Plumb decoded video pixel format from GPU process to renderer.

media/gpu/ipc/client/gpu_video_decode_accelerator_host.cc

Index: media/gpu/ipc/client/gpu_video_decode_accelerator_host.cc
diff --git a/media/gpu/ipc/client/gpu_video_decode_accelerator_host.cc b/media/gpu/ipc/client/gpu_video_decode_accelerator_host.cc
index 43545e7390f1de2f9d4f076bcc42a468841c8685..e280602fc453768f6e4b10b9ac5b184f285ffc28 100644
--- a/media/gpu/ipc/client/gpu_video_decode_accelerator_host.cc
+++ b/media/gpu/ipc/client/gpu_video_decode_accelerator_host.cc
@@ -178,6 +178,10 @@ void GpuVideoDecodeAcceleratorHost::Destroy() {
   delete this;
 }
 
+VideoPixelFormat GpuVideoDecodeAcceleratorHost::GetOutputFormat() const {
+  return picture_buffer_format_;

[miu, 2016/05/12 20:03:02]

Order-of-operations consideration: Should we DCHECK_NE(PIXEL_FORMAT_UNKNOWN,
picture_buffer_format_) to ensure this isn't called before the format is known?
Or, is it normal for the caller to call GetOutputFormat() and expects an
"UNKNOWN" result before picture buffers are provided?

+}
+
 void GpuVideoDecodeAcceleratorHost::OnWillDeleteImpl() {
   DCHECK(CalledOnValidThread());
   impl_ = NULL;
@@ -217,12 +221,14 @@ void GpuVideoDecodeAcceleratorHost::OnBitstreamBufferProcessed(
 }
 
 void GpuVideoDecodeAcceleratorHost::OnProvidePictureBuffer(
+    VideoPixelFormat format,
     uint32_t num_requested_buffers,
     uint32_t textures_per_buffer,
     const gfx::Size& dimensions,
     uint32_t texture_target) {
   DCHECK(CalledOnValidThread());

[miu, 2016/05/12 20:03:02]

Robustness consideration: Should we DCHECK() that the dimensions and format do
not change? (Could this method be called more than once?)

   picture_buffer_dimensions_ = dimensions;
+  picture_buffer_format_ = format;
 
   const int kMaxVideoPlanes = 4;
   if (textures_per_buffer > kMaxVideoPlanes) {