diff options
author | D. Van Assche <dimitri.vanassche@macq.eu> | 2016-08-18 17:33:13 +0200 |
---|---|---|
committer | D. Van Assche <dimitri.vanassche@macq.eu> | 2016-08-18 17:33:13 +0200 |
commit | 57f1aa76857cf0566e05b43b9b2a98a3a6139c8b (patch) | |
tree | 4ed277095302ea0870b764dde32936dad56bb410 /src/kernels/level3/transpose_pad.opencl | |
parent | 7c13bacf129291e3e295ecb6e833788477085fa0 (diff) |
Adapt opencl files for 1.1 OpenCL
In OpenCL 1.1 __kernel has to be before __attribute__, at least with
Vivante compiler.
Diffstat (limited to 'src/kernels/level3/transpose_pad.opencl')
-rw-r--r-- | src/kernels/level3/transpose_pad.opencl | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/kernels/level3/transpose_pad.opencl b/src/kernels/level3/transpose_pad.opencl index 2de0c7bd..63cc6e9a 100644 --- a/src/kernels/level3/transpose_pad.opencl +++ b/src/kernels/level3/transpose_pad.opencl @@ -24,8 +24,8 @@ R"( // Transposes a matrix from source to destination. The output is padded with zero values in case the // destination matrix dimensions are larger than the transposed source matrix dimensions. -__attribute__((reqd_work_group_size(PADTRA_TILE, PADTRA_TILE, 1))) -__kernel void TransposePadMatrix(const int src_one, const int src_two, +__kernel __attribute__((reqd_work_group_size(PADTRA_TILE, PADTRA_TILE, 1))) +void TransposePadMatrix(const int src_one, const int src_two, const int src_ld, const int src_offset, __global const real* restrict src, const int dest_one, const int dest_two, @@ -88,8 +88,8 @@ __kernel void TransposePadMatrix(const int src_one, const int src_two, // Transposes a matrix, while considering possible padding in the source matrix. Data is read from a // padded source matrix, but only the actual data is written back to the transposed destination // matrix. This kernel optionally checks for upper/lower triangular matrices. -__attribute__((reqd_work_group_size(PADTRA_TILE, PADTRA_TILE, 1))) -__kernel void TransposeMatrix(const int src_one, const int src_two, +__kernel __attribute__((reqd_work_group_size(PADTRA_TILE, PADTRA_TILE, 1))) +void TransposeMatrix(const int src_one, const int src_two, const int src_ld, const int src_offset, __global const real* restrict src, const int dest_one, const int dest_two, |