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

CUDA 9 lacks host-side conversions from float=>half. Instead, we must reinterpret_cast<> from cutlass::half_t => half.

Browse Source
This commit is contained in:
akerr 2018-09-29 15:04:20 -07:00
parent 6877595a5e
commit cfe4b933ef
1 changed files with 9 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
examples/02_cutlass_utilities/cutlass_utilities.cu
View File

@ -144,18 +144,18 @@ cudaError_t Cutlass_FP16_SgemmNN(
typename Gemm::Params params; typename Gemm::Params params;
int result = params.initialize( int result = params.initialize(
M, // GEMM M dimension M, // GEMM M dimension
N, // GEMM N dimension N, // GEMM N dimension
K, // GEMM K dimension K, // GEMM K dimension
half(float(alpha)), // scalar alpha - This is a legal conversion from cutlass::half_t to CUDA's half. reinterpret_cast<half const &>(alpha), // scalar alpha - This is a legal conversion from cutlass::half_t to CUDA's half.
A, // matrix A operand A, // matrix A operand
lda, lda,
B, // matrix B operand B, // matrix B operand
ldb, ldb,
half(float(beta)), // scalar beta - This is a legal conversion from cutlass::half_t to CUDA's half. reinterpret_cast<half const &>(beta), // scalar beta - This is a legal conversion from cutlass::half_t to CUDA's half.
C, // source matrix C C, // source matrix C
ldc, 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 ldc
); );