leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 26 Packages Projects Releases Wiki Activity

CUTLASS 3.3.0 (#1167)

Browse Source 
* Release 3.3.0

Adds support for mixed precision GEMMs On Hopper and Ampere
Adds support for < 16B aligned GEMMs on Hopper
Enhancements to EVT
Enhancements to Python interface
Enhancements to Sub-byte type handling in CuTe
Several other bug-fixes and performance improvements.

* minor doc update
This commit is contained in:
Pradeep Ramani
2023-11-02 11:09:05 -04:00
committed by GitHub
parent 922fb5108b
commit c008b4aea8
263 changed files with 16214 additions and 5008 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/profiler/CMakeLists.txt
+6 -1
View File
@@ -97,7 +97,12 @@ install(
RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
)
set(CUTLASS_PROFILER_TEST_COMMAND_OPTIONS_GEMM --operation=Gemm --providers=cutlass --verification-providers=cublas,device --junit-output=test_cutlass_profiler_gemm --print-kernel-before-running=true)
if (CUDA_VERSION VERSION_GREATER_EQUAL 12.3 AND CUDA_VERSION VERSION_LESS 12.4 AND (90a IN_LIST CUTLASS_NVCC_ARCHS_ENABLED OR (90 IN_LIST CUTLASS_NVCC_ARCHS_ENABLED)))
set(CUTLASS_PROFILER_TEST_COMMAND_OPTIONS_GEMM --operation=Gemm --providers=cutlass --verification-providers=cublas,host --junit-output=test_cutlass_profiler_gemm --print-kernel-before-running=true)
else()
set(CUTLASS_PROFILER_TEST_COMMAND_OPTIONS_GEMM --operation=Gemm --providers=cutlass --verification-providers=cublas,device --junit-output=test_cutlass_profiler_gemm --print-kernel-before-running=true)
endif()
set(CUTLASS_PROFILER_TEST_COMMAND_OPTIONS_CONV2D --operation=Conv2d --providers=cutlass --verification-providers=cudnn,device --junit-output=test_cutlass_profiler_conv2d --print-kernel-before-running=true)
set(CUTLASS_PROFILER_TEST_COMMAND_OPTIONS_CONV3D --operation=Conv3d --providers=cutlass --verification-providers=cudnn,device,host --junit-output=test_cutlass_profiler_conv3d --print-kernel-before-running=true)
set(CUTLASS_PROFILER_TEST_COMMAND_OPTIONS_SPGEMM --operation=SparseGemm --providers=cutlass --verification-providers=cublas,device,host --junit-output=test_cutlass_profiler_spgemm --print-kernel-before-running=true)
tools/profiler/src/device_allocation.cu
-1
View File
@@ -2005,7 +2005,6 @@ void DeviceAllocation::write_tensor_csv(
case library::NumericTypeID::kFE5M2:
write_tensor_csv_static_type<float_e5m2_t>(out, *this);
break;
case library::NumericTypeID::kF16:
write_tensor_csv_static_type<half_t>(out, *this);
break;
tools/profiler/src/gemm_operation_profiler.cu
+1 -1
View File
@@ -585,11 +585,11 @@ Status GemmOperationProfiler::initialize_workspace(
workspace_size = underlying_operation->get_device_workspace_size(&gemm_workspace_.configuration,
&gemm_workspace_.arguments);
gemm_workspace_.device_workspace.reset(library::NumericTypeID::kU8, workspace_size);
status = underlying_operation->initialize(
&gemm_workspace_.configuration,
gemm_workspace_.host_workspace.data(),
gemm_workspace_.device_workspace.data());
if (status != Status::kSuccess) {
return status;
}