GitLab

This CL takes VideoFrame-independent code from
VaapiJpegDecodeAccelerator's DecodeTask() and OutputPicture() methods
and puts it in VaapiJpegDecoder::DoDecode(). The idea is for
VaapiJpegDecoder to be responsible for dealing with the VaapiWrapper, so
now part of the initialization of the VaapiJpegDecodeAccelerator is
delegated to VaapiJpegDecoder.

Now, VaapiJpegDecoder::DoDecode() will take encoded data. It will do the
parsing, VAAPI buffer submission, decode execution, and VAImage
extraction. It will then return the decoded output as a ScopedVAImage.
The intent is to have a VaapiJpegDecoder member in two classes:
VaapiJpegDecodeAccelerator (existing) and the eventual implementation of
gpu::ImageDecodeAcceleratorWorker. In the first case, we use the
ScopedVAImage to copy the decoded data into a VideoFrame. In the second
case, we use it to copy the decoded data into a vector.

The unit tests are modified accordingly. Additionally,
::testing::FLAGS_gtest_death_test_style = "threadsafe" is added to two
of the tests due to safety issues when using EXPECT_DCHECK_DEATH when
there are multiple threads.

The reason for introducing the grayscale image is that before, the
DecodeFail test case was using a normal image and poking at the results
of the JPEG parsing so that it failed. Now, the decoder interface
doesn't take a JpegParseResult, so we can't do the same poking. Instead,
we give it a real unsupported image.

Bug: 924310
Test: jpeg_decode_accelerator_unittest passes on nocturne and grunt.
Change-Id: I52d9760d233e63b2831408f60309803984784903
Reviewed-on: https://chromium-review.googlesource.com/c/1427457
Commit-Queue: Andres Calderon Jaramillo <andrescj@chromium.org>
Auto-Submit: Andres Calderon Jaramillo <andrescj@chromium.org>
Reviewed-by: Daniele Castagna <dcastagna@chromium.org>
Reviewed-by: Miguel Casas <mcasas@chromium.org>
Cr-Commit-Position: refs/heads/master@{#634298}