V4L2VDA: use YV12 as output format if processor supports it.

media/gpu/v4l2_video_decode_accelerator.cc

Index: media/gpu/v4l2_video_decode_accelerator.cc
diff --git a/media/gpu/v4l2_video_decode_accelerator.cc b/media/gpu/v4l2_video_decode_accelerator.cc
index 01e650a2f3cf35fdb5581f672b957672d9a2ae06..b4a27d1455fd734ecaf22f7956e8a4235e846d09 100644
--- a/media/gpu/v4l2_video_decode_accelerator.cc
+++ b/media/gpu/v4l2_video_decode_accelerator.cc
@@ -2038,9 +2038,30 @@ uint32_t V4L2VideoDecodeAccelerator::FindImageProcessorInputFormat() {
 }
 
 uint32_t V4L2VideoDecodeAccelerator::FindImageProcessorOutputFormat() {
+  // Prefer YVU420 and NV12 because ArcGpuVideoDecodeAccelerator and drm_gralloc
+  // only support single physical plane. Prefer YVU420 over NV12 because chrome
+  // rendering supports YV12 only. When we use NV12, chrome thinks it is RGBA
+  // and mali handles that transparently knowing it is really NV12.
+  static const uint32_t kPreferredFormats[] = {V4L2_PIX_FMT_YVU420,
+                                               V4L2_PIX_FMT_NV12};
+  struct {

[kcwu, 2016/08/22 08:07:00]

I feel it's unnecessary to use struct. A normal function or even a lambda is
enough.

[wuchengli, 2016/08/22 09:08:04]

Yeah. Lambda is simpler. Done.

+    bool operator()(uint32_t a, uint32_t b) {
+      auto iter_a = std::find(std::begin(kPreferredFormats),
+                              std::end(kPreferredFormats), a);
+      auto iter_b = std::find(std::begin(kPreferredFormats),
+                              std::end(kPreferredFormats), b);
+      return iter_a < iter_b;
+    }
+  } preferred_formats_first;
+
   V4L2ImageProcessor image_processor(image_processor_device_);
   std::vector<uint32_t> processor_output_formats =
       image_processor.GetSupportedOutputFormats();
+
+  // Move the preferred format to the front.
+  std::sort(processor_output_formats.begin(), processor_output_formats.end(),

[kcwu, 2016/08/22 08:07:00]

Just curious, does the ordering of GetSupportedOutputFormats() output mean
something? I'm wondering should we try to keep original order if not in our
preference list?
If so, use stable_sort.

[wuchengli, 2016/08/22 09:08:03]

The order of GetSupportedOutputFormats does not mean anything. It's just the
order returned by the kernel driver. Also, the formats returned from
GetSupportedOutputFormats should not have duplicates.

+            preferred_formats_first);
+
   for (uint32_t processor_output_format : processor_output_formats) {
     if (device_->CanCreateEGLImageFrom(processor_output_format)) {
       DVLOGF(1) << "Image processor output format=" << processor_output_format;