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/convert_triangular.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/convert_triangular.opencl')
| -rw-r--r-- | src/kernels/level3/convert_triangular.opencl | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/kernels/level3/convert_triangular.opencl b/src/kernels/level3/convert_triangular.opencl index a9d5e769..563f719f 100644 --- a/src/kernels/level3/convert_triangular.opencl +++ b/src/kernels/level3/convert_triangular.opencl @@ -20,7 +20,11 @@ R"( // Kernel to populate a squared triangular matrix, given that the triangle which holds the data is // stored as the lower-triangle of the input matrix. This uses the padding kernel's parameters. -__kernel __attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#elif + __kernel __attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) +#endif void TriaLowerToSquared(const int src_dim, const int src_ld, const int src_offset, __global const real* restrict src, @@ -55,7 +59,11 @@ void TriaLowerToSquared(const int src_dim, } // Same as above, but now the matrix' data is stored in the upper-triangle -__kernel __attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#elif + __kernel __attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) +#endif void TriaUpperToSquared(const int src_dim, const int src_ld, const int src_offset, __global const real* restrict src, |