aboutsummaryrefslogtreecommitdiff
path: root/src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp')
-rw-r--r--src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp12
1 files changed, 10 insertions, 2 deletions
diff --git a/src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp
index acb998840b..621c9693fe 100644
--- a/src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCTransposeKernel.cpp
@@ -75,8 +75,16 @@ void GCTransposeKernel::configure(const IGCTensor *input, IGCTensor *output)
build_opts.emplace(("#define TRANSPOSE_4X4"));
num_elems_processed_per_iteration = 4;
#elif defined(TRANSPOSE_8X8) /* TRANSPOSE_4X4 */
- build_opts.emplace(("#define TRANSPOSE_8X8"));
- num_elems_processed_per_iteration = 8;
+ if(w_out != h_out)
+ {
+ build_opts.emplace("#define TRANSPOSE_8X8");
+ num_elems_processed_per_iteration = 8;
+ }
+ else
+ {
+ build_opts.emplace("#define TRANSPOSE_8X8_SQUARE");
+ num_elems_processed_per_iteration = 8;
+ }
#endif /* TRANSPOSE_4X4 */
}