libvmaf cuda - init_fex_cuda: Assertion `0' failed.

[nguyend24]

Hello,

I've been trying to get libvmaf with cuda working. I compiled ffmpeg using the Dockerfile.cuda. Ran into a few issues with that which were resolved with the suggestions here #1328.

I copied the ffmpeg executable out of the docker container along with the needed libraries to use in a Debian WSL environment.

However, it's throwing an error
code: 218; description: CUDA_ERROR_INVALID_PTX
ffvmaf: ../src/feature/cuda/integer_adm_cuda.c:1026: init_fex_cuda: Assertion `0' failed.

Full logs:
vmaf.txt

[kylophone]

This is typically a mismatch between CUDA driver and runtime. This has come up a few times, and I think we will eventually be looking into something like ffnvcodec to help solve issues like this.

[Gui-Yom]

I think the problem comes from the fact that the CUDA Driver API states :

Takes a pointer image and loads the corresponding module module into the current context. The pointer may be obtained by mapping a cubin or PTX or fatbin file, passing a cubin or PTX or fatbin file as a NULL-terminated text string, or incorporating a cubin or fatbin object into the executable resources and using operating system calls such as Windows FindResource() to obtain the pointer.

The ptx files as formatted by xxd do not contain a terminating null byte as required. On my machine, vmaf hangs indefinitely when loading the ptx. I tried adding the null byte and that fixed up things for me. I can share a fix if you want to.

[amachang]

I faced the same issue. And, as Gui Yom said, there certainly is a problem with PTX code strings embedded in C that were not null-terminated. I wrote some workaround codes. Use it if it helps someone.