diff options
Diffstat (limited to 'src/core/CL/kernels/CLIm2ColKernel.cpp')
-rw-r--r-- | src/core/CL/kernels/CLIm2ColKernel.cpp | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLIm2ColKernel.cpp b/src/core/CL/kernels/CLIm2ColKernel.cpp index 54ef23f2a2..e3d8df53e5 100644 --- a/src/core/CL/kernels/CLIm2ColKernel.cpp +++ b/src/core/CL/kernels/CLIm2ColKernel.cpp @@ -192,11 +192,15 @@ Im2ColConfiguration configure_opencl_kernel(const ITensorInfo *input, const Size num_elems_processed_per_iteration = 2; is_padding_required_nchw = false; - // Only the 3x3 case is optimized for NHWC + // Only the 3x3 and 9x9 cases are optimized for NHWC if(kernel_dims == Size2D(3U, 3U)) { kernel_name = "im2col3x3_"; } + else if(kernel_dims == Size2D(9U, 9U)) + { + kernel_name = "im2col9x9_"; + } build_opts.add_option("-DVECTOR_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration)); build_opts.add_option("-DLAST_ACCESSED=" + support::cpp11::to_string(std::max(static_cast<int>(input_channel - num_elems_processed_per_iteration), 0))); |