diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-06-24 14:56:34 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-07-09 09:31:37 +0000 |
commit | 30271c779c36a2abe6995c4454674d92bbc1f91f (patch) | |
tree | 531257ff87cf2cb8d6f3b8da0abe3e6cb77a2a0e /src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp | |
parent | 30dbeef2f46bdd6fe05d25dfa27cb4b2359dced3 (diff) | |
download | ComputeLibrary-30271c779c36a2abe6995c4454674d92bbc1f91f.tar.gz |
COMPMID-2156: Optimized dilated convolution for NEON.
Change-Id: I3a8abe8cc9637c8983d9bd69dcbaee1a15eac8d0
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1492
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp')
-rw-r--r-- | src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp | 29 |
1 files changed, 27 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp b/src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp index cbdb19a067..87d2bfd8e6 100644 --- a/src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp +++ b/src/core/NEON/kernels/convolution/depthwise/impl_fp16_fp16.hpp @@ -30,8 +30,8 @@ * !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! */ #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC -#include "arm_compute/core/NEON/kernels/convolution/common/arm.hpp" -#include "arm_compute/core/NEON/kernels/convolution/depthwise/impl_base.hpp" +#include "arm.hpp" +#include "impl_base.hpp" #pragma once @@ -68,6 +68,31 @@ template < unsigned int KernelRows, unsigned int KernelCols, unsigned int StrideRows, unsigned int StrideCols > +DepthwiseConvolution< + OutputTileRows, OutputTileCols, + KernelRows, KernelCols, StrideRows, StrideCols, + float16_t, float16_t, float16_t +>::DepthwiseConvolution( + int n_batches, int n_input_rows, int n_input_cols, int n_channels, + int n_output_rows, int n_output_cols, + ActivationFunction activation, + unsigned int padding_top, + unsigned int padding_left, + unsigned int padding_bottom, + unsigned int padding_right +) : Base( + n_batches, n_input_rows, n_input_cols, n_channels, + n_output_rows, n_output_cols, activation, + padding_top, padding_left, padding_bottom, padding_right + ) +{ +} + +template < + unsigned int OutputTileRows, unsigned int OutputTileCols, + unsigned int KernelRows, unsigned int KernelCols, + unsigned int StrideRows, unsigned int StrideCols +> template <ActivationFunction Activation> void DepthwiseConvolution< OutputTileRows, OutputTileCols, |