We derive the destination buffer stride from the input stride,
which meant if the image was flipped with a negative stride,
we'd be FFALIGNING a negative number which ends up being huge,
thus making the Vulkan buffer allocation fail and the whole
image transfer fail.
Only found out about this as OpenGL compositors can copy an entire
image with a single call if its flipped, rather than iterate over
each line.
The idea was to allow separate planes to be filtered independently, however,
in hindsight, literaly nothing uses separate per-plane semaphores and it
would only work when each plane is backed by separate device memory.
The specifications are very vague about who has ownership, and in this case,
Vulkan takes ownership of all DMABUF FDs passed to it, causing errors
to occur if someone gave us images for mapping which were meant to be kept.
The old behavior worked with one-way VAAPI and DMABUF imports, but was broken
with clients like wlroots' dmabuf-capture.
There was a recent change in Intel's driver that triggered a driver-internal
error if the semaphore given to the command buffer wasn't initialized.
Given that the specifications require the semaphore to be initialized,
this is within spec. Unlike what's causing it in the first place, which is
that there are no ways to extract/import dma sync objects from DMABUFs,
so we must leave our semaphores bare.
This commit adds the necessary code to initialize and use a Vulkan device
within the hwcontext libavutil framework.
Currently direct mapping to VAAPI and DRM frames is functional, and
transfers to CUDA and native frames are supported.
Lets hope the future Vulkan video decode extension fits well within this
framework.