leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 27 Packages Projects Releases Wiki Activity

v4.2 release. (#2587)

Browse Source 
* Fix default cluster callback values to 1 to avoid profiler failure when these values are not set in command line.

* v4.2 release.
This commit is contained in:
Junkai-Wu
2025-08-22 18:11:24 -04:00
committed by GitHub
parent 11cad1f67b
commit a49a78ffef
351 changed files with 28182 additions and 2032 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/unit/gemm/device/gemm_testbed_3x_ptr_array.hpp
+26
View File
@@ -830,6 +830,32 @@ struct HostCollectiveMainloop<cutlass::gemm::KernelPtrArrayTmaWarpSpecializedCoo
) : Base::HostCollectiveMainloop(check_relative_equality_, init_A_, init_B_, seed_, stride_factor_A_, stride_factor_B_) {}
};
//
// Block Scaled Gemm Input Operands : A , B, scalefactorA, scalefactorB
//
template<
class Gemm,
int SchedulerPipelineStageCount_,
int AccumulatorPipelineStageCount_,
class ElementA_,
class ElementB_
>
struct HostCollectiveMainloop<cutlass::gemm::KernelPtrArrayTmaWarpSpecializedBlockScaledSm103<SchedulerPipelineStageCount_,
AccumulatorPipelineStageCount_>,
Gemm, ElementA_, ElementB_> : public
HostCollectiveMainloop<cutlass::gemm::KernelPtrArrayTmaWarpSpecializedBlockScaledSm100<SchedulerPipelineStageCount_,AccumulatorPipelineStageCount_>,
Gemm, ElementA_, ElementB_> {
using Base = HostCollectiveMainloop<cutlass::gemm::KernelPtrArrayTmaWarpSpecializedBlockScaledSm100<SchedulerPipelineStageCount_,AccumulatorPipelineStageCount_>,
Gemm, ElementA_, ElementB_>;
HostCollectiveMainloop(
CheckEquality check_relative_equality_ = CheckEquality::EXACT,
cutlass::Distribution::Kind init_A_ = cutlass::Distribution::Uniform,
cutlass::Distribution::Kind init_B_ = cutlass::Distribution::Uniform,
uint64_t seed_ = Base::kDefaultSeed,
typename Base::LayoutTagA::Stride stride_factor_A_ = typename Base::LayoutTagA::Stride(),
typename Base::LayoutTagB::Stride stride_factor_B_ = typename Base::LayoutTagB::Stride()
) : Base::HostCollectiveMainloop(check_relative_equality_, init_A_, init_B_, seed_, stride_factor_A_, stride_factor_B_) {}
};
template<class Gemm>
struct HostCollectiveDefaultEpilogue {