aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/vulkan_ffv1.c
Commit message (Collapse)AuthorAgeFilesLines
* vulkan/ffv1dec: fix FFVkSPIRVCompiler leakaverne2025-06-111-35/+23
|
* vulkan/ffv1dec: fix leak in FFVulkanDecodeSharedaverne2025-06-111-0/+2
|
* vulkan_ffv1: pipe through slice decoding statusLynne2025-05-201-17/+43
|
* ffv1enc_vulkan: switch to 2-line cache, unify prediction codeLynne2025-05-201-1/+4
|
* vulkan_ffv1: add cached symbol reader for AMDLynne2025-04-141-1/+6
| | | | | | Speeds up everything on AMD by 3x. This uses 32 local invocations to load state into cache, as well as to do the RCT faster.
* vulkan_ffv1: remove need for scratch data during setupLynne2025-04-141-22/+1
| | | | This saves on some VRAM, but mainly allows for a more unified path.
* vulkan_ffv1: externalize extended lookup checkLynne2025-04-141-0/+6
| | | | 8% speedup on nvidia on 4k.
* ffv1/vulkan: redo context count tracking and quant_table_idx managementLynne2025-04-141-14/+8
| | | | | | | | This commit also makes it possible for the encoder to choose a different quantization table on a per-slice basis, as well as adding this capability to the decoder. Also, this commit fully fixes decoding of context=1 encoded files.
* vulkan_ffv1: cache only 2 lines when decoding RGBLynne2025-04-141-202/+81
| | | | | | | | This reduces the intermediate VRAM used for RGB decoding by a factor of 100x for 6k video. This also speeds the decoder up by 16% for 4k RGB24 and 31% for 6k video. This is equivalent to what the software decoder does, but with less pointers.
* vulkan_ffv1: improve buffer barrier correctness for slice stateLynne2025-04-141-3/+2
| | | | This is likely a nanooptimization, but its more correct.
* vulkan_ffv1: fix reset shader dependenciesLynne2025-04-141-19/+17
| | | | | Without a barrier upfront, the reset shader may read data fields not yet set by the setup shader.
* vulkan_ffv1: fallback to upload if mapping packet fails, fix fallbackLynne2025-04-141-12/+7
| | | | | | The commit which added support for host mapping accidentally broke the original, upload route. For drivers without host-mapping (very few), fix it.
* vulkan_ffv1: allocate just as much memory for slice state as neededLynne2025-04-141-4/+4
| | | | | Rather than always using the maximum allowed slices, just use the number of slices present in this frame.
* vulkan_ffv1: remove unused defineLynne2025-04-141-2/+0
| | | | Leftover debug macro.
* vulkan_ffv1: enable acceleration on IntelLynne2025-04-141-14/+0
| | | | Fixed by previous commit.
* ffv1: add a Vulkan-based decoderLynne2025-03-171-0/+1317
This patch adds a fully-featured level 3 and 4 decoder for FFv1, supporting Golomb and all Range coding variants, all pixel formats, and all features, except for the newly added floating-point formats. On a 6000 Ada, for 3840x2160 bgr0 content at 50Mbps (standard desktop recording), it is able to do 400fps. An Alder Lake with 24 threads can barely do 100fps.
generated by cgit v1.2.3 (git 2.25.1) at 2025-08-07 11:20:56 +0000