CUDA 13.0 apply patch #24533
CUDA 13.0 apply patch #24533
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request applies a patch to ensure compatibility with CUDA 13.0. The changes correctly replace the deprecated cub::Sum and cub::Max with cuda::std::plus and cuda::maximum respectively for CUDA builds, while maintaining the existing cub:: operations for ROCm builds. This is achieved by introducing AddOp and MaxOp type aliases. The implementation is consistent across all modified files and appears correct. I have no specific comments as the changes are well-executed.
|
Have you tried building on CUDA 13.0 with this patch? If it works we can merge this. |
csrc/moe/topk_softmax_kernels.cu
Outdated
| using AddOp = cuda::std::plus<>; | ||
| using MaxOp = cuda::maximum<>; |
There was a problem hiding this comment.
It seems like maximum does not currently work on 12.8 so we should guard on the CUDA version here
|
This pull request has merge conflicts that must be resolved before it can be |
|
Done in #24599 |
2 participants
As suggested by #24464 (comment)