diff options
Diffstat (limited to 'src/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.cpp')
-rw-r--r-- | src/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.cpp | 9 |
1 files changed, 8 insertions, 1 deletions
diff --git a/src/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.cpp index 5c7320aa8d..962f044214 100644 --- a/src/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.cpp +++ b/src/core/GLES_COMPUTE/kernels/GCDirectConvolutionLayerKernel.cpp @@ -191,6 +191,10 @@ void GCDirectConvolutionLayerKernel<kernel_size>::configure(const IGCTensor *inp case DataType::F16: num_elems_read_per_iteration_x = 8; num_elems_written_per_iteration_x = 8; + if(weights->info()->dimension(2) % 2 == 0) + { + options.emplace("#define WEIGHTS_OPTIMIZATION"); + } break; case DataType::F32: @@ -255,7 +259,10 @@ void GCDirectConvolutionLayerKernel<kernel_size>::configure(const IGCTensor *inp switch(weights->info()->data_type()) { case DataType::F16: - weights_access = AccessWindowStatic(weights->info(), 0, 0, kernel_size + 1, kernel_size); + if((weights->info()->dimension(2) % 2 != 0) || (kernel_size != 1)) + { + weights_access = AccessWindowStatic(weights->info(), 0, 0, kernel_size + 1, kernel_size); + } if(_bias != nullptr) { bias_access = AccessWindowStatic(_bias->info(), 0, 0, _bias->info()->dimension(0) + 1, 1); |