diff options
Diffstat (limited to 'src/kernels/level1/xcopy.opencl')
| -rw-r--r-- | src/kernels/level1/xcopy.opencl | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/kernels/level1/xcopy.opencl b/src/kernels/level1/xcopy.opencl index aed80fc2..493197af 100644 --- a/src/kernels/level1/xcopy.opencl +++ b/src/kernels/level1/xcopy.opencl @@ -22,7 +22,11 @@ R"( // ================================================================================================= // Full version of the kernel with offsets and strided accesses -__kernel __attribute__((reqd_work_group_size(WGS, 1, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#else + __kernel __attribute__((reqd_work_group_size(WGS, 1, 1))) +#endif void Xcopy(const int n, const __global real* restrict xgm, const int x_offset, const int x_inc, __global real* ygm, const int y_offset, const int y_inc) { @@ -37,7 +41,11 @@ void Xcopy(const int n, // Faster version of the kernel without offsets and strided accesses. Also assumes that 'n' is // dividable by 'VW', 'WGS' and 'WPT'. -__kernel __attribute__((reqd_work_group_size(WGS, 1, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#else + __kernel __attribute__((reqd_work_group_size(WGS, 1, 1))) +#endif void XcopyFast(const int n, const __global realV* restrict xgm, __global realV* ygm) { |