This causes POLLERR when decoding from meson-vdec on the Libre
Computer Alta (and likely other Amlogic boards). Our requried buffer
counts were already lower than the default 16/20 configuration that
FFmpeg uses, so overriding them was just a memory usage optimization.
This can be overriden with GL_IS_SLOW=0 at runtime
for systems with a fast GPU, though it doesn't make
much difference either way for non-KMSDRM cases.
For slower GPUs, this enables important scaling and
rendering optimizations that can substantially
improve performance (especially on fillrate-limited
GPUs common in SBCs).
- For hwaccel decoders, don't depend on FFmpeg's 4 working surfaces.
The decoders might need one or more of them internally.
- For V4L2M2M decoders, adjust the buffer counts to match our needs
and requirements regarding Pacer-held frames.
NVENC on the 591.59 driver seems to not include the bitstream
restrictions element, which renders our attempt to set the
max_dec_frame_buffering value ineffective. Handle this by
providing inserting default bitstream restrictions if one
is not present.
Every renderer except SDL had opted-in for testing due to various
quirks discovered over the years, so just do away with this option
and test all renderers.
The DRM renderer does this for the out-of-tree v4l2m2m patches to ensure
we get NV12 buffers, but it ends up clobbering our own preference for
DRM_PRIME frames in the process.
This case basically works like a degenerate case of a DRM hwaccel
without DRM master where we just provide EGL export functionality
except that we don't even need a DRM FD in this case.
There are patches floating around the FFmpeg list for this:
https://patchwork.ffmpeg.org/project/ffmpeg/list/?series=12604
SpacemiT K1's BSP has a version of FFmpeg that contains both
v4l2m2m and their own *_stcodec decoders. Not only is their
V4L2 driver for their mvx decoder seemingly totally broken
(output buffers are all zeros), but it doesn't have any of
the DRM_PRIME support work that is maintained out of tree.
The result is we use a broken copyback decoder instead of
a (mostly) working zero-copy decoder because we didn't
include any logic to rank non-hwaccel decoders by pixfmt
like we have with device type for hwaccel decoders.
There are at least 2 out-of-tree OMX decoder implementations
that both lack AV_CODEC_CAP_HARDWARE, so they trick us into
thinking that neither of them is hardware accelerated.
Fixes false decoder warnings when linked against the patched
FFmpeg builds shipping on VisionFive 2 and LicheePi 4A SBCs.
Vulkan H.264/HEVC video decoding isn't supported with Fedora's Mesa binaries due to
patent issues, so only the VAAPI/DRM-PRIME import path was tested locally with ANV.
HDR video is getting tonemapped to my SDR monitor reasonably, so HDR output probably
works with GameScope but I haven't confirmed yet.
See #1117
Cameron Gutman
Andy Grundman
ns6089