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

CUTLASS 3.2.1 (#1113)

Browse Source 
* Updates for 3.2.1 release.

* Minor fix in gemm op profiler for raster order.

* Add scheduler mapping for raster order in the kernels.
This commit is contained in:
ANIKET SHIVAM
2023-09-26 17:24:26 -04:00
committed by GitHub
parent e0aaa3c3b3
commit 90d3b0fb18
428 changed files with 22252 additions and 21761 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/13_two_tensor_op_fusion/fused_two_convs_s8_sm75_rf.cu
-5
View File
@@ -220,7 +220,6 @@ bool run_fused_conv2d_fprop_optimized_s8_sm75_rf_res() {
return pass;
}
int main() {
std::vector<bool (*)()>funcs = {
@@ -229,10 +228,6 @@ int main() {
};
return testRun(75, funcs, "conv int8 RF residency");
}
////////////////////////////////////////////////////////////////////////////////
examples/13_two_tensor_op_fusion/fused_two_convs_s8_sm75_shmem.cu
-8
View File
@@ -39,7 +39,6 @@
#include "device/b2b_implicit_gemm_convolution.h"
#include "b2b_interleaved_conv2d_run.h"
#include "test_run.h"
////////////////////////////////////////////////////////////////////////////////
cutlass::conv::Conv2dProblemSize conv2d_s8_sm75_problem_size_0 (
@@ -219,20 +218,13 @@ bool run_fused_conv2d_fprop_optimized_s8_sm75_shmem() {
return pass;
}
int main() {
std::vector<bool (*)()>funcs = {
&run_nonfused_conv2d_fprop_optimized_s8_sm75,
&run_fused_conv2d_fprop_optimized_s8_sm75_shmem
};
return testRun(75, funcs, "conv int8 shmem staging");
}
////////////////////////////////////////////////////////////////////////////////
examples/13_two_tensor_op_fusion/fused_two_gemms_s8_sm75_rf.cu
-4
View File
@@ -195,7 +195,6 @@ bool run_fused_gemm_s8_rf_res() {
return passed;
}
int main() {
std::vector<bool (*)()>funcs = {
@@ -204,9 +203,6 @@ int main() {
};
return testRun(75, funcs, "gemm int8 RF residency");
}
////////////////////////////////////////////////////////////////////////////////
examples/13_two_tensor_op_fusion/fused_two_gemms_s8_sm75_shmem.cu
-6
View File
@@ -43,7 +43,6 @@
#include "device/b2b_gemm.h"
#include "b2b_interleaved_gemm_run.h"
#include "test_run.h"
////////////////////////////////////////////////////////////////////////////////
cutlass::gemm::GemmCoord gemm_s8_sm75_problem_size_0(128*640, 64, 576);
@@ -197,18 +196,13 @@ bool run_fused_gemm_s8_shmem() {
return passed;
}
int main() {
std::vector<bool (*)()>funcs = {
&run_nonfused_gemm_s8,
&run_fused_gemm_s8_shmem
};
return testRun(75, funcs, "gemm int8 shmem staing");
}
////////////////////////////////////////////////////////////////////////////////
examples/13_two_tensor_op_fusion/threadblock/grouped_threadblock_swizzle.h
-28
View File
@@ -90,34 +90,6 @@ struct GroupedThreadblockSwizzle : detail::GroupedThreadblockSwizzleBase {
}
};
template <
typename ThreadblockShape,
typename LayoutC,
cutlass::gemm::kernel::GroupScheduleMode GroupScheduleMode_ = cutlass::gemm::kernel::GroupScheduleMode::kDeviceOnly,
int PrefetchTileCount = 128,
int ThreadCount = PrefetchTileCount>
struct GemmGroupedThreadblockSwizzle : GroupedThreadblockSwizzle<
cutlass::gemm::kernel::GemmGroupedProblemVisitor<
ThreadblockShape,
GroupScheduleMode_,
PrefetchTileCount,
ThreadCount,
platform::is_same<LayoutC, cutlass::layout::ColumnMajor>::value
>
> {
using Base = GroupedThreadblockSwizzle<cutlass::gemm::kernel::GemmGroupedProblemVisitor<
ThreadblockShape,
GroupScheduleMode_,
PrefetchTileCount,
ThreadCount,
platform::is_same<LayoutC, cutlass::layout::ColumnMajor>::value>>;
CUTLASS_HOST_DEVICE
GemmGroupedThreadblockSwizzle(typename Base::ProblemVisitor::Params& params,
typename Base::ProblemVisitor::SharedStorage& shared_storage,
int block_idx) : Base(params, shared_storage, block_idx) {}
};
template <
typename ThreadblockShape,
typename LayoutC,