diff options
author | Angus, Alexander <aangus@qti.qualcomm.com> | 2023-01-03 10:56:04 -0800 |
---|---|---|
committer | Angus, Alexander <aangus@qti.qualcomm.com> | 2023-01-03 10:56:04 -0800 |
commit | 4f394608a28f419dfd6091c704148d9e638a26f0 (patch) | |
tree | 4c0e042109c4d249ff5b700fc49a862169edec5a /src/kernels/level3/copy_fast.opencl | |
parent | 03cffa83c5f7742f8ec0c5e762bb7048e38952f3 (diff) |
implemented changes to boost Adreno performance according to https://jira-dc.qualcomm.com/jira/browse/OSR-8731
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..e1a815f5 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 +#elif + __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, |