diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2016-09-13 21:14:51 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2016-09-13 21:14:51 +0200 |
commit | f07ac22f5b57d22756d779d2e53620f988d786ee (patch) | |
tree | e8bcbc331683ca6fd807f5a5b83bb05c6e6fed69 /src/kernels/level3/copy_pad.opencl | |
parent | 7c13bacf129291e3e295ecb6e833788477085fa0 (diff) | |
parent | 4b94afda941a86f363064ff02f97e21eb9618794 (diff) |
Merge pull request #99 from CNugteren/development
Update to version 0.9.0
Diffstat (limited to 'src/kernels/level3/copy_pad.opencl')
-rw-r--r-- | src/kernels/level3/copy_pad.opencl | 42 |
1 files changed, 21 insertions, 21 deletions
diff --git a/src/kernels/level3/copy_pad.opencl b/src/kernels/level3/copy_pad.opencl index d276cc60..29480b25 100644 --- a/src/kernels/level3/copy_pad.opencl +++ b/src/kernels/level3/copy_pad.opencl @@ -24,16 +24,16 @@ R"( // Copies a matrix from source to destination. The output is padded with zero values in case the // destination matrix dimensions are larger than the source matrix dimensions. Additionally, the ld // value and offset can be different. -__attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) -__kernel void CopyPadMatrix(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, - const int dest_ld, const int dest_offset, - __global real* dest, - const __constant real* restrict arg_alpha, - const int do_conjugate) { - const real alpha = arg_alpha[0]; +__kernel __attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) +void CopyPadMatrix(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, + const int dest_ld, const int dest_offset, + __global real* dest, + const real_arg arg_alpha, + const int do_conjugate) { + const real alpha = GetRealArg(arg_alpha); // Loops over the work per thread in both dimensions #pragma unroll @@ -65,17 +65,17 @@ __kernel void CopyPadMatrix(const int src_one, const int src_two, // Same as above, but now un-pads a matrix. This kernel reads data from a padded source matrix, but // writes only the actual data back to the destination matrix. Again, the ld value and offset can // be different. -__attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) -__kernel void CopyMatrix(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, - const int dest_ld, const int dest_offset, - __global real* dest, - const __constant real* restrict arg_alpha, - const int upper, const int lower, - const int diagonal_imag_zero) { - const real alpha = arg_alpha[0]; +__kernel __attribute__((reqd_work_group_size(PAD_DIMX, PAD_DIMY, 1))) +void CopyMatrix(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, + const int dest_ld, const int dest_offset, + __global real* dest, + const real_arg arg_alpha, + const int upper, const int lower, + const int diagonal_imag_zero) { + const real alpha = GetRealArg(arg_alpha); // Loops over the work per thread in both dimensions #pragma unroll |