GitLab

We used to pass a vector of DMABuf FDs to the GL/EGL image creation
functions. This was very error-prone, as these vectors could contain
less FDs than there were planes in the buffer, and this information
was used to signal that the last FD contained several planes. The
function receiving these incomplete FDs would then try to figure out the
actual planes layout, with more or less luck, often introducing bugs in
the way.

Solve this issue once and for all by passing all importable buffer
information within the VDAs through NativePixmapHandles: they always
contain as many FDs as there are planes, and also have detailed layout
information. On top of that, they are what we receive when using import
mode (which means we can pass the exact layout information down to
GL/EGL), and we already have a function that creates them from a
VideoFrame. This helps reduce the amount of code, all the while making
it safer and more readable.

BUG=b:146842214
TEST=VDAtest passes on Hana.
TEST=VDAtest passes on Kevin with VD disabled.
TEST=VDAtest passes on Kukui.
TEST=crosvideo.appspot.com plays videos properly on Hana.
TEST=crosvideo.appspot.com plays videos properly on Kevin with VD disabled.
TEST=crosvideo.appspot.com plays videos properly on Kukui.
TEST=Youtube Android plays properly on Hana.
TEST=Youtube Android plays properly on Kevin with VD disabled.
TEST=Youtube Android plays properly on Kukui.

Change-Id: I88a78b786b1c90820221c9f0770a52a339913537
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1982499
Commit-Queue: Alexandre Courbot <acourbot@chromium.org>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#728450}