diff options
| author | Cedric Nugteren <web@cedricnugteren.nl> | 2017-12-03 16:40:36 +0100 |
|---|---|---|
| committer | Cedric Nugteren <web@cedricnugteren.nl> | 2017-12-03 16:40:36 +0100 |
| commit | cf4555d1f44aea9c82b60211b5650b6b77a1226c (patch) | |
| tree | 459676fcf89b85aaab7b014d935d1f5b3ab984fc /src/kernels/level1/xdot.opencl | |
| parent | 0a1a3de58a410f61f3b990537541a633826ea640 (diff) | |
Added GEMM (direct and in-direct) to the pre-processor testing; modified the loops in kernel accordingly
Diffstat (limited to 'src/kernels/level1/xdot.opencl')
| -rw-r--r-- | src/kernels/level1/xdot.opencl | 2 |
1 files changed, 0 insertions, 2 deletions
diff --git a/src/kernels/level1/xdot.opencl b/src/kernels/level1/xdot.opencl index 02f04ea7..1a703d96 100644 --- a/src/kernels/level1/xdot.opencl +++ b/src/kernels/level1/xdot.opencl @@ -55,7 +55,6 @@ void Xdot(const int n, barrier(CLK_LOCAL_MEM_FENCE); // Performs reduction in local memory - #pragma unroll for (int s=WGS1/2; s>0; s=s>>1) { if (lid < s) { Add(lm[lid], lm[lid], lm[lid + s]); @@ -84,7 +83,6 @@ void XdotEpilogue(const __global real* restrict input, barrier(CLK_LOCAL_MEM_FENCE); // Performs reduction in local memory - #pragma unroll for (int s=WGS2/2; s>0; s=s>>1) { if (lid < s) { Add(lm[lid], lm[lid], lm[lid + s]); |