media/gpu/v4l2ip: allocate as many buffers as possible in DMABUF mode
If we are operating in DMABUF mode, then we will try to keep the same frame assigned to the same V4L2 buffer. The number of buffers requested by the client is indicative of the real load, but there is no hard guarantee that we will not receive more frames than that, thus resulting in sub-optimal affinity tracking. Try to prevent this by allocating as many buffers as allowed by the V4L2 API (32) when operating in DMABUF mode. Since these buffers won't have backing memory unless we attach a DMABUF to them, they are virtually free if unused. BUG=b:159688625 TEST=video.DecodeAccel.h264 passes on Kukui. Change-Id: Ibc9ea18835acd4b8d3bc6ffc7393bf0d5d97d04d Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2415908 Commit-Queue: Alexandre Courbot <acourbot@chromium.org> Auto-Submit: Alexandre Courbot <acourbot@chromium.org> Reviewed-by:Fritz Koenig <frkoenig@chromium.org> Cr-Commit-Position: refs/heads/master@{#813110}
Showing
Please register or sign in to comment