diff options
| author | Freddie Liardet <frederick.liardet@arm.com> | 2021-08-03 15:57:32 +0100 |
|---|---|---|
| committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-09-07 11:13:44 +0000 |
| commit | d216f570750b8ccde3754c4aef53fc20a90cb32d (patch) | |
| tree | 83a88d3d4391c6a8ca5dabc73c763e6f0878c595 /src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp | |
| parent | 4e53c5ab47a713ab0ce53d076e2e4cf274fec312 (diff) | |
| download | ComputeLibrary-d216f570750b8ccde3754c4aef53fc20a90cb32d.tar.gz | |
Update cpu depthwise kernels
Resolves: COMPMID-4688
Signed-off-by: Freddie Liardet <frederick.liardet@arm.com>
Change-Id: I9e22f967f5b7ccaebff2fc49f0253f621d62d820
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6030
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp | 40 |
1 files changed, 1 insertions, 39 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp index 04b904275c..d59d6b7e35 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp @@ -22,13 +22,7 @@ * SOFTWARE. */ -#include "arm_gemm.hpp" -#include "src/core/NEON/kernels/arm_gemm/utils.hpp" -#include "src/core/NEON/kernels/assembly/depthwise.hpp" -#include <cstdint> -#include <cstring> - -using namespace arm_gemm; +#include "8b_mla.hpp" size_t generic_get_packed_size( const VLType vec_type, @@ -84,41 +78,9 @@ void generic_pack( } } - -#define ADD_IMPLEMENTATION(ARCH, TYPENAME, TYPE, VEC_TYPE, ACC_DEPTH, KERN_ROWS, KERN_COLS) \ -struct interleave_ ## ARCH ## _ ## TYPENAME ## _ ## KERN_ROWS ## x ## KERN_COLS ## _mla \ -{ \ - static size_t get_packed_size(const DepthwiseArgs &args); \ - static void pack_parameters( \ - unsigned int n_channels, void *outptr, \ - const TYPE *weights, size_t ld_weight_col, size_t ld_weight_row \ - ); \ -}; \ -\ -size_t interleave_ ## ARCH ## _ ## TYPENAME ## _ ## KERN_ROWS ## x ## KERN_COLS ## _mla::get_packed_size(const DepthwiseArgs &args) \ -{ \ - return generic_get_packed_size(VLType::VEC_TYPE, ACC_DEPTH, KERN_ROWS, KERN_COLS, args.input_channels); \ -} \ -\ -void interleave_ ## ARCH ## _ ## TYPENAME ## _ ## KERN_ROWS ## x ## KERN_COLS ## _mla::pack_parameters(unsigned int n_channels, void *outptr, \ - const TYPE *weights, size_t ld_weight_col, size_t ld_weight_row) \ -{ \ - generic_pack(VLType::VEC_TYPE, ACC_DEPTH, KERN_ROWS, KERN_COLS, n_channels, outptr, weights, ld_weight_col, ld_weight_row); \ -} - - namespace arm_conv { namespace depthwise { -#if defined(ARM_COMPUTE_ENABLE_SVE) - -ADD_IMPLEMENTATION(sve, s8q, int8_t, SVE, 2, 3, 3) -ADD_IMPLEMENTATION(sve, s8q, int8_t, SVE, 2, 5, 5) -ADD_IMPLEMENTATION(sve, u8q, uint8_t, SVE, 2, 3, 3) -ADD_IMPLEMENTATION(sve, u8q, uint8_t, SVE, 2, 5, 5) - -#endif // defined(ARM_COMPUTE_ENABLE_SVE) - ADD_IMPLEMENTATION(a64, s8q, int8_t, None, 2, 3, 3) ADD_IMPLEMENTATION(a64, s8q, int8_t, None, 2, 5, 5) ADD_IMPLEMENTATION(a64, u8q, uint8_t, None, 2, 3, 3) |