Enable VDAs to run with IMPORT mode on any devices |
|||
Issue descriptionVDA's output mode is ALLOCATE mode in VDA unittest by default. I tried to run VDA unittest with IMPORT mode, and then resulted the test failed ( crbug.com/879435 ) in * Tegra * Exynos5250 * Exynos5420 * Kepler We should be able to specify IMPORT mode on any devices. Test results: https://stainless.corp.google.com/search?view=matrix&row=model&col=build&first_date=2018-08-25&last_date=2018-09-07&test=video_VideoDecodeAccelerator&status=FAIL&reason=IMPORT&exclude_cts=false&exclude_not_run=false&exclude_non_release=false&exclude_au=true&exclude_acts=true&exclude_retried=true&exclude_non_production=false
,
Sep 7
Hmm, if I recall correctly, deanliao@ used Guado with Kepler for early testing of import mode and it worked. For Exynos, it's just infeasible, due to hardware limitations, because the hardware needs to know all the buffers before it can start streaming. Tegra doesn't have a real V4L2 driver. It emulates V4L2 in userspace using some wrapper library paired with libv4l2. It wasn't possible to get that library to correctly implement other parts of V4L2 decoder interface (initial resolution change AFAIR), so I'd be careful in thinking about import mode, which requires proper DMA-buf support.
,
Sep 12
I asked deanliao@ offline, and he said, """ i915 driver change makes it creates tiled buffer and V4L2VDA can only handle linear buffer. The VDA test code get a tiled buffer which cannot be used for V4L2VDA. It started to fail on R68. My code was developed and committed on R67. It is not an easy fix for Guado to test import mode now. """
,
Sep 23
I think we cannot do anything on currently failed devices. Perhaps, we can fix it on Kepler?
,
Oct 23
|
|||
►
Sign in to add a comment |
|||
Comment 1 by hiroh@chromium.org
, Sep 7