diff options
| author | Angus, Alexander <aangus@qti.qualcomm.com> | 2023-01-03 10:56:04 -0800 |
|---|---|---|
| committer | Angus, Alexander <aangus@qti.qualcomm.com> | 2023-01-03 10:56:04 -0800 |
| commit | 4f394608a28f419dfd6091c704148d9e638a26f0 (patch) | |
| tree | 4c0e042109c4d249ff5b700fc49a862169edec5a /src/kernels/levelx/col2im.opencl | |
| parent | 03cffa83c5f7742f8ec0c5e762bb7048e38952f3 (diff) | |
implemented changes to boost Adreno performance according to https://jira-dc.qualcomm.com/jira/browse/OSR-8731
Diffstat (limited to 'src/kernels/levelx/col2im.opencl')
| -rw-r--r-- | src/kernels/levelx/col2im.opencl | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/kernels/levelx/col2im.opencl b/src/kernels/levelx/col2im.opencl index 425ffbbc..fcc307c6 100644 --- a/src/kernels/levelx/col2im.opencl +++ b/src/kernels/levelx/col2im.opencl @@ -92,7 +92,11 @@ INLINE_FUNC void Xcol2im(const int input_h, const int input_w, const int channel // ================================================================================================= // Kernel flip version of the Xcol2im kernel (for convolution) -__kernel __attribute__((reqd_work_group_size(COPY_DIMX, COPY_DIMY, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#elif + __kernel __attribute__((reqd_work_group_size(COPY_DIMX, COPY_DIMY, 1))) +#endif void Xcol2imKernelFlip(const int input_h, const int input_w, const int channels, const int output_h, const int output_w, const int kernel_h, const int kernel_w, @@ -113,7 +117,11 @@ void Xcol2imKernelFlip(const int input_h, const int input_w, const int channels, } // Normal version of the Xcol2im kernel (for cross-correlation) -__kernel __attribute__((reqd_work_group_size(COPY_DIMX, COPY_DIMY, 1))) +#if RELAX_WORKGROUP_SIZE == 1 + __kernel +#elif + __kernel __attribute__((reqd_work_group_size(COPY_DIMX, COPY_DIMY, 1))) +#endif void Xcol2imKernelNormal(const int input_h, const int input_w, const int channels, const int output_h, const int output_w, const int kernel_h, const int kernel_w, |