void CDVDDemuxClient::ParsePacket(DemuxPacket* pkt)
{
CDemuxStream* st = GetStream(pkt->iStreamId);
if (st == nullptr)
return;
if (st->ExtraSize)
return;
CDemuxStreamClientInternal* stream = dynamic_cast<CDemuxStreamClientInternal*>(st);
if (stream == nullptr ||
stream->m_parser == nullptr)
return;
if (stream->m_context == nullptr)
{
AVCodec *codec = avcodec_find_decoder(st->codec);
if (codec == nullptr)
{
CLog::Log(LOGERROR, "%s - can't find decoder", __FUNCTION__);
stream->DisposeParser();
return;
}
stream->m_context = avcodec_alloc_context3(codec);
if (stream->m_context == nullptr)
{
CLog::Log(LOGERROR, "%s - can't allocate context", __FUNCTION__);
stream->DisposeParser();
return;
}
stream->m_context->time_base.num = 1;
stream->m_context->time_base.den = DVD_TIME_BASE;
}
if (stream->m_parser_split && stream->m_parser->parser->split)
{
int len = stream->m_parser->parser->split(stream->m_context, pkt->pData, pkt->iSize);
if (len > 0 && len < FF_MAX_EXTRADATA_SIZE)
{
if (st->ExtraData)
delete[] (uint8_t*)st->ExtraData;
st->changes++;
st->disabled = false;
st->ExtraSize = len;
st->ExtraData = new uint8_t[len+FF_INPUT_BUFFER_PADDING_SIZE];
memcpy(st->ExtraData, pkt->pData, len);
memset((uint8_t*)st->ExtraData + len, 0 , FF_INPUT_BUFFER_PADDING_SIZE);
stream->m_parser_split = false;
}
}
uint8_t *outbuf = nullptr;
int outbuf_size = 0;
int len = av_parser_parse2(stream->m_parser,
stream->m_context, &outbuf, &outbuf_size,
pkt->pData, pkt->iSize,
(int64_t)(pkt->pts * DVD_TIME_BASE),
(int64_t)(pkt->dts * DVD_TIME_BASE),
0);
// our parse is setup to parse complete frames, so we don't care about outbufs
if(len >= 0)
{
if (stream->m_context->profile != st->profile &&
stream->m_context->profile != FF_PROFILE_UNKNOWN)
{
CLog::Log(LOGDEBUG, "CDVDDemuxClient::ParsePacket - {%d} profile changed from %d to %d", st->uniqueId, st->profile, stream->m_context->profile);
st->profile = stream->m_context->profile;
st->changes++;
st->disabled = false;
}
if (stream->m_context->level != st->level &&
stream->m_context->level != FF_LEVEL_UNKNOWN)
{
CLog::Log(LOGDEBUG, "CDVDDemuxClient::ParsePacket - {%d} level changed from %d to %d", st->uniqueId, st->level, stream->m_context->level);
st->level = stream->m_context->level;
st->changes++;
st->disabled = false;
}
switch (st->type)
{
case STREAM_AUDIO:
{
CDemuxStreamClientInternalTpl<CDemuxStreamAudio>* sta = static_cast<CDemuxStreamClientInternalTpl<CDemuxStreamAudio>*>(st);
if (stream->m_context->channels != sta->iChannels &&
stream->m_context->channels != 0)
{
CLog::Log(LOGDEBUG, "CDVDDemuxClient::ParsePacket - {%d} channels changed from %d to %d", st->uniqueId, sta->iChannels, stream->m_context->channels);
sta->iChannels = stream->m_context->channels;
sta->changes++;
sta->disabled = false;
}
if (stream->m_context->sample_rate != sta->iSampleRate &&
stream->m_context->sample_rate != 0)
{
CLog::Log(LOGDEBUG, "CDVDDemuxClient::ParsePacket - {%d} samplerate changed from %d to %d", st->uniqueId, sta->iSampleRate, stream->m_context->sample_rate);
sta->iSampleRate = stream->m_context->sample_rate;
sta->changes++;
sta->disabled = false;
}
break;
}
case STREAM_VIDEO:
{
CDemuxStreamClientInternalTpl<CDemuxStreamVideo>* stv = static_cast<CDemuxStreamClientInternalTpl<CDemuxStreamVideo>*>(st);
if (stream->m_context->width != stv->iWidth &&
stream->m_context->width != 0)
{
CLog::Log(LOGDEBUG, "CDVDDemuxClient::ParsePacket - {%d} width changed from %d to %d", st->uniqueId, stv->iWidth, stream->m_context->width);
stv->iWidth = stream->m_context->width;
stv->changes++;
stv->disabled = false;
}
if (stream->m_context->height != stv->iHeight &&
stream->m_context->height != 0)
{
CLog::Log(LOGDEBUG, "CDVDDemuxClient::ParsePacket - {%d} height changed from %d to %d", st->uniqueId, stv->iHeight, stream->m_context->height);
stv->iHeight = stream->m_context->height;
stv->changes++;
stv->disabled = false;
}
break;
}
default:
break;
}
}
else
CLog::Log(LOGDEBUG, "%s - parser returned error %d", __FUNCTION__, len);
return;
}
STDMETHODIMP CDecAvcodec::Decode(const BYTE *buffer, int buflen, REFERENCE_TIME rtStartIn, REFERENCE_TIME rtStopIn, BOOL bSyncPoint, BOOL bDiscontinuity)
{
int got_picture = 0;
int used_bytes = 0;
BOOL bParserFrame = FALSE;
BOOL bFlush = (buffer == NULL);
BOOL bEndOfSequence = FALSE;
AVPacket avpkt;
av_init_packet(&avpkt);
if (m_pAVCtx->active_thread_type & FF_THREAD_FRAME) {
if (!m_bFFReordering) {
m_tcThreadBuffer[m_CurrentThread].rtStart = rtStartIn;
m_tcThreadBuffer[m_CurrentThread].rtStop = rtStopIn;
}
m_CurrentThread = (m_CurrentThread + 1) % m_pAVCtx->thread_count;
} else if (m_bBFrameDelay) {
m_tcBFrameDelay[m_nBFramePos].rtStart = rtStartIn;
m_tcBFrameDelay[m_nBFramePos].rtStop = rtStopIn;
m_nBFramePos = !m_nBFramePos;
}
uint8_t *pDataBuffer = NULL;
if (!bFlush && buflen > 0) {
if (!m_bInputPadded && (!(m_pAVCtx->active_thread_type & FF_THREAD_FRAME) || m_pParser)) {
// Copy bitstream into temporary buffer to ensure overread protection
// Verify buffer size
if (buflen > m_nFFBufferSize) {
m_nFFBufferSize = buflen;
m_pFFBuffer = (BYTE *)av_realloc_f(m_pFFBuffer, m_nFFBufferSize + FF_INPUT_BUFFER_PADDING_SIZE, 1);
if (!m_pFFBuffer) {
m_nFFBufferSize = 0;
return E_OUTOFMEMORY;
}
}
memcpy(m_pFFBuffer, buffer, buflen);
memset(m_pFFBuffer+buflen, 0, FF_INPUT_BUFFER_PADDING_SIZE);
pDataBuffer = m_pFFBuffer;
} else {
pDataBuffer = (uint8_t *)buffer;
}
if (m_nCodecId == AV_CODEC_ID_H264) {
BOOL bRecovered = m_h264RandomAccess.searchRecoveryPoint(pDataBuffer, buflen);
if (!bRecovered) {
return S_OK;
}
} else if (m_nCodecId == AV_CODEC_ID_VP8 && m_bWaitingForKeyFrame) {
if (!(pDataBuffer[0] & 1)) {
DbgLog((LOG_TRACE, 10, L"::Decode(): Found VP8 key-frame, resuming decoding"));
m_bWaitingForKeyFrame = FALSE;
} else {
return S_OK;
}
}
}
while (buflen > 0 || bFlush) {
REFERENCE_TIME rtStart = rtStartIn, rtStop = rtStopIn;
if (!bFlush) {
avpkt.data = pDataBuffer;
avpkt.size = buflen;
avpkt.pts = rtStartIn;
if (rtStartIn != AV_NOPTS_VALUE && rtStopIn != AV_NOPTS_VALUE)
avpkt.duration = (int)(rtStopIn - rtStartIn);
else
avpkt.duration = 0;
avpkt.flags = AV_PKT_FLAG_KEY;
if (m_bHasPalette) {
m_bHasPalette = FALSE;
uint32_t *pal = (uint32_t *)av_packet_new_side_data(&avpkt, AV_PKT_DATA_PALETTE, AVPALETTE_SIZE);
int pal_size = FFMIN((1 << m_pAVCtx->bits_per_coded_sample) << 2, m_pAVCtx->extradata_size);
uint8_t *pal_src = m_pAVCtx->extradata + m_pAVCtx->extradata_size - pal_size;
for (int i = 0; i < pal_size/4; i++)
pal[i] = 0xFF<<24 | AV_RL32(pal_src+4*i);
}
} else {
avpkt.data = NULL;
avpkt.size = 0;
}
// Parse the data if a parser is present
// This is mandatory for MPEG-1/2
if (m_pParser) {
BYTE *pOut = NULL;
int pOut_size = 0;
used_bytes = av_parser_parse2(m_pParser, m_pAVCtx, &pOut, &pOut_size, avpkt.data, avpkt.size, AV_NOPTS_VALUE, AV_NOPTS_VALUE, 0);
if (used_bytes == 0 && pOut_size == 0 && !bFlush) {
DbgLog((LOG_TRACE, 50, L"::Decode() - could not process buffer, starving?"));
break;
}
// Update start time cache
// If more data was read then output, update the cache (incomplete frame)
// If output is bigger, a frame was completed, update the actual rtStart with the cached value, and then overwrite the cache
if (used_bytes > pOut_size) {
if (rtStartIn != AV_NOPTS_VALUE)
m_rtStartCache = rtStartIn;
} else if (used_bytes == pOut_size || ((used_bytes + 9) == pOut_size)) {
// Why +9 above?
// Well, apparently there are some broken MKV muxers that like to mux the MPEG-2 PICTURE_START_CODE block (which is 9 bytes) in the package with the previous frame
// This would cause the frame timestamps to be delayed by one frame exactly, and cause timestamp reordering to go wrong.
// So instead of failing on those samples, lets just assume that 9 bytes are that case exactly.
m_rtStartCache = rtStartIn = AV_NOPTS_VALUE;
} else if (pOut_size > used_bytes) {
rtStart = m_rtStartCache;
m_rtStartCache = rtStartIn;
// The value was used once, don't use it for multiple frames, that ends up in weird timings
rtStartIn = AV_NOPTS_VALUE;
}
bParserFrame = (pOut_size > 0);
if (pOut_size > 0 || bFlush) {
if (pOut && pOut_size > 0) {
if (pOut_size > m_nFFBufferSize2) {
m_nFFBufferSize2 = pOut_size;
m_pFFBuffer2 = (BYTE *)av_realloc_f(m_pFFBuffer2, m_nFFBufferSize2 + FF_INPUT_BUFFER_PADDING_SIZE, 1);
if (!m_pFFBuffer2) {
m_nFFBufferSize2 = 0;
return E_OUTOFMEMORY;
}
}
memcpy(m_pFFBuffer2, pOut, pOut_size);
memset(m_pFFBuffer2+pOut_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
avpkt.data = m_pFFBuffer2;
avpkt.size = pOut_size;
avpkt.pts = rtStart;
avpkt.duration = 0;
const uint8_t *eosmarker = CheckForEndOfSequence(m_nCodecId, avpkt.data, avpkt.size, &m_MpegParserState);
if (eosmarker) {
bEndOfSequence = TRUE;
}
} else {
avpkt.data = NULL;
avpkt.size = 0;
}
int ret2 = avcodec_decode_video2 (m_pAVCtx, m_pFrame, &got_picture, &avpkt);
if (ret2 < 0) {
DbgLog((LOG_TRACE, 50, L"::Decode() - decoding failed despite successfull parsing"));
got_picture = 0;
}
} else {
got_picture = 0;
}
} else {
used_bytes = avcodec_decode_video2 (m_pAVCtx, m_pFrame, &got_picture, &avpkt);
}
if (FAILED(PostDecode())) {
av_frame_unref(m_pFrame);
return E_FAIL;
}
// Decoding of this frame failed ... oh well!
if (used_bytes < 0) {
av_frame_unref(m_pFrame);
return S_OK;
}
// When Frame Threading, we won't know how much data has been consumed, so it by default eats everything.
// In addition, if no data got consumed, and no picture was extracted, the frame probably isn't all that useufl.
// The MJPEB decoder is somewhat buggy and doesn't let us know how much data was consumed really...
if ((!m_pParser && (m_pAVCtx->active_thread_type & FF_THREAD_FRAME || (!got_picture && used_bytes == 0))) || m_bNoBufferConsumption || bFlush) {
buflen = 0;
} else {
buflen -= used_bytes;
pDataBuffer += used_bytes;
}
// Judge frame usability
// This determines if a frame is artifact free and can be delivered
// For H264 this does some wicked magic hidden away in the H264RandomAccess class
// MPEG-2 and VC-1 just wait for a keyframe..
if (m_nCodecId == AV_CODEC_ID_H264 && (bParserFrame || !m_pParser || got_picture)) {
m_h264RandomAccess.judgeFrameUsability(m_pFrame, &got_picture);
} else if (m_bResumeAtKeyFrame) {
if (m_bWaitingForKeyFrame && got_picture) {
if (m_pFrame->key_frame) {
DbgLog((LOG_TRACE, 50, L"::Decode() - Found Key-Frame, resuming decoding at %I64d", m_pFrame->pkt_pts));
m_bWaitingForKeyFrame = FALSE;
} else {
got_picture = 0;
}
}
}
// Handle B-frame delay for frame threading codecs
if ((m_pAVCtx->active_thread_type & FF_THREAD_FRAME) && m_bBFrameDelay) {
m_tcBFrameDelay[m_nBFramePos] = m_tcThreadBuffer[m_CurrentThread];
m_nBFramePos = !m_nBFramePos;
}
if (!got_picture || !m_pFrame->data[0]) {
if (!avpkt.size)
bFlush = FALSE; // End flushing, no more frames
av_frame_unref(m_pFrame);
continue;
}
///////////////////////////////////////////////////////////////////////////////////////////////
// Determine the proper timestamps for the frame, based on different possible flags.
///////////////////////////////////////////////////////////////////////////////////////////////
if (m_bFFReordering) {
rtStart = m_pFrame->pkt_pts;
if (m_pFrame->pkt_duration)
rtStop = m_pFrame->pkt_pts + m_pFrame->pkt_duration;
else
rtStop = AV_NOPTS_VALUE;
} else if (m_bBFrameDelay && m_pAVCtx->has_b_frames) {
rtStart = m_tcBFrameDelay[m_nBFramePos].rtStart;
rtStop = m_tcBFrameDelay[m_nBFramePos].rtStop;
} else if (m_pAVCtx->active_thread_type & FF_THREAD_FRAME) {
unsigned index = m_CurrentThread;
rtStart = m_tcThreadBuffer[index].rtStart;
rtStop = m_tcThreadBuffer[index].rtStop;
}
if (m_bRVDropBFrameTimings && m_pFrame->pict_type == AV_PICTURE_TYPE_B) {
rtStart = AV_NOPTS_VALUE;
}
if (m_bCalculateStopTime)
rtStop = AV_NOPTS_VALUE;
///////////////////////////////////////////////////////////////////////////////////////////////
// All required values collected, deliver the frame
///////////////////////////////////////////////////////////////////////////////////////////////
LAVFrame *pOutFrame = NULL;
AllocateFrame(&pOutFrame);
AVRational display_aspect_ratio;
int64_t num = (int64_t)m_pFrame->sample_aspect_ratio.num * m_pFrame->width;
int64_t den = (int64_t)m_pFrame->sample_aspect_ratio.den * m_pFrame->height;
av_reduce(&display_aspect_ratio.num, &display_aspect_ratio.den, num, den, 1 << 30);
pOutFrame->width = m_pFrame->width;
pOutFrame->height = m_pFrame->height;
pOutFrame->aspect_ratio = display_aspect_ratio;
pOutFrame->repeat = m_pFrame->repeat_pict;
pOutFrame->key_frame = m_pFrame->key_frame;
pOutFrame->frame_type = av_get_picture_type_char(m_pFrame->pict_type);
pOutFrame->ext_format = GetDXVA2ExtendedFlags(m_pAVCtx, m_pFrame);
if (m_pFrame->interlaced_frame || (!m_pAVCtx->progressive_sequence && (m_nCodecId == AV_CODEC_ID_H264 || m_nCodecId == AV_CODEC_ID_MPEG2VIDEO)))
m_iInterlaced = 1;
else if (m_pAVCtx->progressive_sequence)
m_iInterlaced = 0;
pOutFrame->interlaced = (m_pFrame->interlaced_frame || (m_iInterlaced == 1 && m_pSettings->GetDeinterlacingMode() == DeintMode_Aggressive) || m_pSettings->GetDeinterlacingMode() == DeintMode_Force) && !(m_pSettings->GetDeinterlacingMode() == DeintMode_Disable);
LAVDeintFieldOrder fo = m_pSettings->GetDeintFieldOrder();
pOutFrame->tff = (fo == DeintFieldOrder_Auto) ? m_pFrame->top_field_first : (fo == DeintFieldOrder_TopFieldFirst);
pOutFrame->rtStart = rtStart;
pOutFrame->rtStop = rtStop;
PixelFormatMapping map = getPixFmtMapping((AVPixelFormat)m_pFrame->format);
pOutFrame->format = map.lavpixfmt;
pOutFrame->bpp = map.bpp;
if (m_nCodecId == AV_CODEC_ID_MPEG2VIDEO || m_nCodecId == AV_CODEC_ID_MPEG1VIDEO)
pOutFrame->avgFrameDuration = GetFrameDuration();
if (map.conversion) {
ConvertPixFmt(m_pFrame, pOutFrame);
} else {
for (int i = 0; i < 4; i++) {
pOutFrame->data[i] = m_pFrame->data[i];
pOutFrame->stride[i] = m_pFrame->linesize[i];
}
pOutFrame->priv_data = av_frame_alloc();
av_frame_ref((AVFrame *)pOutFrame->priv_data, m_pFrame);
pOutFrame->destruct = lav_avframe_free;
}
if (bEndOfSequence)
pOutFrame->flags |= LAV_FRAME_FLAG_END_OF_SEQUENCE;
if (pOutFrame->format == LAVPixFmt_DXVA2) {
pOutFrame->data[0] = m_pFrame->data[4];
HandleDXVA2Frame(pOutFrame);
} else {
Deliver(pOutFrame);
}
if (bEndOfSequence) {
bEndOfSequence = FALSE;
if (pOutFrame->format == LAVPixFmt_DXVA2) {
HandleDXVA2Frame(m_pCallback->GetFlushFrame());
} else {
Deliver(m_pCallback->GetFlushFrame());
}
}
if (bFlush) {
m_CurrentThread = (m_CurrentThread + 1) % m_pAVCtx->thread_count;
}
av_frame_unref(m_pFrame);
}
return S_OK;
}
HRESULT CLAVAudio::Bitstream(const BYTE *pDataBuffer, int buffsize, int &consumed, HRESULT *hrDeliver)
{
HRESULT hr = S_OK;
int ret = 0;
BOOL bFlush = (pDataBuffer == nullptr);
AVPacket avpkt;
av_init_packet(&avpkt);
avpkt.duration = 1;
consumed = 0;
while (buffsize > 0) {
if (bFlush) buffsize = 0;
BYTE *pOut = nullptr;
int pOut_size = 0;
int used_bytes = av_parser_parse2(m_pParser, m_pAVCtx, &pOut, &pOut_size, pDataBuffer, buffsize, AV_NOPTS_VALUE, AV_NOPTS_VALUE, 0);
if (used_bytes < 0) {
DbgLog((LOG_TRACE, 50, L"::Bitstream() - audio parsing failed (ret: %d)", -used_bytes));
return E_FAIL;
} else if(used_bytes == 0 && pOut_size == 0) {
DbgLog((LOG_TRACE, 50, L"::Bitstream() - could not process buffer, starving?"));
break;
}
// Timestamp cache to compensate for one frame delay the parser might introduce, in case the frames were already perfectly sliced apart
// If we used more (or equal) bytes then was output again, we encountered a new frame, update timestamps
if (used_bytes >= pOut_size && m_bUpdateTimeCache) {
m_rtStartInputCache = m_rtStartInput;
m_rtStopInputCache = m_rtStopInput;
m_bUpdateTimeCache = FALSE;
}
if (!bFlush && used_bytes > 0) {
buffsize -= used_bytes;
pDataBuffer += used_bytes;
consumed += used_bytes;
}
if (pOut_size > 0) {
hr = m_bsParser.Parse(m_nCodecId, pOut, pOut_size, m_pParser->priv_data);
if (FAILED(hr)) {
continue;
}
if (m_nCodecId == AV_CODEC_ID_DTS && !m_bDTSHD && !m_bForceDTSCore && m_bsParser.m_bDTSHD && m_settings.bBitstream[Bitstream_DTSHD]) {
ActivateDTSHDMuxing();
}
avpkt.data = pOut;
avpkt.size = pOut_size;
// Write SPDIF muxed frame
ret = av_write_frame(m_avBSContext, &avpkt);
if(ret < 0) {
DbgLog((LOG_ERROR, 20, "::Bitstream(): av_write_frame returned error code (%d)", -ret));
m_bsOutput.SetSize(0);
continue;
}
m_bUpdateTimeCache = TRUE;
// Set long-time cache to the first timestamp encountered, used by TrueHD and E-AC3 because the S/PDIF muxer caches data internally
// If the current timestamp is not valid, use the last delivery timestamp in m_rtStart
if (m_rtBitstreamCache == AV_NOPTS_VALUE)
m_rtBitstreamCache = m_rtStartInputCache != AV_NOPTS_VALUE ? m_rtStartInputCache : m_rtStart;
// Deliver frame
if (m_bsOutput.GetCount() > 0) {
*hrDeliver = DeliverBitstream(m_nCodecId, m_bsOutput.Ptr(), m_bsOutput.GetCount(), pOut_size, m_rtStartInputCache, m_rtStopInputCache);
m_bsOutput.SetSize(0);
}
}
}
return S_OK;
}
int ff_qsv_process_data(AVCodecContext *avctx, QSVContext *q,
AVFrame *frame, int *got_frame, AVPacket *pkt)
{
uint8_t *dummy_data;
int dummy_size;
int ret;
if (!q->avctx_internal) {
q->avctx_internal = avcodec_alloc_context3(NULL);
if (!q->avctx_internal)
return AVERROR(ENOMEM);
if (avctx->extradata) {
q->avctx_internal->extradata = av_mallocz(avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
if (!q->avctx_internal->extradata)
return AVERROR(ENOMEM);
memcpy(q->avctx_internal->extradata, avctx->extradata,
avctx->extradata_size);
q->avctx_internal->extradata_size = avctx->extradata_size;
}
q->parser = av_parser_init(avctx->codec_id);
if (!q->parser)
return AVERROR(ENOMEM);
q->parser->flags |= PARSER_FLAG_COMPLETE_FRAMES;
q->orig_pix_fmt = AV_PIX_FMT_NONE;
}
if (!pkt->size)
return qsv_decode(avctx, q, frame, got_frame, pkt);
/* we assume the packets are already split properly and want
* just the codec parameters here */
av_parser_parse2(q->parser, q->avctx_internal,
&dummy_data, &dummy_size,
pkt->data, pkt->size, pkt->pts, pkt->dts,
pkt->pos);
/* TODO: flush delayed frames on reinit */
if (q->parser->format != q->orig_pix_fmt ||
q->parser->coded_width != avctx->coded_width ||
q->parser->coded_height != avctx->coded_height) {
mfxSession session = NULL;
enum AVPixelFormat pix_fmts[3] = { AV_PIX_FMT_QSV,
AV_PIX_FMT_NONE,
AV_PIX_FMT_NONE };
enum AVPixelFormat qsv_format;
qsv_format = ff_qsv_map_pixfmt(q->parser->format);
if (qsv_format < 0) {
av_log(avctx, AV_LOG_ERROR,
"Only 8-bit YUV420 streams are supported.\n");
ret = AVERROR(ENOSYS);
goto reinit_fail;
}
q->orig_pix_fmt = q->parser->format;
avctx->pix_fmt = pix_fmts[1] = qsv_format;
avctx->width = q->parser->width;
avctx->height = q->parser->height;
avctx->coded_width = q->parser->coded_width;
avctx->coded_height = q->parser->coded_height;
avctx->level = q->avctx_internal->level;
avctx->profile = q->avctx_internal->profile;
ret = ff_get_format(avctx, pix_fmts);
if (ret < 0)
goto reinit_fail;
avctx->pix_fmt = ret;
if (avctx->hwaccel_context) {
AVQSVContext *user_ctx = avctx->hwaccel_context;
session = user_ctx->session;
q->iopattern = user_ctx->iopattern;
q->ext_buffers = user_ctx->ext_buffers;
q->nb_ext_buffers = user_ctx->nb_ext_buffers;
}
ret = ff_qsv_decode_init(avctx, q, session);
if (ret < 0)
goto reinit_fail;
}
return qsv_decode(avctx, q, frame, got_frame, pkt);
reinit_fail:
q->orig_pix_fmt = q->parser->format = avctx->pix_fmt = AV_PIX_FMT_NONE;
return ret;
}
STDMETHODIMP CDecAvcodec::Decode(const BYTE *buffer, int buflen, REFERENCE_TIME rtStartIn, REFERENCE_TIME rtStopIn, BOOL bSyncPoint, BOOL bDiscontinuity)
{
CheckPointer(m_pAVCtx, E_UNEXPECTED);
int got_picture = 0;
int used_bytes = 0;
BOOL bFlush = (buffer == nullptr);
BOOL bEndOfSequence = FALSE;
AVPacket avpkt;
av_init_packet(&avpkt);
if (m_pAVCtx->active_thread_type & FF_THREAD_FRAME) {
if (!m_bFFReordering) {
m_tcThreadBuffer[m_CurrentThread].rtStart = rtStartIn;
m_tcThreadBuffer[m_CurrentThread].rtStop = rtStopIn;
}
m_CurrentThread = (m_CurrentThread + 1) % m_pAVCtx->thread_count;
} else if (m_bBFrameDelay) {
m_tcBFrameDelay[m_nBFramePos].rtStart = rtStartIn;
m_tcBFrameDelay[m_nBFramePos].rtStop = rtStopIn;
m_nBFramePos = !m_nBFramePos;
}
uint8_t *pDataBuffer = nullptr;
if (!bFlush && buflen > 0) {
if (!m_bInputPadded && (!(m_pAVCtx->active_thread_type & FF_THREAD_FRAME) || m_pParser)) {
// Copy bitstream into temporary buffer to ensure overread protection
// Verify buffer size
if (buflen > m_nFFBufferSize) {
m_nFFBufferSize = buflen;
m_pFFBuffer = (BYTE *)av_realloc_f(m_pFFBuffer, m_nFFBufferSize + FF_INPUT_BUFFER_PADDING_SIZE, 1);
if (!m_pFFBuffer) {
m_nFFBufferSize = 0;
return E_OUTOFMEMORY;
}
}
memcpy(m_pFFBuffer, buffer, buflen);
memset(m_pFFBuffer+buflen, 0, FF_INPUT_BUFFER_PADDING_SIZE);
pDataBuffer = m_pFFBuffer;
} else {
pDataBuffer = (uint8_t *)buffer;
}
if (m_nCodecId == AV_CODEC_ID_VP8 && m_bWaitingForKeyFrame) {
if (!(pDataBuffer[0] & 1)) {
DbgLog((LOG_TRACE, 10, L"::Decode(): Found VP8 key-frame, resuming decoding"));
m_bWaitingForKeyFrame = FALSE;
} else {
return S_OK;
}
}
}
while (buflen > 0 || bFlush) {
REFERENCE_TIME rtStart = rtStartIn, rtStop = rtStopIn;
if (!bFlush) {
avpkt.data = pDataBuffer;
avpkt.size = buflen;
avpkt.pts = rtStartIn;
if (rtStartIn != AV_NOPTS_VALUE && rtStopIn != AV_NOPTS_VALUE)
avpkt.duration = (int)(rtStopIn - rtStartIn);
else
avpkt.duration = 0;
avpkt.flags = AV_PKT_FLAG_KEY;
if (m_bHasPalette) {
m_bHasPalette = FALSE;
uint32_t *pal = (uint32_t *)av_packet_new_side_data(&avpkt, AV_PKT_DATA_PALETTE, AVPALETTE_SIZE);
int pal_size = FFMIN((1 << m_pAVCtx->bits_per_coded_sample) << 2, m_pAVCtx->extradata_size);
uint8_t *pal_src = m_pAVCtx->extradata + m_pAVCtx->extradata_size - pal_size;
for (int i = 0; i < pal_size/4; i++)
pal[i] = 0xFF<<24 | AV_RL32(pal_src+4*i);
}
} else {
avpkt.data = nullptr;
avpkt.size = 0;
}
// Parse the data if a parser is present
// This is mandatory for MPEG-1/2
if (m_pParser) {
BYTE *pOut = nullptr;
int pOut_size = 0;
used_bytes = av_parser_parse2(m_pParser, m_pAVCtx, &pOut, &pOut_size, avpkt.data, avpkt.size, AV_NOPTS_VALUE, AV_NOPTS_VALUE, 0);
if (used_bytes == 0 && pOut_size == 0 && !bFlush) {
DbgLog((LOG_TRACE, 50, L"::Decode() - could not process buffer, starving?"));
break;
} else if (used_bytes > 0) {
buflen -= used_bytes;
pDataBuffer += used_bytes;
}
// Update start time cache
// If more data was read then output, update the cache (incomplete frame)
// If output is bigger, a frame was completed, update the actual rtStart with the cached value, and then overwrite the cache
if (used_bytes > pOut_size) {
if (rtStartIn != AV_NOPTS_VALUE)
m_rtStartCache = rtStartIn;
} else if (used_bytes == pOut_size || ((used_bytes + 9) == pOut_size)) {
// Why +9 above?
// Well, apparently there are some broken MKV muxers that like to mux the MPEG-2 PICTURE_START_CODE block (which is 9 bytes) in the package with the previous frame
// This would cause the frame timestamps to be delayed by one frame exactly, and cause timestamp reordering to go wrong.
// So instead of failing on those samples, lets just assume that 9 bytes are that case exactly.
m_rtStartCache = rtStartIn = AV_NOPTS_VALUE;
} else if (pOut_size > used_bytes) {
rtStart = m_rtStartCache;
m_rtStartCache = rtStartIn;
// The value was used once, don't use it for multiple frames, that ends up in weird timings
rtStartIn = AV_NOPTS_VALUE;
}
if (pOut_size > 0 || bFlush) {
if (pOut && pOut_size > 0) {
if (pOut_size > m_nFFBufferSize2) {
m_nFFBufferSize2 = pOut_size;
m_pFFBuffer2 = (BYTE *)av_realloc_f(m_pFFBuffer2, m_nFFBufferSize2 + FF_INPUT_BUFFER_PADDING_SIZE, 1);
if (!m_pFFBuffer2) {
m_nFFBufferSize2 = 0;
return E_OUTOFMEMORY;
}
}
memcpy(m_pFFBuffer2, pOut, pOut_size);
memset(m_pFFBuffer2+pOut_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
avpkt.data = m_pFFBuffer2;
avpkt.size = pOut_size;
avpkt.pts = rtStart;
avpkt.duration = 0;
const uint8_t *eosmarker = CheckForEndOfSequence(m_nCodecId, avpkt.data, avpkt.size, &m_MpegParserState);
if (eosmarker) {
bEndOfSequence = TRUE;
}
} else {
avpkt.data = nullptr;
avpkt.size = 0;
}
int ret2 = avcodec_decode_video2 (m_pAVCtx, m_pFrame, &got_picture, &avpkt);
if (ret2 < 0) {
DbgLog((LOG_TRACE, 50, L"::Decode() - decoding failed despite successfull parsing"));
got_picture = 0;
}
} else {
got_picture = 0;
}
} else {
used_bytes = avcodec_decode_video2 (m_pAVCtx, m_pFrame, &got_picture, &avpkt);
buflen = 0;
}
if (FAILED(PostDecode())) {
av_frame_unref(m_pFrame);
return E_FAIL;
}
// Decoding of this frame failed ... oh well!
if (used_bytes < 0) {
av_frame_unref(m_pFrame);
return S_OK;
}
// Judge frame usability
// This determines if a frame is artifact free and can be delivered.
if (m_bResumeAtKeyFrame) {
if (m_bWaitingForKeyFrame && got_picture) {
if (m_pFrame->key_frame) {
DbgLog((LOG_TRACE, 50, L"::Decode() - Found Key-Frame, resuming decoding at %I64d", m_pFrame->pkt_pts));
m_bWaitingForKeyFrame = FALSE;
} else {
got_picture = 0;
}
}
}
// Handle B-frame delay for frame threading codecs
if ((m_pAVCtx->active_thread_type & FF_THREAD_FRAME) && m_bBFrameDelay) {
m_tcBFrameDelay[m_nBFramePos] = m_tcThreadBuffer[m_CurrentThread];
m_nBFramePos = !m_nBFramePos;
}
if (!got_picture || !m_pFrame->data[0]) {
if (!avpkt.size)
bFlush = FALSE; // End flushing, no more frames
av_frame_unref(m_pFrame);
continue;
}
///////////////////////////////////////////////////////////////////////////////////////////////
// Determine the proper timestamps for the frame, based on different possible flags.
///////////////////////////////////////////////////////////////////////////////////////////////
if (m_bFFReordering) {
rtStart = m_pFrame->pkt_pts;
if (m_pFrame->pkt_duration)
rtStop = m_pFrame->pkt_pts + m_pFrame->pkt_duration;
else
rtStop = AV_NOPTS_VALUE;
} else if (m_bBFrameDelay && m_pAVCtx->has_b_frames) {
rtStart = m_tcBFrameDelay[m_nBFramePos].rtStart;
rtStop = m_tcBFrameDelay[m_nBFramePos].rtStop;
} else if (m_pAVCtx->active_thread_type & FF_THREAD_FRAME) {
unsigned index = m_CurrentThread;
rtStart = m_tcThreadBuffer[index].rtStart;
rtStop = m_tcThreadBuffer[index].rtStop;
}
if (m_bRVDropBFrameTimings && m_pFrame->pict_type == AV_PICTURE_TYPE_B) {
rtStart = AV_NOPTS_VALUE;
}
if (m_bCalculateStopTime)
rtStop = AV_NOPTS_VALUE;
///////////////////////////////////////////////////////////////////////////////////////////////
// All required values collected, deliver the frame
///////////////////////////////////////////////////////////////////////////////////////////////
LAVFrame *pOutFrame = nullptr;
AllocateFrame(&pOutFrame);
AVRational display_aspect_ratio;
int64_t num = (int64_t)m_pFrame->sample_aspect_ratio.num * m_pFrame->width;
int64_t den = (int64_t)m_pFrame->sample_aspect_ratio.den * m_pFrame->height;
av_reduce(&display_aspect_ratio.num, &display_aspect_ratio.den, num, den, INT_MAX);
pOutFrame->width = m_pFrame->width;
pOutFrame->height = m_pFrame->height;
pOutFrame->aspect_ratio = display_aspect_ratio;
pOutFrame->repeat = m_pFrame->repeat_pict;
pOutFrame->key_frame = m_pFrame->key_frame;
pOutFrame->frame_type = av_get_picture_type_char(m_pFrame->pict_type);
pOutFrame->ext_format = GetDXVA2ExtendedFlags(m_pAVCtx, m_pFrame);
if (m_pFrame->interlaced_frame || (!m_pAVCtx->progressive_sequence && (m_nCodecId == AV_CODEC_ID_H264 || m_nCodecId == AV_CODEC_ID_MPEG2VIDEO)))
m_iInterlaced = 1;
else if (m_pAVCtx->progressive_sequence)
m_iInterlaced = 0;
if ((m_nCodecId == AV_CODEC_ID_H264 || m_nCodecId == AV_CODEC_ID_MPEG2VIDEO) && m_pFrame->repeat_pict)
m_nSoftTelecine = 2;
else if (m_nSoftTelecine > 0)
m_nSoftTelecine--;
// Don't apply aggressive deinterlacing to content that looks soft-telecined, as it would destroy the content
bool bAggressiveFlag = (m_iInterlaced == 1 && m_pSettings->GetDeinterlacingMode() == DeintMode_Aggressive) && !m_nSoftTelecine;
pOutFrame->interlaced = (m_pFrame->interlaced_frame || bAggressiveFlag || m_pSettings->GetDeinterlacingMode() == DeintMode_Force) && !(m_pSettings->GetDeinterlacingMode() == DeintMode_Disable);
LAVDeintFieldOrder fo = m_pSettings->GetDeintFieldOrder();
pOutFrame->tff = (fo == DeintFieldOrder_Auto) ? m_pFrame->top_field_first : (fo == DeintFieldOrder_TopFieldFirst);
pOutFrame->rtStart = rtStart;
pOutFrame->rtStop = rtStop;
PixelFormatMapping map = getPixFmtMapping((AVPixelFormat)m_pFrame->format);
pOutFrame->format = map.lavpixfmt;
pOutFrame->bpp = map.bpp;
if (m_nCodecId == AV_CODEC_ID_MPEG2VIDEO || m_nCodecId == AV_CODEC_ID_MPEG1VIDEO)
pOutFrame->avgFrameDuration = GetFrameDuration();
AVFrameSideData * sdHDR = av_frame_get_side_data(m_pFrame, AV_FRAME_DATA_HDR_MASTERING_INFO);
if (sdHDR) {
if (sdHDR->size == 24) {
MediaSideDataHDR * hdr = (MediaSideDataHDR *)AddLAVFrameSideData(pOutFrame, IID_MediaSideDataHDR, sizeof(MediaSideDataHDR));
if (hdr) {
CByteParser hdrParser(sdHDR->data, sdHDR->size);
for (int i = 0; i < 3; i++)
{
hdr->display_primaries_x[i] = hdrParser.BitRead(16) * 0.00002;
hdr->display_primaries_y[i] = hdrParser.BitRead(16) * 0.00002;
}
hdr->white_point_x = hdrParser.BitRead(16) * 0.00002;
hdr->white_point_y = hdrParser.BitRead(16) * 0.00002;
hdr->max_display_mastering_luminance = hdrParser.BitRead(32) * 0.0001;
hdr->min_display_mastering_luminance = hdrParser.BitRead(32) * 0.0001;
}
}
else {
DbgLog((LOG_TRACE, 10, L"::Decode(): Found HDR data of an unexpected size (%d)", sdHDR->size));
}
}
if (map.conversion) {
ConvertPixFmt(m_pFrame, pOutFrame);
} else {
AVFrame *pFrameRef = av_frame_alloc();
av_frame_ref(pFrameRef, m_pFrame);
for (int i = 0; i < 4; i++) {
pOutFrame->data[i] = pFrameRef->data[i];
pOutFrame->stride[i] = pFrameRef->linesize[i];
}
pOutFrame->priv_data = pFrameRef;
pOutFrame->destruct = lav_avframe_free;
// Check alignment on rawvideo, which can be off depending on the source file
if (m_nCodecId == AV_CODEC_ID_RAWVIDEO) {
for (int i = 0; i < 4; i++) {
if ((intptr_t)pOutFrame->data[i] % 16u || pOutFrame->stride[i] % 16u) {
// copy the frame, its not aligned properly and would crash later
CopyLAVFrameInPlace(pOutFrame);
break;
}
}
}
}
if (bEndOfSequence)
pOutFrame->flags |= LAV_FRAME_FLAG_END_OF_SEQUENCE;
if (pOutFrame->format == LAVPixFmt_DXVA2) {
pOutFrame->data[0] = m_pFrame->data[4];
HandleDXVA2Frame(pOutFrame);
} else {
Deliver(pOutFrame);
}
if (bEndOfSequence) {
bEndOfSequence = FALSE;
if (pOutFrame->format == LAVPixFmt_DXVA2) {
HandleDXVA2Frame(m_pCallback->GetFlushFrame());
} else {
Deliver(m_pCallback->GetFlushFrame());
}
}
if (bFlush) {
m_CurrentThread = (m_CurrentThread + 1) % m_pAVCtx->thread_count;
}
av_frame_unref(m_pFrame);
}
return S_OK;
}
static event_t *
get_packet_a(rtmp_t *r, uint8_t *data, size_t size, int64_t dts,
media_pipe_t *mp)
{
uint8_t flags;
uint8_t type = 0;
enum CodecID id;
if(r->r->m_read.flags & RTMP_READ_SEEKING)
return NULL;
if(size < 2)
return NULL;
flags = *data++;
size--;
switch(flags & 0xf0) {
case 0xa0: id = CODEC_ID_AAC;
type = *data++;
size--;
break;
case 0x20:
id = CODEC_ID_MP3;
break;
default:
return NULL;
}
if(r->acodec == NULL) {
AVCodecContext *ctx;
int parse = 0;
const char *fmt;
switch(id) {
case CODEC_ID_AAC:
if(type != 0 || size < 0)
return NULL;
ctx = avcodec_alloc_context();
ctx->extradata = av_mallocz(size + FF_INPUT_BUFFER_PADDING_SIZE);
memcpy(ctx->extradata, data, size);
ctx->extradata_size = size;
fmt = "AAC";
break;
case CODEC_ID_MP3:
ctx = avcodec_alloc_context();
parse = 1;
fmt = "MP3";
break;
default:
abort();
}
mp_add_track(mp->mp_prop_audio_tracks,
NULL,
"rtmp:1",
fmt,
fmt,
NULL,
NULL,
0);
prop_set_string(mp->mp_prop_audio_track_current, "rtmp:1");
r->acodec = media_codec_create(id, parse, NULL, ctx, NULL, mp);
return NULL;
}
media_codec_t *mc = r->acodec;
dts *= 1000;
if(dts < r->seekpos)
return NULL;
if(mc->parser_ctx == NULL)
return sendpkt(r, &mp->mp_audio, mc,
dts, dts, dts, data, size, 0, MB_AUDIO, 0);
while(size > 0) {
int outlen;
uint8_t *outbuf;
int rlen = av_parser_parse2(mc->parser_ctx,
mc->codec_ctx, &outbuf, &outlen,
data, size, dts, dts, AV_NOPTS_VALUE);
if(outlen) {
event_t *e = sendpkt(r, &mp->mp_audio, mc,
mc->parser_ctx->dts,
mc->parser_ctx->pts, mc->parser_ctx->dts,
outbuf, outlen, 0,
MB_AUDIO, 0);
if(e != NULL)
return e;
}
dts = AV_NOPTS_VALUE;
data += rlen;
size -= rlen;
}
return NULL;
}
