diff options
Diffstat (limited to 'src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp | 10 |
1 files changed, 9 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp b/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp index 265c5074c5..75e6d5e971 100644 --- a/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp +++ b/src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp @@ -108,7 +108,7 @@ void CLDirectConvolutionLayerKernel::configure(const ICLTensor *input, const ICL options.emplace("-DHAS_BIAS"); } - if((gpu_target == GPUTarget::BIFROST) && (kernel_size <= 5) && (kernel_size != 1) && (_conv_stride_x == 1) && (_conv_stride_y == 1) && (input->info()->data_type() == DataType::F32)) + if((gpu_target == GPUTarget::BIFROST) && (kernel_size <= 5) && (_conv_stride_x == 1) && (_conv_stride_y == 1) && (input->info()->data_type() == DataType::F32)) { options.emplace("-DWEIGHTS_DEPTH=" + support::cpp11::to_string(_weights->info()->dimension(2))); @@ -125,6 +125,14 @@ void CLDirectConvolutionLayerKernel::configure(const ICLTensor *input, const ICL switch(kernel_size) { + case 1: + { + num_elems_read_per_iteration_x = 4; + num_elems_read_per_iteration_y = 4; + num_elems_written_per_iteration_x = 4; + num_elems_written_per_iteration_y = 4; + break; + } case 3: { num_elems_read_per_iteration_x = 6; |