diff options
| author | Gard Spreemann <gspr@nonempty.org> | 2023-06-08 11:52:00 +0200 |
|---|---|---|
| committer | Gard Spreemann <gspr@nonempty.org> | 2023-06-08 11:52:00 +0200 |
| commit | 63870a2e60c1bc8bfa7e3672457b551a8e51ffaf (patch) | |
| tree | fe2c0cd5f62e3fbd17e58d3903ec6bb37983f620 /src/kernels/level3/copy_fast.opencl | |
| parent | d31fb141cb597aaf405674621aa25f263aa375e1 (diff) | |
| parent | b0b302889cc786907efb080c4e1beea30d2fa39f (diff) | |
Merge tag '1.6.0' into gspr/post-bookworm
Diffstat (limited to 'src/kernels/level3/copy_fast.opencl')
| -rw-r--r-- | src/kernels/level3/copy_fast.opencl | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/src/kernels/level3/copy_fast.opencl b/src/kernels/level3/copy_fast.opencl index ef8a9017..04e4b2ab 100644 --- a/src/kernels/level3/copy_fast.opencl +++ b/src/kernels/level3/copy_fast.opencl @@ -35,7 +35,11 @@ R"( // Fast copy kernel. Requires 'ld' and the number of threads in dimension 0 to be a multiple of // COPY_VW. Also requires both matrices to be of the same dimensions and without offset. -__kernel __attribute__((reqd_work_group_size(COPY_DIMX, COPY_DIMY, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#else + __kernel __attribute__((reqd_work_group_size(COPY_DIMX, COPY_DIMY, 1))) +#endif void CopyMatrixFast(const int ld, __global const realC* restrict src, __global realC* dest, |