} + +func Put(buf []byte) { + // Safeguard: do not allow invalid buffers. + switch c := cap(buf); uint32(c) { + case Block64Kb: + BlockPool64K.Put(buf[:c]) + case Block256Kb: + BlockPool256K.Put(buf[:c]) + case Block1Mb: + BlockPool1M.Put(buf[:c]) + case Block4Mb: + BlockPool4M.Put(buf[:c]) + case Block8Mb: + BlockPool8M.Put(buf[:c]) + } +} + +type CompressionLevel uint32 + +const Fast CompressionLevel = 0 diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s new file mode 100644 index 0000000..1d00133 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s @@ -0,0 +1,448 @@ +// +build !appengine +// +build gc +// +build !noasm + +#include "go_asm.h" +#include "textflag.h" + +// AX scratch +// BX scratch +// CX literal and match lengths +// DX token, match offset +// +// DI &dst +// SI &src +// R8 &dst + len(dst) +// R9 &src + len(src) +// R11 &dst +// R12 short output end +// R13 short input end +// R14 &dict +// R15 len(dict) + +// func decodeBlock(dst, src, dict []byte) int +TEXT ·decodeBlock(SB), NOSPLIT, $48-80 + MOVQ dst_base+0(FP), DI + MOVQ DI, R11 + MOVQ dst_len+8(FP), R8 + ADDQ DI, R8 + + MOVQ src_base+24(FP), SI + MOVQ src_len+32(FP), R9 + CMPQ R9, $0 + JE err_corrupt + ADDQ SI, R9 + + MOVQ dict_base+48(FP), R14 + MOVQ dict_len+56(FP), R15 + + // shortcut ends + // short output end + MOVQ R8, R12 + SUBQ $32, R12 + // short input end + MOVQ R9, R13 + SUBQ $16, R13 + + XORL CX, CX + +loop: + // token := uint32(src[si]) + MOVBLZX (SI), DX + INCQ SI + + // lit_len = token >> 4 + // if lit_len > 0 + // CX = lit_len + MOVL DX, CX + SHRL $4, CX + + // if lit_len != 0xF + CMPL CX, $0xF + JEQ lit_len_loop + CMPQ DI, R12 + JAE copy_literal + CMPQ SI, R13 + JAE copy_literal + + // copy shortcut + + // A two-stage shortcut for the most common case: + // 1) If the literal length is 0..14, and there is enough space, + // enter the shortcut and copy 16 bytes on behalf of the literals + // (in the fast mode, only 8 bytes can be safely copied this way). + // 2) Further if the match length is 4..18, copy 18 bytes in a similar + // manner; but we ensure that there's enough space in the output for + // those 18 bytes earlier, upon entering the shortcut (in other words, + // there is a combined check for both stages). + + // copy literal + MOVOU (SI), X0 + MOVOU X0, (DI) + ADDQ CX, DI + ADDQ CX, SI + + MOVL DX, CX + ANDL $0xF, CX + + // The second stage: prepare for match copying, decode full info. + // If it doesn't work out, the info won't be wasted. + // offset := uint16(data[:2]) + MOVWLZX (SI), DX + TESTL DX, DX + JE err_corrupt + ADDQ $2, SI + JC err_short_buf + + MOVQ DI, AX + SUBQ DX, AX + JC err_corrupt + CMPQ AX, DI + JA err_short_buf + + // if we can't do the second stage then jump straight to read the + // match length, we already have the offset. + CMPL CX, $0xF + JEQ match_len_loop_pre + CMPL DX, $8 + JLT match_len_loop_pre + CMPQ AX, R11 + JB match_len_loop_pre + + // memcpy(op + 0, match + 0, 8); + MOVQ (AX), BX + MOVQ BX, (DI) + // memcpy(op + 8, match + 8, 8); + MOVQ 8(AX), BX + MOVQ BX, 8(DI) + // memcpy(op +16, match +16, 2); + MOVW 16(AX), BX + MOVW BX, 16(DI) + + LEAQ const_minMatch(DI)(CX*1), DI + + // shortcut complete, load next token + JMP loopcheck + + // Read the rest of the literal length: + // do { BX = src[si++]; lit_len += BX } while (BX == 0xFF). +lit_len_loop: + CMPQ SI, R9 + JAE err_short_buf + + MOVBLZX (SI), BX + INCQ SI + ADDQ BX, CX + + CMPB BX, $0xFF + JE lit_len_loop + +copy_literal: + // bounds check src and dst + MOVQ SI, AX + ADDQ CX, AX + JC err_short_buf + CMPQ AX, R9 + JA err_short_buf + + MOVQ DI, BX + ADDQ CX, BX + JC err_short_buf + CMPQ BX, R8 + JA err_short_buf + + // Copy literals of <=48 bytes through the XMM registers. + CMPQ CX, $48 + JGT memmove_lit + + // if len(dst[di:]) < 48 + MOVQ R8, AX + SUBQ DI, AX + CMPQ AX, $48 + JLT memmove_lit + + // if len(src[si:]) < 48 + MOVQ R9, BX + SUBQ SI, BX + CMPQ BX, $48 + JLT memmove_lit + + MOVOU (SI), X0 + MOVOU 16(SI), X1 + MOVOU 32(SI), X2 + MOVOU X0, (DI) + MOVOU X1, 16(DI) + MOVOU X2, 32(DI) + + ADDQ CX, SI + ADDQ CX, DI + + JMP finish_lit_copy + +memmove_lit: + // memmove(to, from, len) + MOVQ DI, 0(SP) + MOVQ SI, 8(SP) + MOVQ CX, 16(SP) + + // Spill registers. Increment SI, DI now so we don't need to save CX. + ADDQ CX, DI + ADDQ CX, SI + MOVQ DI, 24(SP) + MOVQ SI, 32(SP) + MOVL DX, 40(SP) + + CALL runtime·memmove(SB) + + // restore registers + MOVQ 24(SP), DI + MOVQ 32(SP), SI + MOVL 40(SP), DX + + // recalc initial values + MOVQ dst_base+0(FP), R8 + MOVQ R8, R11 + ADDQ dst_len+8(FP), R8 + MOVQ src_base+24(FP), R9 + ADDQ src_len+32(FP), R9 + MOVQ dict_base+48(FP), R14 + MOVQ dict_len+56(FP), R15 + MOVQ R8, R12 + SUBQ $32, R12 + MOVQ R9, R13 + SUBQ $16, R13 + +finish_lit_copy: + // CX := mLen + // free up DX to use for offset + MOVL DX, CX + ANDL $0xF, CX + + CMPQ SI, R9 + JAE end + + // offset + // si += 2 + // DX := int(src[si-2]) | int(src[si-1])<<8 + ADDQ $2, SI + JC err_short_buf + CMPQ SI, R9 + JA err_short_buf + MOVWQZX -2(SI), DX + + // 0 offset is invalid + TESTL DX, DX + JEQ err_corrupt + +match_len_loop_pre: + // if mlen != 0xF + CMPB CX, $0xF + JNE copy_match + + // do { BX = src[si++]; mlen += BX } while (BX == 0xFF). +match_len_loop: + CMPQ SI, R9 + JAE err_short_buf + + MOVBLZX (SI), BX + INCQ SI + ADDQ BX, CX + + CMPB BX, $0xFF + JE match_len_loop + +copy_match: + ADDQ $const_minMatch, CX + + // check we have match_len bytes left in dst + // di+match_len < len(dst) + MOVQ DI, AX + ADDQ CX, AX + JC err_short_buf + CMPQ AX, R8 + JA err_short_buf + + // DX = offset + // CX = match_len + // BX = &dst + (di - offset) + MOVQ DI, BX + SUBQ DX, BX + + // check BX is within dst + // if BX < &dst + JC copy_match_from_dict + CMPQ BX, R11 + JBE copy_match_from_dict + + // if offset + match_len < di + LEAQ (BX)(CX*1), AX + CMPQ DI, AX + JA copy_interior_match + + // AX := len(dst[:di]) + // MOVQ DI, AX + // SUBQ R11, AX + + // copy 16 bytes at a time + // if di-offset < 16 copy 16-(di-offset) bytes to di + // then do the remaining + +copy_match_loop: + // for match_len >= 0 + // dst[di] = dst[i] + // di++ + // i++ + MOVB (BX), AX + MOVB AX, (DI) + INCQ DI + INCQ BX + DECQ CX + JNZ copy_match_loop + + JMP loopcheck + +copy_interior_match: + CMPQ CX, $16 + JGT memmove_match + + // if len(dst[di:]) < 16 + MOVQ R8, AX + SUBQ DI, AX + CMPQ AX, $16 + JLT memmove_match + + MOVOU (BX), X0 + MOVOU X0, (DI) + + ADDQ CX, DI + XORL CX, CX + JMP loopcheck + +copy_match_from_dict: + // CX = match_len + // BX = &dst + (di - offset) + + // AX = offset - di = dict_bytes_available => count of bytes potentially covered by the dictionary + MOVQ R11, AX + SUBQ BX, AX + + // BX = len(dict) - dict_bytes_available + MOVQ R15, BX + SUBQ AX, BX + JS err_short_dict + + ADDQ R14, BX + + // if match_len > dict_bytes_available, match fits entirely within external dictionary : just copy + CMPQ CX, AX + JLT memmove_match + + // The match stretches over the dictionary and our block + // 1) copy what comes from the dictionary + // AX = dict_bytes_available = copy_size + // BX = &dict_end - copy_size + // CX = match_len + + // memmove(to, from, len) + MOVQ DI, 0(SP) + MOVQ BX, 8(SP) + MOVQ AX, 16(SP) + // store extra stuff we want to recover + // spill + MOVQ DI, 24(SP) + MOVQ SI, 32(SP) + MOVQ CX, 40(SP) + CALL runtime·memmove(SB) + + // restore registers + MOVQ 16(SP), AX // copy_size + MOVQ 24(SP), DI + MOVQ 32(SP), SI + MOVQ 40(SP), CX // match_len + + // recalc initial values + MOVQ dst_base+0(FP), R8 + MOVQ R8, R11 // TODO: make these sensible numbers + ADDQ dst_len+8(FP), R8 + MOVQ src_base+24(FP), R9 + ADDQ src_len+32(FP), R9 + MOVQ dict_base+48(FP), R14 + MOVQ dict_len+56(FP), R15 + MOVQ R8, R12 + SUBQ $32, R12 + MOVQ R9, R13 + SUBQ $16, R13 + + // di+=copy_size + ADDQ AX, DI + + // 2) copy the rest from the current block + // CX = match_len - copy_size = rest_size + SUBQ AX, CX + MOVQ R11, BX + + // check if we have a copy overlap + // AX = &dst + rest_size + MOVQ CX, AX + ADDQ BX, AX + // if &dst + rest_size > di, copy byte by byte + CMPQ AX, DI + + JA copy_match_loop + +memmove_match: + // memmove(to, from, len) + MOVQ DI, 0(SP) + MOVQ BX, 8(SP) + MOVQ CX, 16(SP) + + // Spill registers. Increment DI now so we don't need to save CX. + ADDQ CX, DI + MOVQ DI, 24(SP) + MOVQ SI, 32(SP) + + CALL runtime·memmove(SB) + + // restore registers + MOVQ 24(SP), DI + MOVQ 32(SP), SI + + // recalc initial values + MOVQ dst_base+0(FP), R8 + MOVQ R8, R11 // TODO: make these sensible numbers + ADDQ dst_len+8(FP), R8 + MOVQ src_base+24(FP), R9 + ADDQ src_len+32(FP), R9 + MOVQ R8, R12 + SUBQ $32, R12 + MOVQ R9, R13 + SUBQ $16, R13 + MOVQ dict_base+48(FP), R14 + MOVQ dict_len+56(FP), R15 + XORL CX, CX + +loopcheck: + // for si < len(src) + CMPQ SI, R9 + JB loop + +end: + // Remaining length must be zero. + TESTQ CX, CX + JNE err_corrupt + + SUBQ R11, DI + MOVQ DI, ret+72(FP) + RET + +err_corrupt: + MOVQ $-1, ret+72(FP) + RET + +err_short_buf: + MOVQ $-2, ret+72(FP) + RET + +err_short_dict: + MOVQ $-3, ret+72(FP) + RET diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s new file mode 100644 index 0000000..20b21fc --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s @@ -0,0 +1,231 @@ +// +build gc +// +build !noasm + +#include "go_asm.h" +#include "textflag.h" + +// Register allocation. +#define dst R0 +#define dstorig R1 +#define src R2 +#define dstend R3 +#define srcend R4 +#define match R5 // Match address. +#define dictend R6 +#define token R7 +#define len R8 // Literal and match lengths. +#define offset R7 // Match offset; overlaps with token. +#define tmp1 R9 +#define tmp2 R11 +#define tmp3 R12 + +// func decodeBlock(dst, src, dict []byte) int +TEXT ·decodeBlock(SB), NOFRAME+NOSPLIT, $-4-40 + MOVW dst_base +0(FP), dst + MOVW dst_len +4(FP), dstend + MOVW src_base +12(FP), src + MOVW src_len +16(FP), srcend + + CMP $0, srcend + BEQ shortSrc + + ADD dst, dstend + ADD src, srcend + + MOVW dst, dstorig + +loop: + // Read token. Extract literal length. + MOVBU.P 1(src), token + MOVW token >> 4, len + CMP $15, len + BNE readLitlenDone + +readLitlenLoop: + CMP src, srcend + BEQ shortSrc + MOVBU.P 1(src), tmp1 + ADD.S tmp1, len + BVS shortDst + CMP $255, tmp1 + BEQ readLitlenLoop + +readLitlenDone: + CMP $0, len + BEQ copyLiteralDone + + // Bounds check dst+len and src+len. + ADD.S dst, len, tmp1 + ADD.CC.S src, len, tmp2 + BCS shortSrc + CMP dstend, tmp1 + //BHI shortDst // Uncomment for distinct error codes. + CMP.LS srcend, tmp2 + BHI shortSrc + + // Copy literal. + CMP $4, len + BLO copyLiteralFinish + + // Copy 0-3 bytes until src is aligned. + TST $1, src + MOVBU.NE.P 1(src), tmp1 + MOVB.NE.P tmp1, 1(dst) + SUB.NE $1, len + + TST $2, src + MOVHU.NE.P 2(src), tmp2 + MOVB.NE.P tmp2, 1(dst) + MOVW.NE tmp2 >> 8, tmp1 + MOVB.NE.P tmp1, 1(dst) + SUB.NE $2, len + + B copyLiteralLoopCond + +copyLiteralLoop: + // Aligned load, unaligned write. + MOVW.P 4(src), tmp1 + MOVW tmp1 >> 8, tmp2 + MOVB tmp2, 1(dst) + MOVW tmp1 >> 16, tmp3 + MOVB tmp3, 2(dst) + MOVW tmp1 >> 24, tmp2 + MOVB tmp2, 3(dst) + MOVB.P tmp1, 4(dst) +copyLiteralLoopCond: + // Loop until len-4 < 0. + SUB.S $4, len + BPL copyLiteralLoop + +copyLiteralFinish: + // Copy remaining 0-3 bytes. + // At this point, len may be < 0, but len&3 is still accurate. + TST $1, len + MOVB.NE.P 1(src), tmp3 + MOVB.NE.P tmp3, 1(dst) + TST $2, len + MOVB.NE.P 2(src), tmp1 + MOVB.NE.P tmp1, 2(dst) + MOVB.NE -1(src), tmp2 + MOVB.NE tmp2, -1(dst) + +copyLiteralDone: + // Initial part of match length. + // This frees up the token register for reuse as offset. + AND $15, token, len + + CMP src, srcend + BEQ end + + // Read offset. + ADD.S $2, src + BCS shortSrc + CMP srcend, src + BHI shortSrc + MOVBU -2(src), offset + MOVBU -1(src), tmp1 + ORR.S tmp1 << 8, offset + BEQ corrupt + + // Read rest of match length. + CMP $15, len + BNE readMatchlenDone + +readMatchlenLoop: + CMP src, srcend + BEQ shortSrc + MOVBU.P 1(src), tmp1 + ADD.S tmp1, len + BVS shortDst + CMP $255, tmp1 + BEQ readMatchlenLoop + +readMatchlenDone: + // Bounds check dst+len+minMatch. + ADD.S dst, len, tmp1 + ADD.CC.S $const_minMatch, tmp1 + BCS shortDst + CMP dstend, tmp1 + BHI shortDst + + RSB dst, offset, match + CMP dstorig, match + BGE copyMatch4 + + // match < dstorig means the match starts in the dictionary, + // at len(dict) - offset + (dst - dstorig). + MOVW dict_base+24(FP), match + MOVW dict_len +28(FP), dictend + + ADD $const_minMatch, len + + RSB dst, dstorig, tmp1 + RSB dictend, offset, tmp2 + ADD.S tmp2, tmp1 + BMI shortDict + ADD match, dictend + ADD tmp1, match + +copyDict: + MOVBU.P 1(match), tmp1 + MOVB.P tmp1, 1(dst) + SUB.S $1, len + CMP.NE match, dictend + BNE copyDict + + // If the match extends beyond the dictionary, the rest is at dstorig. + CMP $0, len + BEQ copyMatchDone + MOVW dstorig, match + B copyMatch + + // Copy a regular match. + // Since len+minMatch is at least four, we can do a 4× unrolled + // byte copy loop. Using MOVW instead of four byte loads is faster, + // but to remain portable we'd have to align match first, which is + // too expensive. By alternating loads and stores, we also handle + // the case offset < 4. +copyMatch4: + SUB.S $4, len + MOVBU.P 4(match), tmp1 + MOVB.P tmp1, 4(dst) + MOVBU -3(match), tmp2 + MOVB tmp2, -3(dst) + MOVBU -2(match), tmp3 + MOVB tmp3, -2(dst) + MOVBU -1(match), tmp1 + MOVB tmp1, -1(dst) + BPL copyMatch4 + + // Restore len, which is now negative. + ADD.S $4, len + BEQ copyMatchDone + +copyMatch: + // Finish with a byte-at-a-time copy. + SUB.S $1, len + MOVBU.P 1(match), tmp2 + MOVB.P tmp2, 1(dst) + BNE copyMatch + +copyMatchDone: + CMP src, srcend + BNE loop + +end: + CMP $0, len + BNE corrupt + SUB dstorig, dst, tmp1 + MOVW tmp1, ret+36(FP) + RET + + // The error cases have distinct labels so we can put different + // return codes here when debugging, or if the error returns need to + // be changed. +shortDict: +shortDst: +shortSrc: +corrupt: + MOVW $-1, tmp1 + MOVW tmp1, ret+36(FP) + RET diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm64.s b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm64.s new file mode 100644 index 0000000..c43e8a8 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm64.s @@ -0,0 +1,230 @@ +// +build gc +// +build !noasm + +// This implementation assumes that strict alignment checking is turned off. +// The Go compiler makes the same assumption. + +#include "go_asm.h" +#include "textflag.h" + +// Register allocation. +#define dst R0 +#define dstorig R1 +#define src R2 +#define dstend R3 +#define dstend16 R4 // dstend - 16 +#define srcend R5 +#define srcend16 R6 // srcend - 16 +#define match R7 // Match address. +#define dict R8 +#define dictlen R9 +#define dictend R10 +#define token R11 +#define len R12 // Literal and match lengths. +#define lenRem R13 +#define offset R14 // Match offset. +#define tmp1 R15 +#define tmp2 R16 +#define tmp3 R17 +#define tmp4 R19 + +// func decodeBlock(dst, src, dict []byte) int +TEXT ·decodeBlock(SB), NOFRAME+NOSPLIT, $0-80 + LDP dst_base+0(FP), (dst, dstend) + ADD dst, dstend + MOVD dst, dstorig + + LDP src_base+24(FP), (src, srcend) + CBZ srcend, shortSrc + ADD src, srcend + + // dstend16 = max(dstend-16, 0) and similarly for srcend16. + SUBS $16, dstend, dstend16 + CSEL LO, ZR, dstend16, dstend16 + SUBS $16, srcend, srcend16 + CSEL LO, ZR, srcend16, srcend16 + + LDP dict_base+48(FP), (dict, dictlen) + ADD dict, dictlen, dictend + +loop: + // Read token. Extract literal length. + MOVBU.P 1(src), token + LSR $4, token, len + CMP $15, len + BNE readLitlenDone + +readLitlenLoop: + CMP src, srcend + BEQ shortSrc + MOVBU.P 1(src), tmp1 + ADDS tmp1, len + BVS shortDst + CMP $255, tmp1 + BEQ readLitlenLoop + +readLitlenDone: + CBZ len, copyLiteralDone + + // Bounds check dst+len and src+len. + ADDS dst, len, tmp1 + BCS shortSrc + ADDS src, len, tmp2 + BCS shortSrc + CMP dstend, tmp1 + BHI shortDst + CMP srcend, tmp2 + BHI shortSrc + + // Copy literal. + SUBS $16, len + BLO copyLiteralShort + +copyLiteralLoop: + LDP.P 16(src), (tmp1, tmp2) + STP.P (tmp1, tmp2), 16(dst) + SUBS $16, len + BPL copyLiteralLoop + + // Copy (final part of) literal of length 0-15. + // If we have >=16 bytes left in src and dst, just copy 16 bytes. +copyLiteralShort: + CMP dstend16, dst + CCMP LO, src, srcend16, $0b0010 // 0010 = preserve carry (LO). + BHS copyLiteralShortEnd + + AND $15, len + + LDP (src), (tmp1, tmp2) + ADD len, src + STP (tmp1, tmp2), (dst) + ADD len, dst + + B copyLiteralDone + + // Safe but slow copy near the end of src, dst. +copyLiteralShortEnd: + TBZ $3, len, 3(PC) + MOVD.P 8(src), tmp1 + MOVD.P tmp1, 8(dst) + TBZ $2, len, 3(PC) + MOVW.P 4(src), tmp2 + MOVW.P tmp2, 4(dst) + TBZ $1, len, 3(PC) + MOVH.P 2(src), tmp3 + MOVH.P tmp3, 2(dst) + TBZ $0, len, 3(PC) + MOVBU.P 1(src), tmp4 + MOVB.P tmp4, 1(dst) + +copyLiteralDone: + // Initial part of match length. + AND $15, token, len + + CMP src, srcend + BEQ end + + // Read offset. + ADDS $2, src + BCS shortSrc + CMP srcend, src + BHI shortSrc + MOVHU -2(src), offset + CBZ offset, corrupt + + // Read rest of match length. + CMP $15, len + BNE readMatchlenDone + +readMatchlenLoop: + CMP src, srcend + BEQ shortSrc + MOVBU.P 1(src), tmp1 + ADDS tmp1, len + BVS shortDst + CMP $255, tmp1 + BEQ readMatchlenLoop + +readMatchlenDone: + ADD $const_minMatch, len + + // Bounds check dst+len. + ADDS dst, len, tmp2 + BCS shortDst + CMP dstend, tmp2 + BHI shortDst + + SUB offset, dst, match + CMP dstorig, match + BHS copyMatchTry8 + + // match < dstorig means the match starts in the dictionary, + // at len(dict) - offset + (dst - dstorig). + SUB dstorig, dst, tmp1 + SUB offset, dictlen, tmp2 + ADDS tmp2, tmp1 + BMI shortDict + ADD dict, tmp1, match + +copyDict: + MOVBU.P 1(match), tmp3 + MOVB.P tmp3, 1(dst) + SUBS $1, len + CCMP NE, dictend, match, $0b0100 // 0100 sets the Z (EQ) flag. + BNE copyDict + + CBZ len, copyMatchDone + + // If the match extends beyond the dictionary, the rest is at dstorig. + // Recompute the offset for the next check. + MOVD dstorig, match + SUB dstorig, dst, offset + +copyMatchTry8: + // Copy doublewords if both len and offset are at least eight. + // A 16-at-a-time loop doesn't provide a further speedup. + CMP $8, len + CCMP HS, offset, $8, $0 + BLO copyMatchLoop1 + + AND $7, len, lenRem + SUB $8, len +copyMatchLoop8: + MOVD.P 8(match), tmp1 + MOVD.P tmp1, 8(dst) + SUBS $8, len + BPL copyMatchLoop8 + + MOVD (match)(len), tmp2 // match+len == match+lenRem-8. + ADD lenRem, dst + MOVD $0, len + MOVD tmp2, -8(dst) + B copyMatchDone + +copyMatchLoop1: + // Byte-at-a-time copy for small offsets. + MOVBU.P 1(match), tmp2 + MOVB.P tmp2, 1(dst) + SUBS $1, len + BNE copyMatchLoop1 + +copyMatchDone: + CMP src, srcend + BNE loop + +end: + CBNZ len, corrupt + SUB dstorig, dst, tmp1 + MOVD tmp1, ret+72(FP) + RET + + // The error cases have distinct labels so we can put different + // return codes here when debugging, or if the error returns need to + // be changed. +shortDict: +shortDst: +shortSrc: +corrupt: + MOVD $-1, tmp1 + MOVD tmp1, ret+72(FP) + RET diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_asm.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_asm.go new file mode 100644 index 0000000..8d9023d --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_asm.go @@ -0,0 +1,10 @@ +//go:build (amd64 || arm || arm64) && !appengine && gc && !noasm +// +build amd64 arm arm64 +// +build !appengine +// +build gc +// +build !noasm + +package lz4block + +//go:noescape +func decodeBlock(dst, src, dict []byte) int diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go new file mode 100644 index 0000000..2010cd7 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go @@ -0,0 +1,136 @@ +//go:build (!amd64 && !arm && !arm64) || appengine || !gc || noasm +// +build !amd64,!arm,!arm64 appengine !gc noasm + +package lz4block + +import ( + "encoding/binary" +) + +func decodeBlock(dst, src, dict []byte) (ret int) { + // Restrict capacities so we don't read or write out of bounds. + dst = dst[:len(dst):len(dst)] + src = src[:len(src):len(src)] + + const hasError = -2 + + if len(src) == 0 { + return hasError + } + + defer func() { + if recover() != nil { + ret = hasError + } + }() + + var si, di uint + for si < uint(len(src)) { + // Literals and match lengths (token). + b := uint(src[si]) + si++ + + // Literals. + if lLen := b >> 4; lLen > 0 { + switch { + case lLen < 0xF && si+16 < uint(len(src)): + // Shortcut 1 + // if we have enough room in src and dst, and the literals length + // is small enough (0..14) then copy all 16 bytes, even if not all + // are part of the literals. + copy(dst[di:], src[si:si+16]) + si += lLen + di += lLen + if mLen := b & 0xF; mLen < 0xF { + // Shortcut 2 + // if the match length (4..18) fits within the literals, then copy + // all 18 bytes, even if not all are part of the literals. + mLen += 4 + if offset := u16(src[si:]); mLen <= offset && offset < di { + i := di - offset + end := i + 18 + copy(dst[di:], dst[i:end]) + si += 2 + di += mLen + continue + } + } + case lLen == 0xF: + for { + x := uint(src[si]) + if lLen += x; int(lLen) < 0 { + return hasError + } + si++ + if x != 0xFF { + break + } + } + fallthrough + default: + copy(dst[di:di+lLen], src[si:si+lLen]) + si += lLen + di += lLen + } + } + + mLen := b & 0xF + if si == uint(len(src)) && mLen == 0 { + break + } else if si >= uint(len(src)) { + return hasError + } + + offset := u16(src[si:]) + if offset == 0 { + return hasError + } + si += 2 + + // Match. + mLen += minMatch + if mLen == minMatch+0xF { + for { + x := uint(src[si]) + if mLen += x; int(mLen) < 0 { + return hasError + } + si++ + if x != 0xFF { + break + } + } + } + + // Copy the match. + if di < offset { + // The match is beyond our block, meaning the first part + // is in the dictionary. + fromDict := dict[uint(len(dict))+di-offset:] + n := uint(copy(dst[di:di+mLen], fromDict)) + di += n + if mLen -= n; mLen == 0 { + continue + } + // We copied n = offset-di bytes from the dictionary, + // then set di = di+n = offset, so the following code + // copies from dst[di-offset:] = dst[0:]. + } + + expanded := dst[di-offset:] + if mLen > offset { + // Efficiently copy the match dst[di-offset:di] into the dst slice. + bytesToCopy := offset * (mLen / offset) + for n := offset; n <= bytesToCopy+offset; n *= 2 { + copy(expanded[n:], expanded[:n]) + } + di += bytesToCopy + mLen -= bytesToCopy + } + di += uint(copy(dst[di:di+mLen], expanded[:mLen])) + } + + return int(di) +} + +func u16(p []byte) uint { return uint(binary.LittleEndian.Uint16(p)) } diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4errors/errors.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4errors/errors.go new file mode 100644 index 0000000..710ea42 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4errors/errors.go @@ -0,0 +1,19 @@ +package lz4errors + +type Error string + +func (e Error) Error() string { return string(e) } + +const ( + ErrInvalidSourceShortBuffer Error = "lz4: invalid source or destination buffer too short" + ErrInvalidFrame Error = "lz4: bad magic number" + ErrInternalUnhandledState Error = "lz4: unhandled state" + ErrInvalidHeaderChecksum Error = "lz4: invalid header checksum" + ErrInvalidBlockChecksum Error = "lz4: invalid block checksum" + ErrInvalidFrameChecksum Error = "lz4: invalid frame checksum" + ErrOptionInvalidCompressionLevel Error = "lz4: invalid compression level" + ErrOptionClosedOrError Error = "lz4: cannot apply options on closed or in error object" + ErrOptionInvalidBlockSize Error = "lz4: invalid block size" + ErrOptionNotApplicable Error = "lz4: option not applicable" + ErrWriterNotClosed Error = "lz4: writer not closed" +) diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go new file mode 100644 index 0000000..459086f --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go @@ -0,0 +1,350 @@ +package lz4stream + +import ( + "encoding/binary" + "fmt" + "io" + "sync" + + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" + "github.com/pierrec/lz4/v4/internal/xxh32" +) + +type Blocks struct { + Block *FrameDataBlock + Blocks chan chan *FrameDataBlock + mu sync.Mutex + err error +} + +func (b *Blocks) initW(f *Frame, dst io.Writer, num int) { + if num == 1 { + b.Blocks = nil + b.Block = NewFrameDataBlock(f) + return + } + b.Block = nil + if cap(b.Blocks) != num { + b.Blocks = make(chan chan *FrameDataBlock, num) + } + // goroutine managing concurrent block compression goroutines. + go func() { + // Process next block compression item. + for c := range b.Blocks { + // Read the next compressed block result. + // Waiting here ensures that the blocks are output in the order they were sent. + // The incoming channel is always closed as it indicates to the caller that + // the block has been processed. + block := <-c + if block == nil { + // Notify the block compression routine that we are done with its result. + // This is used when a sentinel block is sent to terminate the compression. + close(c) + return + } + // Do not attempt to write the block upon any previous failure. + if b.err == nil { + // Write the block. + if err := block.Write(f, dst); err != nil { + // Keep the first error. + b.err = err + // All pending compression goroutines need to shut down, so we need to keep going. + } + } + close(c) + } + }() +} + +func (b *Blocks) close(f *Frame, num int) error { + if num == 1 { + if b.Block != nil { + b.Block.Close(f) + } + err := b.err + b.err = nil + return err + } + if b.Blocks == nil { + err := b.err + b.err = nil + return err + } + c := make(chan *FrameDataBlock) + b.Blocks <- c + c <- nil + <-c + err := b.err + b.err = nil + return err +} + +// ErrorR returns any error set while uncompressing a stream. +func (b *Blocks) ErrorR() error { + b.mu.Lock() + defer b.mu.Unlock() + return b.err +} + +// initR returns a channel that streams the uncompressed blocks if in concurrent +// mode and no error. When the channel is closed, check for any error with b.ErrorR. +// +// If not in concurrent mode, the uncompressed block is b.Block and the returned error +// needs to be checked. +func (b *Blocks) initR(f *Frame, num int, src io.Reader) (chan []byte, error) { + size := f.Descriptor.Flags.BlockSizeIndex() + if num == 1 { + b.Blocks = nil + b.Block = NewFrameDataBlock(f) + return nil, nil + } + b.Block = nil + blocks := make(chan chan []byte, num) + // data receives the uncompressed blocks. + data := make(chan []byte) + // Read blocks from the source sequentially + // and uncompress them concurrently. + + // In legacy mode, accrue the uncompress sizes in cum. + var cum uint32 + go func() { + var cumx uint32 + var err error + for b.ErrorR() == nil { + block := NewFrameDataBlock(f) + cumx, err = block.Read(f, src, 0) + if err != nil { + block.Close(f) + break + } + // Recheck for an error as reading may be slow and uncompressing is expensive. + if b.ErrorR() != nil { + block.Close(f) + break + } + c := make(chan []byte) + blocks <- c + go func() { + defer block.Close(f) + data, err := block.Uncompress(f, size.Get(), nil, false) + if err != nil { + b.closeR(err) + // Close the block channel to indicate an error. + close(c) + } else { + c <- data + } + }() + } + // End the collection loop and the data channel. + c := make(chan []byte) + blocks <- c + c <- nil // signal the collection loop that we are done + <-c // wait for the collect loop to complete + if f.isLegacy() && cum == cumx { + err = io.EOF + } + b.closeR(err) + close(data) + }() + // Collect the uncompressed blocks and make them available + // on the returned channel. + go func(leg bool) { + defer close(blocks) + skipBlocks := false + for c := range blocks { + buf, ok := <-c + if !ok { + // A closed channel indicates an error. + // All remaining channels should be discarded. + skipBlocks = true + continue + } + if buf == nil { + // Signal to end the loop. + close(c) + return + } + if skipBlocks { + // A previous error has occurred, skipping remaining channels. + continue + } + // Perform checksum now as the blocks are received in order. + if f.Descriptor.Flags.ContentChecksum() { + _, _ = f.checksum.Write(buf) + } + if leg { + cum += uint32(len(buf)) + } + data <- buf + close(c) + } + }(f.isLegacy()) + return data, nil +} + +// closeR safely sets the error on b if not already set. +func (b *Blocks) closeR(err error) { + b.mu.Lock() + if b.err == nil { + b.err = err + } + b.mu.Unlock() +} + +func NewFrameDataBlock(f *Frame) *FrameDataBlock { + buf := f.Descriptor.Flags.BlockSizeIndex().Get() + return &FrameDataBlock{Data: buf, data: buf} +} + +type FrameDataBlock struct { + Size DataBlockSize + Data []byte // compressed or uncompressed data (.data or .src) + Checksum uint32 + data []byte // buffer for compressed data + src []byte // uncompressed data + err error // used in concurrent mode +} + +func (b *FrameDataBlock) Close(f *Frame) { + b.Size = 0 + b.Checksum = 0 + b.err = nil + if b.data != nil { + // Block was not already closed. + lz4block.Put(b.data) + b.Data = nil + b.data = nil + b.src = nil + } +} + +// Block compression errors are ignored since the buffer is sized appropriately. +func (b *FrameDataBlock) Compress(f *Frame, src []byte, level lz4block.CompressionLevel) *FrameDataBlock { + data := b.data + if f.isLegacy() { + // In legacy mode, the buffer is sized according to CompressBlockBound, + // but only 8Mb is buffered for compression. + src = src[:8<<20] + } else { + data = data[:len(src)] // trigger the incompressible flag in CompressBlock + } + var n int + switch level { + case lz4block.Fast: + n, _ = lz4block.CompressBlock(src, data) + default: + n, _ = lz4block.CompressBlockHC(src, data, level) + } + if n == 0 { + b.Size.UncompressedSet(true) + b.Data = src + } else { + b.Size.UncompressedSet(false) + b.Data = data[:n] + } + b.Size.sizeSet(len(b.Data)) + b.src = src // keep track of the source for content checksum + + if f.Descriptor.Flags.BlockChecksum() { + b.Checksum = xxh32.ChecksumZero(src) + } + return b +} + +func (b *FrameDataBlock) Write(f *Frame, dst io.Writer) error { + // Write is called in the same order as blocks are compressed, + // so content checksum must be done here. + if f.Descriptor.Flags.ContentChecksum() { + _, _ = f.checksum.Write(b.src) + } + buf := f.buf[:] + binary.LittleEndian.PutUint32(buf, uint32(b.Size)) + if _, err := dst.Write(buf[:4]); err != nil { + return err + } + + if _, err := dst.Write(b.Data); err != nil { + return err + } + + if b.Checksum == 0 { + return nil + } + binary.LittleEndian.PutUint32(buf, b.Checksum) + _, err := dst.Write(buf[:4]) + return err +} + +// Read updates b with the next block data, size and checksum if available. +func (b *FrameDataBlock) Read(f *Frame, src io.Reader, cum uint32) (uint32, error) { + x, err := f.readUint32(src) + if err != nil { + return 0, err + } + if f.isLegacy() { + switch x { + case frameMagicLegacy: + // Concatenated legacy frame. + return b.Read(f, src, cum) + case cum: + // Only works in non concurrent mode, for concurrent mode + // it is handled separately. + // Linux kernel format appends the total uncompressed size at the end. + return 0, io.EOF + } + } else if x == 0 { + // Marker for end of stream. + return 0, io.EOF + } + b.Size = DataBlockSize(x) + + size := b.Size.size() + if size > cap(b.data) { + return x, lz4errors.ErrOptionInvalidBlockSize + } + b.data = b.data[:size] + if _, err := io.ReadFull(src, b.data); err != nil { + return x, err + } + if f.Descriptor.Flags.BlockChecksum() { + sum, err := f.readUint32(src) + if err != nil { + return 0, err + } + b.Checksum = sum + } + return x, nil +} + +func (b *FrameDataBlock) Uncompress(f *Frame, dst, dict []byte, sum bool) ([]byte, error) { + if b.Size.Uncompressed() { + n := copy(dst, b.data) + dst = dst[:n] + } else { + n, err := lz4block.UncompressBlock(b.data, dst, dict) + if err != nil { + return nil, err + } + dst = dst[:n] + } + if f.Descriptor.Flags.BlockChecksum() { + if c := xxh32.ChecksumZero(dst); c != b.Checksum { + err := fmt.Errorf("%w: got %x; expected %x", lz4errors.ErrInvalidBlockChecksum, c, b.Checksum) + return nil, err + } + } + if sum && f.Descriptor.Flags.ContentChecksum() { + _, _ = f.checksum.Write(dst) + } + return dst, nil +} + +func (f *Frame) readUint32(r io.Reader) (x uint32, err error) { + if _, err = io.ReadFull(r, f.buf[:4]); err != nil { + return + } + x = binary.LittleEndian.Uint32(f.buf[:4]) + return +} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go new file mode 100644 index 0000000..18192a9 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go @@ -0,0 +1,204 @@ +// Package lz4stream provides the types that support reading and writing LZ4 data streams. +package lz4stream + +import ( + "encoding/binary" + "fmt" + "io" + "io/ioutil" + + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" + "github.com/pierrec/lz4/v4/internal/xxh32" +) + +//go:generate go run gen.go + +const ( + frameMagic uint32 = 0x184D2204 + frameSkipMagic uint32 = 0x184D2A50 + frameMagicLegacy uint32 = 0x184C2102 +) + +func NewFrame() *Frame { + return &Frame{} +} + +type Frame struct { + buf [15]byte // frame descriptor needs at most 4(magic)+4+8+1=11 bytes + Magic uint32 + Descriptor FrameDescriptor + Blocks Blocks + Checksum uint32 + checksum xxh32.XXHZero +} + +// Reset allows reusing the Frame. +// The Descriptor configuration is not modified. +func (f *Frame) Reset(num int) { + f.Magic = 0 + f.Descriptor.Checksum = 0 + f.Descriptor.ContentSize = 0 + _ = f.Blocks.close(f, num) + f.Checksum = 0 +} + +func (f *Frame) InitW(dst io.Writer, num int, legacy bool) { + if legacy { + f.Magic = frameMagicLegacy + idx := lz4block.Index(lz4block.Block8Mb) + f.Descriptor.Flags.BlockSizeIndexSet(idx) + } else { + f.Magic = frameMagic + f.Descriptor.initW() + } + f.Blocks.initW(f, dst, num) + f.checksum.Reset() +} + +func (f *Frame) CloseW(dst io.Writer, num int) error { + if err := f.Blocks.close(f, num); err != nil { + return err + } + if f.isLegacy() { + return nil + } + buf := f.buf[:0] + // End mark (data block size of uint32(0)). + buf = append(buf, 0, 0, 0, 0) + if f.Descriptor.Flags.ContentChecksum() { + buf = f.checksum.Sum(buf) + } + _, err := dst.Write(buf) + return err +} + +func (f *Frame) isLegacy() bool { + return f.Magic == frameMagicLegacy +} + +func (f *Frame) ParseHeaders(src io.Reader) error { + if f.Magic > 0 { + // Header already read. + return nil + } + +newFrame: + var err error + if f.Magic, err = f.readUint32(src); err != nil { + return err + } + switch m := f.Magic; { + case m == frameMagic || m == frameMagicLegacy: + // All 16 values of frameSkipMagic are valid. + case m>>8 == frameSkipMagic>>8: + skip, err := f.readUint32(src) + if err != nil { + return err + } + if _, err := io.CopyN(ioutil.Discard, src, int64(skip)); err != nil { + return err + } + goto newFrame + default: + return lz4errors.ErrInvalidFrame + } + if err := f.Descriptor.initR(f, src); err != nil { + return err + } + f.checksum.Reset() + return nil +} + +func (f *Frame) InitR(src io.Reader, num int) (chan []byte, error) { + return f.Blocks.initR(f, num, src) +} + +func (f *Frame) CloseR(src io.Reader) (err error) { + if f.isLegacy() { + return nil + } + if !f.Descriptor.Flags.ContentChecksum() { + return nil + } + if f.Checksum, err = f.readUint32(src); err != nil { + return err + } + if c := f.checksum.Sum32(); c != f.Checksum { + return fmt.Errorf("%w: got %x; expected %x", lz4errors.ErrInvalidFrameChecksum, c, f.Checksum) + } + return nil +} + +type FrameDescriptor struct { + Flags DescriptorFlags + ContentSize uint64 + Checksum uint8 +} + +func (fd *FrameDescriptor) initW() { + fd.Flags.VersionSet(1) + fd.Flags.BlockIndependenceSet(true) +} + +func (fd *FrameDescriptor) Write(f *Frame, dst io.Writer) error { + if fd.Checksum > 0 { + // Header already written. + return nil + } + + buf := f.buf[:4] + // Write the magic number here even though it belongs to the Frame. + binary.LittleEndian.PutUint32(buf, f.Magic) + if !f.isLegacy() { + buf = buf[:4+2] + binary.LittleEndian.PutUint16(buf[4:], uint16(fd.Flags)) + + if fd.Flags.Size() { + buf = buf[:4+2+8] + binary.LittleEndian.PutUint64(buf[4+2:], fd.ContentSize) + } + fd.Checksum = descriptorChecksum(buf[4:]) + buf = append(buf, fd.Checksum) + } + + _, err := dst.Write(buf) + return err +} + +func (fd *FrameDescriptor) initR(f *Frame, src io.Reader) error { + if f.isLegacy() { + idx := lz4block.Index(lz4block.Block8Mb) + f.Descriptor.Flags.BlockSizeIndexSet(idx) + return nil + } + // Read the flags and the checksum, hoping that there is not content size. + buf := f.buf[:3] + if _, err := io.ReadFull(src, buf); err != nil { + return err + } + descr := binary.LittleEndian.Uint16(buf) + fd.Flags = DescriptorFlags(descr) + if fd.Flags.Size() { + // Append the 8 missing bytes. + buf = buf[:3+8] + if _, err := io.ReadFull(src, buf[3:]); err != nil { + return err + } + fd.ContentSize = binary.LittleEndian.Uint64(buf[2:]) + } + fd.Checksum = buf[len(buf)-1] // the checksum is the last byte + buf = buf[:len(buf)-1] // all descriptor fields except checksum + if c := descriptorChecksum(buf); fd.Checksum != c { + return fmt.Errorf("%w: got %x; expected %x", lz4errors.ErrInvalidHeaderChecksum, c, fd.Checksum) + } + // Validate the elements that can be. + if idx := fd.Flags.BlockSizeIndex(); !idx.IsValid() { + return lz4errors.ErrOptionInvalidBlockSize + } + return nil +} + +func descriptorChecksum(buf []byte) byte { + return byte(xxh32.ChecksumZero(buf) >> 8) +} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame_gen.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame_gen.go new file mode 100644 index 0000000..d33a6be --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame_gen.go @@ -0,0 +1,103 @@ +// Code generated by `gen.exe`. DO NOT EDIT. + +package lz4stream + +import "github.com/pierrec/lz4/v4/internal/lz4block" + +// DescriptorFlags is defined as follow: +// field bits +// ----- ---- +// _ 2 +// ContentChecksum 1 +// Size 1 +// BlockChecksum 1 +// BlockIndependence 1 +// Version 2 +// _ 4 +// BlockSizeIndex 3 +// _ 1 +type DescriptorFlags uint16 + +// Getters. +func (x DescriptorFlags) ContentChecksum() bool { return x>>2&1 != 0 } +func (x DescriptorFlags) Size() bool { return x>>3&1 != 0 } +func (x DescriptorFlags) BlockChecksum() bool { return x>>4&1 != 0 } +func (x DescriptorFlags) BlockIndependence() bool { return x>>5&1 != 0 } +func (x DescriptorFlags) Version() uint16 { return uint16(x >> 6 & 0x3) } +func (x DescriptorFlags) BlockSizeIndex() lz4block.BlockSizeIndex { + return lz4block.BlockSizeIndex(x >> 12 & 0x7) +} + +// Setters. +func (x *DescriptorFlags) ContentChecksumSet(v bool) *DescriptorFlags { + const b = 1 << 2 + if v { + *x = *x&^b | b + } else { + *x &^= b + } + return x +} +func (x *DescriptorFlags) SizeSet(v bool) *DescriptorFlags { + const b = 1 << 3 + if v { + *x = *x&^b | b + } else { + *x &^= b + } + return x +} +func (x *DescriptorFlags) BlockChecksumSet(v bool) *DescriptorFlags { + const b = 1 << 4 + if v { + *x = *x&^b | b + } else { + *x &^= b + } + return x +} +func (x *DescriptorFlags) BlockIndependenceSet(v bool) *DescriptorFlags { + const b = 1 << 5 + if v { + *x = *x&^b | b + } else { + *x &^= b + } + return x +} +func (x *DescriptorFlags) VersionSet(v uint16) *DescriptorFlags { + *x = *x&^(0x3<<6) | (DescriptorFlags(v) & 0x3 << 6) + return x +} +func (x *DescriptorFlags) BlockSizeIndexSet(v lz4block.BlockSizeIndex) *DescriptorFlags { + *x = *x&^(0x7<<12) | (DescriptorFlags(v) & 0x7 << 12) + return x +} + +// Code generated by `gen.exe`. DO NOT EDIT. + +// DataBlockSize is defined as follow: +// field bits +// ----- ---- +// size 31 +// Uncompressed 1 +type DataBlockSize uint32 + +// Getters. +func (x DataBlockSize) size() int { return int(x & 0x7FFFFFFF) } +func (x DataBlockSize) Uncompressed() bool { return x>>31&1 != 0 } + +// Setters. +func (x *DataBlockSize) sizeSet(v int) *DataBlockSize { + *x = *x&^0x7FFFFFFF | DataBlockSize(v)&0x7FFFFFFF + return x +} +func (x *DataBlockSize) UncompressedSet(v bool) *DataBlockSize { + const b = 1 << 31 + if v { + *x = *x&^b | b + } else { + *x &^= b + } + return x +} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go new file mode 100644 index 0000000..8d3206a --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go @@ -0,0 +1,212 @@ +// Package xxh32 implements the very fast XXH hashing algorithm (32 bits version). +// (https://github.com/Cyan4973/XXH/) +package xxh32 + +import ( + "encoding/binary" +) + +const ( + prime1 uint32 = 2654435761 + prime2 uint32 = 2246822519 + prime3 uint32 = 3266489917 + prime4 uint32 = 668265263 + prime5 uint32 = 374761393 + + primeMask = 0xFFFFFFFF + prime1plus2 = uint32((uint64(prime1) + uint64(prime2)) & primeMask) // 606290984 + prime1minus = uint32((-int64(prime1)) & primeMask) // 1640531535 +) + +// XXHZero represents an xxhash32 object with seed 0. +type XXHZero struct { + v [4]uint32 + totalLen uint64 + buf [16]byte + bufused int +} + +// Sum appends the current hash to b and returns the resulting slice. +// It does not change the underlying hash state. +func (xxh XXHZero) Sum(b []byte) []byte { + h32 := xxh.Sum32() + return append(b, byte(h32), byte(h32>>8), byte(h32>>16), byte(h32>>24)) +} + +// Reset resets the Hash to its initial state. +func (xxh *XXHZero) Reset() { + xxh.v[0] = prime1plus2 + xxh.v[1] = prime2 + xxh.v[2] = 0 + xxh.v[3] = prime1minus + xxh.totalLen = 0 + xxh.bufused = 0 +} + +// Size returns the number of bytes returned by Sum(). +func (xxh *XXHZero) Size() int { + return 4 +} + +// BlockSizeIndex gives the minimum number of bytes accepted by Write(). +func (xxh *XXHZero) BlockSize() int { + return 1 +} + +// Write adds input bytes to the Hash. +// It never returns an error. +func (xxh *XXHZero) Write(input []byte) (int, error) { + if xxh.totalLen == 0 { + xxh.Reset() + } + n := len(input) + m := xxh.bufused + + xxh.totalLen += uint64(n) + + r := len(xxh.buf) - m + if n < r { + copy(xxh.buf[m:], input) + xxh.bufused += len(input) + return n, nil + } + + var buf *[16]byte + if m != 0 { + // some data left from previous update + buf = &xxh.buf + c := copy(buf[m:], input) + n -= c + input = input[c:] + } + update(&xxh.v, buf, input) + xxh.bufused = copy(xxh.buf[:], input[n-n%16:]) + + return n, nil +} + +// Portable version of update. This updates v by processing all of buf +// (if not nil) and all full 16-byte blocks of input. +func updateGo(v *[4]uint32, buf *[16]byte, input []byte) { + // Causes compiler to work directly from registers instead of stack: + v1, v2, v3, v4 := v[0], v[1], v[2], v[3] + + if buf != nil { + v1 = rol13(v1+binary.LittleEndian.Uint32(buf[:])*prime2) * prime1 + v2 = rol13(v2+binary.LittleEndian.Uint32(buf[4:])*prime2) * prime1 + v3 = rol13(v3+binary.LittleEndian.Uint32(buf[8:])*prime2) * prime1 + v4 = rol13(v4+binary.LittleEndian.Uint32(buf[12:])*prime2) * prime1 + } + + for ; len(input) >= 16; input = input[16:] { + sub := input[:16] //BCE hint for compiler + v1 = rol13(v1+binary.LittleEndian.Uint32(sub[:])*prime2) * prime1 + v2 = rol13(v2+binary.LittleEndian.Uint32(sub[4:])*prime2) * prime1 + v3 = rol13(v3+binary.LittleEndian.Uint32(sub[8:])*prime2) * prime1 + v4 = rol13(v4+binary.LittleEndian.Uint32(sub[12:])*prime2) * prime1 + } + v[0], v[1], v[2], v[3] = v1, v2, v3, v4 +} + +// Sum32 returns the 32 bits Hash value. +func (xxh *XXHZero) Sum32() uint32 { + h32 := uint32(xxh.totalLen) + if h32 >= 16 { + h32 += rol1(xxh.v[0]) + rol7(xxh.v[1]) + rol12(xxh.v[2]) + rol18(xxh.v[3]) + } else { + h32 += prime5 + } + + p := 0 + n := xxh.bufused + buf := xxh.buf + for n := n - 4; p <= n; p += 4 { + h32 += binary.LittleEndian.Uint32(buf[p:p+4]) * prime3 + h32 = rol17(h32) * prime4 + } + for ; p < n; p++ { + h32 += uint32(buf[p]) * prime5 + h32 = rol11(h32) * prime1 + } + + h32 ^= h32 >> 15 + h32 *= prime2 + h32 ^= h32 >> 13 + h32 *= prime3 + h32 ^= h32 >> 16 + + return h32 +} + +// Portable version of ChecksumZero. +func checksumZeroGo(input []byte) uint32 { + n := len(input) + h32 := uint32(n) + + if n < 16 { + h32 += prime5 + } else { + v1 := prime1plus2 + v2 := prime2 + v3 := uint32(0) + v4 := prime1minus + p := 0 + for n := n - 16; p <= n; p += 16 { + sub := input[p:][:16] //BCE hint for compiler + v1 = rol13(v1+binary.LittleEndian.Uint32(sub[:])*prime2) * prime1 + v2 = rol13(v2+binary.LittleEndian.Uint32(sub[4:])*prime2) * prime1 + v3 = rol13(v3+binary.LittleEndian.Uint32(sub[8:])*prime2) * prime1 + v4 = rol13(v4+binary.LittleEndian.Uint32(sub[12:])*prime2) * prime1 + } + input = input[p:] + n -= p + h32 += rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + } + + p := 0 + for n := n - 4; p <= n; p += 4 { + h32 += binary.LittleEndian.Uint32(input[p:p+4]) * prime3 + h32 = rol17(h32) * prime4 + } + for p < n { + h32 += uint32(input[p]) * prime5 + h32 = rol11(h32) * prime1 + p++ + } + + h32 ^= h32 >> 15 + h32 *= prime2 + h32 ^= h32 >> 13 + h32 *= prime3 + h32 ^= h32 >> 16 + + return h32 +} + +func rol1(u uint32) uint32 { + return u<<1 | u>>31 +} + +func rol7(u uint32) uint32 { + return u<<7 | u>>25 +} + +func rol11(u uint32) uint32 { + return u<<11 | u>>21 +} + +func rol12(u uint32) uint32 { + return u<<12 | u>>20 +} + +func rol13(u uint32) uint32 { + return u<<13 | u>>19 +} + +func rol17(u uint32) uint32 { + return u<<17 | u>>15 +} + +func rol18(u uint32) uint32 { + return u<<18 | u>>14 +} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go new file mode 100644 index 0000000..0978b26 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go @@ -0,0 +1,11 @@ +// +build !noasm + +package xxh32 + +// ChecksumZero returns the 32-bit hash of input. +// +//go:noescape +func ChecksumZero(input []byte) uint32 + +//go:noescape +func update(v *[4]uint32, buf *[16]byte, input []byte) diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s new file mode 100644 index 0000000..c18ffd5 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s @@ -0,0 +1,251 @@ +// +build !noasm + +#include "go_asm.h" +#include "textflag.h" + +// Register allocation. +#define p R0 +#define n R1 +#define h R2 +#define v1 R2 // Alias for h. +#define v2 R3 +#define v3 R4 +#define v4 R5 +#define x1 R6 +#define x2 R7 +#define x3 R8 +#define x4 R9 + +// We need the primes in registers. The 16-byte loop only uses prime{1,2}. +#define prime1r R11 +#define prime2r R12 +#define prime3r R3 // The rest can alias v{2-4}. +#define prime4r R4 +#define prime5r R5 + +// Update round macros. These read from and increment p. + +#define round16aligned \ + MOVM.IA.W (p), [x1, x2, x3, x4] \ + \ + MULA x1, prime2r, v1, v1 \ + MULA x2, prime2r, v2, v2 \ + MULA x3, prime2r, v3, v3 \ + MULA x4, prime2r, v4, v4 \ + \ + MOVW v1 @> 19, v1 \ + MOVW v2 @> 19, v2 \ + MOVW v3 @> 19, v3 \ + MOVW v4 @> 19, v4 \ + \ + MUL prime1r, v1 \ + MUL prime1r, v2 \ + MUL prime1r, v3 \ + MUL prime1r, v4 \ + +#define round16unaligned \ + MOVBU.P 16(p), x1 \ + MOVBU -15(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -14(p), x3 \ + MOVBU -13(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v1, v1 \ + MOVW v1 @> 19, v1 \ + MUL prime1r, v1 \ + \ + MOVBU -12(p), x1 \ + MOVBU -11(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -10(p), x3 \ + MOVBU -9(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v2, v2 \ + MOVW v2 @> 19, v2 \ + MUL prime1r, v2 \ + \ + MOVBU -8(p), x1 \ + MOVBU -7(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -6(p), x3 \ + MOVBU -5(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v3, v3 \ + MOVW v3 @> 19, v3 \ + MUL prime1r, v3 \ + \ + MOVBU -4(p), x1 \ + MOVBU -3(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -2(p), x3 \ + MOVBU -1(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v4, v4 \ + MOVW v4 @> 19, v4 \ + MUL prime1r, v4 \ + + +// func ChecksumZero([]byte) uint32 +TEXT ·ChecksumZero(SB), NOFRAME|NOSPLIT, $-4-16 + MOVW input_base+0(FP), p + MOVW input_len+4(FP), n + + MOVW $const_prime1, prime1r + MOVW $const_prime2, prime2r + + // Set up h for n < 16. It's tempting to say {ADD prime5, n, h} + // here, but that's a pseudo-op that generates a load through R11. + MOVW $const_prime5, prime5r + ADD prime5r, n, h + CMP $0, n + BEQ end + + // We let n go negative so we can do comparisons with SUB.S + // instead of separate CMP. + SUB.S $16, n + BMI loop16done + + ADD prime1r, prime2r, v1 + MOVW prime2r, v2 + MOVW $0, v3 + RSB $0, prime1r, v4 + + TST $3, p + BNE loop16unaligned + +loop16aligned: + SUB.S $16, n + round16aligned + BPL loop16aligned + B loop16finish + +loop16unaligned: + SUB.S $16, n + round16unaligned + BPL loop16unaligned + +loop16finish: + MOVW v1 @> 31, h + ADD v2 @> 25, h + ADD v3 @> 20, h + ADD v4 @> 14, h + + // h += len(input) with v2 as temporary. + MOVW input_len+4(FP), v2 + ADD v2, h + +loop16done: + ADD $16, n // Restore number of bytes left. + + SUB.S $4, n + MOVW $const_prime3, prime3r + BMI loop4done + MOVW $const_prime4, prime4r + + TST $3, p + BNE loop4unaligned + +loop4aligned: + SUB.S $4, n + + MOVW.P 4(p), x1 + MULA prime3r, x1, h, h + MOVW h @> 15, h + MUL prime4r, h + + BPL loop4aligned + B loop4done + +loop4unaligned: + SUB.S $4, n + + MOVBU.P 4(p), x1 + MOVBU -3(p), x2 + ORR x2 << 8, x1 + MOVBU -2(p), x3 + ORR x3 << 16, x1 + MOVBU -1(p), x4 + ORR x4 << 24, x1 + + MULA prime3r, x1, h, h + MOVW h @> 15, h + MUL prime4r, h + + BPL loop4unaligned + +loop4done: + ADD.S $4, n // Restore number of bytes left. + BEQ end + + MOVW $const_prime5, prime5r + +loop1: + SUB.S $1, n + + MOVBU.P 1(p), x1 + MULA prime5r, x1, h, h + MOVW h @> 21, h + MUL prime1r, h + + BNE loop1 + +end: + MOVW $const_prime3, prime3r + EOR h >> 15, h + MUL prime2r, h + EOR h >> 13, h + MUL prime3r, h + EOR h >> 16, h + + MOVW h, ret+12(FP) + RET + + +// func update(v *[4]uint64, buf *[16]byte, p []byte) +TEXT ·update(SB), NOFRAME|NOSPLIT, $-4-20 + MOVW v+0(FP), p + MOVM.IA (p), [v1, v2, v3, v4] + + MOVW $const_prime1, prime1r + MOVW $const_prime2, prime2r + + // Process buf, if not nil. + MOVW buf+4(FP), p + CMP $0, p + BEQ noBuffered + + round16aligned + +noBuffered: + MOVW input_base +8(FP), p + MOVW input_len +12(FP), n + + SUB.S $16, n + BMI end + + TST $3, p + BNE loop16unaligned + +loop16aligned: + SUB.S $16, n + round16aligned + BPL loop16aligned + B end + +loop16unaligned: + SUB.S $16, n + round16unaligned + BPL loop16unaligned + +end: + MOVW v+0(FP), p + MOVM.IA [v1, v2, v3, v4], (p) + RET diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go new file mode 100644 index 0000000..c96b59b --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go @@ -0,0 +1,10 @@ +// +build !arm noasm + +package xxh32 + +// ChecksumZero returns the 32-bit hash of input. +func ChecksumZero(input []byte) uint32 { return checksumZeroGo(input) } + +func update(v *[4]uint32, buf *[16]byte, input []byte) { + updateGo(v, buf, input) +} diff --git a/vendor/github.com/pierrec/lz4/v4/lz4.go b/vendor/github.com/pierrec/lz4/v4/lz4.go new file mode 100644 index 0000000..a62022e --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/lz4.go @@ -0,0 +1,157 @@ +// Package lz4 implements reading and writing lz4 compressed data. +// +// The package supports both the LZ4 stream format, +// as specified in http://fastcompression.blogspot.fr/2013/04/lz4-streaming-format-final.html, +// and the LZ4 block format, defined at +// http://fastcompression.blogspot.fr/2011/05/lz4-explained.html. +// +// See https://github.com/lz4/lz4 for the reference C implementation. +package lz4 + +import ( + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" +) + +func _() { + // Safety checks for duplicated elements. + var x [1]struct{} + _ = x[lz4block.CompressionLevel(Fast)-lz4block.Fast] + _ = x[Block64Kb-BlockSize(lz4block.Block64Kb)] + _ = x[Block256Kb-BlockSize(lz4block.Block256Kb)] + _ = x[Block1Mb-BlockSize(lz4block.Block1Mb)] + _ = x[Block4Mb-BlockSize(lz4block.Block4Mb)] +} + +// CompressBlockBound returns the maximum size of a given buffer of size n, when not compressible. +func CompressBlockBound(n int) int { + return lz4block.CompressBlockBound(n) +} + +// UncompressBlock uncompresses the source buffer into the destination one, +// and returns the uncompressed size. +// +// The destination buffer must be sized appropriately. +// +// An error is returned if the source data is invalid or the destination buffer is too small. +func UncompressBlock(src, dst []byte) (int, error) { + return lz4block.UncompressBlock(src, dst, nil) +} + +// UncompressBlockWithDict uncompresses the source buffer into the destination one using a +// dictionary, and returns the uncompressed size. +// +// The destination buffer must be sized appropriately. +// +// An error is returned if the source data is invalid or the destination buffer is too small. +func UncompressBlockWithDict(src, dst, dict []byte) (int, error) { + return lz4block.UncompressBlock(src, dst, dict) +} + +// A Compressor compresses data into the LZ4 block format. +// It uses a fast compression algorithm. +// +// A Compressor is not safe for concurrent use by multiple goroutines. +// +// Use a Writer to compress into the LZ4 stream format. +type Compressor struct{ c lz4block.Compressor } + +// CompressBlock compresses the source buffer src into the destination dst. +// +// If compression is successful, the first return value is the size of the +// compressed data, which is always >0. +// +// If dst has length at least CompressBlockBound(len(src)), compression always +// succeeds. Otherwise, the first return value is zero. The error return is +// non-nil if the compressed data does not fit in dst, but it might fit in a +// larger buffer that is still smaller than CompressBlockBound(len(src)). The +// return value (0, nil) means the data is likely incompressible and a buffer +// of length CompressBlockBound(len(src)) should be passed in. +func (c *Compressor) CompressBlock(src, dst []byte) (int, error) { + return c.c.CompressBlock(src, dst) +} + +// CompressBlock compresses the source buffer into the destination one. +// This is the fast version of LZ4 compression and also the default one. +// +// The argument hashTable is scratch space for a hash table used by the +// compressor. If provided, it should have length at least 1<<16. If it is +// shorter (or nil), CompressBlock allocates its own hash table. +// +// The size of the compressed data is returned. +// +// If the destination buffer size is lower than CompressBlockBound and +// the compressed size is 0 and no error, then the data is incompressible. +// +// An error is returned if the destination buffer is too small. + +// CompressBlock is equivalent to Compressor.CompressBlock. +// The final argument is ignored and should be set to nil. +// +// This function is deprecated. Use a Compressor instead. +func CompressBlock(src, dst []byte, _ []int) (int, error) { + return lz4block.CompressBlock(src, dst) +} + +// A CompressorHC compresses data into the LZ4 block format. +// Its compression ratio is potentially better than that of a Compressor, +// but it is also slower and requires more memory. +// +// A Compressor is not safe for concurrent use by multiple goroutines. +// +// Use a Writer to compress into the LZ4 stream format. +type CompressorHC struct { + // Level is the maximum search depth for compression. + // Values <= 0 mean no maximum. + Level CompressionLevel + c lz4block.CompressorHC +} + +// CompressBlock compresses the source buffer src into the destination dst. +// +// If compression is successful, the first return value is the size of the +// compressed data, which is always >0. +// +// If dst has length at least CompressBlockBound(len(src)), compression always +// succeeds. Otherwise, the first return value is zero. The error return is +// non-nil if the compressed data does not fit in dst, but it might fit in a +// larger buffer that is still smaller than CompressBlockBound(len(src)). The +// return value (0, nil) means the data is likely incompressible and a buffer +// of length CompressBlockBound(len(src)) should be passed in. +func (c *CompressorHC) CompressBlock(src, dst []byte) (int, error) { + return c.c.CompressBlock(src, dst, lz4block.CompressionLevel(c.Level)) +} + +// CompressBlockHC is equivalent to CompressorHC.CompressBlock. +// The final two arguments are ignored and should be set to nil. +// +// This function is deprecated. Use a CompressorHC instead. +func CompressBlockHC(src, dst []byte, depth CompressionLevel, _, _ []int) (int, error) { + return lz4block.CompressBlockHC(src, dst, lz4block.CompressionLevel(depth)) +} + +const ( + // ErrInvalidSourceShortBuffer is returned by UncompressBlock or CompressBLock when a compressed + // block is corrupted or the destination buffer is not large enough for the uncompressed data. + ErrInvalidSourceShortBuffer = lz4errors.ErrInvalidSourceShortBuffer + // ErrInvalidFrame is returned when reading an invalid LZ4 archive. + ErrInvalidFrame = lz4errors.ErrInvalidFrame + // ErrInternalUnhandledState is an internal error. + ErrInternalUnhandledState = lz4errors.ErrInternalUnhandledState + // ErrInvalidHeaderChecksum is returned when reading a frame. + ErrInvalidHeaderChecksum = lz4errors.ErrInvalidHeaderChecksum + // ErrInvalidBlockChecksum is returned when reading a frame. + ErrInvalidBlockChecksum = lz4errors.ErrInvalidBlockChecksum + // ErrInvalidFrameChecksum is returned when reading a frame. + ErrInvalidFrameChecksum = lz4errors.ErrInvalidFrameChecksum + // ErrOptionInvalidCompressionLevel is returned when the supplied compression level is invalid. + ErrOptionInvalidCompressionLevel = lz4errors.ErrOptionInvalidCompressionLevel + // ErrOptionClosedOrError is returned when an option is applied to a closed or in error object. + ErrOptionClosedOrError = lz4errors.ErrOptionClosedOrError + // ErrOptionInvalidBlockSize is returned when + ErrOptionInvalidBlockSize = lz4errors.ErrOptionInvalidBlockSize + // ErrOptionNotApplicable is returned when trying to apply an option to an object not supporting it. + ErrOptionNotApplicable = lz4errors.ErrOptionNotApplicable + // ErrWriterNotClosed is returned when attempting to reset an unclosed writer. + ErrWriterNotClosed = lz4errors.ErrWriterNotClosed +) diff --git a/vendor/github.com/pierrec/lz4/v4/options.go b/vendor/github.com/pierrec/lz4/v4/options.go new file mode 100644 index 0000000..46a8738 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/options.go @@ -0,0 +1,214 @@ +package lz4 + +import ( + "fmt" + "reflect" + "runtime" + + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" +) + +//go:generate go run golang.org/x/tools/cmd/stringer -type=BlockSize,CompressionLevel -output options_gen.go + +type ( + applier interface { + Apply(...Option) error + private() + } + // Option defines the parameters to setup an LZ4 Writer or Reader. + Option func(applier) error +) + +// String returns a string representation of the option with its parameter(s). +func (o Option) String() string { + return o(nil).Error() +} + +// Default options. +var ( + DefaultBlockSizeOption = BlockSizeOption(Block4Mb) + DefaultChecksumOption = ChecksumOption(true) + DefaultConcurrency = ConcurrencyOption(1) + defaultOnBlockDone = OnBlockDoneOption(nil) +) + +const ( + Block64Kb BlockSize = 1 << (16 + iota*2) + Block256Kb + Block1Mb + Block4Mb +) + +// BlockSizeIndex defines the size of the blocks to be compressed. +type BlockSize uint32 + +// BlockSizeOption defines the maximum size of compressed blocks (default=Block4Mb). +func BlockSizeOption(size BlockSize) Option { + return func(a applier) error { + switch w := a.(type) { + case nil: + s := fmt.Sprintf("BlockSizeOption(%s)", size) + return lz4errors.Error(s) + case *Writer: + size := uint32(size) + if !lz4block.IsValid(size) { + return fmt.Errorf("%w: %d", lz4errors.ErrOptionInvalidBlockSize, size) + } + w.frame.Descriptor.Flags.BlockSizeIndexSet(lz4block.Index(size)) + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +// BlockChecksumOption enables or disables block checksum (default=false). +func BlockChecksumOption(flag bool) Option { + return func(a applier) error { + switch w := a.(type) { + case nil: + s := fmt.Sprintf("BlockChecksumOption(%v)", flag) + return lz4errors.Error(s) + case *Writer: + w.frame.Descriptor.Flags.BlockChecksumSet(flag) + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +// ChecksumOption enables/disables all blocks or content checksum (default=true). +func ChecksumOption(flag bool) Option { + return func(a applier) error { + switch w := a.(type) { + case nil: + s := fmt.Sprintf("ChecksumOption(%v)", flag) + return lz4errors.Error(s) + case *Writer: + w.frame.Descriptor.Flags.ContentChecksumSet(flag) + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +// SizeOption sets the size of the original uncompressed data (default=0). It is useful to know the size of the +// whole uncompressed data stream. +func SizeOption(size uint64) Option { + return func(a applier) error { + switch w := a.(type) { + case nil: + s := fmt.Sprintf("SizeOption(%d)", size) + return lz4errors.Error(s) + case *Writer: + w.frame.Descriptor.Flags.SizeSet(size > 0) + w.frame.Descriptor.ContentSize = size + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +// ConcurrencyOption sets the number of go routines used for compression. +// If n <= 0, then the output of runtime.GOMAXPROCS(0) is used. +func ConcurrencyOption(n int) Option { + if n <= 0 { + n = runtime.GOMAXPROCS(0) + } + return func(a applier) error { + switch rw := a.(type) { + case nil: + s := fmt.Sprintf("ConcurrencyOption(%d)", n) + return lz4errors.Error(s) + case *Writer: + rw.num = n + return nil + case *Reader: + rw.num = n + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +// CompressionLevel defines the level of compression to use. The higher the better, but slower, compression. +type CompressionLevel uint32 + +const ( + Fast CompressionLevel = 0 + Level1 CompressionLevel = 1 << (8 + iota) + Level2 + Level3 + Level4 + Level5 + Level6 + Level7 + Level8 + Level9 +) + +// CompressionLevelOption defines the compression level (default=Fast). +func CompressionLevelOption(level CompressionLevel) Option { + return func(a applier) error { + switch w := a.(type) { + case nil: + s := fmt.Sprintf("CompressionLevelOption(%s)", level) + return lz4errors.Error(s) + case *Writer: + switch level { + case Fast, Level1, Level2, Level3, Level4, Level5, Level6, Level7, Level8, Level9: + default: + return fmt.Errorf("%w: %d", lz4errors.ErrOptionInvalidCompressionLevel, level) + } + w.level = lz4block.CompressionLevel(level) + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +func onBlockDone(int) {} + +// OnBlockDoneOption is triggered when a block has been processed. For a Writer, it is when is has been compressed, +// for a Reader, it is when it has been uncompressed. +func OnBlockDoneOption(handler func(size int)) Option { + if handler == nil { + handler = onBlockDone + } + return func(a applier) error { + switch rw := a.(type) { + case nil: + s := fmt.Sprintf("OnBlockDoneOption(%s)", reflect.TypeOf(handler).String()) + return lz4errors.Error(s) + case *Writer: + rw.handler = handler + return nil + case *Reader: + rw.handler = handler + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} + +// LegacyOption provides support for writing LZ4 frames in the legacy format. +// +// See https://github.com/lz4/lz4/blob/dev/doc/lz4_Frame_format.md#legacy-frame. +// +// NB. compressed Linux kernel images use a tweaked LZ4 legacy format where +// the compressed stream is followed by the original (uncompressed) size of +// the kernel (https://events.static.linuxfound.org/sites/events/files/lcjpcojp13_klee.pdf). +// This is also supported as a special case. +func LegacyOption(legacy bool) Option { + return func(a applier) error { + switch rw := a.(type) { + case nil: + s := fmt.Sprintf("LegacyOption(%v)", legacy) + return lz4errors.Error(s) + case *Writer: + rw.legacy = legacy + return nil + } + return lz4errors.ErrOptionNotApplicable + } +} diff --git a/vendor/github.com/pierrec/lz4/v4/options_gen.go b/vendor/github.com/pierrec/lz4/v4/options_gen.go new file mode 100644 index 0000000..2de8149 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/options_gen.go @@ -0,0 +1,92 @@ +// Code generated by "stringer -type=BlockSize,CompressionLevel -output options_gen.go"; DO NOT EDIT. + +package lz4 + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[Block64Kb-65536] + _ = x[Block256Kb-262144] + _ = x[Block1Mb-1048576] + _ = x[Block4Mb-4194304] +} + +const ( + _BlockSize_name_0 = "Block64Kb" + _BlockSize_name_1 = "Block256Kb" + _BlockSize_name_2 = "Block1Mb" + _BlockSize_name_3 = "Block4Mb" +) + +func (i BlockSize) String() string { + switch { + case i == 65536: + return _BlockSize_name_0 + case i == 262144: + return _BlockSize_name_1 + case i == 1048576: + return _BlockSize_name_2 + case i == 4194304: + return _BlockSize_name_3 + default: + return "BlockSize(" + strconv.FormatInt(int64(i), 10) + ")" + } +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[Fast-0] + _ = x[Level1-512] + _ = x[Level2-1024] + _ = x[Level3-2048] + _ = x[Level4-4096] + _ = x[Level5-8192] + _ = x[Level6-16384] + _ = x[Level7-32768] + _ = x[Level8-65536] + _ = x[Level9-131072] +} + +const ( + _CompressionLevel_name_0 = "Fast" + _CompressionLevel_name_1 = "Level1" + _CompressionLevel_name_2 = "Level2" + _CompressionLevel_name_3 = "Level3" + _CompressionLevel_name_4 = "Level4" + _CompressionLevel_name_5 = "Level5" + _CompressionLevel_name_6 = "Level6" + _CompressionLevel_name_7 = "Level7" + _CompressionLevel_name_8 = "Level8" + _CompressionLevel_name_9 = "Level9" +) + +func (i CompressionLevel) String() string { + switch { + case i == 0: + return _CompressionLevel_name_0 + case i == 512: + return _CompressionLevel_name_1 + case i == 1024: + return _CompressionLevel_name_2 + case i == 2048: + return _CompressionLevel_name_3 + case i == 4096: + return _CompressionLevel_name_4 + case i == 8192: + return _CompressionLevel_name_5 + case i == 16384: + return _CompressionLevel_name_6 + case i == 32768: + return _CompressionLevel_name_7 + case i == 65536: + return _CompressionLevel_name_8 + case i == 131072: + return _CompressionLevel_name_9 + default: + return "CompressionLevel(" + strconv.FormatInt(int64(i), 10) + ")" + } +} diff --git a/vendor/github.com/pierrec/lz4/v4/reader.go b/vendor/github.com/pierrec/lz4/v4/reader.go new file mode 100644 index 0000000..275daad --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/reader.go @@ -0,0 +1,275 @@ +package lz4 + +import ( + "bytes" + "io" + + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" + "github.com/pierrec/lz4/v4/internal/lz4stream" +) + +var readerStates = []aState{ + noState: newState, + errorState: newState, + newState: readState, + readState: closedState, + closedState: newState, +} + +// NewReader returns a new LZ4 frame decoder. +func NewReader(r io.Reader) *Reader { + return newReader(r, false) +} + +func newReader(r io.Reader, legacy bool) *Reader { + zr := &Reader{frame: lz4stream.NewFrame()} + zr.state.init(readerStates) + _ = zr.Apply(DefaultConcurrency, defaultOnBlockDone) + zr.Reset(r) + return zr +} + +// Reader allows reading an LZ4 stream. +type Reader struct { + state _State + src io.Reader // source reader + num int // concurrency level + frame *lz4stream.Frame // frame being read + data []byte // block buffer allocated in non concurrent mode + reads chan []byte // pending data + idx int // size of pending data + handler func(int) + cum uint32 + dict []byte +} + +func (*Reader) private() {} + +func (r *Reader) Apply(options ...Option) (err error) { + defer r.state.check(&err) + switch r.state.state { + case newState: + case errorState: + return r.state.err + default: + return lz4errors.ErrOptionClosedOrError + } + for _, o := range options { + if err = o(r); err != nil { + return + } + } + return +} + +// Size returns the size of the underlying uncompressed data, if set in the stream. +func (r *Reader) Size() int { + switch r.state.state { + case readState, closedState: + if r.frame.Descriptor.Flags.Size() { + return int(r.frame.Descriptor.ContentSize) + } + } + return 0 +} + +func (r *Reader) isNotConcurrent() bool { + return r.num == 1 +} + +func (r *Reader) init() error { + err := r.frame.ParseHeaders(r.src) + if err != nil { + return err + } + if !r.frame.Descriptor.Flags.BlockIndependence() { + // We can't decompress dependent blocks concurrently. + // Instead of throwing an error to the user, silently drop concurrency + r.num = 1 + } + data, err := r.frame.InitR(r.src, r.num) + if err != nil { + return err + } + r.reads = data + r.idx = 0 + size := r.frame.Descriptor.Flags.BlockSizeIndex() + r.data = size.Get() + r.cum = 0 + return nil +} + +func (r *Reader) Read(buf []byte) (n int, err error) { + defer r.state.check(&err) + switch r.state.state { + case readState: + case closedState, errorState: + return 0, r.state.err + case newState: + // First initialization. + if err = r.init(); r.state.next(err) { + return + } + default: + return 0, r.state.fail() + } + for len(buf) > 0 { + var bn int + if r.idx == 0 { + if r.isNotConcurrent() { + bn, err = r.read(buf) + } else { + lz4block.Put(r.data) + r.data = <-r.reads + if len(r.data) == 0 { + // No uncompressed data: something went wrong or we are done. + err = r.frame.Blocks.ErrorR() + } + } + switch err { + case nil: + case io.EOF: + if er := r.frame.CloseR(r.src); er != nil { + err = er + } + lz4block.Put(r.data) + r.data = nil + return + default: + return + } + } + if bn == 0 { + // Fill buf with buffered data. + bn = copy(buf, r.data[r.idx:]) + r.idx += bn + if r.idx == len(r.data) { + // All data read, get ready for the next Read. + r.idx = 0 + } + } + buf = buf[bn:] + n += bn + r.handler(bn) + } + return +} + +// read uncompresses the next block as follow: +// - if buf has enough room, the block is uncompressed into it directly +// and the lenght of used space is returned +// - else, the uncompress data is stored in r.data and 0 is returned +func (r *Reader) read(buf []byte) (int, error) { + block := r.frame.Blocks.Block + _, err := block.Read(r.frame, r.src, r.cum) + if err != nil { + return 0, err + } + var direct bool + dst := r.data[:cap(r.data)] + if len(buf) >= len(dst) { + // Uncompress directly into buf. + direct = true + dst = buf + } + dst, err = block.Uncompress(r.frame, dst, r.dict, true) + if err != nil { + return 0, err + } + if !r.frame.Descriptor.Flags.BlockIndependence() { + if len(r.dict)+len(dst) > 128*1024 { + preserveSize := 64*1024 - len(dst) + if preserveSize < 0 { + preserveSize = 0 + } + r.dict = r.dict[len(r.dict)-preserveSize:] + } + r.dict = append(r.dict, dst...) + } + r.cum += uint32(len(dst)) + if direct { + return len(dst), nil + } + r.data = dst + return 0, nil +} + +// Reset clears the state of the Reader r such that it is equivalent to its +// initial state from NewReader, but instead reading from reader. +// No access to reader is performed. +func (r *Reader) Reset(reader io.Reader) { + if r.data != nil { + lz4block.Put(r.data) + r.data = nil + } + r.frame.Reset(r.num) + r.state.reset() + r.src = reader + r.reads = nil +} + +// WriteTo efficiently uncompresses the data from the Reader underlying source to w. +func (r *Reader) WriteTo(w io.Writer) (n int64, err error) { + switch r.state.state { + case closedState, errorState: + return 0, r.state.err + case newState: + if err = r.init(); r.state.next(err) { + return + } + default: + return 0, r.state.fail() + } + defer r.state.nextd(&err) + + var data []byte + if r.isNotConcurrent() { + size := r.frame.Descriptor.Flags.BlockSizeIndex() + data = size.Get() + defer lz4block.Put(data) + } + for { + var bn int + var dst []byte + if r.isNotConcurrent() { + bn, err = r.read(data) + dst = data[:bn] + } else { + lz4block.Put(dst) + dst = <-r.reads + bn = len(dst) + if bn == 0 { + // No uncompressed data: something went wrong or we are done. + err = r.frame.Blocks.ErrorR() + } + } + switch err { + case nil: + case io.EOF: + err = r.frame.CloseR(r.src) + return + default: + return + } + r.handler(bn) + bn, err = w.Write(dst) + n += int64(bn) + if err != nil { + return + } + } +} + +// ValidFrameHeader returns a bool indicating if the given bytes slice matches a LZ4 header. +func ValidFrameHeader(in []byte) (bool, error) { + f := lz4stream.NewFrame() + err := f.ParseHeaders(bytes.NewReader(in)) + if err == nil { + return true, nil + } + if err == lz4errors.ErrInvalidFrame { + return false, nil + } + return false, err +} diff --git a/vendor/github.com/pierrec/lz4/v4/state.go b/vendor/github.com/pierrec/lz4/v4/state.go new file mode 100644 index 0000000..d94f04d --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/state.go @@ -0,0 +1,75 @@ +package lz4 + +import ( + "errors" + "fmt" + "io" + + "github.com/pierrec/lz4/v4/internal/lz4errors" +) + +//go:generate go run golang.org/x/tools/cmd/stringer -type=aState -output state_gen.go + +const ( + noState aState = iota // uninitialized reader + errorState // unrecoverable error encountered + newState // instantiated object + readState // reading data + writeState // writing data + closedState // all done +) + +type ( + aState uint8 + _State struct { + states []aState + state aState + err error + } +) + +func (s *_State) init(states []aState) { + s.states = states + s.state = states[0] +} + +func (s *_State) reset() { + s.state = s.states[0] + s.err = nil +} + +// next sets the state to the next one unless it is passed a non nil error. +// It returns whether or not it is in error. +func (s *_State) next(err error) bool { + if err != nil { + s.err = fmt.Errorf("%s: %w", s.state, err) + s.state = errorState + return true + } + s.state = s.states[s.state] + return false +} + +// nextd is like next but for defers. +func (s *_State) nextd(errp *error) bool { + return errp != nil && s.next(*errp) +} + +// check sets s in error if not already in error and if the error is not nil or io.EOF, +func (s *_State) check(errp *error) { + if s.state == errorState || errp == nil { + return + } + if err := *errp; err != nil { + s.err = fmt.Errorf("%w[%s]", err, s.state) + if !errors.Is(err, io.EOF) { + s.state = errorState + } + } +} + +func (s *_State) fail() error { + s.state = errorState + s.err = fmt.Errorf("%w[%s]", lz4errors.ErrInternalUnhandledState, s.state) + return s.err +} diff --git a/vendor/github.com/pierrec/lz4/v4/state_gen.go b/vendor/github.com/pierrec/lz4/v4/state_gen.go new file mode 100644 index 0000000..75fb828 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/state_gen.go @@ -0,0 +1,28 @@ +// Code generated by "stringer -type=aState -output state_gen.go"; DO NOT EDIT. + +package lz4 + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[noState-0] + _ = x[errorState-1] + _ = x[newState-2] + _ = x[readState-3] + _ = x[writeState-4] + _ = x[closedState-5] +} + +const _aState_name = "noStateerrorStatenewStatereadStatewriteStateclosedState" + +var _aState_index = [...]uint8{0, 7, 17, 25, 34, 44, 55} + +func (i aState) String() string { + if i >= aState(len(_aState_index)-1) { + return "aState(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _aState_name[_aState_index[i]:_aState_index[i+1]] +} diff --git a/vendor/github.com/pierrec/lz4/v4/writer.go b/vendor/github.com/pierrec/lz4/v4/writer.go new file mode 100644 index 0000000..77699f2 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/writer.go @@ -0,0 +1,238 @@ +package lz4 + +import ( + "io" + + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" + "github.com/pierrec/lz4/v4/internal/lz4stream" +) + +var writerStates = []aState{ + noState: newState, + newState: writeState, + writeState: closedState, + closedState: newState, + errorState: newState, +} + +// NewWriter returns a new LZ4 frame encoder. +func NewWriter(w io.Writer) *Writer { + zw := &Writer{frame: lz4stream.NewFrame()} + zw.state.init(writerStates) + _ = zw.Apply(DefaultBlockSizeOption, DefaultChecksumOption, DefaultConcurrency, defaultOnBlockDone) + zw.Reset(w) + return zw +} + +// Writer allows writing an LZ4 stream. +type Writer struct { + state _State + src io.Writer // destination writer + level lz4block.CompressionLevel // how hard to try + num int // concurrency level + frame *lz4stream.Frame // frame being built + data []byte // pending data + idx int // size of pending data + handler func(int) + legacy bool +} + +func (*Writer) private() {} + +func (w *Writer) Apply(options ...Option) (err error) { + defer w.state.check(&err) + switch w.state.state { + case newState: + case errorState: + return w.state.err + default: + return lz4errors.ErrOptionClosedOrError + } + w.Reset(w.src) + for _, o := range options { + if err = o(w); err != nil { + return + } + } + return +} + +func (w *Writer) isNotConcurrent() bool { + return w.num == 1 +} + +// init sets up the Writer when in newState. It does not change the Writer state. +func (w *Writer) init() error { + w.frame.InitW(w.src, w.num, w.legacy) + size := w.frame.Descriptor.Flags.BlockSizeIndex() + w.data = size.Get() + w.idx = 0 + return w.frame.Descriptor.Write(w.frame, w.src) +} + +func (w *Writer) Write(buf []byte) (n int, err error) { + defer w.state.check(&err) + switch w.state.state { + case writeState: + case closedState, errorState: + return 0, w.state.err + case newState: + if err = w.init(); w.state.next(err) { + return + } + default: + return 0, w.state.fail() + } + + zn := len(w.data) + for len(buf) > 0 { + if w.isNotConcurrent() && w.idx == 0 && len(buf) >= zn { + // Avoid a copy as there is enough data for a block. + if err = w.write(buf[:zn], false); err != nil { + return + } + n += zn + buf = buf[zn:] + continue + } + // Accumulate the data to be compressed. + m := copy(w.data[w.idx:], buf) + n += m + w.idx += m + buf = buf[m:] + + if w.idx < len(w.data) { + // Buffer not filled. + return + } + + // Buffer full. + if err = w.write(w.data, true); err != nil { + return + } + if !w.isNotConcurrent() { + size := w.frame.Descriptor.Flags.BlockSizeIndex() + w.data = size.Get() + } + w.idx = 0 + } + return +} + +func (w *Writer) write(data []byte, safe bool) error { + if w.isNotConcurrent() { + block := w.frame.Blocks.Block + err := block.Compress(w.frame, data, w.level).Write(w.frame, w.src) + w.handler(len(block.Data)) + return err + } + c := make(chan *lz4stream.FrameDataBlock) + w.frame.Blocks.Blocks <- c + go func(c chan *lz4stream.FrameDataBlock, data []byte, safe bool) { + b := lz4stream.NewFrameDataBlock(w.frame) + c <- b.Compress(w.frame, data, w.level) + <-c + w.handler(len(b.Data)) + b.Close(w.frame) + if safe { + // safe to put it back as the last usage of it was FrameDataBlock.Write() called before c is closed + lz4block.Put(data) + } + }(c, data, safe) + + return nil +} + +// Flush any buffered data to the underlying writer immediately. +func (w *Writer) Flush() (err error) { + switch w.state.state { + case writeState: + case errorState: + return w.state.err + default: + return nil + } + + if w.idx > 0 { + // Flush pending data, disable w.data freeing as it is done later on. + if err = w.write(w.data[:w.idx], false); err != nil { + return err + } + w.idx = 0 + } + return nil +} + +// Close closes the Writer, flushing any unwritten data to the underlying writer +// without closing it. +func (w *Writer) Close() error { + if err := w.Flush(); err != nil { + return err + } + err := w.frame.CloseW(w.src, w.num) + // It is now safe to free the buffer. + if w.data != nil { + lz4block.Put(w.data) + w.data = nil + } + return err +} + +// Reset clears the state of the Writer w such that it is equivalent to its +// initial state from NewWriter, but instead writing to writer. +// Reset keeps the previous options unless overwritten by the supplied ones. +// No access to writer is performed. +// +// w.Close must be called before Reset or pending data may be dropped. +func (w *Writer) Reset(writer io.Writer) { + w.frame.Reset(w.num) + w.state.reset() + w.src = writer +} + +// ReadFrom efficiently reads from r and compressed into the Writer destination. +func (w *Writer) ReadFrom(r io.Reader) (n int64, err error) { + switch w.state.state { + case closedState, errorState: + return 0, w.state.err + case newState: + if err = w.init(); w.state.next(err) { + return + } + default: + return 0, w.state.fail() + } + defer w.state.check(&err) + + size := w.frame.Descriptor.Flags.BlockSizeIndex() + var done bool + var rn int + data := size.Get() + if w.isNotConcurrent() { + // Keep the same buffer for the whole process. + defer lz4block.Put(data) + } + for !done { + rn, err = io.ReadFull(r, data) + switch err { + case nil: + case io.EOF, io.ErrUnexpectedEOF: // read may be partial + done = true + default: + return + } + n += int64(rn) + err = w.write(data[:rn], true) + if err != nil { + return + } + w.handler(rn) + if !done && !w.isNotConcurrent() { + // The buffer will be returned automatically by go routines (safe=true) + // so get a new one fo the next round. + data = size.Get() + } + } + return +} diff --git a/vendor/github.com/redis/go-redis/v9/.gitignore b/vendor/github.com/redis/go-redis/v9/.gitignore new file mode 100644 index 0000000..00710d5 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/.gitignore @@ -0,0 +1,15 @@ +*.rdb +testdata/* +.idea/ +.DS_Store +*.tar.gz +*.dic +redis8tests.sh +coverage.txt +**/coverage.txt +.vscode +tmp/* +*.test + +# maintenanceNotifications upgrade documentation (temporary) +maintenanceNotifications/docs/ diff --git a/vendor/github.com/redis/go-redis/v9/.golangci.yml b/vendor/github.com/redis/go-redis/v9/.golangci.yml new file mode 100644 index 0000000..872454f --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/.golangci.yml @@ -0,0 +1,34 @@ +version: "2" +run: + timeout: 5m + tests: false +linters: + settings: + staticcheck: + checks: + - all + # Incorrect or missing package comment. + # https://staticcheck.dev/docs/checks/#ST1000 + - -ST1000 + # Omit embedded fields from selector expression. + # https://staticcheck.dev/docs/checks/#QF1008 + - -QF1008 + - -ST1003 + exclusions: + generated: lax + presets: + - comments + - common-false-positives + - legacy + - std-error-handling + paths: + - third_party$ + - builtin$ + - examples$ +formatters: + exclusions: + generated: lax + paths: + - third_party$ + - builtin$ + - examples$ diff --git a/vendor/github.com/redis/go-redis/v9/.prettierrc.yml b/vendor/github.com/redis/go-redis/v9/.prettierrc.yml new file mode 100644 index 0000000..8b7f044 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/.prettierrc.yml @@ -0,0 +1,4 @@ +semi: false +singleQuote: true +proseWrap: always +printWidth: 100 diff --git a/vendor/github.com/redis/go-redis/v9/CONTRIBUTING.md b/vendor/github.com/redis/go-redis/v9/CONTRIBUTING.md new file mode 100644 index 0000000..8c68c52 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/CONTRIBUTING.md @@ -0,0 +1,118 @@ +# Contributing + +## Introduction + +We appreciate your interest in considering contributing to go-redis. +Community contributions mean a lot to us. + +## Contributions we need + +You may already know how you'd like to contribute, whether it's a fix for a bug you +encountered, or a new feature your team wants to use. + +If you don't know where to start, consider improving +documentation, bug triaging, and writing tutorials are all examples of +helpful contributions that mean less work for you. + +## Your First Contribution + +Unsure where to begin contributing? You can start by looking through +[help-wanted +issues](https://github.com/redis/go-redis/issues?q=is%3Aopen+is%3Aissue+label%3ahelp-wanted). + +Never contributed to open source before? Here are a couple of friendly +tutorials: + +- +- + +## Getting Started + +Here's how to get started with your code contribution: + +1. Create your own fork of go-redis +2. Do the changes in your fork +3. If you need a development environment, run `make docker.start`. + +> Note: this clones and builds the docker containers specified in `docker-compose.yml`, to understand more about +> the infrastructure that will be started you can check the `docker-compose.yml`. You also have the possiblity +> to specify the redis image that will be pulled with the env variable `CLIENT_LIBS_TEST_IMAGE`. +> By default the docker image that will be pulled and started is `redislabs/client-libs-test:8.2.1-pre`. +> If you want to test with newer Redis version, using a newer version of `redislabs/client-libs-test` should work out of the box. + +4. While developing, make sure the tests pass by running `make test` (if you have the docker containers running, `make test.ci` may be sufficient). +> Note: `make test` will try to start all containers, run the tests with `make test.ci` and then stop all containers. +5. If you like the change and think the project could use it, send a + pull request + +To see what else is part of the automation, run `invoke -l` + + +## Testing + +### Setting up Docker +To run the tests, you need to have Docker installed and running. If you are using a host OS that does not support +docker host networks out of the box (e.g. Windows, OSX), you need to set up a docker desktop and enable docker host networks. + +### Running tests +Call `make test` to run all tests. + +Continuous Integration uses these same wrappers to run all of these +tests against multiple versions of redis. Feel free to test your +changes against all the go versions supported, as declared by the +[build.yml](./.github/workflows/build.yml) file. + +### Troubleshooting + +If you get any errors when running `make test`, make sure +that you are using supported versions of Docker and go. + +## How to Report a Bug + +### Security Vulnerabilities + +**NOTE**: If you find a security vulnerability, do NOT open an issue. +Email [Redis Open Source ()](mailto:oss@redis.com) instead. + +In order to determine whether you are dealing with a security issue, ask +yourself these two questions: + +- Can I access something that's not mine, or something I shouldn't + have access to? +- Can I disable something for other people? + +If the answer to either of those two questions are *yes*, then you're +probably dealing with a security issue. Note that even if you answer +*no* to both questions, you may still be dealing with a security +issue, so if you're unsure, just email [us](mailto:oss@redis.com). + +### Everything Else + +When filing an issue, make sure to answer these five questions: + +1. What version of go-redis are you using? +2. What version of redis are you using? +3. What did you do? +4. What did you expect to see? +5. What did you see instead? + +## Suggest a feature or enhancement + +If you'd like to contribute a new feature, make sure you check our +issue list to see if someone has already proposed it. Work may already +be underway on the feature you want or we may have rejected a +feature like it already. + +If you don't see anything, open a new issue that describes the feature +you would like and how it should work. + +## Code review process + +The core team regularly looks at pull requests. We will provide +feedback as soon as possible. After receiving our feedback, please respond +within two weeks. After that time, we may close your PR if it isn't +showing any activity. + +## Support + +Maintainers can provide limited support to contributors on discord: https://discord.gg/W4txy5AeKM diff --git a/vendor/github.com/redis/go-redis/v9/LICENSE b/vendor/github.com/redis/go-redis/v9/LICENSE new file mode 100644 index 0000000..f4967db --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/LICENSE @@ -0,0 +1,25 @@ +Copyright (c) 2013 The github.com/redis/go-redis Authors. +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/vendor/github.com/redis/go-redis/v9/Makefile b/vendor/github.com/redis/go-redis/v9/Makefile new file mode 100644 index 0000000..0252a7e --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/Makefile @@ -0,0 +1,87 @@ +GO_MOD_DIRS := $(shell find . -type f -name 'go.mod' -exec dirname {} \; | sort) +REDIS_VERSION ?= 8.2 +RE_CLUSTER ?= false +RCE_DOCKER ?= true +CLIENT_LIBS_TEST_IMAGE ?= redislabs/client-libs-test:8.2.1-pre + +docker.start: + export RE_CLUSTER=$(RE_CLUSTER) && \ + export RCE_DOCKER=$(RCE_DOCKER) && \ + export REDIS_VERSION=$(REDIS_VERSION) && \ + export CLIENT_LIBS_TEST_IMAGE=$(CLIENT_LIBS_TEST_IMAGE) && \ + docker compose --profile all up -d --quiet-pull + +docker.stop: + docker compose --profile all down + +test: + $(MAKE) docker.start + @if [ -z "$(REDIS_VERSION)" ]; then \ + echo "REDIS_VERSION not set, running all tests"; \ + $(MAKE) test.ci; \ + else \ + MAJOR_VERSION=$$(echo "$(REDIS_VERSION)" | cut -d. -f1); \ + if [ "$$MAJOR_VERSION" -ge 8 ]; then \ + echo "REDIS_VERSION $(REDIS_VERSION) >= 8, running all tests"; \ + $(MAKE) test.ci; \ + else \ + echo "REDIS_VERSION $(REDIS_VERSION) < 8, skipping vector_sets tests"; \ + $(MAKE) test.ci.skip-vectorsets; \ + fi; \ + fi + $(MAKE) docker.stop + +test.ci: + set -e; for dir in $(GO_MOD_DIRS); do \ + echo "go test in $${dir}"; \ + (cd "$${dir}" && \ + export RE_CLUSTER=$(RE_CLUSTER) && \ + export RCE_DOCKER=$(RCE_DOCKER) && \ + export REDIS_VERSION=$(REDIS_VERSION) && \ + go mod tidy -compat=1.18 && \ + go vet && \ + go test -v -coverprofile=coverage.txt -covermode=atomic ./... -race -skip Example); \ + done + cd internal/customvet && go build . + go vet -vettool ./internal/customvet/customvet + +test.ci.skip-vectorsets: + set -e; for dir in $(GO_MOD_DIRS); do \ + echo "go test in $${dir} (skipping vector sets)"; \ + (cd "$${dir}" && \ + export RE_CLUSTER=$(RE_CLUSTER) && \ + export RCE_DOCKER=$(RCE_DOCKER) && \ + export REDIS_VERSION=$(REDIS_VERSION) && \ + go mod tidy -compat=1.18 && \ + go vet && \ + go test -v -coverprofile=coverage.txt -covermode=atomic ./... -race \ + -run '^(?!.*(?:VectorSet|vectorset|ExampleClient_vectorset)).*$$' -skip Example); \ + done + cd internal/customvet && go build . + go vet -vettool ./internal/customvet/customvet + +bench: + export RE_CLUSTER=$(RE_CLUSTER) && \ + export RCE_DOCKER=$(RCE_DOCKER) && \ + export REDIS_VERSION=$(REDIS_VERSION) && \ + go test ./... -test.run=NONE -test.bench=. -test.benchmem -skip Example + +.PHONY: all test test.ci test.ci.skip-vectorsets bench fmt + +build: + export RE_CLUSTER=$(RE_CLUSTER) && \ + export RCE_DOCKER=$(RCE_DOCKER) && \ + export REDIS_VERSION=$(REDIS_VERSION) && \ + go build . + +fmt: + gofumpt -w ./ + goimports -w -local github.com/redis/go-redis ./ + +go_mod_tidy: + set -e; for dir in $(GO_MOD_DIRS); do \ + echo "go mod tidy in $${dir}"; \ + (cd "$${dir}" && \ + go get -u ./... && \ + go mod tidy -compat=1.18); \ + done diff --git a/vendor/github.com/redis/go-redis/v9/README.md b/vendor/github.com/redis/go-redis/v9/README.md new file mode 100644 index 0000000..0716403 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/README.md @@ -0,0 +1,461 @@ +# Redis client for Go + +[![build workflow](https://github.com/redis/go-redis/actions/workflows/build.yml/badge.svg)](https://github.com/redis/go-redis/actions) +[![PkgGoDev](https://pkg.go.dev/badge/github.com/redis/go-redis/v9)](https://pkg.go.dev/github.com/redis/go-redis/v9?tab=doc) +[![Documentation](https://img.shields.io/badge/redis-documentation-informational)](https://redis.uptrace.dev/) +[![Go Report Card](https://goreportcard.com/badge/github.com/redis/go-redis/v9)](https://goreportcard.com/report/github.com/redis/go-redis/v9) +[![codecov](https://codecov.io/github/redis/go-redis/graph/badge.svg?token=tsrCZKuSSw)](https://codecov.io/github/redis/go-redis) + +[![Discord](https://img.shields.io/discord/697882427875393627.svg?style=social&logo=discord)](https://discord.gg/W4txy5AeKM) +[![Twitch](https://img.shields.io/twitch/status/redisinc?style=social)](https://www.twitch.tv/redisinc) +[![YouTube](https://img.shields.io/youtube/channel/views/UCD78lHSwYqMlyetR0_P4Vig?style=social)](https://www.youtube.com/redisinc) +[![Twitter](https://img.shields.io/twitter/follow/redisinc?style=social)](https://twitter.com/redisinc) +[![Stack Exchange questions](https://img.shields.io/stackexchange/stackoverflow/t/go-redis?style=social&logo=stackoverflow&label=Stackoverflow)](https://stackoverflow.com/questions/tagged/go-redis) + +> go-redis is the official Redis client library for the Go programming language. It offers a straightforward interface for interacting with Redis servers. + +## Supported versions + +In `go-redis` we are aiming to support the last three releases of Redis. Currently, this means we do support: +- [Redis 7.2](https://raw.githubusercontent.com/redis/redis/7.2/00-RELEASENOTES) - using Redis Stack 7.2 for modules support +- [Redis 7.4](https://raw.githubusercontent.com/redis/redis/7.4/00-RELEASENOTES) - using Redis Stack 7.4 for modules support +- [Redis 8.0](https://raw.githubusercontent.com/redis/redis/8.0/00-RELEASENOTES) - using Redis CE 8.0 where modules are included +- [Redis 8.2](https://raw.githubusercontent.com/redis/redis/8.2/00-RELEASENOTES) - using Redis CE 8.2 where modules are included + +Although the `go.mod` states it requires at minimum `go 1.18`, our CI is configured to run the tests against all three +versions of Redis and latest two versions of Go ([1.23](https://go.dev/doc/devel/release#go1.23.0), +[1.24](https://go.dev/doc/devel/release#go1.24.0)). We observe that some modules related test may not pass with +Redis Stack 7.2 and some commands are changed with Redis CE 8.0. +Please do refer to the documentation and the tests if you experience any issues. We do plan to update the go version +in the `go.mod` to `go 1.24` in one of the next releases. + +## How do I Redis? + +[Learn for free at Redis University](https://university.redis.com/) + +[Build faster with the Redis Launchpad](https://launchpad.redis.com/) + +[Try the Redis Cloud](https://redis.com/try-free/) + +[Dive in developer tutorials](https://developer.redis.com/) + +[Join the Redis community](https://redis.com/community/) + +[Work at Redis](https://redis.com/company/careers/jobs/) + +## Documentation + +- [English](https://redis.uptrace.dev) +- [简体中文](https://redis.uptrace.dev/zh/) + +## Resources + +- [Discussions](https://github.com/redis/go-redis/discussions) +- [Chat](https://discord.gg/W4txy5AeKM) +- [Reference](https://pkg.go.dev/github.com/redis/go-redis/v9) +- [Examples](https://pkg.go.dev/github.com/redis/go-redis/v9#pkg-examples) + +## Ecosystem + +- [Redis Mock](https://github.com/go-redis/redismock) +- [Distributed Locks](https://github.com/bsm/redislock) +- [Redis Cache](https://github.com/go-redis/cache) +- [Rate limiting](https://github.com/go-redis/redis_rate) + +This client also works with [Kvrocks](https://github.com/apache/incubator-kvrocks), a distributed +key value NoSQL database that uses RocksDB as storage engine and is compatible with Redis protocol. + +## Features + +- Redis commands except QUIT and SYNC. +- Automatic connection pooling. +- [StreamingCredentialsProvider (e.g. entra id, oauth)](#1-streaming-credentials-provider-highest-priority) (experimental) +- [Pub/Sub](https://redis.uptrace.dev/guide/go-redis-pubsub.html). +- [Pipelines and transactions](https://redis.uptrace.dev/guide/go-redis-pipelines.html). +- [Scripting](https://redis.uptrace.dev/guide/lua-scripting.html). +- [Redis Sentinel](https://redis.uptrace.dev/guide/go-redis-sentinel.html). +- [Redis Cluster](https://redis.uptrace.dev/guide/go-redis-cluster.html). +- [Redis Ring](https://redis.uptrace.dev/guide/ring.html). +- [Redis Performance Monitoring](https://redis.uptrace.dev/guide/redis-performance-monitoring.html). +- [Redis Probabilistic [RedisStack]](https://redis.io/docs/data-types/probabilistic/) +- [Customizable read and write buffers size.](#custom-buffer-sizes) + +## Installation + +go-redis supports 2 last Go versions and requires a Go version with +[modules](https://github.com/golang/go/wiki/Modules) support. So make sure to initialize a Go +module: + +```shell +go mod init github.com/my/repo +``` + +Then install go-redis/**v9**: + +```shell +go get github.com/redis/go-redis/v9 +``` + +## Quickstart + +```go +import ( + "context" + "fmt" + + "github.com/redis/go-redis/v9" +) + +var ctx = context.Background() + +func ExampleClient() { + rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + Password: "", // no password set + DB: 0, // use default DB + }) + + err := rdb.Set(ctx, "key", "value", 0).Err() + if err != nil { + panic(err) + } + + val, err := rdb.Get(ctx, "key").Result() + if err != nil { + panic(err) + } + fmt.Println("key", val) + + val2, err := rdb.Get(ctx, "key2").Result() + if err == redis.Nil { + fmt.Println("key2 does not exist") + } else if err != nil { + panic(err) + } else { + fmt.Println("key2", val2) + } + // Output: key value + // key2 does not exist +} +``` + +### Authentication + +The Redis client supports multiple ways to provide authentication credentials, with a clear priority order. Here are the available options: + +#### 1. Streaming Credentials Provider (Highest Priority) - Experimental feature + +The streaming credentials provider allows for dynamic credential updates during the connection lifetime. This is particularly useful for managed identity services and token-based authentication. + +```go +type StreamingCredentialsProvider interface { + Subscribe(listener CredentialsListener) (Credentials, UnsubscribeFunc, error) +} + +type CredentialsListener interface { + OnNext(credentials Credentials) // Called when credentials are updated + OnError(err error) // Called when an error occurs +} + +type Credentials interface { + BasicAuth() (username string, password string) + RawCredentials() string +} +``` + +Example usage: +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + StreamingCredentialsProvider: &MyCredentialsProvider{}, +}) +``` + +**Note:** The streaming credentials provider can be used with [go-redis-entraid](https://github.com/redis/go-redis-entraid) to enable Entra ID (formerly Azure AD) authentication. This allows for seamless integration with Azure's managed identity services and token-based authentication. + +Example with Entra ID: +```go +import ( + "github.com/redis/go-redis/v9" + "github.com/redis/go-redis-entraid" +) + +// Create an Entra ID credentials provider +provider := entraid.NewDefaultAzureIdentityProvider() + +// Configure Redis client with Entra ID authentication +rdb := redis.NewClient(&redis.Options{ + Addr: "your-redis-server.redis.cache.windows.net:6380", + StreamingCredentialsProvider: provider, + TLSConfig: &tls.Config{ + MinVersion: tls.VersionTLS12, + }, +}) +``` + +#### 2. Context-based Credentials Provider + +The context-based provider allows credentials to be determined at the time of each operation, using the context. + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + CredentialsProviderContext: func(ctx context.Context) (string, string, error) { + // Return username, password, and any error + return "user", "pass", nil + }, +}) +``` + +#### 3. Regular Credentials Provider + +A simple function-based provider that returns static credentials. + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + CredentialsProvider: func() (string, string) { + // Return username and password + return "user", "pass" + }, +}) +``` + +#### 4. Username/Password Fields (Lowest Priority) + +The most basic way to provide credentials is through the `Username` and `Password` fields in the options. + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + Username: "user", + Password: "pass", +}) +``` + +#### Priority Order + +The client will use credentials in the following priority order: +1. Streaming Credentials Provider (if set) +2. Context-based Credentials Provider (if set) +3. Regular Credentials Provider (if set) +4. Username/Password fields (if set) + +If none of these are set, the client will attempt to connect without authentication. + +### Protocol Version + +The client supports both RESP2 and RESP3 protocols. You can specify the protocol version in the options: + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + Password: "", // no password set + DB: 0, // use default DB + Protocol: 3, // specify 2 for RESP 2 or 3 for RESP 3 +}) +``` + +### Connecting via a redis url + +go-redis also supports connecting via the +[redis uri specification](https://github.com/redis/redis-specifications/tree/master/uri/redis.txt). +The example below demonstrates how the connection can easily be configured using a string, adhering +to this specification. + +```go +import ( + "github.com/redis/go-redis/v9" +) + +func ExampleClient() *redis.Client { + url := "redis://user:password@localhost:6379/0?protocol=3" + opts, err := redis.ParseURL(url) + if err != nil { + panic(err) + } + + return redis.NewClient(opts) +} + +``` + +### Instrument with OpenTelemetry + +```go +import ( + "github.com/redis/go-redis/v9" + "github.com/redis/go-redis/extra/redisotel/v9" + "errors" +) + +func main() { + ... + rdb := redis.NewClient(&redis.Options{...}) + + if err := errors.Join(redisotel.InstrumentTracing(rdb), redisotel.InstrumentMetrics(rdb)); err != nil { + log.Fatal(err) + } +``` + + +### Buffer Size Configuration + +go-redis uses 32KiB read and write buffers by default for optimal performance. For high-throughput applications or large pipelines, you can customize buffer sizes: + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + ReadBufferSize: 1024 * 1024, // 1MiB read buffer + WriteBufferSize: 1024 * 1024, // 1MiB write buffer +}) +``` + +### Advanced Configuration + +go-redis supports extending the client identification phase to allow projects to send their own custom client identification. + +#### Default Client Identification + +By default, go-redis automatically sends the client library name and version during the connection process. This feature is available in redis-server as of version 7.2. As a result, the command is "fire and forget", meaning it should fail silently, in the case that the redis server does not support this feature. + +#### Disabling Identity Verification + +When connection identity verification is not required or needs to be explicitly disabled, a `DisableIdentity` configuration option exists. +Initially there was a typo and the option was named `DisableIndentity` instead of `DisableIdentity`. The misspelled option is marked as Deprecated and will be removed in V10 of this library. +Although both options will work at the moment, the correct option is `DisableIdentity`. The deprecated option will be removed in V10 of this library, so please use the correct option name to avoid any issues. + +To disable verification, set the `DisableIdentity` option to `true` in the Redis client options: + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + Password: "", + DB: 0, + DisableIdentity: true, // Disable set-info on connect +}) +``` + +#### Unstable RESP3 Structures for RediSearch Commands +When integrating Redis with application functionalities using RESP3, it's important to note that some response structures aren't final yet. This is especially true for more complex structures like search and query results. We recommend using RESP2 when using the search and query capabilities, but we plan to stabilize the RESP3-based API-s in the coming versions. You can find more guidance in the upcoming release notes. + +To enable unstable RESP3, set the option in your client configuration: + +```go +redis.NewClient(&redis.Options{ + UnstableResp3: true, + }) +``` +**Note:** When UnstableResp3 mode is enabled, it's necessary to use RawResult() and RawVal() to retrieve a raw data. + Since, raw response is the only option for unstable search commands Val() and Result() calls wouldn't have any affect on them: + +```go +res1, err := client.FTSearchWithArgs(ctx, "txt", "foo bar", &redis.FTSearchOptions{}).RawResult() +val1 := client.FTSearchWithArgs(ctx, "txt", "foo bar", &redis.FTSearchOptions{}).RawVal() +``` + +#### Redis-Search Default Dialect + +In the Redis-Search module, **the default dialect is 2**. If needed, you can explicitly specify a different dialect using the appropriate configuration in your queries. + +**Important**: Be aware that the query dialect may impact the results returned. If needed, you can revert to a different dialect version by passing the desired dialect in the arguments of the command you want to execute. +For example: +``` + res2, err := rdb.FTSearchWithArgs(ctx, + "idx:bicycle", + "@pickup_zone:[CONTAINS $bike]", + &redis.FTSearchOptions{ + Params: map[string]interface{}{ + "bike": "POINT(-0.1278 51.5074)", + }, + DialectVersion: 3, + }, + ).Result() +``` +You can find further details in the [query dialect documentation](https://redis.io/docs/latest/develop/interact/search-and-query/advanced-concepts/dialects/). + +#### Custom buffer sizes +Prior to v9.12, the buffer size was the default go value of 4096 bytes. Starting from v9.12, +go-redis uses 32KiB read and write buffers by default for optimal performance. +For high-throughput applications or large pipelines, you can customize buffer sizes: + +```go +rdb := redis.NewClient(&redis.Options{ + Addr: "localhost:6379", + ReadBufferSize: 1024 * 1024, // 1MiB read buffer + WriteBufferSize: 1024 * 1024, // 1MiB write buffer +}) +``` + +**Important**: If you experience any issues with the default buffer sizes, please try setting them to the go default of 4096 bytes. + +## Contributing +We welcome contributions to the go-redis library! If you have a bug fix, feature request, or improvement, please open an issue or pull request on GitHub. +We appreciate your help in making go-redis better for everyone. +If you are interested in contributing to the go-redis library, please check out our [contributing guidelines](CONTRIBUTING.md) for more information on how to get started. + +## Look and feel + +Some corner cases: + +```go +// SET key value EX 10 NX +set, err := rdb.SetNX(ctx, "key", "value", 10*time.Second).Result() + +// SET key value keepttl NX +set, err := rdb.SetNX(ctx, "key", "value", redis.KeepTTL).Result() + +// SORT list LIMIT 0 2 ASC +vals, err := rdb.Sort(ctx, "list", &redis.Sort{Offset: 0, Count: 2, Order: "ASC"}).Result() + +// ZRANGEBYSCORE zset -inf +inf WITHSCORES LIMIT 0 2 +vals, err := rdb.ZRangeByScoreWithScores(ctx, "zset", &redis.ZRangeBy{ + Min: "-inf", + Max: "+inf", + Offset: 0, + Count: 2, +}).Result() + +// ZINTERSTORE out 2 zset1 zset2 WEIGHTS 2 3 AGGREGATE SUM +vals, err := rdb.ZInterStore(ctx, "out", &redis.ZStore{ + Keys: []string{"zset1", "zset2"}, + Weights: []int64{2, 3} +}).Result() + +// EVAL "return {KEYS[1],ARGV[1]}" 1 "key" "hello" +vals, err := rdb.Eval(ctx, "return {KEYS[1],ARGV[1]}", []string{"key"}, "hello").Result() + +// custom command +res, err := rdb.Do(ctx, "set", "key", "value").Result() +``` + + +## Run the test + +Recommended to use Docker, just need to run: +```shell +make test +``` + +## See also + +- [Golang ORM](https://bun.uptrace.dev) for PostgreSQL, MySQL, MSSQL, and SQLite +- [Golang PostgreSQL](https://bun.uptrace.dev/postgres/) +- [Golang HTTP router](https://bunrouter.uptrace.dev/) +- [Golang ClickHouse ORM](https://github.com/uptrace/go-clickhouse) + +## Contributors + +> The go-redis project was originally initiated by :star: [**uptrace/uptrace**](https://github.com/uptrace/uptrace). +> Uptrace is an open-source APM tool that supports distributed tracing, metrics, and logs. You can +> use it to monitor applications and set up automatic alerts to receive notifications via email, +> Slack, Telegram, and others. +> +> See [OpenTelemetry](https://github.com/redis/go-redis/tree/master/example/otel) example which +> demonstrates how you can use Uptrace to monitor go-redis. + +Thanks to all the people who already contributed! + + + + diff --git a/vendor/github.com/redis/go-redis/v9/RELEASE-NOTES.md b/vendor/github.com/redis/go-redis/v9/RELEASE-NOTES.md new file mode 100644 index 0000000..f977018 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/RELEASE-NOTES.md @@ -0,0 +1,589 @@ +# Release Notes + +# 9.16.0 (2025-10-23) + +## 🚀 Highlights + +### Maintenance Notifications Support + +This release introduces comprehensive support for Redis maintenance notifications, enabling applications to handle server maintenance events gracefully. The new `maintnotifications` package provides: + +- **RESP3 Push Notifications**: Full support for Redis RESP3 protocol push notifications +- **Connection Handoff**: Automatic connection migration during server maintenance with configurable retry policies and circuit breakers +- **Graceful Degradation**: Configurable timeout relaxation during maintenance windows to prevent false failures +- **Event-Driven Architecture**: Background workers with on-demand scaling for efficient handoff processing +- **Production-Ready**: Comprehensive E2E testing framework and monitoring capabilities + +For detailed usage examples and configuration options, see the [maintenance notifications documentation](maintnotifications/README.md). + +## ✨ New Features + +- **Trace Filtering**: Add support for filtering traces for specific commands, including pipeline operations and dial operations ([#3519](https://github.com/redis/go-redis/pull/3519), [#3550](https://github.com/redis/go-redis/pull/3550)) + - New `TraceCmdFilter` option to selectively trace commands + - Reduces overhead by excluding high-frequency or low-value commands from traces + +## 🐛 Bug Fixes + +- **Pipeline Error Handling**: Fix issue where pipeline repeatedly sets the same error ([#3525](https://github.com/redis/go-redis/pull/3525)) +- **Connection Pool**: Ensure re-authentication does not interfere with connection handoff operations ([#3547](https://github.com/redis/go-redis/pull/3547)) + +## 🔧 Improvements + +- **Hash Commands**: Update hash command implementations ([#3523](https://github.com/redis/go-redis/pull/3523)) +- **OpenTelemetry**: Use `metric.WithAttributeSet` to avoid unnecessary attribute copying in redisotel ([#3552](https://github.com/redis/go-redis/pull/3552)) + +## 📚 Documentation + +- **Cluster Client**: Add explanation for why `MaxRetries` is disabled for `ClusterClient` ([#3551](https://github.com/redis/go-redis/pull/3551)) + +## 🧪 Testing & Infrastructure + +- **E2E Testing**: Upgrade E2E testing framework with improved reliability and coverage ([#3541](https://github.com/redis/go-redis/pull/3541)) +- **Release Process**: Improved resiliency of the release process ([#3530](https://github.com/redis/go-redis/pull/3530)) + +## 📦 Dependencies + +- Bump `rojopolis/spellcheck-github-actions` from 0.51.0 to 0.52.0 ([#3520](https://github.com/redis/go-redis/pull/3520)) +- Bump `github/codeql-action` from 3 to 4 ([#3544](https://github.com/redis/go-redis/pull/3544)) + +## 👥 Contributors + +We'd like to thank all the contributors who worked on this release! + +[@ndyakov](https://github.com/ndyakov), [@htemelski-redis](https://github.com/htemelski-redis), [@Sovietaced](https://github.com/Sovietaced), [@Udhayarajan](https://github.com/Udhayarajan), [@boekkooi-impossiblecloud](https://github.com/boekkooi-impossiblecloud), [@Pika-Gopher](https://github.com/Pika-Gopher), [@cxljs](https://github.com/cxljs), [@huiyifyj](https://github.com/huiyifyj), [@omid-h70](https://github.com/omid-h70) + +--- + +**Full Changelog**: https://github.com/redis/go-redis/compare/v9.14.0...v9.16.0 + + +# 9.15.0 was accidentally released. Please use version 9.16.0 instead. + +# 9.15.0-beta.3 (2025-09-26) + +## Highlights +This beta release includes a pre-production version of processing push notifications and hitless upgrades. + +# Changes + +- chore: Update hash_commands.go ([#3523](https://github.com/redis/go-redis/pull/3523)) + +## 🚀 New Features + +- feat: RESP3 notifications support & Hitless notifications handling ([#3418](https://github.com/redis/go-redis/pull/3418)) + +## 🐛 Bug Fixes + +- fix: pipeline repeatedly sets the error ([#3525](https://github.com/redis/go-redis/pull/3525)) + +## 🧰 Maintenance + +- chore(deps): bump rojopolis/spellcheck-github-actions from 0.51.0 to 0.52.0 ([#3520](https://github.com/redis/go-redis/pull/3520)) +- feat(e2e-testing): maintnotifications e2e and refactor ([#3526](https://github.com/redis/go-redis/pull/3526)) +- feat(tag.sh): Improved resiliency of the release process ([#3530](https://github.com/redis/go-redis/pull/3530)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@cxljs](https://github.com/cxljs), [@ndyakov](https://github.com/ndyakov), [@htemelski-redis](https://github.com/htemelski-redis), and [@omid-h70](https://github.com/omid-h70) + + +# 9.15.0-beta.1 (2025-09-10) + +## Highlights +This beta release includes a pre-production version of processing push notifications and hitless upgrades. + +### Hitless Upgrades +Hitless upgrades is a major new feature that allows for zero-downtime upgrades in Redis clusters. +You can find more information in the [Hitless Upgrades documentation](https://github.com/redis/go-redis/tree/master/hitless). + +# Changes + +## 🚀 New Features +- [CAE-1088] & [CAE-1072] feat: RESP3 notifications support & Hitless notifications handling ([#3418](https://github.com/redis/go-redis/pull/3418)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@ndyakov](https://github.com/ndyakov), [@htemelski-redis](https://github.com/htemelski-redis), [@ofekshenawa](https://github.com/ofekshenawa) + + +# 9.14.0 (2025-09-10) + +## Highlights +- Added batch process method to the pipeline ([#3510](https://github.com/redis/go-redis/pull/3510)) + +# Changes + +## 🚀 New Features + +- Added batch process method to the pipeline ([#3510](https://github.com/redis/go-redis/pull/3510)) + +## 🐛 Bug Fixes + +- fix: SetErr on Cmd if the command cannot be queued correctly in multi/exec ([#3509](https://github.com/redis/go-redis/pull/3509)) + +## 🧰 Maintenance + +- Updates release drafter config to exclude dependabot ([#3511](https://github.com/redis/go-redis/pull/3511)) +- chore(deps): bump actions/setup-go from 5 to 6 ([#3504](https://github.com/redis/go-redis/pull/3504)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@elena-kolevska](https://github.com/elena-kolevksa), [@htemelski-redis](https://github.com/htemelski-redis) and [@ndyakov](https://github.com/ndyakov) + + +# 9.13.0 (2025-09-03) + +## Highlights +- Pipeliner expose queued commands ([#3496](https://github.com/redis/go-redis/pull/3496)) +- Ensure that JSON.GET returns Nil response ([#3470](https://github.com/redis/go-redis/pull/3470)) +- Fixes on Read and Write buffer sizes and UniversalOptions + +## Changes +- Pipeliner expose queued commands ([#3496](https://github.com/redis/go-redis/pull/3496)) +- fix(test): fix a timing issue in pubsub test ([#3498](https://github.com/redis/go-redis/pull/3498)) +- Allow users to enable read-write splitting in failover mode. ([#3482](https://github.com/redis/go-redis/pull/3482)) +- Set the read/write buffer size of the sentinel client to 4KiB ([#3476](https://github.com/redis/go-redis/pull/3476)) + +## 🚀 New Features + +- fix(otel): register wait metrics ([#3499](https://github.com/redis/go-redis/pull/3499)) +- Support subscriptions against cluster slave nodes ([#3480](https://github.com/redis/go-redis/pull/3480)) +- Add wait metrics to otel ([#3493](https://github.com/redis/go-redis/pull/3493)) +- Clean failing timeout implementation ([#3472](https://github.com/redis/go-redis/pull/3472)) + +## 🐛 Bug Fixes + +- Do not assume that all non-IP hosts are loopbacks ([#3085](https://github.com/redis/go-redis/pull/3085)) +- Ensure that JSON.GET returns Nil response ([#3470](https://github.com/redis/go-redis/pull/3470)) + +## 🧰 Maintenance + +- fix(otel): register wait metrics ([#3499](https://github.com/redis/go-redis/pull/3499)) +- fix(make test): Add default env in makefile ([#3491](https://github.com/redis/go-redis/pull/3491)) +- Update the introduction to running tests in README.md ([#3495](https://github.com/redis/go-redis/pull/3495)) +- test: Add comprehensive edge case tests for IncrByFloat command ([#3477](https://github.com/redis/go-redis/pull/3477)) +- Set the default read/write buffer size of Redis connection to 32KiB ([#3483](https://github.com/redis/go-redis/pull/3483)) +- Bumps test image to 8.2.1-pre ([#3478](https://github.com/redis/go-redis/pull/3478)) +- fix UniversalOptions miss ReadBufferSize and WriteBufferSize options ([#3485](https://github.com/redis/go-redis/pull/3485)) +- chore(deps): bump actions/checkout from 4 to 5 ([#3484](https://github.com/redis/go-redis/pull/3484)) +- Removes dry run for stale issues policy ([#3471](https://github.com/redis/go-redis/pull/3471)) +- Update otel metrics URL ([#3474](https://github.com/redis/go-redis/pull/3474)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@LINKIWI](https://github.com/LINKIWI), [@cxljs](https://github.com/cxljs), [@cybersmeashish](https://github.com/cybersmeashish), [@elena-kolevska](https://github.com/elena-kolevska), [@htemelski-redis](https://github.com/htemelski-redis), [@mwhooker](https://github.com/mwhooker), [@ndyakov](https://github.com/ndyakov), [@ofekshenawa](https://github.com/ofekshenawa), [@suever](https://github.com/suever) + + +# 9.12.1 (2025-08-11) +## 🚀 Highlights +In the last version (9.12.0) the client introduced bigger write and read buffer sized. The default value we set was 512KiB. +However, users reported that this is too big for most use cases and can lead to high memory usage. +In this version the default value is changed to 256KiB. The `README.md` was updated to reflect the +correct default value and include a note that the default value can be changed. + +## 🐛 Bug Fixes + +- fix(options): Add buffer sizes to failover. Update README ([#3468](https://github.com/redis/go-redis/pull/3468)) + +## 🧰 Maintenance + +- fix(options): Add buffer sizes to failover. Update README ([#3468](https://github.com/redis/go-redis/pull/3468)) +- chore: update & fix otel example ([#3466](https://github.com/redis/go-redis/pull/3466)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@ndyakov](https://github.com/ndyakov) and [@vmihailenco](https://github.com/vmihailenco) + +# 9.12.0 (2025-08-05) + +## 🚀 Highlights + +- This release includes support for [Redis 8.2](https://redis.io/docs/latest/operate/oss_and_stack/stack-with-enterprise/release-notes/redisce/redisos-8.2-release-notes/). +- Introduces an experimental Query Builders for `FTSearch`, `FTAggregate` and other search commands. +- Adds support for `EPSILON` option in `FT.VSIM`. +- Includes bug fixes and improvements contributed by the community related to ring and [redisotel](https://github.com/redis/go-redis/tree/master/extra/redisotel). + +## Changes +- Improve stale issue workflow ([#3458](https://github.com/redis/go-redis/pull/3458)) +- chore(ci): Add 8.2 rc2 pre build for CI ([#3459](https://github.com/redis/go-redis/pull/3459)) +- Added new stream commands ([#3450](https://github.com/redis/go-redis/pull/3450)) +- feat: Add "skip_verify" to Sentinel ([#3428](https://github.com/redis/go-redis/pull/3428)) +- fix: `errors.Join` requires Go 1.20 or later ([#3442](https://github.com/redis/go-redis/pull/3442)) +- DOC-4344 document quickstart examples ([#3426](https://github.com/redis/go-redis/pull/3426)) +- feat(bitop): add support for the new bitop operations ([#3409](https://github.com/redis/go-redis/pull/3409)) + +## 🚀 New Features + +- feat: recover addIdleConn may occur panic ([#2445](https://github.com/redis/go-redis/pull/2445)) +- feat(ring): specify custom health check func via HeartbeatFn option ([#2940](https://github.com/redis/go-redis/pull/2940)) +- Add Query Builder for RediSearch commands ([#3436](https://github.com/redis/go-redis/pull/3436)) +- add configurable buffer sizes for Redis connections ([#3453](https://github.com/redis/go-redis/pull/3453)) +- Add VAMANA vector type to RediSearch ([#3449](https://github.com/redis/go-redis/pull/3449)) +- VSIM add `EPSILON` option ([#3454](https://github.com/redis/go-redis/pull/3454)) +- Add closing support to otel metrics instrumentation ([#3444](https://github.com/redis/go-redis/pull/3444)) + +## 🐛 Bug Fixes + +- fix(redisotel): fix buggy append in reportPoolStats ([#3122](https://github.com/redis/go-redis/pull/3122)) +- fix(search): return results even if doc is empty ([#3457](https://github.com/redis/go-redis/pull/3457)) +- [ISSUE-3402]: Ring.Pipelined return dial timeout error ([#3403](https://github.com/redis/go-redis/pull/3403)) + +## 🧰 Maintenance + +- Merges stale issues jobs into one job with two steps ([#3463](https://github.com/redis/go-redis/pull/3463)) +- improve code readability ([#3446](https://github.com/redis/go-redis/pull/3446)) +- chore(release): 9.12.0-beta.1 ([#3460](https://github.com/redis/go-redis/pull/3460)) +- DOC-5472 time series doc examples ([#3443](https://github.com/redis/go-redis/pull/3443)) +- Add VAMANA compression algorithm tests ([#3461](https://github.com/redis/go-redis/pull/3461)) +- bumped redis 8.2 version used in the CI/CD ([#3451](https://github.com/redis/go-redis/pull/3451)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@andy-stark-redis](https://github.com/andy-stark-redis), [@cxljs](https://github.com/cxljs), [@elena-kolevska](https://github.com/elena-kolevska), [@htemelski-redis](https://github.com/htemelski-redis), [@jouir](https://github.com/jouir), [@monkey92t](https://github.com/monkey92t), [@ndyakov](https://github.com/ndyakov), [@ofekshenawa](https://github.com/ofekshenawa), [@rokn](https://github.com/rokn), [@smnvdev](https://github.com/smnvdev), [@strobil](https://github.com/strobil) and [@wzy9607](https://github.com/wzy9607) + +## New Contributors +* [@htemelski-redis](https://github.com/htemelski-redis) made their first contribution in [#3409](https://github.com/redis/go-redis/pull/3409) +* [@smnvdev](https://github.com/smnvdev) made their first contribution in [#3403](https://github.com/redis/go-redis/pull/3403) +* [@rokn](https://github.com/rokn) made their first contribution in [#3444](https://github.com/redis/go-redis/pull/3444) + +# 9.11.0 (2025-06-24) + +## 🚀 Highlights + +Fixes TxPipeline to work correctly in cluster scenarios, allowing execution of commands +only in the same slot. + +# Changes + +## 🚀 New Features + +- Set cluster slot for `scan` commands, rather than random ([#2623](https://github.com/redis/go-redis/pull/2623)) +- Add CredentialsProvider field to UniversalOptions ([#2927](https://github.com/redis/go-redis/pull/2927)) +- feat(redisotel): add WithCallerEnabled option ([#3415](https://github.com/redis/go-redis/pull/3415)) + +## 🐛 Bug Fixes + +- fix(txpipeline): keyless commands should take the slot of the keyed ([#3411](https://github.com/redis/go-redis/pull/3411)) +- fix(loading): cache the loaded flag for slave nodes ([#3410](https://github.com/redis/go-redis/pull/3410)) +- fix(txpipeline): should return error on multi/exec on multiple slots ([#3408](https://github.com/redis/go-redis/pull/3408)) +- fix: check if the shard exists to avoid returning nil ([#3396](https://github.com/redis/go-redis/pull/3396)) + +## 🧰 Maintenance + +- feat: optimize connection pool waitTurn ([#3412](https://github.com/redis/go-redis/pull/3412)) +- chore(ci): update CI redis builds ([#3407](https://github.com/redis/go-redis/pull/3407)) +- chore: remove a redundant method from `Ring`, `Client` and `ClusterClient` ([#3401](https://github.com/redis/go-redis/pull/3401)) +- test: refactor TestBasicCredentials using table-driven tests ([#3406](https://github.com/redis/go-redis/pull/3406)) +- perf: reduce unnecessary memory allocation operations ([#3399](https://github.com/redis/go-redis/pull/3399)) +- fix: insert entry during iterating over a map ([#3398](https://github.com/redis/go-redis/pull/3398)) +- DOC-5229 probabilistic data type examples ([#3413](https://github.com/redis/go-redis/pull/3413)) +- chore(deps): bump rojopolis/spellcheck-github-actions from 0.49.0 to 0.51.0 ([#3414](https://github.com/redis/go-redis/pull/3414)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@andy-stark-redis](https://github.com/andy-stark-redis), [@boekkooi-impossiblecloud](https://github.com/boekkooi-impossiblecloud), [@cxljs](https://github.com/cxljs), [@dcherubini](https://github.com/dcherubini), [@dependabot[bot]](https://github.com/apps/dependabot), [@iamamirsalehi](https://github.com/iamamirsalehi), [@ndyakov](https://github.com/ndyakov), [@pete-woods](https://github.com/pete-woods), [@twz915](https://github.com/twz915) and [dependabot[bot]](https://github.com/apps/dependabot) + +# 9.10.0 (2025-06-06) + +## 🚀 Highlights + +`go-redis` now supports [vector sets](https://redis.io/docs/latest/develop/data-types/vector-sets/). This data type is marked +as "in preview" in Redis and its support in `go-redis` is marked as experimental. You can find examples in the documentation and +in the `doctests` folder. + +# Changes + +## 🚀 New Features + +- feat: support vectorset ([#3375](https://github.com/redis/go-redis/pull/3375)) + +## 🧰 Maintenance + +- Add the missing NewFloatSliceResult for testing ([#3393](https://github.com/redis/go-redis/pull/3393)) +- DOC-5078 vector set examples ([#3394](https://github.com/redis/go-redis/pull/3394)) + +## Contributors +We'd like to thank all the contributors who worked on this release! + +[@AndBobsYourUncle](https://github.com/AndBobsYourUncle), [@andy-stark-redis](https://github.com/andy-stark-redis), [@fukua95](https://github.com/fukua95) and [@ndyakov](https://github.com/ndyakov) + + + +# 9.9.0 (2025-05-27) + +## 🚀 Highlights +- **Token-based Authentication**: Added `StreamingCredentialsProvider` for dynamic credential updates (experimental) + - Can be used with [go-redis-entraid](https://github.com/redis/go-redis-entraid) for Azure AD authentication +- **Connection Statistics**: Added connection waiting statistics for better monitoring +- **Failover Improvements**: Added `ParseFailoverURL` for easier failover configuration +- **Ring Client Enhancements**: Added shard access methods for better Pub/Sub management + +## ✨ New Features +- Added `StreamingCredentialsProvider` for token-based authentication ([#3320](https://github.com/redis/go-redis/pull/3320)) + - Supports dynamic credential updates + - Includes connection close hooks + - Note: Currently marked as experimental +- Added `ParseFailoverURL` for parsing failover URLs ([#3362](https://github.com/redis/go-redis/pull/3362)) +- Added connection waiting statistics ([#2804](https://github.com/redis/go-redis/pull/2804)) +- Added new utility functions: + - `ParseFloat` and `MustParseFloat` in public utils package ([#3371](https://github.com/redis/go-redis/pull/3371)) + - Unit tests for `Atoi`, `ParseInt`, `ParseUint`, and `ParseFloat` ([#3377](https://github.com/redis/go-redis/pull/3377)) +- Added Ring client shard access methods: + - `GetShardClients()` to retrieve all active shard clients + - `GetShardClientForKey(key string)` to get the shard client for a specific key ([#3388](https://github.com/redis/go-redis/pull/3388)) + +## 🐛 Bug Fixes +- Fixed routing reads to loading slave nodes ([#3370](https://github.com/redis/go-redis/pull/3370)) +- Added support for nil lag in XINFO GROUPS ([#3369](https://github.com/redis/go-redis/pull/3369)) +- Fixed pool acquisition timeout issues ([#3381](https://github.com/redis/go-redis/pull/3381)) +- Optimized unnecessary copy operations ([#3376](https://github.com/redis/go-redis/pull/3376)) + +## 📚 Documentation +- Updated documentation for XINFO GROUPS with nil lag support ([#3369](https://github.com/redis/go-redis/pull/3369)) +- Added package-level comments for new features + +## ⚡ Performance and Reliability +- Optimized `ReplaceSpaces` function ([#3383](https://github.com/redis/go-redis/pull/3383)) +- Set default value for `Options.Protocol` in `init()` ([#3387](https://github.com/redis/go-redis/pull/3387)) +- Exported pool errors for public consumption ([#3380](https://github.com/redis/go-redis/pull/3380)) + +## 🔧 Dependencies and Infrastructure +- Updated Redis CI to version 8.0.1 ([#3372](https://github.com/redis/go-redis/pull/3372)) +- Updated spellcheck GitHub Actions ([#3389](https://github.com/redis/go-redis/pull/3389)) +- Removed unused parameters ([#3382](https://github.com/redis/go-redis/pull/3382), [#3384](https://github.com/redis/go-redis/pull/3384)) + +## 🧪 Testing +- Added unit tests for pool acquisition timeout ([#3381](https://github.com/redis/go-redis/pull/3381)) +- Added unit tests for utility functions ([#3377](https://github.com/redis/go-redis/pull/3377)) + +## 👥 Contributors + +We would like to thank all the contributors who made this release possible: + +[@ndyakov](https://github.com/ndyakov), [@ofekshenawa](https://github.com/ofekshenawa), [@LINKIWI](https://github.com/LINKIWI), [@iamamirsalehi](https://github.com/iamamirsalehi), [@fukua95](https://github.com/fukua95), [@lzakharov](https://github.com/lzakharov), [@DengY11](https://github.com/DengY11) + +## 📝 Changelog + +For a complete list of changes, see the [full changelog](https://github.com/redis/go-redis/compare/v9.8.0...v9.9.0). + +# 9.8.0 (2025-04-30) + +## 🚀 Highlights +- **Redis 8 Support**: Full compatibility with Redis 8.0, including testing and CI integration +- **Enhanced Hash Operations**: Added support for new hash commands (`HGETDEL`, `HGETEX`, `HSETEX`) and `HSTRLEN` command +- **Search Improvements**: Enabled Search DIALECT 2 by default and added `CountOnly` argument for `FT.Search` + +## ✨ New Features +- Added support for new hash commands: `HGETDEL`, `HGETEX`, `HSETEX` ([#3305](https://github.com/redis/go-redis/pull/3305)) +- Added `HSTRLEN` command for hash operations ([#2843](https://github.com/redis/go-redis/pull/2843)) +- Added `Do` method for raw query by single connection from `pool.Conn()` ([#3182](https://github.com/redis/go-redis/pull/3182)) +- Prevent false-positive marshaling by treating zero time.Time as empty in isEmptyValue ([#3273](https://github.com/redis/go-redis/pull/3273)) +- Added FailoverClusterClient support for Universal client ([#2794](https://github.com/redis/go-redis/pull/2794)) +- Added support for cluster mode with `IsClusterMode` config parameter ([#3255](https://github.com/redis/go-redis/pull/3255)) +- Added client name support in `HELLO` RESP handshake ([#3294](https://github.com/redis/go-redis/pull/3294)) +- **Enabled Search DIALECT 2 by default** ([#3213](https://github.com/redis/go-redis/pull/3213)) +- Added read-only option for failover configurations ([#3281](https://github.com/redis/go-redis/pull/3281)) +- Added `CountOnly` argument for `FT.Search` to use `LIMIT 0 0` ([#3338](https://github.com/redis/go-redis/pull/3338)) +- Added `DB` option support in `NewFailoverClusterClient` ([#3342](https://github.com/redis/go-redis/pull/3342)) +- Added `nil` check for the options when creating a client ([#3363](https://github.com/redis/go-redis/pull/3363)) + +## 🐛 Bug Fixes +- Fixed `PubSub` concurrency safety issues ([#3360](https://github.com/redis/go-redis/pull/3360)) +- Fixed panic caused when argument is `nil` ([#3353](https://github.com/redis/go-redis/pull/3353)) +- Improved error handling when fetching master node from sentinels ([#3349](https://github.com/redis/go-redis/pull/3349)) +- Fixed connection pool timeout issues and increased retries ([#3298](https://github.com/redis/go-redis/pull/3298)) +- Fixed context cancellation error leading to connection spikes on Primary instances ([#3190](https://github.com/redis/go-redis/pull/3190)) +- Fixed RedisCluster client to consider `MASTERDOWN` a retriable error ([#3164](https://github.com/redis/go-redis/pull/3164)) +- Fixed tracing to show complete commands instead of truncated versions ([#3290](https://github.com/redis/go-redis/pull/3290)) +- Fixed OpenTelemetry instrumentation to prevent multiple span reporting ([#3168](https://github.com/redis/go-redis/pull/3168)) +- Fixed `FT.Search` Limit argument and added `CountOnly` argument for limit 0 0 ([#3338](https://github.com/redis/go-redis/pull/3338)) +- Fixed missing command in interface ([#3344](https://github.com/redis/go-redis/pull/3344)) +- Fixed slot calculation for `COUNTKEYSINSLOT` command ([#3327](https://github.com/redis/go-redis/pull/3327)) +- Updated PubSub implementation with correct context ([#3329](https://github.com/redis/go-redis/pull/3329)) + +## 📚 Documentation +- Added hash search examples ([#3357](https://github.com/redis/go-redis/pull/3357)) +- Fixed documentation comments ([#3351](https://github.com/redis/go-redis/pull/3351)) +- Added `CountOnly` search example ([#3345](https://github.com/redis/go-redis/pull/3345)) +- Added examples for list commands: `LLEN`, `LPOP`, `LPUSH`, `LRANGE`, `RPOP`, `RPUSH` ([#3234](https://github.com/redis/go-redis/pull/3234)) +- Added `SADD` and `SMEMBERS` command examples ([#3242](https://github.com/redis/go-redis/pull/3242)) +- Updated `README.md` to use Redis Discord guild ([#3331](https://github.com/redis/go-redis/pull/3331)) +- Updated `HExpire` command documentation ([#3355](https://github.com/redis/go-redis/pull/3355)) +- Featured OpenTelemetry instrumentation more prominently ([#3316](https://github.com/redis/go-redis/pull/3316)) +- Updated `README.md` with additional information ([#310ce55](https://github.com/redis/go-redis/commit/310ce55)) + +## ⚡ Performance and Reliability +- Bound connection pool background dials to configured dial timeout ([#3089](https://github.com/redis/go-redis/pull/3089)) +- Ensured context isn't exhausted via concurrent query ([#3334](https://github.com/redis/go-redis/pull/3334)) + +## 🔧 Dependencies and Infrastructure +- Updated testing image to Redis 8.0-RC2 ([#3361](https://github.com/redis/go-redis/pull/3361)) +- Enabled CI for Redis CE 8.0 ([#3274](https://github.com/redis/go-redis/pull/3274)) +- Updated various dependencies: + - Bumped golangci/golangci-lint-action from 6.5.0 to 7.0.0 ([#3354](https://github.com/redis/go-redis/pull/3354)) + - Bumped rojopolis/spellcheck-github-actions ([#3336](https://github.com/redis/go-redis/pull/3336)) + - Bumped golang.org/x/net in example/otel ([#3308](https://github.com/redis/go-redis/pull/3308)) +- Migrated golangci-lint configuration to v2 format ([#3354](https://github.com/redis/go-redis/pull/3354)) + +## ⚠️ Breaking Changes +- **Enabled Search DIALECT 2 by default** ([#3213](https://github.com/redis/go-redis/pull/3213)) +- Dropped RedisGears (Triggers and Functions) support ([#3321](https://github.com/redis/go-redis/pull/3321)) +- Dropped FT.PROFILE command that was never enabled ([#3323](https://github.com/redis/go-redis/pull/3323)) + +## 🔒 Security +- Fixed network error handling on SETINFO (CVE-2025-29923) ([#3295](https://github.com/redis/go-redis/pull/3295)) + +## 🧪 Testing +- Added integration tests for Redis 8 behavior changes in Redis Search ([#3337](https://github.com/redis/go-redis/pull/3337)) +- Added vector types INT8 and UINT8 tests ([#3299](https://github.com/redis/go-redis/pull/3299)) +- Added test codes for search_commands.go ([#3285](https://github.com/redis/go-redis/pull/3285)) +- Fixed example test sorting ([#3292](https://github.com/redis/go-redis/pull/3292)) + +## 👥 Contributors + +We would like to thank all the contributors who made this release possible: + +[@alexander-menshchikov](https://github.com/alexander-menshchikov), [@EXPEbdodla](https://github.com/EXPEbdodla), [@afti](https://github.com/afti), [@dmaier-redislabs](https://github.com/dmaier-redislabs), [@four_leaf_clover](https://github.com/four_leaf_clover), [@alohaglenn](https://github.com/alohaglenn), [@gh73962](https://github.com/gh73962), [@justinmir](https://github.com/justinmir), [@LINKIWI](https://github.com/LINKIWI), [@liushuangbill](https://github.com/liushuangbill), [@golang88](https://github.com/golang88), [@gnpaone](https://github.com/gnpaone), [@ndyakov](https://github.com/ndyakov), [@nikolaydubina](https://github.com/nikolaydubina), [@oleglacto](https://github.com/oleglacto), [@andy-stark-redis](https://github.com/andy-stark-redis), [@rodneyosodo](https://github.com/rodneyosodo), [@dependabot](https://github.com/dependabot), [@rfyiamcool](https://github.com/rfyiamcool), [@frankxjkuang](https://github.com/frankxjkuang), [@fukua95](https://github.com/fukua95), [@soleymani-milad](https://github.com/soleymani-milad), [@ofekshenawa](https://github.com/ofekshenawa), [@khasanovbi](https://github.com/khasanovbi) + + +# Old Changelog +## Unreleased + +### Changed + +* `go-redis` won't skip span creation if the parent spans is not recording. ([#2980](https://github.com/redis/go-redis/issues/2980)) + Users can use the OpenTelemetry sampler to control the sampling behavior. + For instance, you can use the `ParentBased(NeverSample())` sampler from `go.opentelemetry.io/otel/sdk/trace` to keep + a similar behavior (drop orphan spans) of `go-redis` as before. + +## [9.0.5](https://github.com/redis/go-redis/compare/v9.0.4...v9.0.5) (2023-05-29) + + +### Features + +* Add ACL LOG ([#2536](https://github.com/redis/go-redis/issues/2536)) ([31ba855](https://github.com/redis/go-redis/commit/31ba855ddebc38fbcc69a75d9d4fb769417cf602)) +* add field protocol to setupClusterQueryParams ([#2600](https://github.com/redis/go-redis/issues/2600)) ([840c25c](https://github.com/redis/go-redis/commit/840c25cb6f320501886a82a5e75f47b491e46fbe)) +* add protocol option ([#2598](https://github.com/redis/go-redis/issues/2598)) ([3917988](https://github.com/redis/go-redis/commit/391798880cfb915c4660f6c3ba63e0c1a459e2af)) + + + +## [9.0.4](https://github.com/redis/go-redis/compare/v9.0.3...v9.0.4) (2023-05-01) + + +### Bug Fixes + +* reader float parser ([#2513](https://github.com/redis/go-redis/issues/2513)) ([46f2450](https://github.com/redis/go-redis/commit/46f245075e6e3a8bd8471f9ca67ea95fd675e241)) + + +### Features + +* add client info command ([#2483](https://github.com/redis/go-redis/issues/2483)) ([b8c7317](https://github.com/redis/go-redis/commit/b8c7317cc6af444603731f7017c602347c0ba61e)) +* no longer verify HELLO error messages ([#2515](https://github.com/redis/go-redis/issues/2515)) ([7b4f217](https://github.com/redis/go-redis/commit/7b4f2179cb5dba3d3c6b0c6f10db52b837c912c8)) +* read the structure to increase the judgment of the omitempty op… ([#2529](https://github.com/redis/go-redis/issues/2529)) ([37c057b](https://github.com/redis/go-redis/commit/37c057b8e597c5e8a0e372337f6a8ad27f6030af)) + + + +## [9.0.3](https://github.com/redis/go-redis/compare/v9.0.2...v9.0.3) (2023-04-02) + +### New Features + +- feat(scan): scan time.Time sets the default decoding (#2413) +- Add support for CLUSTER LINKS command (#2504) +- Add support for acl dryrun command (#2502) +- Add support for COMMAND GETKEYS & COMMAND GETKEYSANDFLAGS (#2500) +- Add support for LCS Command (#2480) +- Add support for BZMPOP (#2456) +- Adding support for ZMPOP command (#2408) +- Add support for LMPOP (#2440) +- feat: remove pool unused fields (#2438) +- Expiretime and PExpireTime (#2426) +- Implement `FUNCTION` group of commands (#2475) +- feat(zadd): add ZAddLT and ZAddGT (#2429) +- Add: Support for COMMAND LIST command (#2491) +- Add support for BLMPOP (#2442) +- feat: check pipeline.Do to prevent confusion with Exec (#2517) +- Function stats, function kill, fcall and fcall_ro (#2486) +- feat: Add support for CLUSTER SHARDS command (#2507) +- feat(cmd): support for adding byte,bit parameters to the bitpos command (#2498) + +### Fixed + +- fix: eval api cmd.SetFirstKeyPos (#2501) +- fix: limit the number of connections created (#2441) +- fixed #2462 v9 continue support dragonfly, it's Hello command return "NOAUTH Authentication required" error (#2479) +- Fix for internal/hscan/structmap.go:89:23: undefined: reflect.Pointer (#2458) +- fix: group lag can be null (#2448) + +### Maintenance + +- Updating to the latest version of redis (#2508) +- Allowing for running tests on a port other than the fixed 6380 (#2466) +- redis 7.0.8 in tests (#2450) +- docs: Update redisotel example for v9 (#2425) +- chore: update go mod, Upgrade golang.org/x/net version to 0.7.0 (#2476) +- chore: add Chinese translation (#2436) +- chore(deps): bump github.com/bsm/gomega from 1.20.0 to 1.26.0 (#2421) +- chore(deps): bump github.com/bsm/ginkgo/v2 from 2.5.0 to 2.7.0 (#2420) +- chore(deps): bump actions/setup-go from 3 to 4 (#2495) +- docs: add instructions for the HSet api (#2503) +- docs: add reading lag field comment (#2451) +- test: update go mod before testing(go mod tidy) (#2423) +- docs: fix comment typo (#2505) +- test: remove testify (#2463) +- refactor: change ListElementCmd to KeyValuesCmd. (#2443) +- fix(appendArg): appendArg case special type (#2489) + +## [9.0.2](https://github.com/redis/go-redis/compare/v9.0.1...v9.0.2) (2023-02-01) + +### Features + +* upgrade OpenTelemetry, use the new metrics API. ([#2410](https://github.com/redis/go-redis/issues/2410)) ([e29e42c](https://github.com/redis/go-redis/commit/e29e42cde2755ab910d04185025dc43ce6f59c65)) + +## v9 2023-01-30 + +### Breaking + +- Changed Pipelines to not be thread-safe any more. + +### Added + +- Added support for [RESP3](https://github.com/antirez/RESP3/blob/master/spec.md) protocol. It was + contributed by @monkey92t who has done the majority of work in this release. +- Added `ContextTimeoutEnabled` option that controls whether the client respects context timeouts + and deadlines. See + [Redis Timeouts](https://redis.uptrace.dev/guide/go-redis-debugging.html#timeouts) for details. +- Added `ParseClusterURL` to parse URLs into `ClusterOptions`, for example, + `redis://user:password@localhost:6789?dial_timeout=3&read_timeout=6s&addr=localhost:6790&addr=localhost:6791`. +- Added metrics instrumentation using `redisotel.IstrumentMetrics`. See + [documentation](https://redis.uptrace.dev/guide/go-redis-monitoring.html) +- Added `redis.HasErrorPrefix` to help working with errors. + +### Changed + +- Removed asynchronous cancellation based on the context timeout. It was racy in v8 and is + completely gone in v9. +- Reworked hook interface and added `DialHook`. +- Replaced `redisotel.NewTracingHook` with `redisotel.InstrumentTracing`. See + [example](example/otel) and + [documentation](https://redis.uptrace.dev/guide/go-redis-monitoring.html). +- Replaced `*redis.Z` with `redis.Z` since it is small enough to be passed as value without making + an allocation. +- Renamed the option `MaxConnAge` to `ConnMaxLifetime`. +- Renamed the option `IdleTimeout` to `ConnMaxIdleTime`. +- Removed connection reaper in favor of `MaxIdleConns`. +- Removed `WithContext` since `context.Context` can be passed directly as an arg. +- Removed `Pipeline.Close` since there is no real need to explicitly manage pipeline resources and + it can be safely reused via `sync.Pool` etc. `Pipeline.Discard` is still available if you want to + reset commands for some reason. + +### Fixed + +- Improved and fixed pipeline retries. +- As usually, added support for more commands and fixed some bugs. diff --git a/vendor/github.com/redis/go-redis/v9/RELEASING.md b/vendor/github.com/redis/go-redis/v9/RELEASING.md new file mode 100644 index 0000000..1115db4 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/RELEASING.md @@ -0,0 +1,15 @@ +# Releasing + +1. Run `release.sh` script which updates versions in go.mod files and pushes a new branch to GitHub: + +```shell +TAG=v1.0.0 ./scripts/release.sh +``` + +2. Open a pull request and wait for the build to finish. + +3. Merge the pull request and run `tag.sh` to create tags for packages: + +```shell +TAG=v1.0.0 ./scripts/tag.sh +``` diff --git a/vendor/github.com/redis/go-redis/v9/acl_commands.go b/vendor/github.com/redis/go-redis/v9/acl_commands.go new file mode 100644 index 0000000..9cb800b --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/acl_commands.go @@ -0,0 +1,89 @@ +package redis + +import "context" + +type ACLCmdable interface { + ACLDryRun(ctx context.Context, username string, command ...interface{}) *StringCmd + + ACLLog(ctx context.Context, count int64) *ACLLogCmd + ACLLogReset(ctx context.Context) *StatusCmd + + ACLSetUser(ctx context.Context, username string, rules ...string) *StatusCmd + ACLDelUser(ctx context.Context, username string) *IntCmd + ACLList(ctx context.Context) *StringSliceCmd + + ACLCat(ctx context.Context) *StringSliceCmd + ACLCatArgs(ctx context.Context, options *ACLCatArgs) *StringSliceCmd +} + +type ACLCatArgs struct { + Category string +} + +func (c cmdable) ACLDryRun(ctx context.Context, username string, command ...interface{}) *StringCmd { + args := make([]interface{}, 0, 3+len(command)) + args = append(args, "acl", "dryrun", username) + args = append(args, command...) + cmd := NewStringCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLLog(ctx context.Context, count int64) *ACLLogCmd { + args := make([]interface{}, 0, 3) + args = append(args, "acl", "log") + if count > 0 { + args = append(args, count) + } + cmd := NewACLLogCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLLogReset(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "acl", "log", "reset") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLDelUser(ctx context.Context, username string) *IntCmd { + cmd := NewIntCmd(ctx, "acl", "deluser", username) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLSetUser(ctx context.Context, username string, rules ...string) *StatusCmd { + args := make([]interface{}, 3+len(rules)) + args[0] = "acl" + args[1] = "setuser" + args[2] = username + for i, rule := range rules { + args[i+3] = rule + } + cmd := NewStatusCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLList(ctx context.Context) *StringSliceCmd { + cmd := NewStringSliceCmd(ctx, "acl", "list") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLCat(ctx context.Context) *StringSliceCmd { + cmd := NewStringSliceCmd(ctx, "acl", "cat") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ACLCatArgs(ctx context.Context, options *ACLCatArgs) *StringSliceCmd { + // if there is a category passed, build new cmd, if there isn't - use the ACLCat method + if options != nil && options.Category != "" { + cmd := NewStringSliceCmd(ctx, "acl", "cat", options.Category) + _ = c(ctx, cmd) + return cmd + } + + return c.ACLCat(ctx) +} diff --git a/vendor/github.com/redis/go-redis/v9/adapters.go b/vendor/github.com/redis/go-redis/v9/adapters.go new file mode 100644 index 0000000..4146153 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/adapters.go @@ -0,0 +1,111 @@ +package redis + +import ( + "context" + "errors" + "net" + "time" + + "github.com/redis/go-redis/v9/internal/interfaces" + "github.com/redis/go-redis/v9/push" +) + +// ErrInvalidCommand is returned when an invalid command is passed to ExecuteCommand. +var ErrInvalidCommand = errors.New("invalid command type") + +// ErrInvalidPool is returned when the pool type is not supported. +var ErrInvalidPool = errors.New("invalid pool type") + +// newClientAdapter creates a new client adapter for regular Redis clients. +func newClientAdapter(client *baseClient) interfaces.ClientInterface { + return &clientAdapter{client: client} +} + +// clientAdapter adapts a Redis client to implement interfaces.ClientInterface. +type clientAdapter struct { + client *baseClient +} + +// GetOptions returns the client options. +func (ca *clientAdapter) GetOptions() interfaces.OptionsInterface { + return &optionsAdapter{options: ca.client.opt} +} + +// GetPushProcessor returns the client's push notification processor. +func (ca *clientAdapter) GetPushProcessor() interfaces.NotificationProcessor { + return &pushProcessorAdapter{processor: ca.client.pushProcessor} +} + +// optionsAdapter adapts Redis options to implement interfaces.OptionsInterface. +type optionsAdapter struct { + options *Options +} + +// GetReadTimeout returns the read timeout. +func (oa *optionsAdapter) GetReadTimeout() time.Duration { + return oa.options.ReadTimeout +} + +// GetWriteTimeout returns the write timeout. +func (oa *optionsAdapter) GetWriteTimeout() time.Duration { + return oa.options.WriteTimeout +} + +// GetNetwork returns the network type. +func (oa *optionsAdapter) GetNetwork() string { + return oa.options.Network +} + +// GetAddr returns the connection address. +func (oa *optionsAdapter) GetAddr() string { + return oa.options.Addr +} + +// IsTLSEnabled returns true if TLS is enabled. +func (oa *optionsAdapter) IsTLSEnabled() bool { + return oa.options.TLSConfig != nil +} + +// GetProtocol returns the protocol version. +func (oa *optionsAdapter) GetProtocol() int { + return oa.options.Protocol +} + +// GetPoolSize returns the connection pool size. +func (oa *optionsAdapter) GetPoolSize() int { + return oa.options.PoolSize +} + +// NewDialer returns a new dialer function for the connection. +func (oa *optionsAdapter) NewDialer() func(context.Context) (net.Conn, error) { + baseDialer := oa.options.NewDialer() + return func(ctx context.Context) (net.Conn, error) { + // Extract network and address from the options + network := oa.options.Network + addr := oa.options.Addr + return baseDialer(ctx, network, addr) + } +} + +// pushProcessorAdapter adapts a push.NotificationProcessor to implement interfaces.NotificationProcessor. +type pushProcessorAdapter struct { + processor push.NotificationProcessor +} + +// RegisterHandler registers a handler for a specific push notification name. +func (ppa *pushProcessorAdapter) RegisterHandler(pushNotificationName string, handler interface{}, protected bool) error { + if pushHandler, ok := handler.(push.NotificationHandler); ok { + return ppa.processor.RegisterHandler(pushNotificationName, pushHandler, protected) + } + return errors.New("handler must implement push.NotificationHandler") +} + +// UnregisterHandler removes a handler for a specific push notification name. +func (ppa *pushProcessorAdapter) UnregisterHandler(pushNotificationName string) error { + return ppa.processor.UnregisterHandler(pushNotificationName) +} + +// GetHandler returns the handler for a specific push notification name. +func (ppa *pushProcessorAdapter) GetHandler(pushNotificationName string) interface{} { + return ppa.processor.GetHandler(pushNotificationName) +} diff --git a/vendor/github.com/redis/go-redis/v9/auth/auth.go b/vendor/github.com/redis/go-redis/v9/auth/auth.go new file mode 100644 index 0000000..1f5c802 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/auth/auth.go @@ -0,0 +1,61 @@ +// Package auth package provides authentication-related interfaces and types. +// It also includes a basic implementation of credentials using username and password. +package auth + +// StreamingCredentialsProvider is an interface that defines the methods for a streaming credentials provider. +// It is used to provide credentials for authentication. +// The CredentialsListener is used to receive updates when the credentials change. +type StreamingCredentialsProvider interface { + // Subscribe subscribes to the credentials provider for updates. + // It returns the current credentials, a cancel function to unsubscribe from the provider, + // and an error if any. + // TODO(ndyakov): Should we add context to the Subscribe method? + Subscribe(listener CredentialsListener) (Credentials, UnsubscribeFunc, error) +} + +// UnsubscribeFunc is a function that is used to cancel the subscription to the credentials provider. +// It is used to unsubscribe from the provider when the credentials are no longer needed. +type UnsubscribeFunc func() error + +// CredentialsListener is an interface that defines the methods for a credentials listener. +// It is used to receive updates when the credentials change. +// The OnNext method is called when the credentials change. +// The OnError method is called when an error occurs while requesting the credentials. +type CredentialsListener interface { + OnNext(credentials Credentials) + OnError(err error) +} + +// Credentials is an interface that defines the methods for credentials. +// It is used to provide the credentials for authentication. +type Credentials interface { + // BasicAuth returns the username and password for basic authentication. + BasicAuth() (username string, password string) + // RawCredentials returns the raw credentials as a string. + // This can be used to extract the username and password from the raw credentials or + // additional information if present in the token. + RawCredentials() string +} + +type basicAuth struct { + username string + password string +} + +// RawCredentials returns the raw credentials as a string. +func (b *basicAuth) RawCredentials() string { + return b.username + ":" + b.password +} + +// BasicAuth returns the username and password for basic authentication. +func (b *basicAuth) BasicAuth() (username string, password string) { + return b.username, b.password +} + +// NewBasicCredentials creates a new Credentials object from the given username and password. +func NewBasicCredentials(username, password string) Credentials { + return &basicAuth{ + username: username, + password: password, + } +} diff --git a/vendor/github.com/redis/go-redis/v9/auth/reauth_credentials_listener.go b/vendor/github.com/redis/go-redis/v9/auth/reauth_credentials_listener.go new file mode 100644 index 0000000..f4b3198 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/auth/reauth_credentials_listener.go @@ -0,0 +1,47 @@ +package auth + +// ReAuthCredentialsListener is a struct that implements the CredentialsListener interface. +// It is used to re-authenticate the credentials when they are updated. +// It contains: +// - reAuth: a function that takes the new credentials and returns an error if any. +// - onErr: a function that takes an error and handles it. +type ReAuthCredentialsListener struct { + reAuth func(credentials Credentials) error + onErr func(err error) +} + +// OnNext is called when the credentials are updated. +// It calls the reAuth function with the new credentials. +// If the reAuth function returns an error, it calls the onErr function with the error. +func (c *ReAuthCredentialsListener) OnNext(credentials Credentials) { + if c.reAuth == nil { + return + } + + err := c.reAuth(credentials) + if err != nil { + c.OnError(err) + } +} + +// OnError is called when an error occurs. +// It can be called from both the credentials provider and the reAuth function. +func (c *ReAuthCredentialsListener) OnError(err error) { + if c.onErr == nil { + return + } + + c.onErr(err) +} + +// NewReAuthCredentialsListener creates a new ReAuthCredentialsListener. +// Implements the auth.CredentialsListener interface. +func NewReAuthCredentialsListener(reAuth func(credentials Credentials) error, onErr func(err error)) *ReAuthCredentialsListener { + return &ReAuthCredentialsListener{ + reAuth: reAuth, + onErr: onErr, + } +} + +// Ensure ReAuthCredentialsListener implements the CredentialsListener interface. +var _ CredentialsListener = (*ReAuthCredentialsListener)(nil) \ No newline at end of file diff --git a/vendor/github.com/redis/go-redis/v9/bitmap_commands.go b/vendor/github.com/redis/go-redis/v9/bitmap_commands.go new file mode 100644 index 0000000..4dbc862 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/bitmap_commands.go @@ -0,0 +1,193 @@ +package redis + +import ( + "context" + "errors" +) + +type BitMapCmdable interface { + GetBit(ctx context.Context, key string, offset int64) *IntCmd + SetBit(ctx context.Context, key string, offset int64, value int) *IntCmd + BitCount(ctx context.Context, key string, bitCount *BitCount) *IntCmd + BitOpAnd(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpOr(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpXor(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpDiff(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpDiff1(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpAndOr(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpOne(ctx context.Context, destKey string, keys ...string) *IntCmd + BitOpNot(ctx context.Context, destKey string, key string) *IntCmd + BitPos(ctx context.Context, key string, bit int64, pos ...int64) *IntCmd + BitPosSpan(ctx context.Context, key string, bit int8, start, end int64, span string) *IntCmd + BitField(ctx context.Context, key string, values ...interface{}) *IntSliceCmd + BitFieldRO(ctx context.Context, key string, values ...interface{}) *IntSliceCmd +} + +func (c cmdable) GetBit(ctx context.Context, key string, offset int64) *IntCmd { + cmd := NewIntCmd(ctx, "getbit", key, offset) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) SetBit(ctx context.Context, key string, offset int64, value int) *IntCmd { + cmd := NewIntCmd( + ctx, + "setbit", + key, + offset, + value, + ) + _ = c(ctx, cmd) + return cmd +} + +type BitCount struct { + Start, End int64 + Unit string // BYTE(default) | BIT +} + +const BitCountIndexByte string = "BYTE" +const BitCountIndexBit string = "BIT" + +func (c cmdable) BitCount(ctx context.Context, key string, bitCount *BitCount) *IntCmd { + args := make([]any, 2, 5) + args[0] = "bitcount" + args[1] = key + if bitCount != nil { + args = append(args, bitCount.Start, bitCount.End) + if bitCount.Unit != "" { + if bitCount.Unit != BitCountIndexByte && bitCount.Unit != BitCountIndexBit { + cmd := NewIntCmd(ctx) + cmd.SetErr(errors.New("redis: invalid bitcount index")) + return cmd + } + args = append(args, bitCount.Unit) + } + } + cmd := NewIntCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) bitOp(ctx context.Context, op, destKey string, keys ...string) *IntCmd { + args := make([]interface{}, 3+len(keys)) + args[0] = "bitop" + args[1] = op + args[2] = destKey + for i, key := range keys { + args[3+i] = key + } + cmd := NewIntCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +// BitOpAnd creates a new bitmap in which users are members of all given bitmaps +func (c cmdable) BitOpAnd(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "and", destKey, keys...) +} + +// BitOpOr creates a new bitmap in which users are member of at least one given bitmap +func (c cmdable) BitOpOr(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "or", destKey, keys...) +} + +// BitOpXor creates a new bitmap in which users are the result of XORing all given bitmaps +func (c cmdable) BitOpXor(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "xor", destKey, keys...) +} + +// BitOpNot creates a new bitmap in which users are not members of a given bitmap +func (c cmdable) BitOpNot(ctx context.Context, destKey string, key string) *IntCmd { + return c.bitOp(ctx, "not", destKey, key) +} + +// BitOpDiff creates a new bitmap in which users are members of bitmap X but not of any of bitmaps Y1, Y2, … +// Introduced with Redis 8.2 +func (c cmdable) BitOpDiff(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "diff", destKey, keys...) +} + +// BitOpDiff1 creates a new bitmap in which users are members of one or more of bitmaps Y1, Y2, … but not members of bitmap X +// Introduced with Redis 8.2 +func (c cmdable) BitOpDiff1(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "diff1", destKey, keys...) +} + +// BitOpAndOr creates a new bitmap in which users are members of bitmap X and also members of one or more of bitmaps Y1, Y2, … +// Introduced with Redis 8.2 +func (c cmdable) BitOpAndOr(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "andor", destKey, keys...) +} + +// BitOpOne creates a new bitmap in which users are members of exactly one of the given bitmaps +// Introduced with Redis 8.2 +func (c cmdable) BitOpOne(ctx context.Context, destKey string, keys ...string) *IntCmd { + return c.bitOp(ctx, "one", destKey, keys...) +} + +// BitPos is an API before Redis version 7.0, cmd: bitpos key bit start end +// if you need the `byte | bit` parameter, please use `BitPosSpan`. +func (c cmdable) BitPos(ctx context.Context, key string, bit int64, pos ...int64) *IntCmd { + args := make([]interface{}, 3+len(pos)) + args[0] = "bitpos" + args[1] = key + args[2] = bit + switch len(pos) { + case 0: + case 1: + args[3] = pos[0] + case 2: + args[3] = pos[0] + args[4] = pos[1] + default: + panic("too many arguments") + } + cmd := NewIntCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +// BitPosSpan supports the `byte | bit` parameters in redis version 7.0, +// the bitpos command defaults to using byte type for the `start-end` range, +// which means it counts in bytes from start to end. you can set the value +// of "span" to determine the type of `start-end`. +// span = "bit", cmd: bitpos key bit start end bit +// span = "byte", cmd: bitpos key bit start end byte +func (c cmdable) BitPosSpan(ctx context.Context, key string, bit int8, start, end int64, span string) *IntCmd { + cmd := NewIntCmd(ctx, "bitpos", key, bit, start, end, span) + _ = c(ctx, cmd) + return cmd +} + +// BitField accepts multiple values: +// - BitField("set", "i1", "offset1", "value1","cmd2", "type2", "offset2", "value2") +// - BitField([]string{"cmd1", "type1", "offset1", "value1","cmd2", "type2", "offset2", "value2"}) +// - BitField([]interface{}{"cmd1", "type1", "offset1", "value1","cmd2", "type2", "offset2", "value2"}) +func (c cmdable) BitField(ctx context.Context, key string, values ...interface{}) *IntSliceCmd { + args := make([]interface{}, 2, 2+len(values)) + args[0] = "bitfield" + args[1] = key + args = appendArgs(args, values) + cmd := NewIntSliceCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +// BitFieldRO - Read-only variant of the BITFIELD command. +// It is like the original BITFIELD but only accepts GET subcommand and can safely be used in read-only replicas. +// - BitFieldRO(ctx, key, "", "", "","") +func (c cmdable) BitFieldRO(ctx context.Context, key string, values ...interface{}) *IntSliceCmd { + args := make([]interface{}, 2, 2+len(values)) + args[0] = "BITFIELD_RO" + args[1] = key + if len(values)%2 != 0 { + panic("BitFieldRO: invalid number of arguments, must be even") + } + for i := 0; i < len(values); i += 2 { + args = append(args, "GET", values[i], values[i+1]) + } + cmd := NewIntSliceCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} diff --git a/vendor/github.com/redis/go-redis/v9/cluster_commands.go b/vendor/github.com/redis/go-redis/v9/cluster_commands.go new file mode 100644 index 0000000..4857b01 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/cluster_commands.go @@ -0,0 +1,199 @@ +package redis + +import "context" + +type ClusterCmdable interface { + ClusterMyShardID(ctx context.Context) *StringCmd + ClusterMyID(ctx context.Context) *StringCmd + ClusterSlots(ctx context.Context) *ClusterSlotsCmd + ClusterShards(ctx context.Context) *ClusterShardsCmd + ClusterLinks(ctx context.Context) *ClusterLinksCmd + ClusterNodes(ctx context.Context) *StringCmd + ClusterMeet(ctx context.Context, host, port string) *StatusCmd + ClusterForget(ctx context.Context, nodeID string) *StatusCmd + ClusterReplicate(ctx context.Context, nodeID string) *StatusCmd + ClusterResetSoft(ctx context.Context) *StatusCmd + ClusterResetHard(ctx context.Context) *StatusCmd + ClusterInfo(ctx context.Context) *StringCmd + ClusterKeySlot(ctx context.Context, key string) *IntCmd + ClusterGetKeysInSlot(ctx context.Context, slot int, count int) *StringSliceCmd + ClusterCountFailureReports(ctx context.Context, nodeID string) *IntCmd + ClusterCountKeysInSlot(ctx context.Context, slot int) *IntCmd + ClusterDelSlots(ctx context.Context, slots ...int) *StatusCmd + ClusterDelSlotsRange(ctx context.Context, min, max int) *StatusCmd + ClusterSaveConfig(ctx context.Context) *StatusCmd + ClusterSlaves(ctx context.Context, nodeID string) *StringSliceCmd + ClusterFailover(ctx context.Context) *StatusCmd + ClusterAddSlots(ctx context.Context, slots ...int) *StatusCmd + ClusterAddSlotsRange(ctx context.Context, min, max int) *StatusCmd + ReadOnly(ctx context.Context) *StatusCmd + ReadWrite(ctx context.Context) *StatusCmd +} + +func (c cmdable) ClusterMyShardID(ctx context.Context) *StringCmd { + cmd := NewStringCmd(ctx, "cluster", "myshardid") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterMyID(ctx context.Context) *StringCmd { + cmd := NewStringCmd(ctx, "cluster", "myid") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterSlots(ctx context.Context) *ClusterSlotsCmd { + cmd := NewClusterSlotsCmd(ctx, "cluster", "slots") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterShards(ctx context.Context) *ClusterShardsCmd { + cmd := NewClusterShardsCmd(ctx, "cluster", "shards") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterLinks(ctx context.Context) *ClusterLinksCmd { + cmd := NewClusterLinksCmd(ctx, "cluster", "links") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterNodes(ctx context.Context) *StringCmd { + cmd := NewStringCmd(ctx, "cluster", "nodes") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterMeet(ctx context.Context, host, port string) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "meet", host, port) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterForget(ctx context.Context, nodeID string) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "forget", nodeID) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterReplicate(ctx context.Context, nodeID string) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "replicate", nodeID) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterResetSoft(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "reset", "soft") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterResetHard(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "reset", "hard") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterInfo(ctx context.Context) *StringCmd { + cmd := NewStringCmd(ctx, "cluster", "info") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterKeySlot(ctx context.Context, key string) *IntCmd { + cmd := NewIntCmd(ctx, "cluster", "keyslot", key) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterGetKeysInSlot(ctx context.Context, slot int, count int) *StringSliceCmd { + cmd := NewStringSliceCmd(ctx, "cluster", "getkeysinslot", slot, count) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterCountFailureReports(ctx context.Context, nodeID string) *IntCmd { + cmd := NewIntCmd(ctx, "cluster", "count-failure-reports", nodeID) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterCountKeysInSlot(ctx context.Context, slot int) *IntCmd { + cmd := NewIntCmd(ctx, "cluster", "countkeysinslot", slot) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterDelSlots(ctx context.Context, slots ...int) *StatusCmd { + args := make([]interface{}, 2+len(slots)) + args[0] = "cluster" + args[1] = "delslots" + for i, slot := range slots { + args[2+i] = slot + } + cmd := NewStatusCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterDelSlotsRange(ctx context.Context, min, max int) *StatusCmd { + size := max - min + 1 + slots := make([]int, size) + for i := 0; i < size; i++ { + slots[i] = min + i + } + return c.ClusterDelSlots(ctx, slots...) +} + +func (c cmdable) ClusterSaveConfig(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "saveconfig") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterSlaves(ctx context.Context, nodeID string) *StringSliceCmd { + cmd := NewStringSliceCmd(ctx, "cluster", "slaves", nodeID) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterFailover(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "cluster", "failover") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterAddSlots(ctx context.Context, slots ...int) *StatusCmd { + args := make([]interface{}, 2+len(slots)) + args[0] = "cluster" + args[1] = "addslots" + for i, num := range slots { + args[2+i] = num + } + cmd := NewStatusCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClusterAddSlotsRange(ctx context.Context, min, max int) *StatusCmd { + size := max - min + 1 + slots := make([]int, size) + for i := 0; i < size; i++ { + slots[i] = min + i + } + return c.ClusterAddSlots(ctx, slots...) +} + +func (c cmdable) ReadOnly(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "readonly") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ReadWrite(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "readwrite") + _ = c(ctx, cmd) + return cmd +} diff --git a/vendor/github.com/redis/go-redis/v9/command.go b/vendor/github.com/redis/go-redis/v9/command.go new file mode 100644 index 0000000..d3fb231 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/command.go @@ -0,0 +1,5745 @@ +package redis + +import ( + "bufio" + "context" + "fmt" + "net" + "regexp" + "strconv" + "strings" + "sync" + "time" + + "github.com/redis/go-redis/v9/internal" + "github.com/redis/go-redis/v9/internal/hscan" + "github.com/redis/go-redis/v9/internal/proto" + "github.com/redis/go-redis/v9/internal/util" +) + +// keylessCommands contains Redis commands that have empty key specifications (9th slot empty) +// Only includes core Redis commands, excludes FT.*, ts.*, timeseries.*, search.* and subcommands +var keylessCommands = map[string]struct{}{ + "acl": {}, + "asking": {}, + "auth": {}, + "bgrewriteaof": {}, + "bgsave": {}, + "client": {}, + "cluster": {}, + "config": {}, + "debug": {}, + "discard": {}, + "echo": {}, + "exec": {}, + "failover": {}, + "function": {}, + "hello": {}, + "latency": {}, + "lolwut": {}, + "module": {}, + "monitor": {}, + "multi": {}, + "pfselftest": {}, + "ping": {}, + "psubscribe": {}, + "psync": {}, + "publish": {}, + "pubsub": {}, + "punsubscribe": {}, + "quit": {}, + "readonly": {}, + "readwrite": {}, + "replconf": {}, + "replicaof": {}, + "role": {}, + "save": {}, + "script": {}, + "select": {}, + "shutdown": {}, + "slaveof": {}, + "slowlog": {}, + "subscribe": {}, + "swapdb": {}, + "sync": {}, + "unsubscribe": {}, + "unwatch": {}, +} + +type Cmder interface { + // command name. + // e.g. "set k v ex 10" -> "set", "cluster info" -> "cluster". + Name() string + + // full command name. + // e.g. "set k v ex 10" -> "set", "cluster info" -> "cluster info". + FullName() string + + // all args of the command. + // e.g. "set k v ex 10" -> "[set k v ex 10]". + Args() []interface{} + + // format request and response string. + // e.g. "set k v ex 10" -> "set k v ex 10: OK", "get k" -> "get k: v". + String() string + + stringArg(int) string + firstKeyPos() int8 + SetFirstKeyPos(int8) + + readTimeout() *time.Duration + readReply(rd *proto.Reader) error + readRawReply(rd *proto.Reader) error + SetErr(error) + Err() error +} + +func setCmdsErr(cmds []Cmder, e error) { + for _, cmd := range cmds { + if cmd.Err() == nil { + cmd.SetErr(e) + } + } +} + +func cmdsFirstErr(cmds []Cmder) error { + for _, cmd := range cmds { + if err := cmd.Err(); err != nil { + return err + } + } + return nil +} + +func writeCmds(wr *proto.Writer, cmds []Cmder) error { + for _, cmd := range cmds { + if err := writeCmd(wr, cmd); err != nil { + return err + } + } + return nil +} + +func writeCmd(wr *proto.Writer, cmd Cmder) error { + return wr.WriteArgs(cmd.Args()) +} + +// cmdFirstKeyPos returns the position of the first key in the command's arguments. +// If the command does not have a key, it returns 0. +// TODO: Use the data in CommandInfo to determine the first key position. +func cmdFirstKeyPos(cmd Cmder) int { + if pos := cmd.firstKeyPos(); pos != 0 { + return int(pos) + } + + name := cmd.Name() + + // first check if the command is keyless + if _, ok := keylessCommands[name]; ok { + return 0 + } + + switch name { + case "eval", "evalsha", "eval_ro", "evalsha_ro": + if cmd.stringArg(2) != "0" { + return 3 + } + + return 0 + case "publish": + return 1 + case "memory": + // https://github.com/redis/redis/issues/7493 + if cmd.stringArg(1) == "usage" { + return 2 + } + } + return 1 +} + +func cmdString(cmd Cmder, val interface{}) string { + b := make([]byte, 0, 64) + + for i, arg := range cmd.Args() { + if i > 0 { + b = append(b, ' ') + } + b = internal.AppendArg(b, arg) + } + + if err := cmd.Err(); err != nil { + b = append(b, ": "...) + b = append(b, err.Error()...) + } else if val != nil { + b = append(b, ": "...) + b = internal.AppendArg(b, val) + } + + return util.BytesToString(b) +} + +//------------------------------------------------------------------------------ + +type baseCmd struct { + ctx context.Context + args []interface{} + err error + keyPos int8 + rawVal interface{} + _readTimeout *time.Duration +} + +var _ Cmder = (*Cmd)(nil) + +func (cmd *baseCmd) Name() string { + if len(cmd.args) == 0 { + return "" + } + // Cmd name must be lower cased. + return internal.ToLower(cmd.stringArg(0)) +} + +func (cmd *baseCmd) FullName() string { + switch name := cmd.Name(); name { + case "cluster", "command": + if len(cmd.args) == 1 { + return name + } + if s2, ok := cmd.args[1].(string); ok { + return name + " " + s2 + } + return name + default: + return name + } +} + +func (cmd *baseCmd) Args() []interface{} { + return cmd.args +} + +func (cmd *baseCmd) stringArg(pos int) string { + if pos < 0 || pos >= len(cmd.args) { + return "" + } + arg := cmd.args[pos] + switch v := arg.(type) { + case string: + return v + case []byte: + return string(v) + default: + // TODO: consider using appendArg + return fmt.Sprint(v) + } +} + +func (cmd *baseCmd) firstKeyPos() int8 { + return cmd.keyPos +} + +func (cmd *baseCmd) SetFirstKeyPos(keyPos int8) { + cmd.keyPos = keyPos +} + +func (cmd *baseCmd) SetErr(e error) { + cmd.err = e +} + +func (cmd *baseCmd) Err() error { + return cmd.err +} + +func (cmd *baseCmd) readTimeout() *time.Duration { + return cmd._readTimeout +} + +func (cmd *baseCmd) setReadTimeout(d time.Duration) { + cmd._readTimeout = &d +} + +func (cmd *baseCmd) readRawReply(rd *proto.Reader) (err error) { + cmd.rawVal, err = rd.ReadReply() + return err +} + +//------------------------------------------------------------------------------ + +type Cmd struct { + baseCmd + + val interface{} +} + +func NewCmd(ctx context.Context, args ...interface{}) *Cmd { + return &Cmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *Cmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *Cmd) SetVal(val interface{}) { + cmd.val = val +} + +func (cmd *Cmd) Val() interface{} { + return cmd.val +} + +func (cmd *Cmd) Result() (interface{}, error) { + return cmd.val, cmd.err +} + +func (cmd *Cmd) Text() (string, error) { + if cmd.err != nil { + return "", cmd.err + } + return toString(cmd.val) +} + +func toString(val interface{}) (string, error) { + switch val := val.(type) { + case string: + return val, nil + default: + err := fmt.Errorf("redis: unexpected type=%T for String", val) + return "", err + } +} + +func (cmd *Cmd) Int() (int, error) { + if cmd.err != nil { + return 0, cmd.err + } + switch val := cmd.val.(type) { + case int64: + return int(val), nil + case string: + return strconv.Atoi(val) + default: + err := fmt.Errorf("redis: unexpected type=%T for Int", val) + return 0, err + } +} + +func (cmd *Cmd) Int64() (int64, error) { + if cmd.err != nil { + return 0, cmd.err + } + return toInt64(cmd.val) +} + +func toInt64(val interface{}) (int64, error) { + switch val := val.(type) { + case int64: + return val, nil + case string: + return strconv.ParseInt(val, 10, 64) + default: + err := fmt.Errorf("redis: unexpected type=%T for Int64", val) + return 0, err + } +} + +func (cmd *Cmd) Uint64() (uint64, error) { + if cmd.err != nil { + return 0, cmd.err + } + return toUint64(cmd.val) +} + +func toUint64(val interface{}) (uint64, error) { + switch val := val.(type) { + case int64: + return uint64(val), nil + case string: + return strconv.ParseUint(val, 10, 64) + default: + err := fmt.Errorf("redis: unexpected type=%T for Uint64", val) + return 0, err + } +} + +func (cmd *Cmd) Float32() (float32, error) { + if cmd.err != nil { + return 0, cmd.err + } + return toFloat32(cmd.val) +} + +func toFloat32(val interface{}) (float32, error) { + switch val := val.(type) { + case int64: + return float32(val), nil + case string: + f, err := strconv.ParseFloat(val, 32) + if err != nil { + return 0, err + } + return float32(f), nil + default: + err := fmt.Errorf("redis: unexpected type=%T for Float32", val) + return 0, err + } +} + +func (cmd *Cmd) Float64() (float64, error) { + if cmd.err != nil { + return 0, cmd.err + } + return toFloat64(cmd.val) +} + +func toFloat64(val interface{}) (float64, error) { + switch val := val.(type) { + case int64: + return float64(val), nil + case string: + return strconv.ParseFloat(val, 64) + default: + err := fmt.Errorf("redis: unexpected type=%T for Float64", val) + return 0, err + } +} + +func (cmd *Cmd) Bool() (bool, error) { + if cmd.err != nil { + return false, cmd.err + } + return toBool(cmd.val) +} + +func toBool(val interface{}) (bool, error) { + switch val := val.(type) { + case bool: + return val, nil + case int64: + return val != 0, nil + case string: + return strconv.ParseBool(val) + default: + err := fmt.Errorf("redis: unexpected type=%T for Bool", val) + return false, err + } +} + +func (cmd *Cmd) Slice() ([]interface{}, error) { + if cmd.err != nil { + return nil, cmd.err + } + switch val := cmd.val.(type) { + case []interface{}: + return val, nil + default: + return nil, fmt.Errorf("redis: unexpected type=%T for Slice", val) + } +} + +func (cmd *Cmd) StringSlice() ([]string, error) { + slice, err := cmd.Slice() + if err != nil { + return nil, err + } + + ss := make([]string, len(slice)) + for i, iface := range slice { + val, err := toString(iface) + if err != nil { + return nil, err + } + ss[i] = val + } + return ss, nil +} + +func (cmd *Cmd) Int64Slice() ([]int64, error) { + slice, err := cmd.Slice() + if err != nil { + return nil, err + } + + nums := make([]int64, len(slice)) + for i, iface := range slice { + val, err := toInt64(iface) + if err != nil { + return nil, err + } + nums[i] = val + } + return nums, nil +} + +func (cmd *Cmd) Uint64Slice() ([]uint64, error) { + slice, err := cmd.Slice() + if err != nil { + return nil, err + } + + nums := make([]uint64, len(slice)) + for i, iface := range slice { + val, err := toUint64(iface) + if err != nil { + return nil, err + } + nums[i] = val + } + return nums, nil +} + +func (cmd *Cmd) Float32Slice() ([]float32, error) { + slice, err := cmd.Slice() + if err != nil { + return nil, err + } + + floats := make([]float32, len(slice)) + for i, iface := range slice { + val, err := toFloat32(iface) + if err != nil { + return nil, err + } + floats[i] = val + } + return floats, nil +} + +func (cmd *Cmd) Float64Slice() ([]float64, error) { + slice, err := cmd.Slice() + if err != nil { + return nil, err + } + + floats := make([]float64, len(slice)) + for i, iface := range slice { + val, err := toFloat64(iface) + if err != nil { + return nil, err + } + floats[i] = val + } + return floats, nil +} + +func (cmd *Cmd) BoolSlice() ([]bool, error) { + slice, err := cmd.Slice() + if err != nil { + return nil, err + } + + bools := make([]bool, len(slice)) + for i, iface := range slice { + val, err := toBool(iface) + if err != nil { + return nil, err + } + bools[i] = val + } + return bools, nil +} + +func (cmd *Cmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadReply() + return err +} + +//------------------------------------------------------------------------------ + +type SliceCmd struct { + baseCmd + + val []interface{} +} + +var _ Cmder = (*SliceCmd)(nil) + +func NewSliceCmd(ctx context.Context, args ...interface{}) *SliceCmd { + return &SliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *SliceCmd) SetVal(val []interface{}) { + cmd.val = val +} + +func (cmd *SliceCmd) Val() []interface{} { + return cmd.val +} + +func (cmd *SliceCmd) Result() ([]interface{}, error) { + return cmd.val, cmd.err +} + +func (cmd *SliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +// Scan scans the results from the map into a destination struct. The map keys +// are matched in the Redis struct fields by the `redis:"field"` tag. +func (cmd *SliceCmd) Scan(dst interface{}) error { + if cmd.err != nil { + return cmd.err + } + + // Pass the list of keys and values. + // Skip the first two args for: HMGET key + var args []interface{} + if cmd.args[0] == "hmget" { + args = cmd.args[2:] + } else { + // Otherwise, it's: MGET field field ... + args = cmd.args[1:] + } + + return hscan.Scan(dst, args, cmd.val) +} + +func (cmd *SliceCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadSlice() + return err +} + +//------------------------------------------------------------------------------ + +type StatusCmd struct { + baseCmd + + val string +} + +var _ Cmder = (*StatusCmd)(nil) + +func NewStatusCmd(ctx context.Context, args ...interface{}) *StatusCmd { + return &StatusCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *StatusCmd) SetVal(val string) { + cmd.val = val +} + +func (cmd *StatusCmd) Val() string { + return cmd.val +} + +func (cmd *StatusCmd) Result() (string, error) { + return cmd.val, cmd.err +} + +func (cmd *StatusCmd) Bytes() ([]byte, error) { + return util.StringToBytes(cmd.val), cmd.err +} + +func (cmd *StatusCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *StatusCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadString() + return err +} + +//------------------------------------------------------------------------------ + +type IntCmd struct { + baseCmd + + val int64 +} + +var _ Cmder = (*IntCmd)(nil) + +func NewIntCmd(ctx context.Context, args ...interface{}) *IntCmd { + return &IntCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *IntCmd) SetVal(val int64) { + cmd.val = val +} + +func (cmd *IntCmd) Val() int64 { + return cmd.val +} + +func (cmd *IntCmd) Result() (int64, error) { + return cmd.val, cmd.err +} + +func (cmd *IntCmd) Uint64() (uint64, error) { + return uint64(cmd.val), cmd.err +} + +func (cmd *IntCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *IntCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadInt() + return err +} + +//------------------------------------------------------------------------------ + +type IntSliceCmd struct { + baseCmd + + val []int64 +} + +var _ Cmder = (*IntSliceCmd)(nil) + +func NewIntSliceCmd(ctx context.Context, args ...interface{}) *IntSliceCmd { + return &IntSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *IntSliceCmd) SetVal(val []int64) { + cmd.val = val +} + +func (cmd *IntSliceCmd) Val() []int64 { + return cmd.val +} + +func (cmd *IntSliceCmd) Result() ([]int64, error) { + return cmd.val, cmd.err +} + +func (cmd *IntSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *IntSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]int64, n) + for i := 0; i < len(cmd.val); i++ { + if cmd.val[i], err = rd.ReadInt(); err != nil { + return err + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type DurationCmd struct { + baseCmd + + val time.Duration + precision time.Duration +} + +var _ Cmder = (*DurationCmd)(nil) + +func NewDurationCmd(ctx context.Context, precision time.Duration, args ...interface{}) *DurationCmd { + return &DurationCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + precision: precision, + } +} + +func (cmd *DurationCmd) SetVal(val time.Duration) { + cmd.val = val +} + +func (cmd *DurationCmd) Val() time.Duration { + return cmd.val +} + +func (cmd *DurationCmd) Result() (time.Duration, error) { + return cmd.val, cmd.err +} + +func (cmd *DurationCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *DurationCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadInt() + if err != nil { + return err + } + switch n { + // -2 if the key does not exist + // -1 if the key exists but has no associated expire + case -2, -1: + cmd.val = time.Duration(n) + default: + cmd.val = time.Duration(n) * cmd.precision + } + return nil +} + +//------------------------------------------------------------------------------ + +type TimeCmd struct { + baseCmd + + val time.Time +} + +var _ Cmder = (*TimeCmd)(nil) + +func NewTimeCmd(ctx context.Context, args ...interface{}) *TimeCmd { + return &TimeCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *TimeCmd) SetVal(val time.Time) { + cmd.val = val +} + +func (cmd *TimeCmd) Val() time.Time { + return cmd.val +} + +func (cmd *TimeCmd) Result() (time.Time, error) { + return cmd.val, cmd.err +} + +func (cmd *TimeCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *TimeCmd) readReply(rd *proto.Reader) error { + if err := rd.ReadFixedArrayLen(2); err != nil { + return err + } + second, err := rd.ReadInt() + if err != nil { + return err + } + microsecond, err := rd.ReadInt() + if err != nil { + return err + } + cmd.val = time.Unix(second, microsecond*1000) + return nil +} + +//------------------------------------------------------------------------------ + +type BoolCmd struct { + baseCmd + + val bool +} + +var _ Cmder = (*BoolCmd)(nil) + +func NewBoolCmd(ctx context.Context, args ...interface{}) *BoolCmd { + return &BoolCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *BoolCmd) SetVal(val bool) { + cmd.val = val +} + +func (cmd *BoolCmd) Val() bool { + return cmd.val +} + +func (cmd *BoolCmd) Result() (bool, error) { + return cmd.val, cmd.err +} + +func (cmd *BoolCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *BoolCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadBool() + + // `SET key value NX` returns nil when key already exists. But + // `SETNX key value` returns bool (0/1). So convert nil to bool. + if err == Nil { + cmd.val = false + err = nil + } + return err +} + +//------------------------------------------------------------------------------ + +type StringCmd struct { + baseCmd + + val string +} + +var _ Cmder = (*StringCmd)(nil) + +func NewStringCmd(ctx context.Context, args ...interface{}) *StringCmd { + return &StringCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *StringCmd) SetVal(val string) { + cmd.val = val +} + +func (cmd *StringCmd) Val() string { + return cmd.val +} + +func (cmd *StringCmd) Result() (string, error) { + return cmd.val, cmd.err +} + +func (cmd *StringCmd) Bytes() ([]byte, error) { + return util.StringToBytes(cmd.val), cmd.err +} + +func (cmd *StringCmd) Bool() (bool, error) { + if cmd.err != nil { + return false, cmd.err + } + return strconv.ParseBool(cmd.val) +} + +func (cmd *StringCmd) Int() (int, error) { + if cmd.err != nil { + return 0, cmd.err + } + return strconv.Atoi(cmd.Val()) +} + +func (cmd *StringCmd) Int64() (int64, error) { + if cmd.err != nil { + return 0, cmd.err + } + return strconv.ParseInt(cmd.Val(), 10, 64) +} + +func (cmd *StringCmd) Uint64() (uint64, error) { + if cmd.err != nil { + return 0, cmd.err + } + return strconv.ParseUint(cmd.Val(), 10, 64) +} + +func (cmd *StringCmd) Float32() (float32, error) { + if cmd.err != nil { + return 0, cmd.err + } + f, err := strconv.ParseFloat(cmd.Val(), 32) + if err != nil { + return 0, err + } + return float32(f), nil +} + +func (cmd *StringCmd) Float64() (float64, error) { + if cmd.err != nil { + return 0, cmd.err + } + return strconv.ParseFloat(cmd.Val(), 64) +} + +func (cmd *StringCmd) Time() (time.Time, error) { + if cmd.err != nil { + return time.Time{}, cmd.err + } + return time.Parse(time.RFC3339Nano, cmd.Val()) +} + +func (cmd *StringCmd) Scan(val interface{}) error { + if cmd.err != nil { + return cmd.err + } + return proto.Scan([]byte(cmd.val), val) +} + +func (cmd *StringCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *StringCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadString() + return err +} + +//------------------------------------------------------------------------------ + +type FloatCmd struct { + baseCmd + + val float64 +} + +var _ Cmder = (*FloatCmd)(nil) + +func NewFloatCmd(ctx context.Context, args ...interface{}) *FloatCmd { + return &FloatCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *FloatCmd) SetVal(val float64) { + cmd.val = val +} + +func (cmd *FloatCmd) Val() float64 { + return cmd.val +} + +func (cmd *FloatCmd) Result() (float64, error) { + return cmd.val, cmd.err +} + +func (cmd *FloatCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *FloatCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = rd.ReadFloat() + return err +} + +//------------------------------------------------------------------------------ + +type FloatSliceCmd struct { + baseCmd + + val []float64 +} + +var _ Cmder = (*FloatSliceCmd)(nil) + +func NewFloatSliceCmd(ctx context.Context, args ...interface{}) *FloatSliceCmd { + return &FloatSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *FloatSliceCmd) SetVal(val []float64) { + cmd.val = val +} + +func (cmd *FloatSliceCmd) Val() []float64 { + return cmd.val +} + +func (cmd *FloatSliceCmd) Result() ([]float64, error) { + return cmd.val, cmd.err +} + +func (cmd *FloatSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *FloatSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make([]float64, n) + for i := 0; i < len(cmd.val); i++ { + switch num, err := rd.ReadFloat(); { + case err == Nil: + cmd.val[i] = 0 + case err != nil: + return err + default: + cmd.val[i] = num + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type StringSliceCmd struct { + baseCmd + + val []string +} + +var _ Cmder = (*StringSliceCmd)(nil) + +func NewStringSliceCmd(ctx context.Context, args ...interface{}) *StringSliceCmd { + return &StringSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *StringSliceCmd) SetVal(val []string) { + cmd.val = val +} + +func (cmd *StringSliceCmd) Val() []string { + return cmd.val +} + +func (cmd *StringSliceCmd) Result() ([]string, error) { + return cmd.val, cmd.err +} + +func (cmd *StringSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *StringSliceCmd) ScanSlice(container interface{}) error { + return proto.ScanSlice(cmd.Val(), container) +} + +func (cmd *StringSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]string, n) + for i := 0; i < len(cmd.val); i++ { + switch s, err := rd.ReadString(); { + case err == Nil: + cmd.val[i] = "" + case err != nil: + return err + default: + cmd.val[i] = s + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type KeyValue struct { + Key string + Value string +} + +type KeyValueSliceCmd struct { + baseCmd + + val []KeyValue +} + +var _ Cmder = (*KeyValueSliceCmd)(nil) + +func NewKeyValueSliceCmd(ctx context.Context, args ...interface{}) *KeyValueSliceCmd { + return &KeyValueSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *KeyValueSliceCmd) SetVal(val []KeyValue) { + cmd.val = val +} + +func (cmd *KeyValueSliceCmd) Val() []KeyValue { + return cmd.val +} + +func (cmd *KeyValueSliceCmd) Result() ([]KeyValue, error) { + return cmd.val, cmd.err +} + +func (cmd *KeyValueSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +// Many commands will respond to two formats: +// 1. 1) "one" +// 2. (double) 1 +// 2. 1) "two" +// 2. (double) 2 +// +// OR: +// 1. "two" +// 2. (double) 2 +// 3. "one" +// 4. (double) 1 +func (cmd *KeyValueSliceCmd) readReply(rd *proto.Reader) error { // nolint:dupl + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + // If the n is 0, can't continue reading. + if n == 0 { + cmd.val = make([]KeyValue, 0) + return nil + } + + typ, err := rd.PeekReplyType() + if err != nil { + return err + } + array := typ == proto.RespArray + + if array { + cmd.val = make([]KeyValue, n) + } else { + cmd.val = make([]KeyValue, n/2) + } + + for i := 0; i < len(cmd.val); i++ { + if array { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + } + + if cmd.val[i].Key, err = rd.ReadString(); err != nil { + return err + } + + if cmd.val[i].Value, err = rd.ReadString(); err != nil { + return err + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type BoolSliceCmd struct { + baseCmd + + val []bool +} + +var _ Cmder = (*BoolSliceCmd)(nil) + +func NewBoolSliceCmd(ctx context.Context, args ...interface{}) *BoolSliceCmd { + return &BoolSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *BoolSliceCmd) SetVal(val []bool) { + cmd.val = val +} + +func (cmd *BoolSliceCmd) Val() []bool { + return cmd.val +} + +func (cmd *BoolSliceCmd) Result() ([]bool, error) { + return cmd.val, cmd.err +} + +func (cmd *BoolSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *BoolSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]bool, n) + for i := 0; i < len(cmd.val); i++ { + if cmd.val[i], err = rd.ReadBool(); err != nil { + return err + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type MapStringStringCmd struct { + baseCmd + + val map[string]string +} + +var _ Cmder = (*MapStringStringCmd)(nil) + +func NewMapStringStringCmd(ctx context.Context, args ...interface{}) *MapStringStringCmd { + return &MapStringStringCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapStringStringCmd) Val() map[string]string { + return cmd.val +} + +func (cmd *MapStringStringCmd) SetVal(val map[string]string) { + cmd.val = val +} + +func (cmd *MapStringStringCmd) Result() (map[string]string, error) { + return cmd.val, cmd.err +} + +func (cmd *MapStringStringCmd) String() string { + return cmdString(cmd, cmd.val) +} + +// Scan scans the results from the map into a destination struct. The map keys +// are matched in the Redis struct fields by the `redis:"field"` tag. +func (cmd *MapStringStringCmd) Scan(dest interface{}) error { + if cmd.err != nil { + return cmd.err + } + + strct, err := hscan.Struct(dest) + if err != nil { + return err + } + + for k, v := range cmd.val { + if err := strct.Scan(k, v); err != nil { + return err + } + } + + return nil +} + +func (cmd *MapStringStringCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + + cmd.val = make(map[string]string, n) + for i := 0; i < n; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + value, err := rd.ReadString() + if err != nil { + return err + } + + cmd.val[key] = value + } + return nil +} + +//------------------------------------------------------------------------------ + +type MapStringIntCmd struct { + baseCmd + + val map[string]int64 +} + +var _ Cmder = (*MapStringIntCmd)(nil) + +func NewMapStringIntCmd(ctx context.Context, args ...interface{}) *MapStringIntCmd { + return &MapStringIntCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapStringIntCmd) SetVal(val map[string]int64) { + cmd.val = val +} + +func (cmd *MapStringIntCmd) Val() map[string]int64 { + return cmd.val +} + +func (cmd *MapStringIntCmd) Result() (map[string]int64, error) { + return cmd.val, cmd.err +} + +func (cmd *MapStringIntCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *MapStringIntCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + + cmd.val = make(map[string]int64, n) + for i := 0; i < n; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + nn, err := rd.ReadInt() + if err != nil { + return err + } + cmd.val[key] = nn + } + return nil +} + +// ------------------------------------------------------------------------------ +type MapStringSliceInterfaceCmd struct { + baseCmd + val map[string][]interface{} +} + +func NewMapStringSliceInterfaceCmd(ctx context.Context, args ...interface{}) *MapStringSliceInterfaceCmd { + return &MapStringSliceInterfaceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapStringSliceInterfaceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *MapStringSliceInterfaceCmd) SetVal(val map[string][]interface{}) { + cmd.val = val +} + +func (cmd *MapStringSliceInterfaceCmd) Result() (map[string][]interface{}, error) { + return cmd.val, cmd.err +} + +func (cmd *MapStringSliceInterfaceCmd) Val() map[string][]interface{} { + return cmd.val +} + +func (cmd *MapStringSliceInterfaceCmd) readReply(rd *proto.Reader) (err error) { + readType, err := rd.PeekReplyType() + if err != nil { + return err + } + + cmd.val = make(map[string][]interface{}) + + switch readType { + case proto.RespMap: + n, err := rd.ReadMapLen() + if err != nil { + return err + } + for i := 0; i < n; i++ { + k, err := rd.ReadString() + if err != nil { + return err + } + nn, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val[k] = make([]interface{}, nn) + for j := 0; j < nn; j++ { + value, err := rd.ReadReply() + if err != nil { + return err + } + cmd.val[k][j] = value + } + } + case proto.RespArray: + // RESP2 response + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + for i := 0; i < n; i++ { + // Each entry in this array is itself an array with key details + itemLen, err := rd.ReadArrayLen() + if err != nil { + return err + } + + key, err := rd.ReadString() + if err != nil { + return err + } + cmd.val[key] = make([]interface{}, 0, itemLen-1) + for j := 1; j < itemLen; j++ { + // Read the inner array for timestamp-value pairs + data, err := rd.ReadReply() + if err != nil { + return err + } + cmd.val[key] = append(cmd.val[key], data) + } + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type StringStructMapCmd struct { + baseCmd + + val map[string]struct{} +} + +var _ Cmder = (*StringStructMapCmd)(nil) + +func NewStringStructMapCmd(ctx context.Context, args ...interface{}) *StringStructMapCmd { + return &StringStructMapCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *StringStructMapCmd) SetVal(val map[string]struct{}) { + cmd.val = val +} + +func (cmd *StringStructMapCmd) Val() map[string]struct{} { + return cmd.val +} + +func (cmd *StringStructMapCmd) Result() (map[string]struct{}, error) { + return cmd.val, cmd.err +} + +func (cmd *StringStructMapCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *StringStructMapCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make(map[string]struct{}, n) + for i := 0; i < n; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + cmd.val[key] = struct{}{} + } + return nil +} + +//------------------------------------------------------------------------------ + +type XMessage struct { + ID string + Values map[string]interface{} +} + +type XMessageSliceCmd struct { + baseCmd + + val []XMessage +} + +var _ Cmder = (*XMessageSliceCmd)(nil) + +func NewXMessageSliceCmd(ctx context.Context, args ...interface{}) *XMessageSliceCmd { + return &XMessageSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XMessageSliceCmd) SetVal(val []XMessage) { + cmd.val = val +} + +func (cmd *XMessageSliceCmd) Val() []XMessage { + return cmd.val +} + +func (cmd *XMessageSliceCmd) Result() ([]XMessage, error) { + return cmd.val, cmd.err +} + +func (cmd *XMessageSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XMessageSliceCmd) readReply(rd *proto.Reader) (err error) { + cmd.val, err = readXMessageSlice(rd) + return err +} + +func readXMessageSlice(rd *proto.Reader) ([]XMessage, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + msgs := make([]XMessage, n) + for i := 0; i < len(msgs); i++ { + if msgs[i], err = readXMessage(rd); err != nil { + return nil, err + } + } + return msgs, nil +} + +func readXMessage(rd *proto.Reader) (XMessage, error) { + if err := rd.ReadFixedArrayLen(2); err != nil { + return XMessage{}, err + } + + id, err := rd.ReadString() + if err != nil { + return XMessage{}, err + } + + v, err := stringInterfaceMapParser(rd) + if err != nil { + if err != proto.Nil { + return XMessage{}, err + } + } + + return XMessage{ + ID: id, + Values: v, + }, nil +} + +func stringInterfaceMapParser(rd *proto.Reader) (map[string]interface{}, error) { + n, err := rd.ReadMapLen() + if err != nil { + return nil, err + } + + m := make(map[string]interface{}, n) + for i := 0; i < n; i++ { + key, err := rd.ReadString() + if err != nil { + return nil, err + } + + value, err := rd.ReadString() + if err != nil { + return nil, err + } + + m[key] = value + } + return m, nil +} + +//------------------------------------------------------------------------------ + +type XStream struct { + Stream string + Messages []XMessage +} + +type XStreamSliceCmd struct { + baseCmd + + val []XStream +} + +var _ Cmder = (*XStreamSliceCmd)(nil) + +func NewXStreamSliceCmd(ctx context.Context, args ...interface{}) *XStreamSliceCmd { + return &XStreamSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XStreamSliceCmd) SetVal(val []XStream) { + cmd.val = val +} + +func (cmd *XStreamSliceCmd) Val() []XStream { + return cmd.val +} + +func (cmd *XStreamSliceCmd) Result() ([]XStream, error) { + return cmd.val, cmd.err +} + +func (cmd *XStreamSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XStreamSliceCmd) readReply(rd *proto.Reader) error { + typ, err := rd.PeekReplyType() + if err != nil { + return err + } + + var n int + if typ == proto.RespMap { + n, err = rd.ReadMapLen() + } else { + n, err = rd.ReadArrayLen() + } + if err != nil { + return err + } + cmd.val = make([]XStream, n) + for i := 0; i < len(cmd.val); i++ { + if typ != proto.RespMap { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + } + if cmd.val[i].Stream, err = rd.ReadString(); err != nil { + return err + } + if cmd.val[i].Messages, err = readXMessageSlice(rd); err != nil { + return err + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type XPending struct { + Count int64 + Lower string + Higher string + Consumers map[string]int64 +} + +type XPendingCmd struct { + baseCmd + val *XPending +} + +var _ Cmder = (*XPendingCmd)(nil) + +func NewXPendingCmd(ctx context.Context, args ...interface{}) *XPendingCmd { + return &XPendingCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XPendingCmd) SetVal(val *XPending) { + cmd.val = val +} + +func (cmd *XPendingCmd) Val() *XPending { + return cmd.val +} + +func (cmd *XPendingCmd) Result() (*XPending, error) { + return cmd.val, cmd.err +} + +func (cmd *XPendingCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XPendingCmd) readReply(rd *proto.Reader) error { + var err error + if err = rd.ReadFixedArrayLen(4); err != nil { + return err + } + cmd.val = &XPending{} + + if cmd.val.Count, err = rd.ReadInt(); err != nil { + return err + } + + if cmd.val.Lower, err = rd.ReadString(); err != nil && err != Nil { + return err + } + + if cmd.val.Higher, err = rd.ReadString(); err != nil && err != Nil { + return err + } + + n, err := rd.ReadArrayLen() + if err != nil && err != Nil { + return err + } + cmd.val.Consumers = make(map[string]int64, n) + for i := 0; i < n; i++ { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + + consumerName, err := rd.ReadString() + if err != nil { + return err + } + consumerPending, err := rd.ReadInt() + if err != nil { + return err + } + cmd.val.Consumers[consumerName] = consumerPending + } + return nil +} + +//------------------------------------------------------------------------------ + +type XPendingExt struct { + ID string + Consumer string + Idle time.Duration + RetryCount int64 +} + +type XPendingExtCmd struct { + baseCmd + val []XPendingExt +} + +var _ Cmder = (*XPendingExtCmd)(nil) + +func NewXPendingExtCmd(ctx context.Context, args ...interface{}) *XPendingExtCmd { + return &XPendingExtCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XPendingExtCmd) SetVal(val []XPendingExt) { + cmd.val = val +} + +func (cmd *XPendingExtCmd) Val() []XPendingExt { + return cmd.val +} + +func (cmd *XPendingExtCmd) Result() ([]XPendingExt, error) { + return cmd.val, cmd.err +} + +func (cmd *XPendingExtCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XPendingExtCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]XPendingExt, n) + + for i := 0; i < len(cmd.val); i++ { + if err = rd.ReadFixedArrayLen(4); err != nil { + return err + } + + if cmd.val[i].ID, err = rd.ReadString(); err != nil { + return err + } + + if cmd.val[i].Consumer, err = rd.ReadString(); err != nil && err != Nil { + return err + } + + idle, err := rd.ReadInt() + if err != nil && err != Nil { + return err + } + cmd.val[i].Idle = time.Duration(idle) * time.Millisecond + + if cmd.val[i].RetryCount, err = rd.ReadInt(); err != nil && err != Nil { + return err + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type XAutoClaimCmd struct { + baseCmd + + start string + val []XMessage +} + +var _ Cmder = (*XAutoClaimCmd)(nil) + +func NewXAutoClaimCmd(ctx context.Context, args ...interface{}) *XAutoClaimCmd { + return &XAutoClaimCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XAutoClaimCmd) SetVal(val []XMessage, start string) { + cmd.val = val + cmd.start = start +} + +func (cmd *XAutoClaimCmd) Val() (messages []XMessage, start string) { + return cmd.val, cmd.start +} + +func (cmd *XAutoClaimCmd) Result() (messages []XMessage, start string, err error) { + return cmd.val, cmd.start, cmd.err +} + +func (cmd *XAutoClaimCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XAutoClaimCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + switch n { + case 2, // Redis 6 + 3: // Redis 7: + // ok + default: + return fmt.Errorf("redis: got %d elements in XAutoClaim reply, wanted 2/3", n) + } + + cmd.start, err = rd.ReadString() + if err != nil { + return err + } + + cmd.val, err = readXMessageSlice(rd) + if err != nil { + return err + } + + if n >= 3 { + if err := rd.DiscardNext(); err != nil { + return err + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type XAutoClaimJustIDCmd struct { + baseCmd + + start string + val []string +} + +var _ Cmder = (*XAutoClaimJustIDCmd)(nil) + +func NewXAutoClaimJustIDCmd(ctx context.Context, args ...interface{}) *XAutoClaimJustIDCmd { + return &XAutoClaimJustIDCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XAutoClaimJustIDCmd) SetVal(val []string, start string) { + cmd.val = val + cmd.start = start +} + +func (cmd *XAutoClaimJustIDCmd) Val() (ids []string, start string) { + return cmd.val, cmd.start +} + +func (cmd *XAutoClaimJustIDCmd) Result() (ids []string, start string, err error) { + return cmd.val, cmd.start, cmd.err +} + +func (cmd *XAutoClaimJustIDCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XAutoClaimJustIDCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + switch n { + case 2, // Redis 6 + 3: // Redis 7: + // ok + default: + return fmt.Errorf("redis: got %d elements in XAutoClaimJustID reply, wanted 2/3", n) + } + + cmd.start, err = rd.ReadString() + if err != nil { + return err + } + + nn, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make([]string, nn) + for i := 0; i < nn; i++ { + cmd.val[i], err = rd.ReadString() + if err != nil { + return err + } + } + + if n >= 3 { + if err := rd.DiscardNext(); err != nil { + return err + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type XInfoConsumersCmd struct { + baseCmd + val []XInfoConsumer +} + +type XInfoConsumer struct { + Name string + Pending int64 + Idle time.Duration + Inactive time.Duration +} + +var _ Cmder = (*XInfoConsumersCmd)(nil) + +func NewXInfoConsumersCmd(ctx context.Context, stream string, group string) *XInfoConsumersCmd { + return &XInfoConsumersCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: []interface{}{"xinfo", "consumers", stream, group}, + }, + } +} + +func (cmd *XInfoConsumersCmd) SetVal(val []XInfoConsumer) { + cmd.val = val +} + +func (cmd *XInfoConsumersCmd) Val() []XInfoConsumer { + return cmd.val +} + +func (cmd *XInfoConsumersCmd) Result() ([]XInfoConsumer, error) { + return cmd.val, cmd.err +} + +func (cmd *XInfoConsumersCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XInfoConsumersCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]XInfoConsumer, n) + + for i := 0; i < len(cmd.val); i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return err + } + + var key string + for f := 0; f < nn; f++ { + key, err = rd.ReadString() + if err != nil { + return err + } + + switch key { + case "name": + cmd.val[i].Name, err = rd.ReadString() + case "pending": + cmd.val[i].Pending, err = rd.ReadInt() + case "idle": + var idle int64 + idle, err = rd.ReadInt() + cmd.val[i].Idle = time.Duration(idle) * time.Millisecond + case "inactive": + var inactive int64 + inactive, err = rd.ReadInt() + cmd.val[i].Inactive = time.Duration(inactive) * time.Millisecond + default: + return fmt.Errorf("redis: unexpected content %s in XINFO CONSUMERS reply", key) + } + if err != nil { + return err + } + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type XInfoGroupsCmd struct { + baseCmd + val []XInfoGroup +} + +type XInfoGroup struct { + Name string + Consumers int64 + Pending int64 + LastDeliveredID string + EntriesRead int64 + // Lag represents the number of pending messages in the stream not yet + // delivered to this consumer group. Returns -1 when the lag cannot be determined. + Lag int64 +} + +var _ Cmder = (*XInfoGroupsCmd)(nil) + +func NewXInfoGroupsCmd(ctx context.Context, stream string) *XInfoGroupsCmd { + return &XInfoGroupsCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: []interface{}{"xinfo", "groups", stream}, + }, + } +} + +func (cmd *XInfoGroupsCmd) SetVal(val []XInfoGroup) { + cmd.val = val +} + +func (cmd *XInfoGroupsCmd) Val() []XInfoGroup { + return cmd.val +} + +func (cmd *XInfoGroupsCmd) Result() ([]XInfoGroup, error) { + return cmd.val, cmd.err +} + +func (cmd *XInfoGroupsCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XInfoGroupsCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]XInfoGroup, n) + + for i := 0; i < len(cmd.val); i++ { + group := &cmd.val[i] + + nn, err := rd.ReadMapLen() + if err != nil { + return err + } + + var key string + for j := 0; j < nn; j++ { + key, err = rd.ReadString() + if err != nil { + return err + } + + switch key { + case "name": + group.Name, err = rd.ReadString() + if err != nil { + return err + } + case "consumers": + group.Consumers, err = rd.ReadInt() + if err != nil { + return err + } + case "pending": + group.Pending, err = rd.ReadInt() + if err != nil { + return err + } + case "last-delivered-id": + group.LastDeliveredID, err = rd.ReadString() + if err != nil { + return err + } + case "entries-read": + group.EntriesRead, err = rd.ReadInt() + if err != nil && err != Nil { + return err + } + case "lag": + group.Lag, err = rd.ReadInt() + + // lag: the number of entries in the stream that are still waiting to be delivered + // to the group's consumers, or a NULL(Nil) when that number can't be determined. + // In that case, we return -1. + if err != nil && err != Nil { + return err + } else if err == Nil { + group.Lag = -1 + } + default: + return fmt.Errorf("redis: unexpected key %q in XINFO GROUPS reply", key) + } + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type XInfoStreamCmd struct { + baseCmd + val *XInfoStream +} + +type XInfoStream struct { + Length int64 + RadixTreeKeys int64 + RadixTreeNodes int64 + Groups int64 + LastGeneratedID string + MaxDeletedEntryID string + EntriesAdded int64 + FirstEntry XMessage + LastEntry XMessage + RecordedFirstEntryID string +} + +var _ Cmder = (*XInfoStreamCmd)(nil) + +func NewXInfoStreamCmd(ctx context.Context, stream string) *XInfoStreamCmd { + return &XInfoStreamCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: []interface{}{"xinfo", "stream", stream}, + }, + } +} + +func (cmd *XInfoStreamCmd) SetVal(val *XInfoStream) { + cmd.val = val +} + +func (cmd *XInfoStreamCmd) Val() *XInfoStream { + return cmd.val +} + +func (cmd *XInfoStreamCmd) Result() (*XInfoStream, error) { + return cmd.val, cmd.err +} + +func (cmd *XInfoStreamCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XInfoStreamCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + cmd.val = &XInfoStream{} + + for i := 0; i < n; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + switch key { + case "length": + cmd.val.Length, err = rd.ReadInt() + if err != nil { + return err + } + case "radix-tree-keys": + cmd.val.RadixTreeKeys, err = rd.ReadInt() + if err != nil { + return err + } + case "radix-tree-nodes": + cmd.val.RadixTreeNodes, err = rd.ReadInt() + if err != nil { + return err + } + case "groups": + cmd.val.Groups, err = rd.ReadInt() + if err != nil { + return err + } + case "last-generated-id": + cmd.val.LastGeneratedID, err = rd.ReadString() + if err != nil { + return err + } + case "max-deleted-entry-id": + cmd.val.MaxDeletedEntryID, err = rd.ReadString() + if err != nil { + return err + } + case "entries-added": + cmd.val.EntriesAdded, err = rd.ReadInt() + if err != nil { + return err + } + case "first-entry": + cmd.val.FirstEntry, err = readXMessage(rd) + if err != nil && err != Nil { + return err + } + case "last-entry": + cmd.val.LastEntry, err = readXMessage(rd) + if err != nil && err != Nil { + return err + } + case "recorded-first-entry-id": + cmd.val.RecordedFirstEntryID, err = rd.ReadString() + if err != nil { + return err + } + default: + return fmt.Errorf("redis: unexpected key %q in XINFO STREAM reply", key) + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type XInfoStreamFullCmd struct { + baseCmd + val *XInfoStreamFull +} + +type XInfoStreamFull struct { + Length int64 + RadixTreeKeys int64 + RadixTreeNodes int64 + LastGeneratedID string + MaxDeletedEntryID string + EntriesAdded int64 + Entries []XMessage + Groups []XInfoStreamGroup + RecordedFirstEntryID string +} + +type XInfoStreamGroup struct { + Name string + LastDeliveredID string + EntriesRead int64 + Lag int64 + PelCount int64 + Pending []XInfoStreamGroupPending + Consumers []XInfoStreamConsumer +} + +type XInfoStreamGroupPending struct { + ID string + Consumer string + DeliveryTime time.Time + DeliveryCount int64 +} + +type XInfoStreamConsumer struct { + Name string + SeenTime time.Time + ActiveTime time.Time + PelCount int64 + Pending []XInfoStreamConsumerPending +} + +type XInfoStreamConsumerPending struct { + ID string + DeliveryTime time.Time + DeliveryCount int64 +} + +var _ Cmder = (*XInfoStreamFullCmd)(nil) + +func NewXInfoStreamFullCmd(ctx context.Context, args ...interface{}) *XInfoStreamFullCmd { + return &XInfoStreamFullCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *XInfoStreamFullCmd) SetVal(val *XInfoStreamFull) { + cmd.val = val +} + +func (cmd *XInfoStreamFullCmd) Val() *XInfoStreamFull { + return cmd.val +} + +func (cmd *XInfoStreamFullCmd) Result() (*XInfoStreamFull, error) { + return cmd.val, cmd.err +} + +func (cmd *XInfoStreamFullCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *XInfoStreamFullCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + + cmd.val = &XInfoStreamFull{} + + for i := 0; i < n; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + switch key { + case "length": + cmd.val.Length, err = rd.ReadInt() + if err != nil { + return err + } + case "radix-tree-keys": + cmd.val.RadixTreeKeys, err = rd.ReadInt() + if err != nil { + return err + } + case "radix-tree-nodes": + cmd.val.RadixTreeNodes, err = rd.ReadInt() + if err != nil { + return err + } + case "last-generated-id": + cmd.val.LastGeneratedID, err = rd.ReadString() + if err != nil { + return err + } + case "entries-added": + cmd.val.EntriesAdded, err = rd.ReadInt() + if err != nil { + return err + } + case "entries": + cmd.val.Entries, err = readXMessageSlice(rd) + if err != nil { + return err + } + case "groups": + cmd.val.Groups, err = readStreamGroups(rd) + if err != nil { + return err + } + case "max-deleted-entry-id": + cmd.val.MaxDeletedEntryID, err = rd.ReadString() + if err != nil { + return err + } + case "recorded-first-entry-id": + cmd.val.RecordedFirstEntryID, err = rd.ReadString() + if err != nil { + return err + } + default: + return fmt.Errorf("redis: unexpected key %q in XINFO STREAM FULL reply", key) + } + } + return nil +} + +func readStreamGroups(rd *proto.Reader) ([]XInfoStreamGroup, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + groups := make([]XInfoStreamGroup, 0, n) + for i := 0; i < n; i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return nil, err + } + + group := XInfoStreamGroup{} + + for j := 0; j < nn; j++ { + key, err := rd.ReadString() + if err != nil { + return nil, err + } + + switch key { + case "name": + group.Name, err = rd.ReadString() + if err != nil { + return nil, err + } + case "last-delivered-id": + group.LastDeliveredID, err = rd.ReadString() + if err != nil { + return nil, err + } + case "entries-read": + group.EntriesRead, err = rd.ReadInt() + if err != nil && err != Nil { + return nil, err + } + case "lag": + // lag: the number of entries in the stream that are still waiting to be delivered + // to the group's consumers, or a NULL(Nil) when that number can't be determined. + group.Lag, err = rd.ReadInt() + if err != nil && err != Nil { + return nil, err + } + case "pel-count": + group.PelCount, err = rd.ReadInt() + if err != nil { + return nil, err + } + case "pending": + group.Pending, err = readXInfoStreamGroupPending(rd) + if err != nil { + return nil, err + } + case "consumers": + group.Consumers, err = readXInfoStreamConsumers(rd) + if err != nil { + return nil, err + } + default: + return nil, fmt.Errorf("redis: unexpected key %q in XINFO STREAM FULL reply", key) + } + } + + groups = append(groups, group) + } + + return groups, nil +} + +func readXInfoStreamGroupPending(rd *proto.Reader) ([]XInfoStreamGroupPending, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + pending := make([]XInfoStreamGroupPending, 0, n) + + for i := 0; i < n; i++ { + if err = rd.ReadFixedArrayLen(4); err != nil { + return nil, err + } + + p := XInfoStreamGroupPending{} + + p.ID, err = rd.ReadString() + if err != nil { + return nil, err + } + + p.Consumer, err = rd.ReadString() + if err != nil { + return nil, err + } + + delivery, err := rd.ReadInt() + if err != nil { + return nil, err + } + p.DeliveryTime = time.Unix(delivery/1000, delivery%1000*int64(time.Millisecond)) + + p.DeliveryCount, err = rd.ReadInt() + if err != nil { + return nil, err + } + + pending = append(pending, p) + } + + return pending, nil +} + +func readXInfoStreamConsumers(rd *proto.Reader) ([]XInfoStreamConsumer, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + consumers := make([]XInfoStreamConsumer, 0, n) + + for i := 0; i < n; i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return nil, err + } + + c := XInfoStreamConsumer{} + + for f := 0; f < nn; f++ { + cKey, err := rd.ReadString() + if err != nil { + return nil, err + } + + switch cKey { + case "name": + c.Name, err = rd.ReadString() + case "seen-time": + seen, err := rd.ReadInt() + if err != nil { + return nil, err + } + c.SeenTime = time.UnixMilli(seen) + case "active-time": + active, err := rd.ReadInt() + if err != nil { + return nil, err + } + c.ActiveTime = time.UnixMilli(active) + case "pel-count": + c.PelCount, err = rd.ReadInt() + case "pending": + pendingNumber, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + c.Pending = make([]XInfoStreamConsumerPending, 0, pendingNumber) + + for pn := 0; pn < pendingNumber; pn++ { + if err = rd.ReadFixedArrayLen(3); err != nil { + return nil, err + } + + p := XInfoStreamConsumerPending{} + + p.ID, err = rd.ReadString() + if err != nil { + return nil, err + } + + delivery, err := rd.ReadInt() + if err != nil { + return nil, err + } + p.DeliveryTime = time.Unix(delivery/1000, delivery%1000*int64(time.Millisecond)) + + p.DeliveryCount, err = rd.ReadInt() + if err != nil { + return nil, err + } + + c.Pending = append(c.Pending, p) + } + default: + return nil, fmt.Errorf("redis: unexpected content %s "+ + "in XINFO STREAM FULL reply", cKey) + } + if err != nil { + return nil, err + } + } + consumers = append(consumers, c) + } + + return consumers, nil +} + +//------------------------------------------------------------------------------ + +type ZSliceCmd struct { + baseCmd + + val []Z +} + +var _ Cmder = (*ZSliceCmd)(nil) + +func NewZSliceCmd(ctx context.Context, args ...interface{}) *ZSliceCmd { + return &ZSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ZSliceCmd) SetVal(val []Z) { + cmd.val = val +} + +func (cmd *ZSliceCmd) Val() []Z { + return cmd.val +} + +func (cmd *ZSliceCmd) Result() ([]Z, error) { + return cmd.val, cmd.err +} + +func (cmd *ZSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ZSliceCmd) readReply(rd *proto.Reader) error { // nolint:dupl + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + // If the n is 0, can't continue reading. + if n == 0 { + cmd.val = make([]Z, 0) + return nil + } + + typ, err := rd.PeekReplyType() + if err != nil { + return err + } + array := typ == proto.RespArray + + if array { + cmd.val = make([]Z, n) + } else { + cmd.val = make([]Z, n/2) + } + + for i := 0; i < len(cmd.val); i++ { + if array { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + } + + if cmd.val[i].Member, err = rd.ReadString(); err != nil { + return err + } + + if cmd.val[i].Score, err = rd.ReadFloat(); err != nil { + return err + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type ZWithKeyCmd struct { + baseCmd + + val *ZWithKey +} + +var _ Cmder = (*ZWithKeyCmd)(nil) + +func NewZWithKeyCmd(ctx context.Context, args ...interface{}) *ZWithKeyCmd { + return &ZWithKeyCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ZWithKeyCmd) SetVal(val *ZWithKey) { + cmd.val = val +} + +func (cmd *ZWithKeyCmd) Val() *ZWithKey { + return cmd.val +} + +func (cmd *ZWithKeyCmd) Result() (*ZWithKey, error) { + return cmd.val, cmd.err +} + +func (cmd *ZWithKeyCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ZWithKeyCmd) readReply(rd *proto.Reader) (err error) { + if err = rd.ReadFixedArrayLen(3); err != nil { + return err + } + cmd.val = &ZWithKey{} + + if cmd.val.Key, err = rd.ReadString(); err != nil { + return err + } + if cmd.val.Member, err = rd.ReadString(); err != nil { + return err + } + if cmd.val.Score, err = rd.ReadFloat(); err != nil { + return err + } + + return nil +} + +//------------------------------------------------------------------------------ + +type ScanCmd struct { + baseCmd + + page []string + cursor uint64 + + process cmdable +} + +var _ Cmder = (*ScanCmd)(nil) + +func NewScanCmd(ctx context.Context, process cmdable, args ...interface{}) *ScanCmd { + return &ScanCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + process: process, + } +} + +func (cmd *ScanCmd) SetVal(page []string, cursor uint64) { + cmd.page = page + cmd.cursor = cursor +} + +func (cmd *ScanCmd) Val() (keys []string, cursor uint64) { + return cmd.page, cmd.cursor +} + +func (cmd *ScanCmd) Result() (keys []string, cursor uint64, err error) { + return cmd.page, cmd.cursor, cmd.err +} + +func (cmd *ScanCmd) String() string { + return cmdString(cmd, cmd.page) +} + +func (cmd *ScanCmd) readReply(rd *proto.Reader) error { + if err := rd.ReadFixedArrayLen(2); err != nil { + return err + } + + cursor, err := rd.ReadUint() + if err != nil { + return err + } + cmd.cursor = cursor + + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.page = make([]string, n) + + for i := 0; i < len(cmd.page); i++ { + if cmd.page[i], err = rd.ReadString(); err != nil { + return err + } + } + return nil +} + +// Iterator creates a new ScanIterator. +func (cmd *ScanCmd) Iterator() *ScanIterator { + return &ScanIterator{ + cmd: cmd, + } +} + +//------------------------------------------------------------------------------ + +type ClusterNode struct { + ID string + Addr string + NetworkingMetadata map[string]string +} + +type ClusterSlot struct { + Start int + End int + Nodes []ClusterNode +} + +type ClusterSlotsCmd struct { + baseCmd + + val []ClusterSlot +} + +var _ Cmder = (*ClusterSlotsCmd)(nil) + +func NewClusterSlotsCmd(ctx context.Context, args ...interface{}) *ClusterSlotsCmd { + return &ClusterSlotsCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ClusterSlotsCmd) SetVal(val []ClusterSlot) { + cmd.val = val +} + +func (cmd *ClusterSlotsCmd) Val() []ClusterSlot { + return cmd.val +} + +func (cmd *ClusterSlotsCmd) Result() ([]ClusterSlot, error) { + return cmd.val, cmd.err +} + +func (cmd *ClusterSlotsCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ClusterSlotsCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]ClusterSlot, n) + + for i := 0; i < len(cmd.val); i++ { + n, err = rd.ReadArrayLen() + if err != nil { + return err + } + if n < 2 { + return fmt.Errorf("redis: got %d elements in cluster info, expected at least 2", n) + } + + start, err := rd.ReadInt() + if err != nil { + return err + } + + end, err := rd.ReadInt() + if err != nil { + return err + } + + // subtract start and end. + nodes := make([]ClusterNode, n-2) + + for j := 0; j < len(nodes); j++ { + nn, err := rd.ReadArrayLen() + if err != nil { + return err + } + if nn < 2 || nn > 4 { + return fmt.Errorf("got %d elements in cluster info address, expected 2, 3, or 4", n) + } + + ip, err := rd.ReadString() + if err != nil { + return err + } + + port, err := rd.ReadString() + if err != nil { + return err + } + + nodes[j].Addr = net.JoinHostPort(ip, port) + + if nn >= 3 { + id, err := rd.ReadString() + if err != nil { + return err + } + nodes[j].ID = id + } + + if nn >= 4 { + metadataLength, err := rd.ReadMapLen() + if err != nil { + return err + } + + networkingMetadata := make(map[string]string, metadataLength) + + for i := 0; i < metadataLength; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + value, err := rd.ReadString() + if err != nil { + return err + } + networkingMetadata[key] = value + } + + nodes[j].NetworkingMetadata = networkingMetadata + } + } + + cmd.val[i] = ClusterSlot{ + Start: int(start), + End: int(end), + Nodes: nodes, + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +// GeoLocation is used with GeoAdd to add geospatial location. +type GeoLocation struct { + Name string + Longitude, Latitude, Dist float64 + GeoHash int64 +} + +// GeoRadiusQuery is used with GeoRadius to query geospatial index. +type GeoRadiusQuery struct { + Radius float64 + // Can be m, km, ft, or mi. Default is km. + Unit string + WithCoord bool + WithDist bool + WithGeoHash bool + Count int + // Can be ASC or DESC. Default is no sort order. + Sort string + Store string + StoreDist string + + // WithCoord+WithDist+WithGeoHash + withLen int +} + +type GeoLocationCmd struct { + baseCmd + + q *GeoRadiusQuery + locations []GeoLocation +} + +var _ Cmder = (*GeoLocationCmd)(nil) + +func NewGeoLocationCmd(ctx context.Context, q *GeoRadiusQuery, args ...interface{}) *GeoLocationCmd { + return &GeoLocationCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: geoLocationArgs(q, args...), + }, + q: q, + } +} + +func geoLocationArgs(q *GeoRadiusQuery, args ...interface{}) []interface{} { + args = append(args, q.Radius) + if q.Unit != "" { + args = append(args, q.Unit) + } else { + args = append(args, "km") + } + if q.WithCoord { + args = append(args, "withcoord") + q.withLen++ + } + if q.WithDist { + args = append(args, "withdist") + q.withLen++ + } + if q.WithGeoHash { + args = append(args, "withhash") + q.withLen++ + } + if q.Count > 0 { + args = append(args, "count", q.Count) + } + if q.Sort != "" { + args = append(args, q.Sort) + } + if q.Store != "" { + args = append(args, "store") + args = append(args, q.Store) + } + if q.StoreDist != "" { + args = append(args, "storedist") + args = append(args, q.StoreDist) + } + return args +} + +func (cmd *GeoLocationCmd) SetVal(locations []GeoLocation) { + cmd.locations = locations +} + +func (cmd *GeoLocationCmd) Val() []GeoLocation { + return cmd.locations +} + +func (cmd *GeoLocationCmd) Result() ([]GeoLocation, error) { + return cmd.locations, cmd.err +} + +func (cmd *GeoLocationCmd) String() string { + return cmdString(cmd, cmd.locations) +} + +func (cmd *GeoLocationCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.locations = make([]GeoLocation, n) + + for i := 0; i < len(cmd.locations); i++ { + // only name + if cmd.q.withLen == 0 { + if cmd.locations[i].Name, err = rd.ReadString(); err != nil { + return err + } + continue + } + + // +name + if err = rd.ReadFixedArrayLen(cmd.q.withLen + 1); err != nil { + return err + } + + if cmd.locations[i].Name, err = rd.ReadString(); err != nil { + return err + } + if cmd.q.WithDist { + if cmd.locations[i].Dist, err = rd.ReadFloat(); err != nil { + return err + } + } + if cmd.q.WithGeoHash { + if cmd.locations[i].GeoHash, err = rd.ReadInt(); err != nil { + return err + } + } + if cmd.q.WithCoord { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + if cmd.locations[i].Longitude, err = rd.ReadFloat(); err != nil { + return err + } + if cmd.locations[i].Latitude, err = rd.ReadFloat(); err != nil { + return err + } + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +// GeoSearchQuery is used for GEOSearch/GEOSearchStore command query. +type GeoSearchQuery struct { + Member string + + // Latitude and Longitude when using FromLonLat option. + Longitude float64 + Latitude float64 + + // Distance and unit when using ByRadius option. + // Can use m, km, ft, or mi. Default is km. + Radius float64 + RadiusUnit string + + // Height, width and unit when using ByBox option. + // Can be m, km, ft, or mi. Default is km. + BoxWidth float64 + BoxHeight float64 + BoxUnit string + + // Can be ASC or DESC. Default is no sort order. + Sort string + Count int + CountAny bool +} + +type GeoSearchLocationQuery struct { + GeoSearchQuery + + WithCoord bool + WithDist bool + WithHash bool +} + +type GeoSearchStoreQuery struct { + GeoSearchQuery + + // When using the StoreDist option, the command stores the items in a + // sorted set populated with their distance from the center of the circle or box, + // as a floating-point number, in the same unit specified for that shape. + StoreDist bool +} + +func geoSearchLocationArgs(q *GeoSearchLocationQuery, args []interface{}) []interface{} { + args = geoSearchArgs(&q.GeoSearchQuery, args) + + if q.WithCoord { + args = append(args, "withcoord") + } + if q.WithDist { + args = append(args, "withdist") + } + if q.WithHash { + args = append(args, "withhash") + } + + return args +} + +func geoSearchArgs(q *GeoSearchQuery, args []interface{}) []interface{} { + if q.Member != "" { + args = append(args, "frommember", q.Member) + } else { + args = append(args, "fromlonlat", q.Longitude, q.Latitude) + } + + if q.Radius > 0 { + if q.RadiusUnit == "" { + q.RadiusUnit = "km" + } + args = append(args, "byradius", q.Radius, q.RadiusUnit) + } else { + if q.BoxUnit == "" { + q.BoxUnit = "km" + } + args = append(args, "bybox", q.BoxWidth, q.BoxHeight, q.BoxUnit) + } + + if q.Sort != "" { + args = append(args, q.Sort) + } + + if q.Count > 0 { + args = append(args, "count", q.Count) + if q.CountAny { + args = append(args, "any") + } + } + + return args +} + +type GeoSearchLocationCmd struct { + baseCmd + + opt *GeoSearchLocationQuery + val []GeoLocation +} + +var _ Cmder = (*GeoSearchLocationCmd)(nil) + +func NewGeoSearchLocationCmd( + ctx context.Context, opt *GeoSearchLocationQuery, args ...interface{}, +) *GeoSearchLocationCmd { + return &GeoSearchLocationCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + opt: opt, + } +} + +func (cmd *GeoSearchLocationCmd) SetVal(val []GeoLocation) { + cmd.val = val +} + +func (cmd *GeoSearchLocationCmd) Val() []GeoLocation { + return cmd.val +} + +func (cmd *GeoSearchLocationCmd) Result() ([]GeoLocation, error) { + return cmd.val, cmd.err +} + +func (cmd *GeoSearchLocationCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *GeoSearchLocationCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make([]GeoLocation, n) + for i := 0; i < n; i++ { + _, err = rd.ReadArrayLen() + if err != nil { + return err + } + + var loc GeoLocation + + loc.Name, err = rd.ReadString() + if err != nil { + return err + } + if cmd.opt.WithDist { + loc.Dist, err = rd.ReadFloat() + if err != nil { + return err + } + } + if cmd.opt.WithHash { + loc.GeoHash, err = rd.ReadInt() + if err != nil { + return err + } + } + if cmd.opt.WithCoord { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + loc.Longitude, err = rd.ReadFloat() + if err != nil { + return err + } + loc.Latitude, err = rd.ReadFloat() + if err != nil { + return err + } + } + + cmd.val[i] = loc + } + + return nil +} + +//------------------------------------------------------------------------------ + +type GeoPos struct { + Longitude, Latitude float64 +} + +type GeoPosCmd struct { + baseCmd + + val []*GeoPos +} + +var _ Cmder = (*GeoPosCmd)(nil) + +func NewGeoPosCmd(ctx context.Context, args ...interface{}) *GeoPosCmd { + return &GeoPosCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *GeoPosCmd) SetVal(val []*GeoPos) { + cmd.val = val +} + +func (cmd *GeoPosCmd) Val() []*GeoPos { + return cmd.val +} + +func (cmd *GeoPosCmd) Result() ([]*GeoPos, error) { + return cmd.val, cmd.err +} + +func (cmd *GeoPosCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *GeoPosCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]*GeoPos, n) + + for i := 0; i < len(cmd.val); i++ { + err = rd.ReadFixedArrayLen(2) + if err != nil { + if err == Nil { + cmd.val[i] = nil + continue + } + return err + } + + longitude, err := rd.ReadFloat() + if err != nil { + return err + } + latitude, err := rd.ReadFloat() + if err != nil { + return err + } + + cmd.val[i] = &GeoPos{ + Longitude: longitude, + Latitude: latitude, + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type CommandInfo struct { + Name string + Arity int8 + Flags []string + ACLFlags []string + FirstKeyPos int8 + LastKeyPos int8 + StepCount int8 + ReadOnly bool +} + +type CommandsInfoCmd struct { + baseCmd + + val map[string]*CommandInfo +} + +var _ Cmder = (*CommandsInfoCmd)(nil) + +func NewCommandsInfoCmd(ctx context.Context, args ...interface{}) *CommandsInfoCmd { + return &CommandsInfoCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *CommandsInfoCmd) SetVal(val map[string]*CommandInfo) { + cmd.val = val +} + +func (cmd *CommandsInfoCmd) Val() map[string]*CommandInfo { + return cmd.val +} + +func (cmd *CommandsInfoCmd) Result() (map[string]*CommandInfo, error) { + return cmd.val, cmd.err +} + +func (cmd *CommandsInfoCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *CommandsInfoCmd) readReply(rd *proto.Reader) error { + const numArgRedis5 = 6 + const numArgRedis6 = 7 + const numArgRedis7 = 10 + + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make(map[string]*CommandInfo, n) + + for i := 0; i < n; i++ { + nn, err := rd.ReadArrayLen() + if err != nil { + return err + } + + switch nn { + case numArgRedis5, numArgRedis6, numArgRedis7: + // ok + default: + return fmt.Errorf("redis: got %d elements in COMMAND reply, wanted 6/7/10", nn) + } + + cmdInfo := &CommandInfo{} + if cmdInfo.Name, err = rd.ReadString(); err != nil { + return err + } + + arity, err := rd.ReadInt() + if err != nil { + return err + } + cmdInfo.Arity = int8(arity) + + flagLen, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmdInfo.Flags = make([]string, flagLen) + for f := 0; f < len(cmdInfo.Flags); f++ { + switch s, err := rd.ReadString(); { + case err == Nil: + cmdInfo.Flags[f] = "" + case err != nil: + return err + default: + if !cmdInfo.ReadOnly && s == "readonly" { + cmdInfo.ReadOnly = true + } + cmdInfo.Flags[f] = s + } + } + + firstKeyPos, err := rd.ReadInt() + if err != nil { + return err + } + cmdInfo.FirstKeyPos = int8(firstKeyPos) + + lastKeyPos, err := rd.ReadInt() + if err != nil { + return err + } + cmdInfo.LastKeyPos = int8(lastKeyPos) + + stepCount, err := rd.ReadInt() + if err != nil { + return err + } + cmdInfo.StepCount = int8(stepCount) + + if nn >= numArgRedis6 { + aclFlagLen, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmdInfo.ACLFlags = make([]string, aclFlagLen) + for f := 0; f < len(cmdInfo.ACLFlags); f++ { + switch s, err := rd.ReadString(); { + case err == Nil: + cmdInfo.ACLFlags[f] = "" + case err != nil: + return err + default: + cmdInfo.ACLFlags[f] = s + } + } + } + + if nn >= numArgRedis7 { + if err := rd.DiscardNext(); err != nil { + return err + } + if err := rd.DiscardNext(); err != nil { + return err + } + if err := rd.DiscardNext(); err != nil { + return err + } + } + + cmd.val[cmdInfo.Name] = cmdInfo + } + + return nil +} + +//------------------------------------------------------------------------------ + +type cmdsInfoCache struct { + fn func(ctx context.Context) (map[string]*CommandInfo, error) + + once internal.Once + cmds map[string]*CommandInfo +} + +func newCmdsInfoCache(fn func(ctx context.Context) (map[string]*CommandInfo, error)) *cmdsInfoCache { + return &cmdsInfoCache{ + fn: fn, + } +} + +func (c *cmdsInfoCache) Get(ctx context.Context) (map[string]*CommandInfo, error) { + err := c.once.Do(func() error { + cmds, err := c.fn(ctx) + if err != nil { + return err + } + + lowerCmds := make(map[string]*CommandInfo, len(cmds)) + + // Extensions have cmd names in upper case. Convert them to lower case. + for k, v := range cmds { + lowerCmds[internal.ToLower(k)] = v + } + + c.cmds = lowerCmds + return nil + }) + return c.cmds, err +} + +//------------------------------------------------------------------------------ + +type SlowLog struct { + ID int64 + Time time.Time + Duration time.Duration + Args []string + // These are also optional fields emitted only by Redis 4.0 or greater: + // https://redis.io/commands/slowlog#output-format + ClientAddr string + ClientName string +} + +type SlowLogCmd struct { + baseCmd + + val []SlowLog +} + +var _ Cmder = (*SlowLogCmd)(nil) + +func NewSlowLogCmd(ctx context.Context, args ...interface{}) *SlowLogCmd { + return &SlowLogCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *SlowLogCmd) SetVal(val []SlowLog) { + cmd.val = val +} + +func (cmd *SlowLogCmd) Val() []SlowLog { + return cmd.val +} + +func (cmd *SlowLogCmd) Result() ([]SlowLog, error) { + return cmd.val, cmd.err +} + +func (cmd *SlowLogCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *SlowLogCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]SlowLog, n) + + for i := 0; i < len(cmd.val); i++ { + nn, err := rd.ReadArrayLen() + if err != nil { + return err + } + if nn < 4 { + return fmt.Errorf("redis: got %d elements in slowlog get, expected at least 4", nn) + } + + if cmd.val[i].ID, err = rd.ReadInt(); err != nil { + return err + } + + createdAt, err := rd.ReadInt() + if err != nil { + return err + } + cmd.val[i].Time = time.Unix(createdAt, 0) + + costs, err := rd.ReadInt() + if err != nil { + return err + } + cmd.val[i].Duration = time.Duration(costs) * time.Microsecond + + cmdLen, err := rd.ReadArrayLen() + if err != nil { + return err + } + if cmdLen < 1 { + return fmt.Errorf("redis: got %d elements commands reply in slowlog get, expected at least 1", cmdLen) + } + + cmd.val[i].Args = make([]string, cmdLen) + for f := 0; f < len(cmd.val[i].Args); f++ { + cmd.val[i].Args[f], err = rd.ReadString() + if err != nil { + return err + } + } + + if nn >= 5 { + if cmd.val[i].ClientAddr, err = rd.ReadString(); err != nil { + return err + } + } + + if nn >= 6 { + if cmd.val[i].ClientName, err = rd.ReadString(); err != nil { + return err + } + } + } + + return nil +} + +//----------------------------------------------------------------------- + +type MapStringInterfaceCmd struct { + baseCmd + + val map[string]interface{} +} + +var _ Cmder = (*MapStringInterfaceCmd)(nil) + +func NewMapStringInterfaceCmd(ctx context.Context, args ...interface{}) *MapStringInterfaceCmd { + return &MapStringInterfaceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapStringInterfaceCmd) SetVal(val map[string]interface{}) { + cmd.val = val +} + +func (cmd *MapStringInterfaceCmd) Val() map[string]interface{} { + return cmd.val +} + +func (cmd *MapStringInterfaceCmd) Result() (map[string]interface{}, error) { + return cmd.val, cmd.err +} + +func (cmd *MapStringInterfaceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *MapStringInterfaceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + + cmd.val = make(map[string]interface{}, n) + for i := 0; i < n; i++ { + k, err := rd.ReadString() + if err != nil { + return err + } + v, err := rd.ReadReply() + if err != nil { + if err == Nil { + cmd.val[k] = Nil + continue + } + if err, ok := err.(proto.RedisError); ok { + cmd.val[k] = err + continue + } + return err + } + cmd.val[k] = v + } + return nil +} + +//----------------------------------------------------------------------- + +type MapStringStringSliceCmd struct { + baseCmd + + val []map[string]string +} + +var _ Cmder = (*MapStringStringSliceCmd)(nil) + +func NewMapStringStringSliceCmd(ctx context.Context, args ...interface{}) *MapStringStringSliceCmd { + return &MapStringStringSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapStringStringSliceCmd) SetVal(val []map[string]string) { + cmd.val = val +} + +func (cmd *MapStringStringSliceCmd) Val() []map[string]string { + return cmd.val +} + +func (cmd *MapStringStringSliceCmd) Result() ([]map[string]string, error) { + return cmd.val, cmd.err +} + +func (cmd *MapStringStringSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *MapStringStringSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make([]map[string]string, n) + for i := 0; i < n; i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return err + } + cmd.val[i] = make(map[string]string, nn) + for f := 0; f < nn; f++ { + k, err := rd.ReadString() + if err != nil { + return err + } + + v, err := rd.ReadString() + if err != nil { + return err + } + cmd.val[i][k] = v + } + } + return nil +} + +// ----------------------------------------------------------------------- + +// MapMapStringInterfaceCmd represents a command that returns a map of strings to interface{}. +type MapMapStringInterfaceCmd struct { + baseCmd + val map[string]interface{} +} + +func NewMapMapStringInterfaceCmd(ctx context.Context, args ...interface{}) *MapMapStringInterfaceCmd { + return &MapMapStringInterfaceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapMapStringInterfaceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *MapMapStringInterfaceCmd) SetVal(val map[string]interface{}) { + cmd.val = val +} + +func (cmd *MapMapStringInterfaceCmd) Result() (map[string]interface{}, error) { + return cmd.val, cmd.err +} + +func (cmd *MapMapStringInterfaceCmd) Val() map[string]interface{} { + return cmd.val +} + +// readReply will try to parse the reply from the proto.Reader for both resp2 and resp3 +func (cmd *MapMapStringInterfaceCmd) readReply(rd *proto.Reader) (err error) { + data, err := rd.ReadReply() + if err != nil { + return err + } + resultMap := map[string]interface{}{} + + switch midResponse := data.(type) { + case map[interface{}]interface{}: // resp3 will return map + for k, v := range midResponse { + stringKey, ok := k.(string) + if !ok { + return fmt.Errorf("redis: invalid map key %#v", k) + } + resultMap[stringKey] = v + } + case []interface{}: // resp2 will return array of arrays + n := len(midResponse) + for i := 0; i < n; i++ { + finalArr, ok := midResponse[i].([]interface{}) // final array that we need to transform to map + if !ok { + return fmt.Errorf("redis: unexpected response %#v", data) + } + m := len(finalArr) + if m%2 != 0 { // since this should be map, keys should be even number + return fmt.Errorf("redis: unexpected response %#v", data) + } + + for j := 0; j < m; j += 2 { + stringKey, ok := finalArr[j].(string) // the first one + if !ok { + return fmt.Errorf("redis: invalid map key %#v", finalArr[i]) + } + resultMap[stringKey] = finalArr[j+1] // second one is value + } + } + default: + return fmt.Errorf("redis: unexpected response %#v", data) + } + + cmd.val = resultMap + return nil +} + +//----------------------------------------------------------------------- + +type MapStringInterfaceSliceCmd struct { + baseCmd + + val []map[string]interface{} +} + +var _ Cmder = (*MapStringInterfaceSliceCmd)(nil) + +func NewMapStringInterfaceSliceCmd(ctx context.Context, args ...interface{}) *MapStringInterfaceSliceCmd { + return &MapStringInterfaceSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *MapStringInterfaceSliceCmd) SetVal(val []map[string]interface{}) { + cmd.val = val +} + +func (cmd *MapStringInterfaceSliceCmd) Val() []map[string]interface{} { + return cmd.val +} + +func (cmd *MapStringInterfaceSliceCmd) Result() ([]map[string]interface{}, error) { + return cmd.val, cmd.err +} + +func (cmd *MapStringInterfaceSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *MapStringInterfaceSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make([]map[string]interface{}, n) + for i := 0; i < n; i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return err + } + cmd.val[i] = make(map[string]interface{}, nn) + for f := 0; f < nn; f++ { + k, err := rd.ReadString() + if err != nil { + return err + } + v, err := rd.ReadReply() + if err != nil { + if err != Nil { + return err + } + } + cmd.val[i][k] = v + } + } + return nil +} + +//------------------------------------------------------------------------------ + +type KeyValuesCmd struct { + baseCmd + + key string + val []string +} + +var _ Cmder = (*KeyValuesCmd)(nil) + +func NewKeyValuesCmd(ctx context.Context, args ...interface{}) *KeyValuesCmd { + return &KeyValuesCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *KeyValuesCmd) SetVal(key string, val []string) { + cmd.key = key + cmd.val = val +} + +func (cmd *KeyValuesCmd) Val() (string, []string) { + return cmd.key, cmd.val +} + +func (cmd *KeyValuesCmd) Result() (string, []string, error) { + return cmd.key, cmd.val, cmd.err +} + +func (cmd *KeyValuesCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *KeyValuesCmd) readReply(rd *proto.Reader) (err error) { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + + cmd.key, err = rd.ReadString() + if err != nil { + return err + } + + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]string, n) + for i := 0; i < n; i++ { + cmd.val[i], err = rd.ReadString() + if err != nil { + return err + } + } + + return nil +} + +//------------------------------------------------------------------------------ + +type ZSliceWithKeyCmd struct { + baseCmd + + key string + val []Z +} + +var _ Cmder = (*ZSliceWithKeyCmd)(nil) + +func NewZSliceWithKeyCmd(ctx context.Context, args ...interface{}) *ZSliceWithKeyCmd { + return &ZSliceWithKeyCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ZSliceWithKeyCmd) SetVal(key string, val []Z) { + cmd.key = key + cmd.val = val +} + +func (cmd *ZSliceWithKeyCmd) Val() (string, []Z) { + return cmd.key, cmd.val +} + +func (cmd *ZSliceWithKeyCmd) Result() (string, []Z, error) { + return cmd.key, cmd.val, cmd.err +} + +func (cmd *ZSliceWithKeyCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ZSliceWithKeyCmd) readReply(rd *proto.Reader) (err error) { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + + cmd.key, err = rd.ReadString() + if err != nil { + return err + } + + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + typ, err := rd.PeekReplyType() + if err != nil { + return err + } + array := typ == proto.RespArray + + if array { + cmd.val = make([]Z, n) + } else { + cmd.val = make([]Z, n/2) + } + + for i := 0; i < len(cmd.val); i++ { + if array { + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + } + + if cmd.val[i].Member, err = rd.ReadString(); err != nil { + return err + } + + if cmd.val[i].Score, err = rd.ReadFloat(); err != nil { + return err + } + } + + return nil +} + +type Function struct { + Name string + Description string + Flags []string +} + +type Library struct { + Name string + Engine string + Functions []Function + Code string +} + +type FunctionListCmd struct { + baseCmd + + val []Library +} + +var _ Cmder = (*FunctionListCmd)(nil) + +func NewFunctionListCmd(ctx context.Context, args ...interface{}) *FunctionListCmd { + return &FunctionListCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *FunctionListCmd) SetVal(val []Library) { + cmd.val = val +} + +func (cmd *FunctionListCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *FunctionListCmd) Val() []Library { + return cmd.val +} + +func (cmd *FunctionListCmd) Result() ([]Library, error) { + return cmd.val, cmd.err +} + +func (cmd *FunctionListCmd) First() (*Library, error) { + if cmd.err != nil { + return nil, cmd.err + } + if len(cmd.val) > 0 { + return &cmd.val[0], nil + } + return nil, Nil +} + +func (cmd *FunctionListCmd) readReply(rd *proto.Reader) (err error) { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + libraries := make([]Library, n) + for i := 0; i < n; i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return err + } + + library := Library{} + for f := 0; f < nn; f++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + switch key { + case "library_name": + library.Name, err = rd.ReadString() + case "engine": + library.Engine, err = rd.ReadString() + case "functions": + library.Functions, err = cmd.readFunctions(rd) + case "library_code": + library.Code, err = rd.ReadString() + default: + return fmt.Errorf("redis: function list unexpected key %s", key) + } + + if err != nil { + return err + } + } + + libraries[i] = library + } + cmd.val = libraries + return nil +} + +func (cmd *FunctionListCmd) readFunctions(rd *proto.Reader) ([]Function, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + functions := make([]Function, n) + for i := 0; i < n; i++ { + nn, err := rd.ReadMapLen() + if err != nil { + return nil, err + } + + function := Function{} + for f := 0; f < nn; f++ { + key, err := rd.ReadString() + if err != nil { + return nil, err + } + + switch key { + case "name": + if function.Name, err = rd.ReadString(); err != nil { + return nil, err + } + case "description": + if function.Description, err = rd.ReadString(); err != nil && err != Nil { + return nil, err + } + case "flags": + // resp set + nx, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + function.Flags = make([]string, nx) + for j := 0; j < nx; j++ { + if function.Flags[j], err = rd.ReadString(); err != nil { + return nil, err + } + } + default: + return nil, fmt.Errorf("redis: function list unexpected key %s", key) + } + } + + functions[i] = function + } + return functions, nil +} + +// FunctionStats contains information about the scripts currently executing on the server, and the available engines +// - Engines: +// Statistics about the engine like number of functions and number of libraries +// - RunningScript: +// The script currently running on the shard we're connecting to. +// For Redis Enterprise and Redis Cloud, this represents the +// function with the longest running time, across all the running functions, on all shards +// - RunningScripts +// All scripts currently running in a Redis Enterprise clustered database. +// Only available on Redis Enterprise +type FunctionStats struct { + Engines []Engine + isRunning bool + rs RunningScript + allrs []RunningScript +} + +func (fs *FunctionStats) Running() bool { + return fs.isRunning +} + +func (fs *FunctionStats) RunningScript() (RunningScript, bool) { + return fs.rs, fs.isRunning +} + +// AllRunningScripts returns all scripts currently running in a Redis Enterprise clustered database. +// Only available on Redis Enterprise +func (fs *FunctionStats) AllRunningScripts() []RunningScript { + return fs.allrs +} + +type RunningScript struct { + Name string + Command []string + Duration time.Duration +} + +type Engine struct { + Language string + LibrariesCount int64 + FunctionsCount int64 +} + +type FunctionStatsCmd struct { + baseCmd + val FunctionStats +} + +var _ Cmder = (*FunctionStatsCmd)(nil) + +func NewFunctionStatsCmd(ctx context.Context, args ...interface{}) *FunctionStatsCmd { + return &FunctionStatsCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *FunctionStatsCmd) SetVal(val FunctionStats) { + cmd.val = val +} + +func (cmd *FunctionStatsCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *FunctionStatsCmd) Val() FunctionStats { + return cmd.val +} + +func (cmd *FunctionStatsCmd) Result() (FunctionStats, error) { + return cmd.val, cmd.err +} + +func (cmd *FunctionStatsCmd) readReply(rd *proto.Reader) (err error) { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + + var key string + var result FunctionStats + for f := 0; f < n; f++ { + key, err = rd.ReadString() + if err != nil { + return err + } + + switch key { + case "running_script": + result.rs, result.isRunning, err = cmd.readRunningScript(rd) + case "engines": + result.Engines, err = cmd.readEngines(rd) + case "all_running_scripts": // Redis Enterprise only + result.allrs, result.isRunning, err = cmd.readRunningScripts(rd) + default: + return fmt.Errorf("redis: function stats unexpected key %s", key) + } + + if err != nil { + return err + } + } + + cmd.val = result + return nil +} + +func (cmd *FunctionStatsCmd) readRunningScript(rd *proto.Reader) (RunningScript, bool, error) { + err := rd.ReadFixedMapLen(3) + if err != nil { + if err == Nil { + return RunningScript{}, false, nil + } + return RunningScript{}, false, err + } + + var runningScript RunningScript + for i := 0; i < 3; i++ { + key, err := rd.ReadString() + if err != nil { + return RunningScript{}, false, err + } + + switch key { + case "name": + runningScript.Name, err = rd.ReadString() + case "duration_ms": + runningScript.Duration, err = cmd.readDuration(rd) + case "command": + runningScript.Command, err = cmd.readCommand(rd) + default: + return RunningScript{}, false, fmt.Errorf("redis: function stats unexpected running_script key %s", key) + } + + if err != nil { + return RunningScript{}, false, err + } + } + + return runningScript, true, nil +} + +func (cmd *FunctionStatsCmd) readEngines(rd *proto.Reader) ([]Engine, error) { + n, err := rd.ReadMapLen() + if err != nil { + return nil, err + } + + engines := make([]Engine, 0, n) + for i := 0; i < n; i++ { + engine := Engine{} + engine.Language, err = rd.ReadString() + if err != nil { + return nil, err + } + + err = rd.ReadFixedMapLen(2) + if err != nil { + return nil, fmt.Errorf("redis: function stats unexpected %s engine map length", engine.Language) + } + + for i := 0; i < 2; i++ { + key, err := rd.ReadString() + switch key { + case "libraries_count": + engine.LibrariesCount, err = rd.ReadInt() + case "functions_count": + engine.FunctionsCount, err = rd.ReadInt() + } + if err != nil { + return nil, err + } + } + + engines = append(engines, engine) + } + return engines, nil +} + +func (cmd *FunctionStatsCmd) readDuration(rd *proto.Reader) (time.Duration, error) { + t, err := rd.ReadInt() + if err != nil { + return time.Duration(0), err + } + return time.Duration(t) * time.Millisecond, nil +} + +func (cmd *FunctionStatsCmd) readCommand(rd *proto.Reader) ([]string, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + command := make([]string, 0, n) + for i := 0; i < n; i++ { + x, err := rd.ReadString() + if err != nil { + return nil, err + } + command = append(command, x) + } + + return command, nil +} + +func (cmd *FunctionStatsCmd) readRunningScripts(rd *proto.Reader) ([]RunningScript, bool, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, false, err + } + + runningScripts := make([]RunningScript, 0, n) + for i := 0; i < n; i++ { + rs, _, err := cmd.readRunningScript(rd) + if err != nil { + return nil, false, err + } + runningScripts = append(runningScripts, rs) + } + + return runningScripts, len(runningScripts) > 0, nil +} + +//------------------------------------------------------------------------------ + +// LCSQuery is a parameter used for the LCS command +type LCSQuery struct { + Key1 string + Key2 string + Len bool + Idx bool + MinMatchLen int + WithMatchLen bool +} + +// LCSMatch is the result set of the LCS command. +type LCSMatch struct { + MatchString string + Matches []LCSMatchedPosition + Len int64 +} + +type LCSMatchedPosition struct { + Key1 LCSPosition + Key2 LCSPosition + + // only for withMatchLen is true + MatchLen int64 +} + +type LCSPosition struct { + Start int64 + End int64 +} + +type LCSCmd struct { + baseCmd + + // 1: match string + // 2: match len + // 3: match idx LCSMatch + readType uint8 + val *LCSMatch +} + +func NewLCSCmd(ctx context.Context, q *LCSQuery) *LCSCmd { + args := make([]interface{}, 3, 7) + args[0] = "lcs" + args[1] = q.Key1 + args[2] = q.Key2 + + cmd := &LCSCmd{readType: 1} + if q.Len { + cmd.readType = 2 + args = append(args, "len") + } else if q.Idx { + cmd.readType = 3 + args = append(args, "idx") + if q.MinMatchLen != 0 { + args = append(args, "minmatchlen", q.MinMatchLen) + } + if q.WithMatchLen { + args = append(args, "withmatchlen") + } + } + cmd.baseCmd = baseCmd{ + ctx: ctx, + args: args, + } + + return cmd +} + +func (cmd *LCSCmd) SetVal(val *LCSMatch) { + cmd.val = val +} + +func (cmd *LCSCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *LCSCmd) Val() *LCSMatch { + return cmd.val +} + +func (cmd *LCSCmd) Result() (*LCSMatch, error) { + return cmd.val, cmd.err +} + +func (cmd *LCSCmd) readReply(rd *proto.Reader) (err error) { + lcs := &LCSMatch{} + switch cmd.readType { + case 1: + // match string + if lcs.MatchString, err = rd.ReadString(); err != nil { + return err + } + case 2: + // match len + if lcs.Len, err = rd.ReadInt(); err != nil { + return err + } + case 3: + // read LCSMatch + if err = rd.ReadFixedMapLen(2); err != nil { + return err + } + + // read matches or len field + for i := 0; i < 2; i++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + switch key { + case "matches": + // read array of matched positions + if lcs.Matches, err = cmd.readMatchedPositions(rd); err != nil { + return err + } + case "len": + // read match length + if lcs.Len, err = rd.ReadInt(); err != nil { + return err + } + } + } + } + + cmd.val = lcs + return nil +} + +func (cmd *LCSCmd) readMatchedPositions(rd *proto.Reader) ([]LCSMatchedPosition, error) { + n, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + positions := make([]LCSMatchedPosition, n) + for i := 0; i < n; i++ { + pn, err := rd.ReadArrayLen() + if err != nil { + return nil, err + } + + if positions[i].Key1, err = cmd.readPosition(rd); err != nil { + return nil, err + } + if positions[i].Key2, err = cmd.readPosition(rd); err != nil { + return nil, err + } + + // read match length if WithMatchLen is true + if pn > 2 { + if positions[i].MatchLen, err = rd.ReadInt(); err != nil { + return nil, err + } + } + } + + return positions, nil +} + +func (cmd *LCSCmd) readPosition(rd *proto.Reader) (pos LCSPosition, err error) { + if err = rd.ReadFixedArrayLen(2); err != nil { + return pos, err + } + if pos.Start, err = rd.ReadInt(); err != nil { + return pos, err + } + if pos.End, err = rd.ReadInt(); err != nil { + return pos, err + } + + return pos, nil +} + +// ------------------------------------------------------------------------ + +type KeyFlags struct { + Key string + Flags []string +} + +type KeyFlagsCmd struct { + baseCmd + + val []KeyFlags +} + +var _ Cmder = (*KeyFlagsCmd)(nil) + +func NewKeyFlagsCmd(ctx context.Context, args ...interface{}) *KeyFlagsCmd { + return &KeyFlagsCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *KeyFlagsCmd) SetVal(val []KeyFlags) { + cmd.val = val +} + +func (cmd *KeyFlagsCmd) Val() []KeyFlags { + return cmd.val +} + +func (cmd *KeyFlagsCmd) Result() ([]KeyFlags, error) { + return cmd.val, cmd.err +} + +func (cmd *KeyFlagsCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *KeyFlagsCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + if n == 0 { + cmd.val = make([]KeyFlags, 0) + return nil + } + + cmd.val = make([]KeyFlags, n) + + for i := 0; i < len(cmd.val); i++ { + + if err = rd.ReadFixedArrayLen(2); err != nil { + return err + } + + if cmd.val[i].Key, err = rd.ReadString(); err != nil { + return err + } + flagsLen, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val[i].Flags = make([]string, flagsLen) + + for j := 0; j < flagsLen; j++ { + if cmd.val[i].Flags[j], err = rd.ReadString(); err != nil { + return err + } + } + } + + return nil +} + +// --------------------------------------------------------------------------------------------------- + +type ClusterLink struct { + Direction string + Node string + CreateTime int64 + Events string + SendBufferAllocated int64 + SendBufferUsed int64 +} + +type ClusterLinksCmd struct { + baseCmd + + val []ClusterLink +} + +var _ Cmder = (*ClusterLinksCmd)(nil) + +func NewClusterLinksCmd(ctx context.Context, args ...interface{}) *ClusterLinksCmd { + return &ClusterLinksCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ClusterLinksCmd) SetVal(val []ClusterLink) { + cmd.val = val +} + +func (cmd *ClusterLinksCmd) Val() []ClusterLink { + return cmd.val +} + +func (cmd *ClusterLinksCmd) Result() ([]ClusterLink, error) { + return cmd.val, cmd.err +} + +func (cmd *ClusterLinksCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ClusterLinksCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]ClusterLink, n) + + for i := 0; i < len(cmd.val); i++ { + m, err := rd.ReadMapLen() + if err != nil { + return err + } + + for j := 0; j < m; j++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + switch key { + case "direction": + cmd.val[i].Direction, err = rd.ReadString() + case "node": + cmd.val[i].Node, err = rd.ReadString() + case "create-time": + cmd.val[i].CreateTime, err = rd.ReadInt() + case "events": + cmd.val[i].Events, err = rd.ReadString() + case "send-buffer-allocated": + cmd.val[i].SendBufferAllocated, err = rd.ReadInt() + case "send-buffer-used": + cmd.val[i].SendBufferUsed, err = rd.ReadInt() + default: + return fmt.Errorf("redis: unexpected key %q in CLUSTER LINKS reply", key) + } + + if err != nil { + return err + } + } + } + + return nil +} + +// ------------------------------------------------------------------------------------------------------------------ + +type SlotRange struct { + Start int64 + End int64 +} + +type Node struct { + ID string + Endpoint string + IP string + Hostname string + Port int64 + TLSPort int64 + Role string + ReplicationOffset int64 + Health string +} + +type ClusterShard struct { + Slots []SlotRange + Nodes []Node +} + +type ClusterShardsCmd struct { + baseCmd + + val []ClusterShard +} + +var _ Cmder = (*ClusterShardsCmd)(nil) + +func NewClusterShardsCmd(ctx context.Context, args ...interface{}) *ClusterShardsCmd { + return &ClusterShardsCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ClusterShardsCmd) SetVal(val []ClusterShard) { + cmd.val = val +} + +func (cmd *ClusterShardsCmd) Val() []ClusterShard { + return cmd.val +} + +func (cmd *ClusterShardsCmd) Result() ([]ClusterShard, error) { + return cmd.val, cmd.err +} + +func (cmd *ClusterShardsCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ClusterShardsCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val = make([]ClusterShard, n) + + for i := 0; i < n; i++ { + m, err := rd.ReadMapLen() + if err != nil { + return err + } + + for j := 0; j < m; j++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + switch key { + case "slots": + l, err := rd.ReadArrayLen() + if err != nil { + return err + } + for k := 0; k < l; k += 2 { + start, err := rd.ReadInt() + if err != nil { + return err + } + + end, err := rd.ReadInt() + if err != nil { + return err + } + + cmd.val[i].Slots = append(cmd.val[i].Slots, SlotRange{Start: start, End: end}) + } + case "nodes": + nodesLen, err := rd.ReadArrayLen() + if err != nil { + return err + } + cmd.val[i].Nodes = make([]Node, nodesLen) + for k := 0; k < nodesLen; k++ { + nodeMapLen, err := rd.ReadMapLen() + if err != nil { + return err + } + + for l := 0; l < nodeMapLen; l++ { + nodeKey, err := rd.ReadString() + if err != nil { + return err + } + + switch nodeKey { + case "id": + cmd.val[i].Nodes[k].ID, err = rd.ReadString() + case "endpoint": + cmd.val[i].Nodes[k].Endpoint, err = rd.ReadString() + case "ip": + cmd.val[i].Nodes[k].IP, err = rd.ReadString() + case "hostname": + cmd.val[i].Nodes[k].Hostname, err = rd.ReadString() + case "port": + cmd.val[i].Nodes[k].Port, err = rd.ReadInt() + case "tls-port": + cmd.val[i].Nodes[k].TLSPort, err = rd.ReadInt() + case "role": + cmd.val[i].Nodes[k].Role, err = rd.ReadString() + case "replication-offset": + cmd.val[i].Nodes[k].ReplicationOffset, err = rd.ReadInt() + case "health": + cmd.val[i].Nodes[k].Health, err = rd.ReadString() + default: + return fmt.Errorf("redis: unexpected key %q in CLUSTER SHARDS node reply", nodeKey) + } + + if err != nil { + return err + } + } + } + default: + return fmt.Errorf("redis: unexpected key %q in CLUSTER SHARDS reply", key) + } + } + } + + return nil +} + +// ----------------------------------------- + +type RankScore struct { + Rank int64 + Score float64 +} + +type RankWithScoreCmd struct { + baseCmd + + val RankScore +} + +var _ Cmder = (*RankWithScoreCmd)(nil) + +func NewRankWithScoreCmd(ctx context.Context, args ...interface{}) *RankWithScoreCmd { + return &RankWithScoreCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *RankWithScoreCmd) SetVal(val RankScore) { + cmd.val = val +} + +func (cmd *RankWithScoreCmd) Val() RankScore { + return cmd.val +} + +func (cmd *RankWithScoreCmd) Result() (RankScore, error) { + return cmd.val, cmd.err +} + +func (cmd *RankWithScoreCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *RankWithScoreCmd) readReply(rd *proto.Reader) error { + if err := rd.ReadFixedArrayLen(2); err != nil { + return err + } + + rank, err := rd.ReadInt() + if err != nil { + return err + } + + score, err := rd.ReadFloat() + if err != nil { + return err + } + + cmd.val = RankScore{Rank: rank, Score: score} + + return nil +} + +// -------------------------------------------------------------------------------------------------- + +// ClientFlags is redis-server client flags, copy from redis/src/server.h (redis 7.0) +type ClientFlags uint64 + +const ( + ClientSlave ClientFlags = 1 << 0 /* This client is a replica */ + ClientMaster ClientFlags = 1 << 1 /* This client is a master */ + ClientMonitor ClientFlags = 1 << 2 /* This client is a slave monitor, see MONITOR */ + ClientMulti ClientFlags = 1 << 3 /* This client is in a MULTI context */ + ClientBlocked ClientFlags = 1 << 4 /* The client is waiting in a blocking operation */ + ClientDirtyCAS ClientFlags = 1 << 5 /* Watched keys modified. EXEC will fail. */ + ClientCloseAfterReply ClientFlags = 1 << 6 /* Close after writing entire reply. */ + ClientUnBlocked ClientFlags = 1 << 7 /* This client was unblocked and is stored in server.unblocked_clients */ + ClientScript ClientFlags = 1 << 8 /* This is a non-connected client used by Lua */ + ClientAsking ClientFlags = 1 << 9 /* Client issued the ASKING command */ + ClientCloseASAP ClientFlags = 1 << 10 /* Close this client ASAP */ + ClientUnixSocket ClientFlags = 1 << 11 /* Client connected via Unix domain socket */ + ClientDirtyExec ClientFlags = 1 << 12 /* EXEC will fail for errors while queueing */ + ClientMasterForceReply ClientFlags = 1 << 13 /* Queue replies even if is master */ + ClientForceAOF ClientFlags = 1 << 14 /* Force AOF propagation of current cmd. */ + ClientForceRepl ClientFlags = 1 << 15 /* Force replication of current cmd. */ + ClientPrePSync ClientFlags = 1 << 16 /* Instance don't understand PSYNC. */ + ClientReadOnly ClientFlags = 1 << 17 /* Cluster client is in read-only state. */ + ClientPubSub ClientFlags = 1 << 18 /* Client is in Pub/Sub mode. */ + ClientPreventAOFProp ClientFlags = 1 << 19 /* Don't propagate to AOF. */ + ClientPreventReplProp ClientFlags = 1 << 20 /* Don't propagate to slaves. */ + ClientPreventProp ClientFlags = ClientPreventAOFProp | ClientPreventReplProp + ClientPendingWrite ClientFlags = 1 << 21 /* Client has output to send but a-write handler is yet not installed. */ + ClientReplyOff ClientFlags = 1 << 22 /* Don't send replies to client. */ + ClientReplySkipNext ClientFlags = 1 << 23 /* Set ClientREPLY_SKIP for next cmd */ + ClientReplySkip ClientFlags = 1 << 24 /* Don't send just this reply. */ + ClientLuaDebug ClientFlags = 1 << 25 /* Run EVAL in debug mode. */ + ClientLuaDebugSync ClientFlags = 1 << 26 /* EVAL debugging without fork() */ + ClientModule ClientFlags = 1 << 27 /* Non connected client used by some module. */ + ClientProtected ClientFlags = 1 << 28 /* Client should not be freed for now. */ + ClientExecutingCommand ClientFlags = 1 << 29 /* Indicates that the client is currently in the process of handling + a command. usually this will be marked only during call() + however, blocked clients might have this flag kept until they + will try to reprocess the command. */ + ClientPendingCommand ClientFlags = 1 << 30 /* Indicates the client has a fully * parsed command ready for execution. */ + ClientTracking ClientFlags = 1 << 31 /* Client enabled keys tracking in order to perform client side caching. */ + ClientTrackingBrokenRedir ClientFlags = 1 << 32 /* Target client is invalid. */ + ClientTrackingBCAST ClientFlags = 1 << 33 /* Tracking in BCAST mode. */ + ClientTrackingOptIn ClientFlags = 1 << 34 /* Tracking in opt-in mode. */ + ClientTrackingOptOut ClientFlags = 1 << 35 /* Tracking in opt-out mode. */ + ClientTrackingCaching ClientFlags = 1 << 36 /* CACHING yes/no was given, depending on optin/optout mode. */ + ClientTrackingNoLoop ClientFlags = 1 << 37 /* Don't send invalidation messages about writes performed by myself.*/ + ClientInTimeoutTable ClientFlags = 1 << 38 /* This client is in the timeout table. */ + ClientProtocolError ClientFlags = 1 << 39 /* Protocol error chatting with it. */ + ClientCloseAfterCommand ClientFlags = 1 << 40 /* Close after executing commands * and writing entire reply. */ + ClientDenyBlocking ClientFlags = 1 << 41 /* Indicate that the client should not be blocked. currently, turned on inside MULTI, Lua, RM_Call, and AOF client */ + ClientReplRDBOnly ClientFlags = 1 << 42 /* This client is a replica that only wants RDB without replication buffer. */ + ClientNoEvict ClientFlags = 1 << 43 /* This client is protected against client memory eviction. */ + ClientAllowOOM ClientFlags = 1 << 44 /* Client used by RM_Call is allowed to fully execute scripts even when in OOM */ + ClientNoTouch ClientFlags = 1 << 45 /* This client will not touch LFU/LRU stats. */ + ClientPushing ClientFlags = 1 << 46 /* This client is pushing notifications. */ +) + +// ClientInfo is redis-server ClientInfo, not go-redis *Client +type ClientInfo struct { + ID int64 // redis version 2.8.12, a unique 64-bit client ID + Addr string // address/port of the client + LAddr string // address/port of local address client connected to (bind address) + FD int64 // file descriptor corresponding to the socket + Name string // the name set by the client with CLIENT SETNAME + Age time.Duration // total duration of the connection in seconds + Idle time.Duration // idle time of the connection in seconds + Flags ClientFlags // client flags (see below) + DB int // current database ID + Sub int // number of channel subscriptions + PSub int // number of pattern matching subscriptions + SSub int // redis version 7.0.3, number of shard channel subscriptions + Multi int // number of commands in a MULTI/EXEC context + Watch int // redis version 7.4 RC1, number of keys this client is currently watching. + QueryBuf int // qbuf, query buffer length (0 means no query pending) + QueryBufFree int // qbuf-free, free space of the query buffer (0 means the buffer is full) + ArgvMem int // incomplete arguments for the next command (already extracted from query buffer) + MultiMem int // redis version 7.0, memory is used up by buffered multi commands + BufferSize int // rbs, usable size of buffer + BufferPeak int // rbp, peak used size of buffer in last 5 sec interval + OutputBufferLength int // obl, output buffer length + OutputListLength int // oll, output list length (replies are queued in this list when the buffer is full) + OutputMemory int // omem, output buffer memory usage + TotalMemory int // tot-mem, total memory consumed by this client in its various buffers + TotalNetIn int // tot-net-in, total network input + TotalNetOut int // tot-net-out, total network output + TotalCmds int // tot-cmds, total number of commands processed + IoThread int // io-thread id + Events string // file descriptor events (see below) + LastCmd string // cmd, last command played + User string // the authenticated username of the client + Redir int64 // client id of current client tracking redirection + Resp int // redis version 7.0, client RESP protocol version + LibName string // redis version 7.2, client library name + LibVer string // redis version 7.2, client library version +} + +type ClientInfoCmd struct { + baseCmd + + val *ClientInfo +} + +var _ Cmder = (*ClientInfoCmd)(nil) + +func NewClientInfoCmd(ctx context.Context, args ...interface{}) *ClientInfoCmd { + return &ClientInfoCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ClientInfoCmd) SetVal(val *ClientInfo) { + cmd.val = val +} + +func (cmd *ClientInfoCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ClientInfoCmd) Val() *ClientInfo { + return cmd.val +} + +func (cmd *ClientInfoCmd) Result() (*ClientInfo, error) { + return cmd.val, cmd.err +} + +func (cmd *ClientInfoCmd) readReply(rd *proto.Reader) (err error) { + txt, err := rd.ReadString() + if err != nil { + return err + } + + // sds o = catClientInfoString(sdsempty(), c); + // o = sdscatlen(o,"\n",1); + // addReplyVerbatim(c,o,sdslen(o),"txt"); + // sdsfree(o); + cmd.val, err = parseClientInfo(strings.TrimSpace(txt)) + return err +} + +// fmt.Sscanf() cannot handle null values +func parseClientInfo(txt string) (info *ClientInfo, err error) { + info = &ClientInfo{} + for _, s := range strings.Split(txt, " ") { + kv := strings.Split(s, "=") + if len(kv) != 2 { + return nil, fmt.Errorf("redis: unexpected client info data (%s)", s) + } + key, val := kv[0], kv[1] + + switch key { + case "id": + info.ID, err = strconv.ParseInt(val, 10, 64) + case "addr": + info.Addr = val + case "laddr": + info.LAddr = val + case "fd": + info.FD, err = strconv.ParseInt(val, 10, 64) + case "name": + info.Name = val + case "age": + var age int + if age, err = strconv.Atoi(val); err == nil { + info.Age = time.Duration(age) * time.Second + } + case "idle": + var idle int + if idle, err = strconv.Atoi(val); err == nil { + info.Idle = time.Duration(idle) * time.Second + } + case "flags": + if val == "N" { + break + } + + for i := 0; i < len(val); i++ { + switch val[i] { + case 'S': + info.Flags |= ClientSlave + case 'O': + info.Flags |= ClientSlave | ClientMonitor + case 'M': + info.Flags |= ClientMaster + case 'P': + info.Flags |= ClientPubSub + case 'x': + info.Flags |= ClientMulti + case 'b': + info.Flags |= ClientBlocked + case 't': + info.Flags |= ClientTracking + case 'R': + info.Flags |= ClientTrackingBrokenRedir + case 'B': + info.Flags |= ClientTrackingBCAST + case 'd': + info.Flags |= ClientDirtyCAS + case 'c': + info.Flags |= ClientCloseAfterCommand + case 'u': + info.Flags |= ClientUnBlocked + case 'A': + info.Flags |= ClientCloseASAP + case 'U': + info.Flags |= ClientUnixSocket + case 'r': + info.Flags |= ClientReadOnly + case 'e': + info.Flags |= ClientNoEvict + case 'T': + info.Flags |= ClientNoTouch + default: + return nil, fmt.Errorf("redis: unexpected client info flags(%s)", string(val[i])) + } + } + case "db": + info.DB, err = strconv.Atoi(val) + case "sub": + info.Sub, err = strconv.Atoi(val) + case "psub": + info.PSub, err = strconv.Atoi(val) + case "ssub": + info.SSub, err = strconv.Atoi(val) + case "multi": + info.Multi, err = strconv.Atoi(val) + case "watch": + info.Watch, err = strconv.Atoi(val) + case "qbuf": + info.QueryBuf, err = strconv.Atoi(val) + case "qbuf-free": + info.QueryBufFree, err = strconv.Atoi(val) + case "argv-mem": + info.ArgvMem, err = strconv.Atoi(val) + case "multi-mem": + info.MultiMem, err = strconv.Atoi(val) + case "rbs": + info.BufferSize, err = strconv.Atoi(val) + case "rbp": + info.BufferPeak, err = strconv.Atoi(val) + case "obl": + info.OutputBufferLength, err = strconv.Atoi(val) + case "oll": + info.OutputListLength, err = strconv.Atoi(val) + case "omem": + info.OutputMemory, err = strconv.Atoi(val) + case "tot-mem": + info.TotalMemory, err = strconv.Atoi(val) + case "tot-net-in": + info.TotalNetIn, err = strconv.Atoi(val) + case "tot-net-out": + info.TotalNetOut, err = strconv.Atoi(val) + case "tot-cmds": + info.TotalCmds, err = strconv.Atoi(val) + case "events": + info.Events = val + case "cmd": + info.LastCmd = val + case "user": + info.User = val + case "redir": + info.Redir, err = strconv.ParseInt(val, 10, 64) + case "resp": + info.Resp, err = strconv.Atoi(val) + case "lib-name": + info.LibName = val + case "lib-ver": + info.LibVer = val + case "io-thread": + info.IoThread, err = strconv.Atoi(val) + default: + return nil, fmt.Errorf("redis: unexpected client info key(%s)", key) + } + + if err != nil { + return nil, err + } + } + + return info, nil +} + +// ------------------------------------------- + +type ACLLogEntry struct { + Count int64 + Reason string + Context string + Object string + Username string + AgeSeconds float64 + ClientInfo *ClientInfo + EntryID int64 + TimestampCreated int64 + TimestampLastUpdated int64 +} + +type ACLLogCmd struct { + baseCmd + + val []*ACLLogEntry +} + +var _ Cmder = (*ACLLogCmd)(nil) + +func NewACLLogCmd(ctx context.Context, args ...interface{}) *ACLLogCmd { + return &ACLLogCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *ACLLogCmd) SetVal(val []*ACLLogEntry) { + cmd.val = val +} + +func (cmd *ACLLogCmd) Val() []*ACLLogEntry { + return cmd.val +} + +func (cmd *ACLLogCmd) Result() ([]*ACLLogEntry, error) { + return cmd.val, cmd.err +} + +func (cmd *ACLLogCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *ACLLogCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadArrayLen() + if err != nil { + return err + } + + cmd.val = make([]*ACLLogEntry, n) + for i := 0; i < n; i++ { + cmd.val[i] = &ACLLogEntry{} + entry := cmd.val[i] + respLen, err := rd.ReadMapLen() + if err != nil { + return err + } + for j := 0; j < respLen; j++ { + key, err := rd.ReadString() + if err != nil { + return err + } + + switch key { + case "count": + entry.Count, err = rd.ReadInt() + case "reason": + entry.Reason, err = rd.ReadString() + case "context": + entry.Context, err = rd.ReadString() + case "object": + entry.Object, err = rd.ReadString() + case "username": + entry.Username, err = rd.ReadString() + case "age-seconds": + entry.AgeSeconds, err = rd.ReadFloat() + case "client-info": + txt, err := rd.ReadString() + if err != nil { + return err + } + entry.ClientInfo, err = parseClientInfo(strings.TrimSpace(txt)) + if err != nil { + return err + } + case "entry-id": + entry.EntryID, err = rd.ReadInt() + case "timestamp-created": + entry.TimestampCreated, err = rd.ReadInt() + case "timestamp-last-updated": + entry.TimestampLastUpdated, err = rd.ReadInt() + default: + return fmt.Errorf("redis: unexpected key %q in ACL LOG reply", key) + } + + if err != nil { + return err + } + } + } + + return nil +} + +// LibraryInfo holds the library info. +type LibraryInfo struct { + LibName *string + LibVer *string +} + +// WithLibraryName returns a valid LibraryInfo with library name only. +func WithLibraryName(libName string) LibraryInfo { + return LibraryInfo{LibName: &libName} +} + +// WithLibraryVersion returns a valid LibraryInfo with library version only. +func WithLibraryVersion(libVer string) LibraryInfo { + return LibraryInfo{LibVer: &libVer} +} + +// ------------------------------------------- + +type InfoCmd struct { + baseCmd + val map[string]map[string]string +} + +var _ Cmder = (*InfoCmd)(nil) + +func NewInfoCmd(ctx context.Context, args ...interface{}) *InfoCmd { + return &InfoCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *InfoCmd) SetVal(val map[string]map[string]string) { + cmd.val = val +} + +func (cmd *InfoCmd) Val() map[string]map[string]string { + return cmd.val +} + +func (cmd *InfoCmd) Result() (map[string]map[string]string, error) { + return cmd.val, cmd.err +} + +func (cmd *InfoCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *InfoCmd) readReply(rd *proto.Reader) error { + val, err := rd.ReadString() + if err != nil { + return err + } + + section := "" + scanner := bufio.NewScanner(strings.NewReader(val)) + for scanner.Scan() { + line := scanner.Text() + if strings.HasPrefix(line, "#") { + if cmd.val == nil { + cmd.val = make(map[string]map[string]string) + } + section = strings.TrimPrefix(line, "# ") + cmd.val[section] = make(map[string]string) + } else if line != "" { + if section == "Modules" { + moduleRe := regexp.MustCompile(`module:name=(.+?),(.+)$`) + kv := moduleRe.FindStringSubmatch(line) + if len(kv) == 3 { + cmd.val[section][kv[1]] = kv[2] + } + } else { + kv := strings.SplitN(line, ":", 2) + if len(kv) == 2 { + cmd.val[section][kv[0]] = kv[1] + } + } + } + } + + return nil +} + +func (cmd *InfoCmd) Item(section, key string) string { + if cmd.val == nil { + return "" + } else if cmd.val[section] == nil { + return "" + } else { + return cmd.val[section][key] + } +} + +type MonitorStatus int + +const ( + monitorStatusIdle MonitorStatus = iota + monitorStatusStart + monitorStatusStop +) + +type MonitorCmd struct { + baseCmd + ch chan string + status MonitorStatus + mu sync.Mutex +} + +func newMonitorCmd(ctx context.Context, ch chan string) *MonitorCmd { + return &MonitorCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: []interface{}{"monitor"}, + }, + ch: ch, + status: monitorStatusIdle, + mu: sync.Mutex{}, + } +} + +func (cmd *MonitorCmd) String() string { + return cmdString(cmd, nil) +} + +func (cmd *MonitorCmd) readReply(rd *proto.Reader) error { + ctx, cancel := context.WithCancel(cmd.ctx) + go func(ctx context.Context) { + for { + select { + case <-ctx.Done(): + return + default: + err := cmd.readMonitor(rd, cancel) + if err != nil { + cmd.err = err + return + } + } + } + }(ctx) + return nil +} + +func (cmd *MonitorCmd) readMonitor(rd *proto.Reader, cancel context.CancelFunc) error { + for { + cmd.mu.Lock() + st := cmd.status + pk, _ := rd.Peek(1) + cmd.mu.Unlock() + if len(pk) != 0 && st == monitorStatusStart { + cmd.mu.Lock() + line, err := rd.ReadString() + cmd.mu.Unlock() + if err != nil { + return err + } + cmd.ch <- line + } + if st == monitorStatusStop { + cancel() + break + } + } + return nil +} + +func (cmd *MonitorCmd) Start() { + cmd.mu.Lock() + defer cmd.mu.Unlock() + cmd.status = monitorStatusStart +} + +func (cmd *MonitorCmd) Stop() { + cmd.mu.Lock() + defer cmd.mu.Unlock() + cmd.status = monitorStatusStop +} + +type VectorScoreSliceCmd struct { + baseCmd + + val []VectorScore +} + +var _ Cmder = (*VectorScoreSliceCmd)(nil) + +func NewVectorInfoSliceCmd(ctx context.Context, args ...any) *VectorScoreSliceCmd { + return &VectorScoreSliceCmd{ + baseCmd: baseCmd{ + ctx: ctx, + args: args, + }, + } +} + +func (cmd *VectorScoreSliceCmd) SetVal(val []VectorScore) { + cmd.val = val +} + +func (cmd *VectorScoreSliceCmd) Val() []VectorScore { + return cmd.val +} + +func (cmd *VectorScoreSliceCmd) Result() ([]VectorScore, error) { + return cmd.val, cmd.err +} + +func (cmd *VectorScoreSliceCmd) String() string { + return cmdString(cmd, cmd.val) +} + +func (cmd *VectorScoreSliceCmd) readReply(rd *proto.Reader) error { + n, err := rd.ReadMapLen() + if err != nil { + return err + } + + cmd.val = make([]VectorScore, n) + for i := 0; i < n; i++ { + name, err := rd.ReadString() + if err != nil { + return err + } + cmd.val[i].Name = name + + score, err := rd.ReadFloat() + if err != nil { + return err + } + cmd.val[i].Score = score + } + return nil +} diff --git a/vendor/github.com/redis/go-redis/v9/commands.go b/vendor/github.com/redis/go-redis/v9/commands.go new file mode 100644 index 0000000..04235a2 --- /dev/null +++ b/vendor/github.com/redis/go-redis/v9/commands.go @@ -0,0 +1,752 @@ +package redis + +import ( + "context" + "encoding" + "errors" + "fmt" + "io" + "net" + "reflect" + "runtime" + "strings" + "time" + + "github.com/redis/go-redis/v9/internal" +) + +// KeepTTL is a Redis KEEPTTL option to keep existing TTL, it requires your redis-server version >= 6.0, +// otherwise you will receive an error: (error) ERR syntax error. +// For example: +// +// rdb.Set(ctx, key, value, redis.KeepTTL) +const KeepTTL = -1 + +func usePrecise(dur time.Duration) bool { + return dur < time.Second || dur%time.Second != 0 +} + +func formatMs(ctx context.Context, dur time.Duration) int64 { + if dur > 0 && dur < time.Millisecond { + internal.Logger.Printf( + ctx, + "specified duration is %s, but minimal supported value is %s - truncating to 1ms", + dur, time.Millisecond, + ) + return 1 + } + return int64(dur / time.Millisecond) +} + +func formatSec(ctx context.Context, dur time.Duration) int64 { + if dur > 0 && dur < time.Second { + internal.Logger.Printf( + ctx, + "specified duration is %s, but minimal supported value is %s - truncating to 1s", + dur, time.Second, + ) + return 1 + } + return int64(dur / time.Second) +} + +func appendArgs(dst, src []interface{}) []interface{} { + if len(src) == 1 { + return appendArg(dst, src[0]) + } + + dst = append(dst, src...) + return dst +} + +func appendArg(dst []interface{}, arg interface{}) []interface{} { + switch arg := arg.(type) { + case []string: + for _, s := range arg { + dst = append(dst, s) + } + return dst + case []interface{}: + dst = append(dst, arg...) + return dst + case map[string]interface{}: + for k, v := range arg { + dst = append(dst, k, v) + } + return dst + case map[string]string: + for k, v := range arg { + dst = append(dst, k, v) + } + return dst + case time.Time, time.Duration, encoding.BinaryMarshaler, net.IP: + return append(dst, arg) + case nil: + return dst + default: + // scan struct field + v := reflect.ValueOf(arg) + if v.Type().Kind() == reflect.Ptr { + if v.IsNil() { + // error: arg is not a valid object + return dst + } + v = v.Elem() + } + + if v.Type().Kind() == reflect.Struct { + return appendStructField(dst, v) + } + + return append(dst, arg) + } +} + +// appendStructField appends the field and value held by the structure v to dst, and returns the appended dst. +func appendStructField(dst []interface{}, v reflect.Value) []interface{} { + typ := v.Type() + for i := 0; i < typ.NumField(); i++ { + tag := typ.Field(i).Tag.Get("redis") + if tag == "" || tag == "-" { + continue + } + name, opt, _ := strings.Cut(tag, ",") + if name == "" { + continue + } + + field := v.Field(i) + + // miss field + if omitEmpty(opt) && isEmptyValue(field) { + continue + } + + if field.CanInterface() { + dst = append(dst, name, field.Interface()) + } + } + + return dst +} + +func omitEmpty(opt string) bool { + for opt != "" { + var name string + name, opt, _ = strings.Cut(opt, ",") + if name == "omitempty" { + return true + } + } + return false +} + +func isEmptyValue(v reflect.Value) bool { + switch v.Kind() { + case reflect.Array, reflect.Map, reflect.Slice, reflect.String: + return v.Len() == 0 + case reflect.Bool: + return !v.Bool() + case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64: + return v.Int() == 0 + case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr: + return v.Uint() == 0 + case reflect.Float32, reflect.Float64: + return v.Float() == 0 + case reflect.Interface, reflect.Pointer: + return v.IsNil() + case reflect.Struct: + if v.Type() == reflect.TypeOf(time.Time{}) { + return v.IsZero() + } + // Only supports the struct time.Time, + // subsequent iterations will follow the func Scan support decoder. + } + return false +} + +type Cmdable interface { + Pipeline() Pipeliner + Pipelined(ctx context.Context, fn func(Pipeliner) error) ([]Cmder, error) + + TxPipelined(ctx context.Context, fn func(Pipeliner) error) ([]Cmder, error) + TxPipeline() Pipeliner + + Command(ctx context.Context) *CommandsInfoCmd + CommandList(ctx context.Context, filter *FilterBy) *StringSliceCmd + CommandGetKeys(ctx context.Context, commands ...interface{}) *StringSliceCmd + CommandGetKeysAndFlags(ctx context.Context, commands ...interface{}) *KeyFlagsCmd + ClientGetName(ctx context.Context) *StringCmd + Echo(ctx context.Context, message interface{}) *StringCmd + Ping(ctx context.Context) *StatusCmd + Quit(ctx context.Context) *StatusCmd + Unlink(ctx context.Context, keys ...string) *IntCmd + + BgRewriteAOF(ctx context.Context) *StatusCmd + BgSave(ctx context.Context) *StatusCmd + ClientKill(ctx context.Context, ipPort string) *StatusCmd + ClientKillByFilter(ctx context.Context, keys ...string) *IntCmd + ClientList(ctx context.Context) *StringCmd + ClientInfo(ctx context.Context) *ClientInfoCmd + ClientPause(ctx context.Context, dur time.Duration) *BoolCmd + ClientUnpause(ctx context.Context) *BoolCmd + ClientID(ctx context.Context) *IntCmd + ClientUnblock(ctx context.Context, id int64) *IntCmd + ClientUnblockWithError(ctx context.Context, id int64) *IntCmd + ClientMaintNotifications(ctx context.Context, enabled bool, endpointType string) *StatusCmd + ConfigGet(ctx context.Context, parameter string) *MapStringStringCmd + ConfigResetStat(ctx context.Context) *StatusCmd + ConfigSet(ctx context.Context, parameter, value string) *StatusCmd + ConfigRewrite(ctx context.Context) *StatusCmd + DBSize(ctx context.Context) *IntCmd + FlushAll(ctx context.Context) *StatusCmd + FlushAllAsync(ctx context.Context) *StatusCmd + FlushDB(ctx context.Context) *StatusCmd + FlushDBAsync(ctx context.Context) *StatusCmd + Info(ctx context.Context, section ...string) *StringCmd + LastSave(ctx context.Context) *IntCmd + Save(ctx context.Context) *StatusCmd + Shutdown(ctx context.Context) *StatusCmd + ShutdownSave(ctx context.Context) *StatusCmd + ShutdownNoSave(ctx context.Context) *StatusCmd + SlaveOf(ctx context.Context, host, port string) *StatusCmd + SlowLogGet(ctx context.Context, num int64) *SlowLogCmd + Time(ctx context.Context) *TimeCmd + DebugObject(ctx context.Context, key string) *StringCmd + MemoryUsage(ctx context.Context, key string, samples ...int) *IntCmd + + ModuleLoadex(ctx context.Context, conf *ModuleLoadexConfig) *StringCmd + + ACLCmdable + BitMapCmdable + ClusterCmdable + GenericCmdable + GeoCmdable + HashCmdable + HyperLogLogCmdable + ListCmdable + ProbabilisticCmdable + PubSubCmdable + ScriptingFunctionsCmdable + SearchCmdable + SetCmdable + SortedSetCmdable + StringCmdable + StreamCmdable + TimeseriesCmdable + JSONCmdable + VectorSetCmdable +} + +type StatefulCmdable interface { + Cmdable + Auth(ctx context.Context, password string) *StatusCmd + AuthACL(ctx context.Context, username, password string) *StatusCmd + Select(ctx context.Context, index int) *StatusCmd + SwapDB(ctx context.Context, index1, index2 int) *StatusCmd + ClientSetName(ctx context.Context, name string) *BoolCmd + ClientSetInfo(ctx context.Context, info LibraryInfo) *StatusCmd + Hello(ctx context.Context, ver int, username, password, clientName string) *MapStringInterfaceCmd +} + +var ( + _ Cmdable = (*Client)(nil) + _ Cmdable = (*Tx)(nil) + _ Cmdable = (*Ring)(nil) + _ Cmdable = (*ClusterClient)(nil) + _ Cmdable = (*Pipeline)(nil) +) + +type cmdable func(ctx context.Context, cmd Cmder) error + +type statefulCmdable func(ctx context.Context, cmd Cmder) error + +//------------------------------------------------------------------------------ + +func (c statefulCmdable) Auth(ctx context.Context, password string) *StatusCmd { + cmd := NewStatusCmd(ctx, "auth", password) + _ = c(ctx, cmd) + return cmd +} + +// AuthACL Perform an AUTH command, using the given user and pass. +// Should be used to authenticate the current connection with one of the connections defined in the ACL list +// when connecting to a Redis 6.0 instance, or greater, that is using the Redis ACL system. +func (c statefulCmdable) AuthACL(ctx context.Context, username, password string) *StatusCmd { + cmd := NewStatusCmd(ctx, "auth", username, password) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) Wait(ctx context.Context, numSlaves int, timeout time.Duration) *IntCmd { + cmd := NewIntCmd(ctx, "wait", numSlaves, int(timeout/time.Millisecond)) + cmd.setReadTimeout(timeout) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) WaitAOF(ctx context.Context, numLocal, numSlaves int, timeout time.Duration) *IntCmd { + cmd := NewIntCmd(ctx, "waitAOF", numLocal, numSlaves, int(timeout/time.Millisecond)) + cmd.setReadTimeout(timeout) + _ = c(ctx, cmd) + return cmd +} + +func (c statefulCmdable) Select(ctx context.Context, index int) *StatusCmd { + cmd := NewStatusCmd(ctx, "select", index) + _ = c(ctx, cmd) + return cmd +} + +func (c statefulCmdable) SwapDB(ctx context.Context, index1, index2 int) *StatusCmd { + cmd := NewStatusCmd(ctx, "swapdb", index1, index2) + _ = c(ctx, cmd) + return cmd +} + +// ClientSetName assigns a name to the connection. +func (c statefulCmdable) ClientSetName(ctx context.Context, name string) *BoolCmd { + cmd := NewBoolCmd(ctx, "client", "setname", name) + _ = c(ctx, cmd) + return cmd +} + +// ClientSetInfo sends a CLIENT SETINFO command with the provided info. +func (c statefulCmdable) ClientSetInfo(ctx context.Context, info LibraryInfo) *StatusCmd { + err := info.Validate() + if err != nil { + panic(err.Error()) + } + + var cmd *StatusCmd + if info.LibName != nil { + libName := fmt.Sprintf("go-redis(%s,%s)", *info.LibName, internal.ReplaceSpaces(runtime.Version())) + cmd = NewStatusCmd(ctx, "client", "setinfo", "LIB-NAME", libName) + } else { + cmd = NewStatusCmd(ctx, "client", "setinfo", "LIB-VER", *info.LibVer) + } + + _ = c(ctx, cmd) + return cmd +} + +// Validate checks if only one field in the struct is non-nil. +func (info LibraryInfo) Validate() error { + if info.LibName != nil && info.LibVer != nil { + return errors.New("both LibName and LibVer cannot be set at the same time") + } + if info.LibName == nil && info.LibVer == nil { + return errors.New("at least one of LibName and LibVer should be set") + } + return nil +} + +// Hello sets the resp protocol used. +func (c statefulCmdable) Hello(ctx context.Context, + ver int, username, password, clientName string, +) *MapStringInterfaceCmd { + args := make([]interface{}, 0, 7) + args = append(args, "hello", ver) + if password != "" { + if username != "" { + args = append(args, "auth", username, password) + } else { + args = append(args, "auth", "default", password) + } + } + if clientName != "" { + args = append(args, "setname", clientName) + } + cmd := NewMapStringInterfaceCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +//------------------------------------------------------------------------------ + +func (c cmdable) Command(ctx context.Context) *CommandsInfoCmd { + cmd := NewCommandsInfoCmd(ctx, "command") + _ = c(ctx, cmd) + return cmd +} + +// FilterBy is used for the `CommandList` command parameter. +type FilterBy struct { + Module string + ACLCat string + Pattern string +} + +func (c cmdable) CommandList(ctx context.Context, filter *FilterBy) *StringSliceCmd { + args := make([]interface{}, 0, 5) + args = append(args, "command", "list") + if filter != nil { + if filter.Module != "" { + args = append(args, "filterby", "module", filter.Module) + } else if filter.ACLCat != "" { + args = append(args, "filterby", "aclcat", filter.ACLCat) + } else if filter.Pattern != "" { + args = append(args, "filterby", "pattern", filter.Pattern) + } + } + cmd := NewStringSliceCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) CommandGetKeys(ctx context.Context, commands ...interface{}) *StringSliceCmd { + args := make([]interface{}, 2+len(commands)) + args[0] = "command" + args[1] = "getkeys" + copy(args[2:], commands) + cmd := NewStringSliceCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) CommandGetKeysAndFlags(ctx context.Context, commands ...interface{}) *KeyFlagsCmd { + args := make([]interface{}, 2+len(commands)) + args[0] = "command" + args[1] = "getkeysandflags" + copy(args[2:], commands) + cmd := NewKeyFlagsCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +// ClientGetName returns the name of the connection. +func (c cmdable) ClientGetName(ctx context.Context) *StringCmd { + cmd := NewStringCmd(ctx, "client", "getname") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) Echo(ctx context.Context, message interface{}) *StringCmd { + cmd := NewStringCmd(ctx, "echo", message) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) Ping(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "ping") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) Do(ctx context.Context, args ...interface{}) *Cmd { + cmd := NewCmd(ctx, args...) + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) Quit(_ context.Context) *StatusCmd { + panic("not implemented") +} + +//------------------------------------------------------------------------------ + +func (c cmdable) BgRewriteAOF(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "bgrewriteaof") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) BgSave(ctx context.Context) *StatusCmd { + cmd := NewStatusCmd(ctx, "bgsave") + _ = c(ctx, cmd) + return cmd +} + +func (c cmdable) ClientKill(ctx context.Context, ipPort string) *StatusCmd { + cmd := NewStatusCmd(ctx, "client", "kill", ipPort) + _ = c(ctx, cmd) + return cmd +} + +// ClientKillByFilter is new style syntax, while the ClientKill is old +// +// CLIENT KILL