[GH-ISSUE #202] CU_TENSOR_MAP_SWIZZLE_128B_ATOM_32B causes CUDA versions below 12.8 to fail to compile. #117

Closed
opened 2026-10-26 21:12:57 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @wiryls on GitHub (Sep 30, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/202

According to CUDA 12.6.1 document and CUDA 12.8.0 document, enum CU_TENSOR_MAP_SWIZZLE_128B_ATOM_32B appears to have been added in 12.8.

Changes in #198 make CUDA versions < 12.8 to fail to compile.

github.com/deepseek-ai/DeepGEMM@594953acce/csrc/jit_kernels/impls/runtime_utils.hpp (L70-L76)

Originally created by @wiryls on GitHub (Sep 30, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/202 According to [CUDA 12.6.1 document](https://docs.nvidia.com/cuda/archive/12.6.1/cuda-driver-api/group__CUDA__TENSOR__MEMORY.html) and [CUDA 12.8.0 document](https://docs.nvidia.com/cuda/archive/12.8.0/cuda-driver-api/group__CUDA__TENSOR__MEMORY.html), enum `CU_TENSOR_MAP_SWIZZLE_128B_ATOM_32B` appears to have been added in 12.8. Changes in #198 make CUDA versions < 12.8 to fail to compile. https://github.com/deepseek-ai/DeepGEMM/blob/594953acce41793ae00a1233eb516044d604bcb6/csrc/jit_kernels/impls/runtime_utils.hpp#L70-L76
Author
Owner

@LyricZhao commented on GitHub (Oct 1, 2025):

Fixed in github.com/deepseek-ai/DeepGEMM@07b82fb8cd. Thanks!

<!-- gh-comment-id:3356084329 --> @LyricZhao commented on GitHub (Oct 1, 2025): Fixed in https://github.com/deepseek-ai/DeepGEMM/commit/07b82fb8cd035330b8849ebc42ef9d6df2dfdde4. Thanks!
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepGEMM#117
No description provided.