Skip to content

Conversation

@Brainiarc7
Copy link

This patch is a fix to #1379, wherein:

  1. The CUDA library stubs are added to ldconfig's cache.
  2. The build type for libvmaf is set to shared.
  3. The included models are also copied to /usr/local/share/model for re-usability with custom FFmpeg pipelines as needed.
  4. ldconfig is called up immediately after libvmaf build so that FFmpeg can detect it.

This patch is a fix to Netflix#1379, wherein:

1. The CUDA library stubs are added to `ldconfig`'s cache.
2. The build type for `libvmaf` is set to `shared`.
3. The included models are also copied to `/usr/local/share/model` for re-usability with custom FFmpeg pipelines as needed.
4. `ldconfig` is called up immediately after `libvmaf` build so that `FFmpeg` can detect it.
@nilfm99 nilfm99 requested a review from kylophone July 15, 2024 15:53
@guyoun
Copy link
Contributor

guyoun commented Jul 22, 2024

I think the FFmpeg build failure is due to the missing libcuda.so.1 library file.

The vmaf is linked with libcuda.so.1 because of SONAME of libcuda.so is libcuda.so.1
When enable libvmaf of FFmpeg, it requires libcuda.so.1 but this file does not exist when building the Docker image.
I am unsure why the SONAME of /usr/local/cuda/lib64/stubs/libcuda.so is libcuda.so.1.
According to this issue on the NVIDIA GitHub repository, there are no plans to fix this.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants