leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

v3.9 update (#2213)

Browse Source 
Co-authored-by: yuzhai <yuzhai@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-04-03 02:10:16 -04:00
committed by GitHub
co-authored by yuzhai
parent 6f4921858b
commit 79fc51f4b8
72 changed files with 19875 additions and 459 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/experimental/distributed/device/dist_gemm_universal_wrapper.hpp
+1 -1
View File
@@ -340,7 +340,7 @@ public:
base_args.epilogue.thread,
reinterpret_cast<const ElementC*>(tensor_c_iter.data()),
tensor_c_iter.stride(),
reinterpret_cast<const ElementD*>(tensor_d_iter.data()),
reinterpret_cast<ElementD*>(tensor_d_iter.data()),
tensor_d_iter.stride()
};
include/cutlass/experimental/distributed/kernel/dist_gemm_kernel_wrapper.hpp
+1 -1
View File
@@ -82,7 +82,7 @@ struct DistributedGemmKernelWrapper<
using BaseArguments = typename BaseKernel::Arguments;
using BaseParams = typename BaseKernel::Params;
static_assert(BaseKernel::ArchTag::kMinComputeCapability == 90, "DistGEMM only supports Hopper GEMMs for now.");
//static_assert(BaseKernel::ArchTag::kMinComputeCapability == 90, "DistGEMM only supports Hopper GEMMs for now.");
static_assert(not cute::is_same_v<typename BaseKernel::ElementC, void>, "DistributedGEMM epilogues must have a source.");
using ElementFlag = uint32_t;