Merge pull request #921 from trav90/media-work

Use new decode API with recent FFmpeg version

4 files changed, 72 insertions, 10 deletions

diff --git a/dom/media/platforms/ffmpeg/FFmpegLibWrapper.cpp b/dom/media/platforms/ffmpeg/FFmpegLibWrapper.cpp
index e1c326818..426e9f74b 100644
--- a/dom/media/platforms/ffmpeg/FFmpegLibWrapper.cpp
+++ b/dom/media/platforms/ffmpeg/FFmpegLibWrapper.cpp
@@ -145,6 +145,8 @@ FFmpegLibWrapper::Link()
   AV_FUNC(avcodec_alloc_frame, (AV_FUNC_53 | AV_FUNC_54))
   AV_FUNC(avcodec_get_frame_defaults, (AV_FUNC_53 | AV_FUNC_54))
   AV_FUNC(avcodec_free_frame, AV_FUNC_54)
+  AV_FUNC(avcodec_send_packet, AV_FUNC_58)
+  AV_FUNC(avcodec_receive_frame, AV_FUNC_58)
   AV_FUNC(av_log_set_level, AV_FUNC_AVUTIL_ALL)
   AV_FUNC(av_malloc, AV_FUNC_AVUTIL_ALL)
   AV_FUNC(av_freep, AV_FUNC_AVUTIL_ALL)
diff --git a/dom/media/platforms/ffmpeg/FFmpegLibWrapper.h b/dom/media/platforms/ffmpeg/FFmpegLibWrapper.h
index c6c43a4ae..b968edd32 100644
--- a/dom/media/platforms/ffmpeg/FFmpegLibWrapper.h
+++ b/dom/media/platforms/ffmpeg/FFmpegLibWrapper.h
@@ -71,6 +71,10 @@ struct FFmpegLibWrapper
   // libavcodec v54 only
   void (*avcodec_free_frame)(AVFrame** frame);
 
+  // libavcodec v58 and later only
+  int (*avcodec_send_packet)(AVCodecContext* avctx, const AVPacket* avpkt);
+  int (*avcodec_receive_frame)(AVCodecContext* avctx, AVFrame* frame);
+
   // libavutil
   void (*av_log_set_level)(int level);
   void*	(*av_malloc)(size_t size);
diff --git a/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.cpp b/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.cpp
index aec1e9136..f3101e44c 100644
--- a/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.cpp
+++ b/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.cpp
@@ -174,12 +174,15 @@ FFmpegVideoDecoder<LIBAV_VER>::DoDecode(MediaRawData* aSample, bool* aGotFrame)
   uint8_t* inputData = const_cast<uint8_t*>(aSample->Data());
   size_t inputSize = aSample->Size();
 
-#if LIBAVCODEC_VERSION_MAJOR >= 54
+#if LIBAVCODEC_VERSION_MAJOR >= 54 && LIBAVCODEC_VERSION_MAJOR < 58
   if (inputSize && mCodecParser && (mCodecID == AV_CODEC_ID_VP8
-#if LIBAVCODEC_VERSION_MAJOR >= 55
+#if LIBAVCODEC_VERSION_MAJOR >= 55 && LIBAVCODEC_VERSION_MAJOR < 58
       || mCodecID == AV_CODEC_ID_VP9
 #endif
-      )) {
+      ))
+#endif
+#if LIBAVCODEC_VERSION_MAJOR >= 54
+  {
     while (inputSize) {
       uint8_t* data = inputData;
       int size = inputSize;
@@ -224,6 +227,48 @@ FFmpegVideoDecoder<LIBAV_VER>::DoDecode(MediaRawData* aSample,
   packet.flags = aSample->mKeyframe ? AV_PKT_FLAG_KEY : 0;
   packet.pos = aSample->mOffset;
 
+#if LIBAVCODEC_VERSION_MAJOR >= 58
+  packet.duration = aSample->mDuration;
+  int res = mLib->avcodec_send_packet(mCodecContext, &packet);
+  if (res < 0) {
+    // In theory, avcodec_send_packet could sent -EAGAIN should its internal
+    // buffers be full. In practice this can't happen as we only feed one frame
+    // at a time, and we immediately call avcodec_receive_frame right after.
+    FFMPEG_LOG("avcodec_send_packet error: %d", res);
+    return MediaResult(NS_ERROR_DOM_MEDIA_DECODE_ERR,
+                       RESULT_DETAIL("avcodec_send_packet error: %d", res));
+  }
+
+  if (aGotFrame) {
+    *aGotFrame = false;
+  }
+  do {
+    if (!PrepareFrame()) {
+      NS_WARNING("FFmpeg h264 decoder failed to allocate frame.");
+      return MediaResult(NS_ERROR_OUT_OF_MEMORY, __func__);
+    }
+    res = mLib->avcodec_receive_frame(mCodecContext, mFrame);
+    if (res == int(AVERROR_EOF)) {
+      return NS_ERROR_DOM_MEDIA_END_OF_STREAM;
+    }
+    if (res == AVERROR(EAGAIN)) {
+      return NS_OK;
+    }
+    if (res < 0) {
+      FFMPEG_LOG("avcodec_receive_frame error: %d", res);
+      return MediaResult(NS_ERROR_DOM_MEDIA_DECODE_ERR,
+                         RESULT_DETAIL("avcodec_receive_frame error: %d", res));
+    }
+    MediaResult rv = CreateImage(mFrame->pkt_pos, mFrame->pkt_pts,
+                                 mFrame->pkt_duration);
+    if (NS_FAILED(rv)) {
+      return rv;
+    }
+    if (aGotFrame) {
+      *aGotFrame = true;
+    }
+  } while (true);
+#else
   // LibAV provides no API to retrieve the decoded sample's duration.
   // (FFmpeg >= 1.0 provides av_frame_get_pkt_duration)
   // As such we instead use a map using the dts as key that we will retrieve
@@ -276,8 +321,21 @@ FFmpegVideoDecoder<LIBAV_VER>::DoDecode(MediaRawData* aSample,
     // against the map becoming extremely big.
     mDurationMap.Clear();
   }
+
+  MediaResult rv = CreateImage(aSample->mOffset, pts, duration);
+  if (NS_SUCCEEDED(rv) && aGotFrame) {
+    *aGotFrame = true;
+  }
+  return rv;
+#endif
+}
+
+MediaResult
+FFmpegVideoDecoder<LIBAV_VER>::CreateImage(int64_t aOffset, int64_t aPts,
+                                           int64_t aDuration)
+{
   FFMPEG_LOG("Got one frame output with pts=%lld dts=%lld duration=%lld opaque=%lld",
-              pts, mFrame->pkt_dts, duration, mCodecContext->reordered_opaque);
+              aPts, mFrame->pkt_dts, aDuration, mCodecContext->reordered_opaque);
 
   VideoData::YCbCrBuffer b;
   b.mPlanes[0].mData = mFrame->data[0];
@@ -317,9 +375,9 @@ FFmpegVideoDecoder<LIBAV_VER>::DoDecode(MediaRawData* aSample,
   RefPtr<VideoData> v =
     VideoData::CreateAndCopyData(mInfo,
                                   mImageContainer,
-                                  aSample->mOffset,
-                                  pts,
-                                  duration,
+                                  aOffset,
+                                  aPts,
+                                  aDuration,
                                   b,
                                   !!mFrame->key_frame,
                                   -1,
@@ -331,9 +389,6 @@ FFmpegVideoDecoder<LIBAV_VER>::DoDecode(MediaRawData* aSample,
                        RESULT_DETAIL("image allocation error"));
   }
   mCallback->Output(v);
-  if (aGotFrame) {
-    *aGotFrame = true;
-  }
   return NS_OK;
 }
 
diff --git a/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.h b/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.h
index 786df0da1..49a55e8a6 100644
--- a/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.h
+++ b/dom/media/platforms/ffmpeg/FFmpegVideoDecoder.h
@@ -49,6 +49,7 @@ private:
   MediaResult DoDecode(MediaRawData* aSample) override;
   MediaResult DoDecode(MediaRawData* aSample, bool* aGotFrame);
   MediaResult DoDecode(MediaRawData* aSample, uint8_t* aData, int aSize, bool* aGotFrame);
+  MediaResult CreateImage(int64_t aOffset, int64_t aPts, int64_t aDuration);
   void ProcessDrain() override;
   void ProcessFlush() override;
   void OutputDelayedFrames();