func (id *Capture) Read(frame *C.AVFrame) error {
if frame == (*C.AVFrame)(null) {
return fmt.Errorf("buffer error")
}
pkt := C.AVPacket{}
C.av_init_packet(&pkt)
defer C.av_free_packet(&pkt)
if C.av_read_frame(id.context, &pkt) < 0 {
return fmt.Errorf("read frame error")
}
if int(pkt.stream_index) != id.index {
return fmt.Errorf("not video frame")
}
got_frame := C.int(0)
if C.avcodec_decode_video2(id.codec, id.frame, &got_frame, &pkt) < 0 {
return fmt.Errorf("decode frame error")
}
if got_frame != 0 {
if C.sws_scale(id.sws, (**C.uint8_t)(&(id.frame.data[0])), &id.frame.linesize[0], 0, id.codec.height, &frame.data[0], &frame.linesize[0]) >= 0 {
return nil
}
return fmt.Errorf("scale error")
}
return fmt.Errorf("no frame out")
}
////Scale the image slice in srcSlice and put the resulting scaled slice in the image in dst.
func SwsScale(ctxt *Context, src *uint8, str int, y, h int, d *uint8, ds int) int {
cctxt := (*C.struct_SwsContext)(unsafe.Pointer(ctxt))
csrc := (*C.uint8_t)(unsafe.Pointer(src))
cstr := (*C.int)(unsafe.Pointer(&str))
cd := (*C.uint8_t)(unsafe.Pointer(d))
cds := (*C.int)(unsafe.Pointer(&ds))
return int(C.sws_scale(cctxt, &csrc, cstr, C.int(y), C.int(h), &cd, cds))
}
func (this *SwsCtx) Scale(src *Frame, dst *Frame) {
C.sws_scale(
this.swsCtx,
(**C.uint8_t)(unsafe.Pointer(&src.avFrame.data)),
(*_Ctype_int)(unsafe.Pointer(&src.avFrame.linesize)),
0,
C.int(src.Height()),
(**C.uint8_t)(unsafe.Pointer(&dst.avFrame.data)),
(*_Ctype_int)(unsafe.Pointer(&dst.avFrame.linesize)))
}
func (g *Generator) NextFrame() (image.Image, int64, error) {
img := image.NewRGBA(image.Rect(0, 0, g.Width, g.Height))
frame := C.av_frame_alloc()
var pkt C.struct_AVPacket
var frameFinished C.int
for C.av_read_frame(g.avfContext, &pkt) == 0 {
if int(pkt.stream_index) != g.vStreamIndex {
C.av_free_packet(&pkt)
continue
}
if C.avcodec_decode_video2(g.avcContext, frame, &frameFinished, &pkt) <= 0 {
C.av_free_packet(&pkt)
return nil, 0, errors.New("can't decode frame")
}
C.av_free_packet(&pkt)
if frameFinished == 0 {
continue
}
ctx := C.sws_getContext(
C.int(g.Width),
C.int(g.Height),
g.avcContext.pix_fmt,
C.int(g.Width),
C.int(g.Height),
C.PIX_FMT_RGBA,
C.SWS_BICUBIC,
nil,
nil,
nil,
)
if ctx == nil {
return nil, 0, errors.New("can't allocate scaling context")
}
srcSlice := (**C.uint8_t)(&frame.data[0])
srcStride := (*C.int)(&frame.linesize[0])
dst := (**C.uint8_t)(unsafe.Pointer(&img.Pix))
dstStride := (*C.int)(unsafe.Pointer(&[1]int{img.Stride}))
C.sws_scale(
ctx,
srcSlice,
srcStride,
C.int(0),
g.avcContext.height,
dst,
dstStride,
)
break
}
timestamp := int64(C.av_frame_get_best_effort_timestamp(frame))
return img, timestamp, nil
}
func sws_scale(ctx *SwsContext, src *Frame, trg *Frame) int {
in_data := (**C.uint8_t)(unsafe.Pointer(&src.avframe.data))
in_line := (*_Ctype_int)(unsafe.Pointer(&src.avframe.linesize))
out_data := (**C.uint8_t)(unsafe.Pointer(&trg.avframe.data))
out_line := (*_Ctype_int)(unsafe.Pointer(&trg.avframe.linesize))
result := int(C.sws_scale(ctx.sws,
in_data,
in_line,
0,
C.int(src.height),
out_data,
out_line))
return result
}
func (this *SwsCtx) Scale(src *Frame, dst *Frame) {
C.sws_scale(
this.swsCtx,
(**C.uint8_t)(unsafe.Pointer(&src.avFrame.data)),
(*_Ctype_int)(unsafe.Pointer(&src.avFrame.linesize)),
0,
C.int(src.Height()),
(**C.uint8_t)(unsafe.Pointer(&dst.avFrame.data)),
(*_Ctype_int)(unsafe.Pointer(&dst.avFrame.linesize)))
defer func() {
if r := recover(); r != nil {
buf := make([]byte, 1<<16)
runtime.Stack(buf, false)
reason := fmt.Sprintf("%v: %s", r, buf)
fmt.Println("Runtime failure, reason -> %s", reason)
}
}()
}
func (e *Encoder) WriteFrame() error {
e._frame.pts = C.int64_t(e._context.frame_number)
var input_data [3]*C.uint8_t
var input_linesize [3]C.int
switch im := e.im.(type) {
case *image.RGBA:
bpp := 4
input_data = [3]*C.uint8_t{ptr(im.Pix)}
input_linesize = [3]C.int{C.int(e.im.Bounds().Dx() * bpp)}
case *image.NRGBA:
bpp := 4
input_data = [3]*C.uint8_t{ptr(im.Pix)}
input_linesize = [3]C.int{C.int(e.im.Bounds().Dx() * bpp)}
default:
panic("Unknown input image type")
}
// Perform scaling from input type to output type
C.sws_scale(e._swscontext, &input_data[0], &input_linesize[0],
0, e._context.height,
&e._frame.data[0], &e._frame.linesize[0])
outsize := C.avcodec_encode_video(e._context, ptr(e._outbuf),
C.int(len(e._outbuf)), e._frame)
if outsize == 0 {
return nil
}
n, err := e.Output.Write(e._outbuf[:outsize])
if err != nil {
return err
}
if n < int(outsize) {
return fmt.Errorf("Short write, expected %d, wrote %d", outsize, n)
}
return nil
}
// Scale a YUVImage and return the new YUVImage
func Scale(src *jpeg.YUVImage, opts ScaleOptions) (*jpeg.YUVImage, error) {
// Figure out what format we're dealing with
var srcFmt, dstFmt int32
var flags C.int
flags = C.SWS_FULL_CHR_H_INT | C.int(opts.Filter) | C.SWS_ACCURATE_RND
components := 3
var dst jpeg.YUVImage
dstFmt = C.PIX_FMT_YUV444P
dst.Format = jpeg.YUV444
switch src.Format {
case jpeg.YUV444:
srcFmt = C.PIX_FMT_YUV444P
flags |= C.SWS_FULL_CHR_H_INP
case jpeg.YUV422:
srcFmt = C.PIX_FMT_YUV422P
case jpeg.YUV440:
srcFmt = C.PIX_FMT_YUV440P
case jpeg.YUV420:
srcFmt = C.PIX_FMT_YUV420P
case jpeg.Grayscale:
srcFmt = C.PIX_FMT_GRAY8
dstFmt = C.PIX_FMT_GRAY8
components = 1
dst.Format = jpeg.Grayscale
}
// swscale can't handle images smaller than this; pad them
paddedDstWidth := opts.DstWidth
paddedSrcWidth := src.Width
padFactor := 1
for paddedDstWidth < 8 || paddedSrcWidth < 4 {
paddedDstWidth *= 2
paddedSrcWidth *= 2
padFactor *= 2
}
// Get the SWS context
sws := C.sws_getContext(C.int(paddedSrcWidth), C.int(src.Height), srcFmt,
C.int(paddedDstWidth), C.int(opts.DstHeight), dstFmt,
flags, nil, nil, nil)
if sws == nil {
return nil, errors.New("sws_getContext failed")
}
defer C.sws_freeContext(sws)
// We only need 3 planes, but libswscale is stupid and checks the alignment
// of all 4 pointers... better give it a dummy one.
var srcYUVPtr [4](*uint8)
var dstYUVPtr [4](*uint8)
var srcStrides [4](C.int)
var dstStrides [4](C.int)
dst.Width = opts.DstWidth
dst.Height = opts.DstHeight
dstStride := pad(paddedDstWidth, jpeg.AlignSize)
dstFinalPaddedWidth := pad(opts.DstWidth, jpeg.AlignSize)
dstPaddedHeight := pad(opts.DstHeight, jpeg.AlignSize)
// Allocate image planes and pointers
for i := 0; i < components; i++ {
dst.Stride[i] = dstStride
dst.Data[i] = make([]byte, dstStride*dstPaddedHeight)
dstYUVPtr[i] = (*uint8)(unsafe.Pointer(&dst.Data[i][0]))
dstStrides[i] = C.int(dstStride)
// apply horizontal padding if image is too small
if padFactor > 1 {
planeWidth := src.PlaneWidth(i)
paddedWidth := planeWidth * padFactor
planeHeight := src.PlaneHeight(i)
paddedStride := pad(paddedWidth, jpeg.AlignSize)
newData := make([]uint8, paddedStride*planeHeight)
for y := 0; y < planeHeight; y++ {
copy(newData[y*paddedStride:], src.Data[i][y*src.Stride[i]:y*src.Stride[i]+planeWidth])
pixel := src.Data[i][y*src.Stride[i]+planeWidth-1]
for x := planeWidth; x < paddedWidth; x++ {
newData[y*paddedStride+x] = pixel
}
}
srcStrides[i] = C.int(paddedStride)
srcYUVPtr[i] = &newData[0]
} else {
srcStrides[i] = C.int(src.Stride[i])
srcYUVPtr[i] = (*uint8)(unsafe.Pointer(&src.Data[i][0]))
}
}
C.sws_scale(sws, (**C.uint8_t)(unsafe.Pointer(&srcYUVPtr[0])), &srcStrides[0], 0, C.int(src.Height),
(**C.uint8_t)(unsafe.Pointer(&dstYUVPtr[0])), &dstStrides[0])
// Replicate the last column and row of pixels as padding, which is typical
// behavior prior to JPEG compression
for i := 0; i < components; i++ {
for y := 0; y < dst.Height; y++ {
pixel := dst.Data[i][y*dstStride+dst.Width-1]
for x := dst.Width; x < dstFinalPaddedWidth; x++ {
dst.Data[i][y*dstStride+x] = pixel
}
}
lastRow := dst.Data[i][dstStride*(dst.Height-1) : dstStride*dst.Height]
for y := dst.Height; y < dstPaddedHeight; y++ {
copy(dst.Data[i][y*dstStride:], lastRow)
}
}
return &dst, nil
}
// ImageWxH returns a screenshot at the ts milliseconds, scaled to the specified width and height.
func (g *Generator) ImageWxH(ts int64, width, height int) (image.Image, error) {
img := image.NewRGBA(image.Rect(0, 0, width, height))
frame := C.av_frame_alloc()
defer C.av_frame_free(&frame)
frameNum := C.av_rescale(
C.int64_t(ts),
C.int64_t(g.streams[g.vStreamIndex].time_base.den),
C.int64_t(g.streams[g.vStreamIndex].time_base.num),
) / 1000
if C.avformat_seek_file(
g.avfContext,
C.int(g.vStreamIndex),
0,
frameNum,
frameNum,
C.AVSEEK_FLAG_FRAME,
) < 0 {
return nil, errors.New("can't seek to timestamp")
}
C.avcodec_flush_buffers(g.avcContext)
var pkt C.struct_AVPacket
var frameFinished C.int
for C.av_read_frame(g.avfContext, &pkt) == 0 {
if int(pkt.stream_index) != g.vStreamIndex {
C.av_free_packet(&pkt)
continue
}
if C.avcodec_decode_video2(g.avcContext, frame, &frameFinished, &pkt) <= 0 {
C.av_free_packet(&pkt)
return nil, errors.New("can't decode frame")
}
C.av_free_packet(&pkt)
if frameFinished == 0 || pkt.dts < frameNum {
continue
}
ctx := C.sws_getContext(
C.int(g.Width),
C.int(g.Height),
g.avcContext.pix_fmt,
C.int(width),
C.int(height),
C.PIX_FMT_RGBA,
C.SWS_BICUBIC,
nil,
nil,
nil,
)
if ctx == nil {
return nil, errors.New("can't allocate scaling context")
}
srcSlice := (**C.uint8_t)(&frame.data[0])
srcStride := (*C.int)(&frame.linesize[0])
dst := (**C.uint8_t)(unsafe.Pointer(&img.Pix))
dstStride := (*C.int)(unsafe.Pointer(&[1]int{img.Stride}))
C.sws_scale(
ctx,
srcSlice,
srcStride,
0,
g.avcContext.height,
dst,
dstStride,
)
break
}
return img, nil
}
func (self *Decoder) Start(videoStream, audioStream *C.AVStream,
scaleWidth, scaleHeight C.int) *Decoder {
self.running = true
self.duration = time.Duration(self.FormatContext.duration * C.AV_TIME_BASE / 1000)
vCodecCtx := videoStream.codec
aCodecCtx := audioStream.codec
self.durationPerSample = time.Second / time.Duration(aCodecCtx.sample_rate)
// frame pool
poolSize := 16
pool := make(chan *C.AVFrame, poolSize)
self.pool = pool
numBytes := C.size_t(C.avpicture_get_size(C.PIX_FMT_YUV420P, scaleWidth, scaleHeight))
for i := 0; i < poolSize; i++ {
frame := C.av_frame_alloc()
self.frames = append(self.frames, frame)
buffer := (*C.uint8_t)(unsafe.Pointer(C.av_malloc(numBytes)))
self.buffers = append(self.buffers, buffer)
C.avpicture_fill((*C.AVPicture)(unsafe.Pointer(frame)), buffer, C.PIX_FMT_YUV420P,
scaleWidth, scaleHeight)
pool <- frame
}
// decode
self.frameChan = make(chan *C.AVFrame, 512)
go func() {
runtime.LockOSThread()
// scale context
scaleContext := C.sws_getCachedContext(nil, vCodecCtx.width, vCodecCtx.height, vCodecCtx.pix_fmt,
scaleWidth, scaleHeight, C.PIX_FMT_YUV420P, C.SWS_LANCZOS, nil, nil, nil)
if scaleContext == nil {
log.Fatal("get scale context failed")
}
// resample context
resampleContext := C.swr_alloc_set_opts(nil,
C.AV_CH_LAYOUT_STEREO, C.AV_SAMPLE_FMT_FLT, aCodecCtx.sample_rate,
C.int64_t(aCodecCtx.channel_layout), aCodecCtx.sample_fmt, aCodecCtx.sample_rate,
0, nil)
if resampleContext == nil {
log.Fatal("get resample context failed")
}
C.swr_init(resampleContext)
var packet C.AVPacket
var frameFinished C.int
var pts int64
var packetTime time.Duration
vFrame := C.av_frame_alloc()
aFrame := C.av_frame_alloc()
videoIndex := videoStream.index
audioIndex := audioStream.index
resampleBuffer := (*C.uint8_t)(C.av_malloc(4096 * 8))
resampleBufferp := &resampleBuffer
self.Timer = NewTimer()
// decode
for self.running {
// seek
if self.seekTarget > 0 {
if C.av_seek_frame(self.FormatContext, -1,
C.int64_t(float64(self.seekNext)/float64(time.Second)*float64(C.AV_TIME_BASE)),
C.AVSEEK_FLAG_BACKWARD) < 0 {
log.Fatal("seek error")
}
for _, codecCtx := range self.openedCodecs {
C.avcodec_flush_buffers(codecCtx)
}
p("frame seek done\n")
}
read_packet:
// read packet
C.av_free_packet(&packet)
if C.av_read_frame(self.FormatContext, &packet) < 0 { // read packet
log.Fatal("read frame error") //TODO stop gracefully
}
// get packet time
if packet.dts != C.AV_NOPTS_VALUE {
pts = int64(packet.dts)
} else {
pts = 0
}
if packet.stream_index == videoIndex {
packetTime = time.Duration(float64(pts) * float64(C.av_q2d(videoStream.time_base)) * float64(time.Second))
} else if packet.stream_index == audioIndex {
packetTime = time.Duration(float64(pts) * float64(C.av_q2d(audioStream.time_base)) * float64(time.Second))
} else { // ignore packet
goto read_packet
}
p("packet time %v at timer time %v\n", packetTime, self.Timer.Now())
// check seek
if self.seekTarget > 0 && packetTime > 0 { // if packet time cannot determined, skip
if packetTime < self.seekTarget { // seek again
self.seekNext += self.seekStep
p("seek again %v\n", self.seekNext)
} else { // seek ok
p("seek ok\n")
self.seekTarget = 0
self.Timer.Jump(packetTime - self.Timer.Now())
}
}
// decode
if packet.stream_index == videoIndex { // decode video
if C.avcodec_decode_video2(vCodecCtx, vFrame, &frameFinished, &packet) < 0 {
continue // bad packet
}
if frameFinished <= 0 {
goto read_packet // frame not complete
}
bufFrame := <-pool // get frame buffer
C.sws_scale(scaleContext, // scale
&vFrame.data[0], &vFrame.linesize[0], 0, vCodecCtx.height,
&bufFrame.data[0], &bufFrame.linesize[0])
bufFrame.pts = C.int64_t(packetTime) // set packet time
self.frameChan <- bufFrame // push to queue
p("video frame %v\n", packetTime)
} else if packet.stream_index == audioIndex { // decode audio
decode_audio_packet:
l := C.avcodec_decode_audio4(aCodecCtx, aFrame, &frameFinished, &packet)
if l < 0 {
continue // bad packet
}
if frameFinished <= 0 {
goto read_packet // frame not complete
}
if frameFinished > 0 { // frame finished
n := C.swr_convert(resampleContext, resampleBufferp, 4096,
aFrame.extended_data, aFrame.nb_samples)
if n != aFrame.nb_samples {
log.Fatal("audio resample failed")
}
self.audioFrames <- &AudioFrame{
time: packetTime,
data: C.GoBytes(unsafe.Pointer(resampleBuffer), n*8),
}
}
if l != packet.size { // multiple frame packet
packet.size -= l
packet.data = (*C.uint8_t)(unsafe.Pointer(uintptr(unsafe.Pointer(packet.data)) + uintptr(l)))
goto decode_audio_packet
}
p("audio frame %v\n", packetTime)
} else { // other stream
goto read_packet
}
}
}()
// sync video
go func() {
maxDelta := (time.Second / time.Duration(C.av_q2d(videoStream.r_frame_rate)/10))
for frame := range self.frameChan {
delta := time.Duration(frame.pts) - self.Timer.Now()
p("video frame %v, delta %v, max delta %v\n", time.Duration(frame.pts), delta, maxDelta)
if delta > 0 {
if delta > maxDelta {
self.Timer.Jump(delta)
print("timer jumped\n")
} else {
time.Sleep(delta)
}
} else if delta < 0 { // drop frame
self.RecycleFrame(frame)
continue
}
self.timedFrames <- frame
}
}()
return self
}