media/gpu: switch v4l2_jpeg_decode_accelerator to use multi-planar APIs

media/gpu/v4l2_jpeg_decode_accelerator.h

Index: media/gpu/v4l2_jpeg_decode_accelerator.h
diff --git a/media/gpu/v4l2_jpeg_decode_accelerator.h b/media/gpu/v4l2_jpeg_decode_accelerator.h
index b2824946b9a812bc15c0bc6d8d67071692024417..6321b8f5a6aebf0fc0cbea71c4a74f5b7b38b499 100644
--- a/media/gpu/v4l2_jpeg_decode_accelerator.h
+++ b/media/gpu/v4l2_jpeg_decode_accelerator.h
@@ -42,12 +42,20 @@ class MEDIA_GPU_EXPORT V4L2JpegDecodeAccelerator
   bool IsSupported() override;
 
  private:
+  // Input pixel format is alwasy V4L2_PIX_FMT_JPEG.
+  static const uint32_t kInputPlanes = 1;
+  // Ouput buffer can have as many as three planes.

[wuchengli, 2016/12/15 09:18:25]

It looks like the output is always three planes, not one or two. s/can have as
many as/have/

s/three planes/three physical planes/

[jcliang, 2016/12/15 14:55:47]

Done.

I was thinking there are 2-plane and 4-plane pixel format as well, but for now
it's always 3-plane.

+  static const uint32_t kOutputPlanes = 3;
+  // Maximum number of buffer planes we are able to handle.
+  static const uint32_t kMaxBufferPlanes = 3;

[wuchengli, 2016/12/15 09:18:25]

We are hard-coding the number of planes to be always 3 (kOutputPlanes).
kMaxBufferPlanes can be removed.

[jcliang, 2016/12/15 14:55:47]

Done.

+
   // Record for input/output buffers.
   struct BufferRecord {
     BufferRecord();
     ~BufferRecord();
-    void* address;  // mmap() address.
-    size_t length;  // mmap() length.
+    void* address[kMaxBufferPlanes];  // mmap() address.
+    size_t length[kMaxBufferPlanes];  // mmap() length.
+    size_t num_planes;                // Number of mapped planes.
 
     // Set true during QBUF and DQBUF. |address| will be accessed by hardware.
     bool at_device;
@@ -82,6 +90,11 @@ class MEDIA_GPU_EXPORT V4L2JpegDecodeAccelerator
   void DestroyInputBuffers();
   void DestroyOutputBuffers();
 

[wuchengli, 2016/12/15 09:18:25]

Document this function. What's the output format? What's the supported input
formats?

This is not copy anymore. Use a better name. ConvertOutputImage?

[jcliang, 2016/12/15 14:55:47]

Done.

+  bool CopyOutputImage(const uint32_t src_pixelformat,
+                       const BufferRecord& src_output_record,
+                       const gfx::Size& src_coded_size,
+                       const scoped_refptr<VideoFrame>& dst_frame);
+
   // Return the number of input/output buffers enqueued to the device.
   size_t InputBufferQueuedCount();
   size_t OutputBufferQueuedCount();
@@ -153,6 +166,8 @@ class MEDIA_GPU_EXPORT V4L2JpegDecodeAccelerator
   std::queue<linked_ptr<JobRecord>> input_jobs_;
   std::queue<linked_ptr<JobRecord>> running_jobs_;
 
+  // V4L2 buffer type of the input buffer.
+  v4l2_buf_type input_buf_type_;
   // Input queue state.
   bool input_streamon_;
   // Mapping of int index to an input buffer record.
@@ -161,6 +176,8 @@ class MEDIA_GPU_EXPORT V4L2JpegDecodeAccelerator
   // ordering.
   std::vector<int> free_input_buffers_;
 
+  // V4L2 buffer type of the output buffer.
+  v4l2_buf_type output_buf_type_;
   // Output queue state.
   bool output_streamon_;
   // Mapping of int index to an output buffer record.