GitLab

Originally, VideoDecoderPipeline delegates VD tasks to the client
thread. Each VD implementation then post the tasks to decoder thread.
Then VD implementations post the callback back to the client thread
when the task is done. In the future, ImageProcessor will also operate
on the decoder thread. We will pass the task and output frames by:

  VDPipeline (client)
  -> VD (client -> decoder -> client)
  -> VDPipeline (client)
  -> IP (client -> decoder -> client)
  -> VDPipeline (client)
  -> Converter (client -> GPU -> client)
  -> VDPipeline (client)
  Total 6 thread hopping.

This CL now posts the VD task on the decoder thread. We don't need to
hop thread inside each VD implementation. With this CL, we could save
2 thread hopping.

  VDPipeline (client -> decoder)
  -> VD (decoder)
  -> VDPipeline (decoder)
  -> IP (decoder)
  -> VDPipeline (decoder)
  -> Converter (decoder -> GPU -> decoder)
  -> VDPipeline (decoder -> client)
  Total 4 thread hopping.

BUG=chromium:998413
TEST=run video_decode_accelerator_tests on Eve and Kevin
     with dcheck_always_on = true

Change-Id: I7367663258adfd0f46ddca25493faff4633ae034
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1777449
Commit-Queue: Chih-Yu Huang <akahuang@chromium.org>
Reviewed-by: David Staessens <dstaessens@chromium.org>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Cr-Commit-Position: refs/heads/master@{#707704}