summaryrefslogtreecommitdiff
path: root/src/kernels/level3/invert_diagonal_blocks_part1.opencl
diff options
context:
space:
mode:
authorAngus, Alexander <aangus@qti.qualcomm.com>2023-01-03 10:56:04 -0800
committerAngus, Alexander <aangus@qti.qualcomm.com>2023-01-03 10:56:04 -0800
commit4f394608a28f419dfd6091c704148d9e638a26f0 (patch)
tree4c0e042109c4d249ff5b700fc49a862169edec5a /src/kernels/level3/invert_diagonal_blocks_part1.opencl
parent03cffa83c5f7742f8ec0c5e762bb7048e38952f3 (diff)
implemented changes to boost Adreno performance according to https://jira-dc.qualcomm.com/jira/browse/OSR-8731
Diffstat (limited to 'src/kernels/level3/invert_diagonal_blocks_part1.opencl')
-rw-r--r--src/kernels/level3/invert_diagonal_blocks_part1.opencl6
1 files changed, 5 insertions, 1 deletions
diff --git a/src/kernels/level3/invert_diagonal_blocks_part1.opencl b/src/kernels/level3/invert_diagonal_blocks_part1.opencl
index c1f96bd7..3df477d1 100644
--- a/src/kernels/level3/invert_diagonal_blocks_part1.opencl
+++ b/src/kernels/level3/invert_diagonal_blocks_part1.opencl
@@ -82,7 +82,11 @@ R"(
// =================================================================================================
// Inverts a diagonal block of INTERNAL_BLOCK_SIZE by INTERNAL_BLOCK_SIZE elements in a larger matrix
-__kernel __attribute__((reqd_work_group_size(INTERNAL_BLOCK_SIZE, 1, 1)))
+#if RELAX_WORKGROUP_SIZE == 1
+ __kernel
+#elif
+ __kernel __attribute__((reqd_work_group_size(INTERNAL_BLOCK_SIZE, 1, 1)))
+#endif
void InvertDiagonalBlock(const int n, __global const real* restrict src, const int src_offset, const int src_ld,
__global real* restrict dest, const int outer_block_size,
const int unit_diagonal, const int is_upper)