diff options
| author | Cedric Nugteren <web@cedricnugteren.nl> | 2017-12-09 14:09:13 +0100 |
|---|---|---|
| committer | Cedric Nugteren <web@cedricnugteren.nl> | 2017-12-09 14:09:13 +0100 |
| commit | 23e3a85f2c328d4a23db2fca5d1d89d78163711f (patch) | |
| tree | 02b8dd5364d958184c45c9bfdb2c28e38d72b24e /src/kernels/level3/xgemm_batched.opencl | |
| parent | d9df62b7942bb8af5fd385b8545aceb1d8b578f3 (diff) | |
Reformatted GEMM kernel to support array-to-register promotion
Diffstat (limited to 'src/kernels/level3/xgemm_batched.opencl')
| -rw-r--r-- | src/kernels/level3/xgemm_batched.opencl | 14 |
1 files changed, 5 insertions, 9 deletions
diff --git a/src/kernels/level3/xgemm_batched.opencl b/src/kernels/level3/xgemm_batched.opencl index c7bf10d5..372f910b 100644 --- a/src/kernels/level3/xgemm_batched.opencl +++ b/src/kernels/level3/xgemm_batched.opencl @@ -46,20 +46,16 @@ void XgemmBatched(const int kSizeM, const int kSizeN, const int kSizeK, __local realN blm[KWG * NWG/VWN]; #endif - // Computes the matrix-multiplication and stores the result in register memory - realM cpm[NWI][MWI/VWM]; + // Computes the matrix-multiplication and stores the result in global memory #if SA == 1 && SB == 1 - XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, cpm, alm, blm); + XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, alpha, beta, alm, blm); #elif SA == 1 - XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, cpm, alm); + XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, alpha, beta, alm); #elif SB == 1 - XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, cpm, blm); + XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, alpha, beta, blm); #else - XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, cpm); + XgemmBody(kSizeM, kSizeN, kSizeK, agm_, bgm_, cgm_, alpha, beta); #endif - - // Stores an MWG * NWG tile of results and performs the multiplication with alpha and beta - StoreResults(cgm_, cpm, kSizeM, alpha, beta); } // ================================================================================================= |