Skip to content

Commit

Permalink
音声処理でのメモリリークを解放。AVFrameのメモリ解放について誤解していた。
Browse files Browse the repository at this point in the history 
AVFrame構造体が残ってしまうことで、パケット数の多いtrueHDなどでメモリリークが顕著に見られていた。
av_frame_free ... AVFrame構造体自体も解放。
av_frame_unref ... bufferの参照カウントを減算するのみ。
  • Loading branch information
@rigaya
rigaya committed Dec 10, 2019
1 parent cd48dd6 commit 63d06c8
Show file tree
Hide file tree
Showing 2 changed files with 19 additions and 16 deletions.
33 changes: 18 additions & 15 deletions QSVPipeline/rgy_output_avcodec.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2477,8 +2477,8 @@ void RGYOutputAvcodec::WriteNextPacketProcessed(AVPktMuxData *pktData, int64_t *
*writtenDts = pktData->dts;
}

vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> RGYOutputAvcodec::AudioDecodePacket(AVMuxAudio *muxAudio, AVPacket *pkt) {
vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> decodedFrames;
vector<unique_ptr<AVFrame, RGYAVDeleter<AVFrame>>> RGYOutputAvcodec::AudioDecodePacket(AVMuxAudio *muxAudio, AVPacket *pkt) {
vector<unique_ptr<AVFrame, RGYAVDeleter<AVFrame>>> decodedFrames;
if (muxAudio->decodeError > muxAudio->ignoreDecodeError) {
return std::move(decodedFrames);
}
Expand All @@ -2491,7 +2491,7 @@ vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> RGYOutputAvcodec::AudioDe
int64_t recieved_samples = 0;
int recv_ret = 0;
for (;;) {
unique_ptr<AVFrame, decltype(&av_frame_unref)> receivedData(nullptr, av_frame_unref);
unique_ptr<AVFrame, RGYAVDeleter<AVFrame>> receivedData(nullptr, RGYAVDeleter<AVFrame>(av_frame_free));
int send_ret = 0;
//必ず一度はパケットを送る
if (!sent_packet || pkt->size > 0) {
Expand All @@ -2516,7 +2516,7 @@ vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> RGYOutputAvcodec::AudioDe
AddMessage(RGY_LOG_ERROR, _T("failed to send packet to audio decoder: %s.\n"), qsv_av_err2str(send_ret).c_str());
muxAudio->decodeError++;
} else {
receivedData = unique_ptr<AVFrame, decltype(&av_frame_unref)>(av_frame_alloc(), av_frame_unref);
receivedData = unique_ptr<AVFrame, RGYAVDeleter<AVFrame>>(av_frame_alloc(), RGYAVDeleter<AVFrame>(av_frame_free));
recv_ret = avcodec_receive_frame(muxAudio->outCodecDecodeCtx, receivedData.get());
if (recv_ret == AVERROR(EAGAIN) //もっとパケットを送る必要がある
|| recv_ret == AVERROR_EOF) { //最後まで読み込んだ
Expand Down Expand Up @@ -2554,7 +2554,7 @@ vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> RGYOutputAvcodec::AudioDe
if (muxAudio->decodeError <= muxAudio->ignoreDecodeError) {
#if 0
//デコードエラーを無視する場合、入力パケットのサイズ分、無音を挿入する
unique_ptr<AVFrame, decltype(&av_frame_unref)> silentFrame(av_frame_alloc(), av_frame_unref);
unique_ptr<AVFrame, RGYAVDeleter<AVFrame>> silentFrame(av_frame_alloc(), RGYAVDeleter<AVFrame>(av_frame_free));
AVRational samplerate = { 1, muxAudio->outCodecDecodeCtx->sample_rate };
silentFrame->nb_samples = (int)av_rescale_q(pktInInfo.duration, muxAudio->streamIn->time_base, samplerate);
silentFrame->channels = muxAudio->outCodecDecodeCtx->channels;
Expand Down Expand Up @@ -2584,7 +2584,7 @@ vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> RGYOutputAvcodec::AudioDe
//音声をフィルタ
vector<AVPktMuxData> RGYOutputAvcodec::AudioFilterFrame(vector<AVPktMuxData> inputFrames) {
vector<AVPktMuxData> outputFrames;
for (const auto& pktData : inputFrames) {
for (auto& pktData : inputFrames) {
AVMuxAudio *muxAudio = pktData.muxAudio;
if (pktData.muxAudio->filterGraph == nullptr) {
//フィルタリングなし
Expand All @@ -2598,16 +2598,19 @@ vector<AVPktMuxData> RGYOutputAvcodec::AudioFilterFrame(vector<AVPktMuxData> inp
break;
}
}
//フィルターチェーンにフレームを追加
if (av_buffersrc_add_frame_flags(muxAudio->filterBufferSrcCtx, pktData.frame, AV_BUFFERSRC_FLAG_PUSH) < 0) {
AddMessage(RGY_LOG_ERROR, _T("failed to feed the audio filtergraph\n"));
m_Mux.format.streamError = true;
av_frame_unref(pktData.frame);
break;
{ //フィルターチェーンにフレームを追加
auto ret = av_buffersrc_add_frame_flags(muxAudio->filterBufferSrcCtx, pktData.frame, AV_BUFFERSRC_FLAG_PUSH);
// AVFrame構造体の破棄
av_frame_free(&pktData.frame);
if (ret < 0) {
AddMessage(RGY_LOG_ERROR, _T("failed to feed the audio filtergraph\n"));
m_Mux.format.streamError = true;
break;
}
}
for (;;) {
unique_ptr<AVFrame, decltype(&av_frame_unref)> filteredFrame(av_frame_alloc(), av_frame_unref);
int ret = av_buffersink_get_frame_flags(muxAudio->filterBufferSinkCtx, filteredFrame.get(), AV_BUFFERSINK_FLAG_NO_REQUEST);
unique_ptr<AVFrame, RGYAVDeleter<AVFrame>> filteredFrame(av_frame_alloc(), RGYAVDeleter<AVFrame>(av_frame_free));
auto ret = av_buffersink_get_frame_flags(muxAudio->filterBufferSinkCtx, filteredFrame.get(), AV_BUFFERSINK_FLAG_NO_REQUEST);
if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF) {
break;
}
Expand Down Expand Up @@ -3073,7 +3076,7 @@ RGY_ERR RGYOutputAvcodec::WriteNextAudioFrame(AVPktMuxData *pktData) {
return RGY_ERR_UNSUPPORTED;
}
auto encPktDatas = AudioEncodeFrame(pktData->muxAudio, pktData->frame);
av_frame_unref(pktData->frame);
av_frame_free(&pktData->frame);
#if ENABLE_AVCODEC_AUDPROCESS_THREAD
if (m_Mux.thread.thAudProcess.joinable()) {
for (auto& pktMux : encPktDatas) {
Expand Down
2 changes: 1 addition & 1 deletion QSVPipeline/rgy_output_avcodec.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -474,7 +474,7 @@ class RGYOutputAvcodec : public RGYOutput
void AudioFlushStream(AVMuxAudio *muxAudio, int64_t *writtenDts);

//音声をデコード
vector<unique_ptr<AVFrame, decltype(&av_frame_unref)>> AudioDecodePacket(AVMuxAudio *muxAudio, AVPacket *pkt);
vector<unique_ptr<AVFrame, RGYAVDeleter<AVFrame>>> AudioDecodePacket(AVMuxAudio *muxAudio, AVPacket *pkt);

//音声をエンコード
vector<AVPktMuxData> AudioEncodeFrame(AVMuxAudio *muxAudio, AVFrame *frame);
Expand Down

0 comments on commit 63d06c8

Please sign in to comment.