diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2018-06-13 14:05:54 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:53:57 +0000 |
commit | f1c2bf0971dd1c996da149faf3dd669d566074c7 (patch) | |
tree | 802b3ce5198c3209d77fc6b603c209023fe45650 /src/runtime/CL | |
parent | 89a2b571cfc0ea87c26ba8b1ed1ab87d13244f0e (diff) | |
download | ComputeLibrary-f1c2bf0971dd1c996da149faf3dd669d566074c7.tar.gz |
COMPMID-1201 - Implementing Winograd Convolution Layer 1x3 and 3x1 kernels on OpenCL
Change-Id: I39667bab49daa4da009694163274a59fd3574c73
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/137595
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/runtime/CL')
-rw-r--r-- | src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp | 22 |
1 files changed, 19 insertions, 3 deletions
diff --git a/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp b/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp index 49753ad080..11714fac41 100644 --- a/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLWinogradConvolutionLayer.cpp @@ -37,11 +37,27 @@ Size2D winograd_output_tile(const Size2D &input_dims, const Size2D &kernel_dims) { Size2D output_tile = Size2D{}; - if(kernel_dims == Size2D(3U, 3U)) + const unsigned int kernel_max_dim = std::max(kernel_dims.width, kernel_dims.height); + + // Check if the input spatial dimensions are smaller than 4 + const bool is_input_lt4 = (input_dims.width <= 4 && input_dims.height <= 4); + + if(kernel_max_dim == 3U) { - output_tile = (input_dims.width <= 4 && input_dims.height <= 4) ? Size2D(2U, 2U) : Size2D(4U, 4U); + if(kernel_dims == Size2D(3U, 3U)) + { + output_tile = is_input_lt4 ? Size2D(2U, 2U) : Size2D(4U, 4U); + } + else if(kernel_dims == Size2D(3U, 1U)) + { + output_tile = is_input_lt4 ? Size2D(2U, 1U) : Size2D(4U, 1U); + } + else + { + output_tile = is_input_lt4 ? Size2D(1U, 2U) : Size2D(1U, 4U); + } } - else if(kernel_dims == Size2D(5U, 5U)) + else if(kernel_max_dim == 5U) { output_tile = Size2D(4U, 4U); } |