cutlass 3.9 update (#2255)

* cutlass 3.9 update

* rebase

* fixes out of shared memory for blockwise Blackwell

* doc format

* fix issue 2253

* disable host ref by default

* fix sm120 smem capacity

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-04-24 15:42:40 -04:00
committed by GitHub
co-authored by yuzhai Haicheng Wu
parent 8e345c5c5b
commit 331a1f5b3f
143 changed files with 18089 additions and 5935 deletions
+3
View File
@@ -37,6 +37,7 @@
// Profiler includes
#include "cutlass/profiler/block_scaled_gemm_operation_profiler.h"
#include "cutlass/profiler/blockwise_gemm_operation_profiler.h"
#include "cutlass/profiler/conv2d_operation_profiler.h"
#include "cutlass/profiler/conv3d_operation_profiler.h"
#include "cutlass/profiler/cutlass_profiler.h"
@@ -64,6 +65,8 @@ CutlassProfiler::CutlassProfiler(
operation_profilers_.emplace_back(new BlockScaledGemmOperationProfiler(options));
operation_profilers_.emplace_back(new BlockwiseGemmOperationProfiler(options));
operation_profilers_.emplace_back(new SparseGemmOperationProfiler(options));
operation_profilers_.emplace_back(new Conv2dOperationProfiler(options));