leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

CUTLASS 1.2

Browse Source
This commit is contained in:
akerr
2018-10-26 14:38:46 -07:00
parent 2332df492e
commit 74df0331f2
97 changed files with 11301 additions and 632 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/02_cutlass_utilities/cutlass_utilities.cu
+10 -9
View File
@@ -103,6 +103,7 @@
// Defines cutlass::reference::host::Gemm()
#include "tools/util/reference/host/gemm.h"
#pragma warning( disable : 4503)
///////////////////////////////////////////////////////////////////////////////////////////////////
/// Define a CUTLASS GEMM template and launch a GEMM kernel.
@@ -144,18 +145,18 @@ cudaError_t Cutlass_FP16_SgemmNN(
typename Gemm::Params params;
int result = params.initialize(
M, // GEMM M dimension
N, // GEMM N dimension
K, // GEMM K dimension
reinterpret_cast<half const &>(alpha), // scalar alpha - This is a legal conversion from cutlass::half_t to CUDA's half.
A, // matrix A operand
M, // GEMM M dimension
N, // GEMM N dimension
K, // GEMM K dimension
reinterpret_cast<half const &>(alpha), // scalar alpha
A, // matrix A operand
lda,
B, // matrix B operand
B, // matrix B operand
ldb,
reinterpret_cast<half const &>(beta), // scalar beta - This is a legal conversion from cutlass::half_t to CUDA's half.
C, // source matrix C
reinterpret_cast<half const &>(beta), // scalar beta
C, // source matrix C
ldc,
C, // destination matrix C (may be different memory than source C matrix)
C, // destination matrix C (may be different memory than source C matrix)
ldc
);