Comment 21 for bug 1406725

After today's investigation, I think the cachelines that are appearing as corruption are not from an lines that should have been invalidated, but are the flushed lines from the next frame. IE, we're releasing the buffer too early, and the flushed lines from the backbuffer render are appearing as corruption in the previous frame. I've been able to use the sync extensions from EGL_KHR_fence_sync to stabilize cpu rendering and make sure that OES_EGL_image_external is synchronized. Will continue testing before saying its figured out, but seems to be improved.

It might take a bit more work to come up with the proper patch that makes both android and mesa happy though.