biondizzle/DeepGEMM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Make various updates and fixes (#198)

Browse Source
This commit is contained in:
Ray Wang
2025-09-25 16:19:07 +08:00
committed by GitHub
parent 79f48ee15a
commit 3f71de7aa9
45 changed files with 3281 additions and 1060 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
csrc/apis/layout.hpp
View File

@@ -56,14 +56,14 @@ static torch::Tensor transform_k_grouped_sf_into_required_layout(const torch::Te
// FP32 on SM90
if (sf.scalar_type() == torch::kFloat and arch_major == 9)
DG_HOST_UNREACHABLE("Unimplemented");
return get_mn_major_tma_aligned_tensor(sf);
// FP32 on SM100
if (sf.scalar_type() == torch::kFloat and arch_major == 10)
return get_k_grouped_mn_major_tma_aligned_packed_ue8m0_tensor(sf, ks_tensor, ks);
// INT on SM100
if (sf.scalar_type() == torch::kFloat and arch_major == 10)
if (sf.scalar_type() == torch::kInt and arch_major == 10)
DG_HOST_UNREACHABLE("Unimplemented");
DG_HOST_UNREACHABLE("Unknown cases");