diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2016-09-21 21:32:18 +0200 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2016-09-21 21:32:18 +0200 |
commit | 6aa652d6ea2389744195ae5cd19321325b2d71aa (patch) | |
tree | 58243cb4fbebb094c543028124412718cdeb7c97 /src/kernels/level1/xdot.opencl | |
parent | 5004a435ff984bba0dff0147a5c4f6a04d703562 (diff) | |
parent | b1929d8ce7022cacbd1812d62098ebd0681bc1ef (diff) |
Merge branch 'development' into gemm_direct
Diffstat (limited to 'src/kernels/level1/xdot.opencl')
-rw-r--r-- | src/kernels/level1/xdot.opencl | 16 |
1 files changed, 8 insertions, 8 deletions
diff --git a/src/kernels/level1/xdot.opencl b/src/kernels/level1/xdot.opencl index e13eb3c1..02f04ea7 100644 --- a/src/kernels/level1/xdot.opencl +++ b/src/kernels/level1/xdot.opencl @@ -30,11 +30,11 @@ R"( // ================================================================================================= // The main reduction kernel, performing the multiplication and the majority of the sum operation -__attribute__((reqd_work_group_size(WGS1, 1, 1))) -__kernel void Xdot(const int n, - const __global real* restrict xgm, const int x_offset, const int x_inc, - const __global real* restrict ygm, const int y_offset, const int y_inc, - __global real* output, const int do_conjugate) { +__kernel __attribute__((reqd_work_group_size(WGS1, 1, 1))) +void Xdot(const int n, + const __global real* restrict xgm, const int x_offset, const int x_inc, + const __global real* restrict ygm, const int y_offset, const int y_inc, + __global real* output, const int do_conjugate) { __local real lm[WGS1]; const int lid = get_local_id(0); const int wgid = get_group_id(0); @@ -73,9 +73,9 @@ __kernel void Xdot(const int n, // The epilogue reduction kernel, performing the final bit of the sum operation. This kernel has to // be launched with a single workgroup only. -__attribute__((reqd_work_group_size(WGS2, 1, 1))) -__kernel void XdotEpilogue(const __global real* restrict input, - __global real* dot, const int dot_offset) { +__kernel __attribute__((reqd_work_group_size(WGS2, 1, 1))) +void XdotEpilogue(const __global real* restrict input, + __global real* dot, const int dot_offset) { __local real lm[WGS2]; const int lid = get_local_id(0); |