summaryrefslogtreecommitdiff
path: root/src/kernels/level3/transpose_fast.opencl
diff options
context:
space:
mode:
Diffstat (limited to 'src/kernels/level3/transpose_fast.opencl')
-rw-r--r--src/kernels/level3/transpose_fast.opencl6
1 files changed, 5 insertions, 1 deletions
diff --git a/src/kernels/level3/transpose_fast.opencl b/src/kernels/level3/transpose_fast.opencl
index 1b9fca45..e89984cc 100644
--- a/src/kernels/level3/transpose_fast.opencl
+++ b/src/kernels/level3/transpose_fast.opencl
@@ -36,7 +36,11 @@ R"(
// Transposes and copies a matrix. Requires both matrices to be of the same dimensions and without
// offset. A more general version is available in 'padtranspose.opencl'.
-__kernel __attribute__((reqd_work_group_size(TRA_DIM, TRA_DIM, 1)))
+#if RELAX_WORKGROUP_SIZE == 1
+ __kernel
+#elif
+ __kernel __attribute__((reqd_work_group_size(TRA_DIM, TRA_DIM, 1)))
+#endif
void TransposeMatrixFast(const int ld,
__global const realT* restrict src,
__global realT* dest,