diff options
Diffstat (limited to 'src/kernels/level1/xnrm2.opencl')
| -rw-r--r-- | src/kernels/level1/xnrm2.opencl | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/kernels/level1/xnrm2.opencl b/src/kernels/level1/xnrm2.opencl index 6a81c150..36ea49b4 100644 --- a/src/kernels/level1/xnrm2.opencl +++ b/src/kernels/level1/xnrm2.opencl @@ -30,7 +30,11 @@ R"( // ================================================================================================= // The main reduction kernel, performing the multiplication and the majority of the operation -__kernel __attribute__((reqd_work_group_size(WGS1, 1, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#else + __kernel __attribute__((reqd_work_group_size(WGS1, 1, 1))) +#endif void Xnrm2(const int n, const __global real* restrict xgm, const int x_offset, const int x_inc, __global real* output) { @@ -71,7 +75,11 @@ void Xnrm2(const int n, // The epilogue reduction kernel, performing the final bit of the operation. This kernel has to // be launched with a single workgroup only. -__kernel __attribute__((reqd_work_group_size(WGS2, 1, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#else + __kernel __attribute__((reqd_work_group_size(WGS2, 1, 1))) +#endif void Xnrm2Epilogue(const __global real* restrict input, __global real* nrm2, const int nrm2_offset) { __local real lm[WGS2]; |