diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2017-07-08 17:12:16 +0200 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2017-07-08 17:12:16 +0200 |
commit | 442c31dd508c573023594a803160ddb69d4929f2 (patch) | |
tree | 55474d09086481117204626b27cbec4ee465be9a /src/kernels/level3/xgemm_part2.opencl | |
parent | 75c0e861b842dbd08def5e55696fd79d713afc96 (diff) |
Made the inline keyword in kernels optional currently only enabled for NVIDIA and ARM GPUs
Diffstat (limited to 'src/kernels/level3/xgemm_part2.opencl')
-rw-r--r-- | src/kernels/level3/xgemm_part2.opencl | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/kernels/level3/xgemm_part2.opencl b/src/kernels/level3/xgemm_part2.opencl index e8234a29..06fafc8f 100644 --- a/src/kernels/level3/xgemm_part2.opencl +++ b/src/kernels/level3/xgemm_part2.opencl @@ -18,7 +18,7 @@ R"( // ================================================================================================= // The vectorised multiply-add function -inline realM MultiplyAddVector(realM cvec, const realM avec, const real bval) { +INLINE_FUNC realM MultiplyAddVector(realM cvec, const realM avec, const real bval) { #if USE_VECTOR_MAD == 1 cvec += avec * bval; #else @@ -64,7 +64,7 @@ inline realM MultiplyAddVector(realM cvec, const realM avec, const real bval) { } // Performs the actual computation: Cpm += Apm * Bpm -inline void MultiplyAccumulate(realM cpm[NWI][MWI/VWM], realM apm[MWI/VWM], realN bpm[NWI/VWN]) { +INLINE_FUNC void MultiplyAccumulate(realM cpm[NWI][MWI/VWM], realM apm[MWI/VWM], realN bpm[NWI/VWN]) { #pragma unroll for (int ni=0; ni<NWI/VWN; ++ni) { #pragma unroll @@ -115,8 +115,8 @@ inline void MultiplyAccumulate(realM cpm[NWI][MWI/VWM], realM apm[MWI/VWM], real // Merges the results in Cpm with the global array in Cgm. This also performs the multiplication // with the constants: Cgm = alpha*A*B + beta*Cgm = alpha*Cpm + beta*Cgm -inline void StoreResults(__global realM* cgm, realM cpm[NWI][MWI/VWM], const int kSizeM, - const real alpha, const real beta) { +INLINE_FUNC void StoreResults(__global realM* cgm, realM cpm[NWI][MWI/VWM], const int kSizeM, + const real alpha, const real beta) { #pragma unroll for (int ni=0; ni<NWI; ++ni) { #pragma unroll |