summaryrefslogtreecommitdiff
path: root/src/kernels/level2/xtrsv.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/level2/xtrsv.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/level2/xtrsv.opencl')
-rw-r--r--src/kernels/level2/xtrsv.opencl12
1 files changed, 10 insertions, 2 deletions
diff --git a/src/kernels/level2/xtrsv.opencl b/src/kernels/level2/xtrsv.opencl
index e7b6ae79..e3b5418c 100644
--- a/src/kernels/level2/xtrsv.opencl
+++ b/src/kernels/level2/xtrsv.opencl
@@ -39,7 +39,11 @@ void FillVector(const int n, const int inc, const int offset,
// =================================================================================================
-__kernel __attribute__((reqd_work_group_size(TRSV_BLOCK_SIZE, 1, 1)))
+#if RELAX_WORKGROUP_SIZE == 1
+ __kernel
+#elif
+ __kernel __attribute__((reqd_work_group_size(TRSV_BLOCK_SIZE, 1, 1)))
+#endif
void trsv_forward(int n,
const __global real *A, const int a_offset, int a_ld,
__global real *b, const int b_offset, int b_inc,
@@ -87,7 +91,11 @@ void trsv_forward(int n,
}
}
-__kernel __attribute__((reqd_work_group_size(TRSV_BLOCK_SIZE, 1, 1)))
+#if RELAX_WORKGROUP_SIZE == 1
+ __kernel
+#elif
+ __kernel __attribute__((reqd_work_group_size(TRSV_BLOCK_SIZE, 1, 1)))
+#endif
void trsv_backward(int n,
const __global real *A, const int a_offset, int a_ld,
__global real *b, const int b_offset, int b_inc,