Skip to content

Update fp8_gemm_blockwise_kernel.cu check cuda version #73581

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lshpku merged 4 commits into from
Jun 27, 2025
Merged

Update fp8_gemm_blockwise_kernel.cu check cuda version #73581

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
327053b
Fix
co63oc Jun 24, 2025
b1c636c
Merge branch 'develop' into f47
co63oc Jun 26, 2025
9c359c4
Fix
co63oc Jun 26, 2025
ed53baa
Merge branch 'develop' into f47
co63oc Jun 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions paddle/phi/kernels/legacy/gpu/fp8_gemm_blockwise_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,7 +156,7 @@ void cublas_gemm_blockwise_impl(const DenseTensor& A,
PADDLE_CUDABLAS_CHECK(phi::dynload::cublasLtMatmulDescCreate(
&operationDesc, CUBLAS_COMPUTE_32F, CUDA_R_32F));

#if CUBLAS_VERSION >= 120804 && CUDA_VERSION >= 12060
#if CUBLAS_VERSION >= 120901 && CUDA_VERSION >= 12090
Copy link
Contributor

@lshpku lshpku Jun 26, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

感谢你发现这个问题！该特性是12.8.5引入的，原代码确实有疏忽，但是限制成12.9也不合适，如果你想继续合入这个PR，可以找找cublas官方文档依据，然后修改这里的版本

Copy link
Contributor Author

@co63oc co63oc Jun 26, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

使用apt-get 默认安装的12.8.4， 没有更新的版本，那安装12.8.5是怎么安装
image

PR这里先按12.8.5修改

// Setup scaling for A and B
cublasLtMatmulMatrixScale_t A_scale_mode, B_scale_mode;
// Note: in cuBLAS term, tensor name A and B are swapped.
Expand Down Expand Up @@ -187,7 +187,7 @@ void cublas_gemm_blockwise_impl(const DenseTensor& A,
sizeof(B_scale_mode)));
#else
PADDLE_THROW(phi::errors::InvalidArgument(
"Sub-channel FP8 GEMM requires CUDA 12.8 and cuBLAS 12.8.4 or later."));
"Sub-channel FP8 GEMM requires CUDA 12.9 and cuBLAS 12.9.1 or later."));
#endif

// setup transa and transb
Expand Down
Loading