Motown: Ffmpeg video decoder

services/media/framework_ffmpeg/ffmpeg_video_decoder.cc

Index: services/media/framework_ffmpeg/ffmpeg_video_decoder.cc
diff --git a/services/media/framework_ffmpeg/ffmpeg_video_decoder.cc b/services/media/framework_ffmpeg/ffmpeg_video_decoder.cc
index 312d5ddbdf258e25deac95c33d72f4d67e42c30e..c1794edd5176764d35359435ef554a03c04622a7 100644
--- a/services/media/framework_ffmpeg/ffmpeg_video_decoder.cc
+++ b/services/media/framework_ffmpeg/ffmpeg_video_decoder.cc
@@ -2,8 +2,14 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+#include <algorithm>
+
 #include "base/logging.h"
+#include "services/media/framework_ffmpeg/ffmpeg_formatting.h"
 #include "services/media/framework_ffmpeg/ffmpeg_video_decoder.h"
+extern "C" {
+#include "third_party/ffmpeg/libavutil/imgutils.h"
+}
 
 namespace mojo {
 namespace media {
@@ -11,6 +17,10 @@ namespace media {
 FfmpegVideoDecoder::FfmpegVideoDecoder(AvCodecContextPtr av_codec_context)
     : FfmpegDecoderBase(std::move(av_codec_context)) {
   DCHECK(context());
+
+  context()->opaque = this;
+  context()->get_buffer2 = AllocateBufferForAvFrame;
+  context()->refcounted_frames = 1;
 }
 
 FfmpegVideoDecoder::~FfmpegVideoDecoder() {}
@@ -19,31 +29,54 @@ int FfmpegVideoDecoder::Decode(const AVPacket& av_packet,
                                const ffmpeg::AvFramePtr& av_frame_ptr,
                                PayloadAllocator* allocator,
                                bool* frame_decoded_out) {
-  DCHECK(av_frame_ptr);
   DCHECK(allocator);
   DCHECK(frame_decoded_out);
   DCHECK(context());
+  DCHECK(av_frame_ptr);
+
+  DCHECK(av_packet.pts != AV_NOPTS_VALUE);
+
+  // Use the provided allocator (for allocations in AllocateBufferForAvFrame).
+  allocator_ = allocator;
+
+  // We put the pts here so it can be recovered later in CreateOutputPacket.
+  // Ffmpeg deals with the frame ordering issues.
+  context()->reordered_opaque = av_packet.pts;
 
   int frame_decoded = 0;
   int input_bytes_used = avcodec_decode_video2(
       context().get(), av_frame_ptr.get(), &frame_decoded, &av_packet);
   *frame_decoded_out = frame_decoded != 0;
+
+  // We're done with this allocator.
+  allocator_ = nullptr;
+
   return input_bytes_used;
 }
 
+void FfmpegVideoDecoder::Flush() {
+  FfmpegDecoderBase::Flush();
+  next_pts_ = Packet::kUnknownPts;
+}
+
 PacketPtr FfmpegVideoDecoder::CreateOutputPacket(const AVFrame& av_frame,
                                                  PayloadAllocator* allocator) {
   DCHECK(allocator);
 
-  // End of stream is indicated when we're draining and produce no packet.
-  // TODO(dalesat): This is just a copy of the audio version.
-  return Packet::Create(av_frame.pts, false, packet_size_, av_frame.data[0],
-                        allocator);
+  // Recover the pts deposited in Decode.
+  next_pts_ = av_frame.reordered_opaque;
+
+  AvBufferContext* av_buffer_context =
+      reinterpret_cast<AvBufferContext*>(av_buffer_get_opaque(av_frame.buf[0]));
+
+  return Packet::Create(
+      next_pts_,
+      false,  // The base class is responsible for end-of-stream.
+      av_buffer_context->size(), av_buffer_context->Release(), allocator);
 }
 
 PacketPtr FfmpegVideoDecoder::CreateOutputEndOfStreamPacket() {
-  // TODO(dalesat): Presentation time for this packet.
-  return Packet::CreateEndOfStream(0);
+  return Packet::CreateEndOfStream(next_pts_);
 }
 
 int FfmpegVideoDecoder::AllocateBufferForAvFrame(
@@ -52,51 +85,107 @@ int FfmpegVideoDecoder::AllocateBufferForAvFrame(
     int flags) {
   // It's important to use av_codec_context here rather than context(),
   // because av_codec_context is different for different threads when we're
-  // decoding on multiple threads. If this code is moved to an instance method,
-  // be sure to avoid using context().
-
-  // TODO(dalesat): Not sure why/if this is needed.
-  // int result = av_image_check_size(
-  //    av_codec_context->width,
-  //    av_codec_context->height,
-  //    0,
-  //    NULL);
-  // if (result < 0) {
-  //  DCHECK(false) << "av_image_check_size failed";
-  //  return result;
-  //}
-
-  // TODO(dalesat): Not sure why this is needed.
-  int coded_width =
-      std::max(av_codec_context->width, av_codec_context->coded_width);
-  int coded_height =
-      std::max(av_codec_context->height, av_codec_context->coded_height);
-  DCHECK_EQ(coded_width, av_codec_context->coded_width)
-      << "coded width is less than width";
-  DCHECK_EQ(coded_height, av_codec_context->coded_height)
-      << "coded height is less than height";
-
-  // TODO(dalesat): Fill in av_frame->data and av_frame->data for each plane.
-
-  av_frame->width = coded_width;
-  av_frame->height = coded_height;
+  // decoding on multiple threads. Be sure to avoid using self->context().
+
+  // CODEC_CAP_DR1 is required in order to do allocation this way.
+  DCHECK(av_codec_context->codec->capabilities & CODEC_CAP_DR1);
+
+  FfmpegVideoDecoder* self =
+      reinterpret_cast<FfmpegVideoDecoder*>(av_codec_context->opaque);
+  DCHECK(self);
+  DCHECK(self->allocator_);
+
+  Extent visible_size(av_codec_context->width, av_codec_context->height);
+  const int result =
+      av_image_check_size(visible_size.width(), visible_size.height(), 0, NULL);
+  if (result < 0) {
+    return result;
+  }
+
+  // FFmpeg has specific requirements on the allocation size of the frame.  The
+  // following logic replicates FFmpeg's allocation strategy to ensure buffers
+  // are not overread / overwritten.  See ff_init_buffer_info() for details.
+
+  // When lowres is non-zero, dimensions should be divided by 2^(lowres), but
+  // since we don't use this, just DCHECK that it's zero.
+  DCHECK_EQ(av_codec_context->lowres, 0);
+  Extent coded_size(
+      std::max(visible_size.width(),
+               static_cast<size_t>(av_codec_context->coded_width)),
+      std::max(visible_size.height(),
+               static_cast<size_t>(av_codec_context->coded_height)));
+
+  VideoStreamType::FrameLayout frame_layout;
+
+  VideoStreamType::InfoForPixelFormat(
+      PixelFormatFromAVPixelFormat(av_codec_context->pix_fmt))
+      .BuildFrameLayout(coded_size, &frame_layout);
+
+  AvBufferContext* av_buffer_context =
+      new AvBufferContext(frame_layout.size, self->allocator_);
+  uint8_t* buffer = av_buffer_context->buffer();
+
+  // TODO(dalesat): For investigation purposes only...remove one day.
+  if (self->first_frame_) {
+    self->first_frame_ = false;
+    self->colorspace_ = av_codec_context->colorspace;
+    self->coded_size_ = coded_size;
+  } else {
+    if (av_codec_context->colorspace != self->colorspace_) {
+      LOG(WARNING) << " colorspace changed to " << av_codec_context->colorspace
+                   << std::endl;
+    }
+    if (coded_size.width() != self->coded_size_.width()) {
+      LOG(WARNING) << " coded_size width changed to " << coded_size.width()
+                   << std::endl;
+    }
+    if (coded_size.height() != self->coded_size_.height()) {
+      LOG(WARNING) << " coded_size height changed to " << coded_size.height()
+                   << std::endl;
+    }
+    self->colorspace_ = av_codec_context->colorspace;
+    self->coded_size_ = coded_size;
+  }
+
+  if (buffer == nullptr) {
+    LOG(ERROR) << "failed to allocate buffer of size " << frame_layout.size;
+    return -1;
+  }
+
+  // Decoders require a zeroed buffer.
+  std::memset(buffer, 0, frame_layout.size);
+
+  for (size_t plane = 0; plane < frame_layout.plane_count; ++plane) {
+    av_frame->data[plane] = buffer + frame_layout.plane_offset_for_plane(plane);
+    av_frame->linesize[plane] = frame_layout.line_stride_for_plane(plane);
+  }
+
+  // TODO(dalesat): Do we need to attach colorspace info to the packet?
+
+  av_frame->width = coded_size.width();
+  av_frame->height = coded_size.height();
   av_frame->format = av_codec_context->pix_fmt;
   av_frame->reordered_opaque = av_codec_context->reordered_opaque;
 
-  av_frame->buf[0] = av_buffer_create(
-      av_frame->data[0],  // Because this is the first chunk in the buffer.
-      0,                  // TODO(dalesat): Provide this.
-      ReleaseBufferForAvFrame,
-      nullptr,  // opaque
-      0);       // flags
+  DCHECK(av_frame->data[0] == buffer);
+  av_frame->buf[0] =
+      av_buffer_create(buffer, av_buffer_context->size(),
+                       ReleaseBufferForAvFrame, av_buffer_context,
+                       0);  // flags
 
   return 0;
 }
 
 void FfmpegVideoDecoder::ReleaseBufferForAvFrame(void* opaque,
                                                  uint8_t* buffer) {
-  // Nothing to do.
-  // TODO(dalesat): Can we get rid of this method altogether?
+  AvBufferContext* av_buffer_context =
+      reinterpret_cast<AvBufferContext*>(opaque);
+  DCHECK(av_buffer_context);
+  // Either this buffer has already been released to someone else's ownership,
+  // or it's the same as the buffer parameter.
+  DCHECK(av_buffer_context->buffer() == nullptr ||
+         av_buffer_context->buffer() == buffer);
+  delete av_buffer_context;
 }
 
 }  // namespace media