Cameron Gutman
83d770cdba
Prefer BT2020_YCC to allow chroma subsampling
2026-01-04 22:16:52 -06:00
Cameron Gutman
25e2d39e26
Fix copy-paste mistake in 8726c1
2026-01-04 21:52:12 -06:00
Cameron Gutman
bd32370b6d
Fix libdrm return value vs errno confusion
...
libdrm is inconsistent between returning -1 with an errno set
and just returning -errno as the return value. Figuring out which
behavior a given libdrm function has require reading the source :/
2026-01-04 21:50:40 -06:00
Cameron Gutman
8726c1c7d0
Bring back atomic async page flip support
...
Now that we're referencing the AVFrame to keep the buffers from
being reused during scanout, this should hopefully work better.
2026-01-04 19:34:13 -06:00
Cameron Gutman
34881599f5
Reference the AVFrame while the backing DMA-BUFs are used for scanout
...
This prevents tearing and other artifacts caused by the decoder
writing to a DMA-BUF that is currently being displayed.
2026-01-04 18:47:00 -06:00
Cameron Gutman
de0fb72424
av_buffer_create() copies the pointer not the data
2026-01-04 18:24:20 -06:00
Cameron Gutman
46c76534bc
Use AVBufferRefs to keep EGLImages and DRM FDs around for the lifetime of the frame
2026-01-04 18:06:24 -06:00
Cameron Gutman
b8c3658c42
Skip modifier check if the frame doesn't have any modifiers
2026-01-03 02:53:44 -06:00
Cameron Gutman
d3fb52c5a0
Check modifiers support in IN_FORMATS property
...
As it turns out, there *are* in fact hardware vendors dumb enough
to ship V4L2 decoders that output buffers which are unsupported by
any plane on their display hardware. By checking modifiers, we can
still fall back to GLES on GL_IS_SLOW=1 builds in this case.
2026-01-03 02:43:15 -06:00
Cameron Gutman
2972022ced
Remove async and non-blocking page flip support
...
It doesn't work well in practice for a few major reasons:
- It can result in the decoder writing on DMA-BUFs being used by the display
- It can generate page flip events that collide with Qt/SDL
- It can stall the decoder if the DMA-BUF has a fence on it from the display driver
2026-01-03 00:41:14 -06:00
Cameron Gutman
c0846155c3
Fix page flipping bugs
2026-01-02 20:01:54 -06:00
Cameron Gutman
ac6dadffc6
Use page flip events to correctly synchronize with async flips
2026-01-02 19:47:43 -06:00
Cameron Gutman
745ac34b15
Improve page flipping and modifier handling in DrmRenderer
2026-01-02 18:44:18 -06:00
Cameron Gutman
5f89636ed7
Improve robustness of DRM property value handling
2026-01-02 02:06:11 -06:00
Cameron Gutman
46d58e55a3
Add macro for printing FourCC values
2026-01-02 01:46:20 -06:00
Cameron Gutman
be460970a8
Remove use of DMA_BUF_IOCTL_SYNC
...
It's only necessary on "real" DMA-BUFs, not dumb buffers.
2026-01-02 01:30:03 -06:00
Cameron Gutman
d50ba06321
Implement DRM atomic renderer with overlay support
...
Fixes #1733
2026-01-02 01:24:35 -06:00
Cameron Gutman
b108684edc
Handle missing bitstream restrictions element in H.264 SPS fixup
...
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.
2025-12-30 17:12:37 -06:00
Cameron Gutman
b6d10a34c8
Update blocking swap buffer opt-out comment
2025-12-30 15:43:31 -06:00
Cameron Gutman
232ff0b3c6
Revert "Remove opt-out of blocking swap buffers logic for KMSDRM"
...
This significantly hurts performance on slower GPUs by preventing
the pipelining required when frametimes exceed the video frame
interval.
This reverts commit c26e437f40
2025-12-30 15:39:47 -06:00
Cameron Gutman
c26e437f40
Remove opt-out of blocking swap buffers logic for KMSDRM
...
We don't set SDL_HINT_VIDEO_DOUBLE_BUFFER since b1c77ff
2025-12-29 16:51:03 -06:00
Cameron Gutman
e89f63b8fb
Skip unnecessary glClear() on X11
2025-12-29 16:50:15 -06:00
Cameron Gutman
93dc6d6b60
Convert Metal shaders to FP16
2025-12-29 01:45:49 -06:00
Cameron Gutman
3e5aa9b127
Simplify EGLImageFactory and remove caching logic for now
...
The platforms that would most benefit (embedded V4L2 decoders)
either don't use frame pooling or don't synchronize with
modified DMA-BUFs unless eglCreateImage() is called each time.
2025-12-28 17:54:31 -06:00
Cameron Gutman
085480e9f4
Add EGLImage caching to improve performance
2025-12-28 17:08:45 -06:00
Cameron Gutman
c3ce0918b3
Optimize blending in DXVA2 and D3D11VA renderers
2025-12-28 13:53:13 -06:00
Cameron Gutman
c5b7a9c679
Optimize EGLRenderer with overlay VAOs and reduced GL_BLEND usage
2025-12-28 13:14:16 -06:00
Cameron Gutman
f1451a0db3
Fix UB due to missing shader attribute location binding
...
Mesa drivers seem to generally bind attributes in order, but the
proprietary PowerVR driver (pvr) doesn't and renders incorrectly
as a result.
2025-12-27 23:56:17 -06:00
Cameron Gutman
8a5f34fb01
Remove logic to force the OpenGL ES library to load
...
This isn't necessary now that we turn on EGL everywhere,
and it breaks OpenGL ES 2.0 support too.
2025-12-27 23:11:53 -06:00
Cameron Gutman
d1f43ca258
Use ESSL 1.0 for EGLRenderer shaders
...
We still ostensibly support GLES 2.0 GPUs.
2025-12-27 22:56:20 -06:00
Cameron Gutman
249c29b74c
Allow GL_IS_SLOW and VULKAN_IS_SLOW to be overridden at runtime
2025-12-27 21:40:32 -06:00
Cameron Gutman
1cf207de07
Allow overriding FFmpeg AVOptions via environment variable
2025-12-27 21:07:04 -06:00
Cameron Gutman
4f08d52754
Enable DRM_PRIME frames for the TH1520
2025-12-27 15:40:59 -06:00
Cameron Gutman
41ad3c0938
Add environment variable override helper function
...
This allows FORCE_QT_GLES and SEPARATE_TEST_DECODER to override both true and false.
2025-12-27 15:33:39 -06:00
Cameron Gutman
85ea28287e
Add workaround for AV1 reinit bug and quiet some debug output
2025-12-23 00:59:19 -06:00
Cameron Gutman
a6f8901a18
Reuse the test decoder to improve startup performance
2025-12-23 00:32:36 -06:00
Cameron Gutman
d501a627f0
Test all renderers before use
...
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.
2025-12-22 23:58:26 -06:00
Cameron Gutman
526d815930
Don't fall back to the generic hwaccel for DRM_PRIME formats
2025-12-22 15:06:20 -06:00
Cameron Gutman
3425fec33d
Use EGL+GLES workaround for Nvidia X11
...
We can avoid disabling EGL entirely by forcing Qt to use GLES,
which is not impacted by the black window issue. This lets us
simplify back to EGL everywhere.
2025-12-14 16:21:32 -06:00
Cameron Gutman
de844f2a0c
Deprioritize the Nvidia VAAPI driver if Vulkan or CUDA are available
2025-12-11 20:13:14 -06:00
Cameron Gutman
54163e30d0
Disable EGL on Nvidia XWayland environments too
...
It renders a black window, just like native X11.
2025-12-11 20:13:14 -06:00
Cameron Gutman
dbcc6a9089
Improve initialization performance when using Vulkan Video decoding
2025-12-07 17:57:22 -06:00
Cameron Gutman
9c6d2bab19
Use EGL_VENDOR instead of NV-CONTROL for Nvidia detection
...
The latter will still match on a hybrid system when the iGPU's EGL
implementation is in use. This scenario works properly with EGL
rendering in Qt and SDL, so we want to allow that.
2025-12-06 11:20:03 -06:00
Cameron Gutman
c92dc2495e
Don't use vaPutSurface() if there are no valid subpicture formats
2025-12-05 23:52:41 -06:00
Cameron Gutman
1c24aada0a
Never use X11 EGL on Nvidia proprietary drivers
...
Fixes #1751
2025-12-05 23:10:31 -06:00
Cameron Gutman
58f9b2a83b
Use opaque black in ClearRenderTargetView()
...
Intel's old Gen9 performance guide says it's a little faster.
2025-12-01 22:39:28 -06:00
Cameron Gutman
d3c23b55dc
Disable the VAAPI RFI latency workaround by default
...
I can no longer reproduce this issue on the same system running
Ubuntu 24.04 with the core22-based edge channel snap anymore.
2025-11-30 22:23:09 -06:00
Cameron Gutman
f0821d9543
Use D3D11_COPY_DISCARD when updating our video texture
...
This ensures the driver knows it doesn't have to synchronize with the previous texture data.
2025-11-24 18:49:19 -06:00
Cameron Gutman
4164a222ae
Consolidate D3D11 constant buffers now that all shaders need a CSC constant buffer
2025-11-24 18:47:14 -06:00
Cameron Gutman
245cd7f1a8
Don't trim the alignment padding with CopySubresourceRegion()
...
The shader already handles trimming for the bound output texture case.
2025-11-24 01:44:51 -06:00
