diff options
author | Michalis Spyrou <michalis.spyrou@arm.com> | 2020-10-21 00:04:14 +0100 |
---|---|---|
committer | Michalis Spyrou <michalis.spyrou@arm.com> | 2020-11-03 15:10:47 +0000 |
commit | ebcebf1dee7f8314976b1e0cabd62b4cf893d765 (patch) | |
tree | 95d3e691a0e88a3e213a1d30446a9224497f2055 /arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | |
parent | da4b1b2055d96aaf73704eb9b0b82d74dc2d699c (diff) | |
download | ComputeLibrary-ebcebf1dee7f8314976b1e0cabd62b4cf893d765.tar.gz |
COMPMID-3638: Move NEON kernels
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Change-Id: Ieed3e4bc8be7fef80c90c5094599b477a56fc473
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4285
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h')
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | 46 |
1 files changed, 24 insertions, 22 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h index 01720f05fa..cb1d6bd782 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h @@ -25,15 +25,6 @@ #define ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H #include "NEActivationLayer.h" -#include "arm_compute/core/NEON/INEKernel.h" -#include "arm_compute/core/NEON/kernels/NEConvertQuantizedSignednessKernel.h" -#include "arm_compute/core/NEON/kernels/NEConvertQuantizedSignednessKernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionOutputStageKernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h" #include "arm_compute/runtime/IFunction.h" #include "arm_compute/runtime/IMemoryManager.h" #include "arm_compute/runtime/MemoryGroup.h" @@ -45,6 +36,15 @@ namespace arm_compute { class ITensor; +class NEConvertQuantizedSignednessKernel; +class NEConvertQuantizedSignednessKernel; +class NEGEMMInterleave4x4Kernel; +class NEGEMMLowpMatrixMultiplyKernel; +class NEGEMMLowpOffsetContributionKernel; +class NEGEMMLowpOffsetContributionOutputStageKernel; +class NEGEMMLowpMatrixAReductionKernel; +class NEGEMMLowpMatrixBReductionKernel; +class NEGEMMTranspose1xWKernel; /** Basic function to execute GEMMLowpMatrixMultiplyCore on NEON. This function calls the following NEON kernels if the DOT product instruction is not available: * @@ -72,6 +72,8 @@ public: NEGEMMLowpMatrixMultiplyCore &operator=(const NEGEMMLowpMatrixMultiplyCore &) = delete; /** Default move assignment operator */ NEGEMMLowpMatrixMultiplyCore &operator=(NEGEMMLowpMatrixMultiplyCore &&) = default; + /** Default destructor */ + ~NEGEMMLowpMatrixMultiplyCore(); /** Initialise the kernel's inputs, output * * @note GEMM_LOWP: low precision GEMM kernel @@ -111,19 +113,19 @@ public: void prepare() override; private: - MemoryGroup _memory_group; - IWeightsManager *_weights_manager; - NEGEMMAssemblyDispatch _asm_glue; - NEGEMMLowpMatrixMultiplyKernel _mm_kernel; - NEGEMMInterleave4x4Kernel _mtx_a_reshape_kernel; - NEGEMMTranspose1xWKernel _mtx_b_reshape_kernel; - NEGEMMLowpMatrixAReductionKernel _mtx_a_reduction_kernel; - NEGEMMLowpMatrixBReductionKernel _mtx_b_reduction_kernel; - NEGEMMLowpOffsetContributionKernel _offset_contribution_kernel; - NEGEMMLowpOffsetContributionOutputStageKernel _offset_contribution_output_stage_kernel; - NEActivationLayer _activation_func; - NEConvertQuantizedSignednessKernel _convert_to_signed_asymm; - NEConvertQuantizedSignednessKernel _convert_from_signed_asymm; + MemoryGroup _memory_group; + IWeightsManager *_weights_manager; + NEGEMMAssemblyDispatch _asm_glue; + std::unique_ptr<NEGEMMLowpMatrixMultiplyKernel> _mm_kernel; + std::unique_ptr<NEGEMMInterleave4x4Kernel> _mtx_a_reshape_kernel; + std::unique_ptr<NEGEMMTranspose1xWKernel> _mtx_b_reshape_kernel; + std::unique_ptr<NEGEMMLowpMatrixAReductionKernel> _mtx_a_reduction_kernel; + std::unique_ptr<NEGEMMLowpMatrixBReductionKernel> _mtx_b_reduction_kernel; + std::unique_ptr<NEGEMMLowpOffsetContributionKernel> _offset_contribution_kernel; + std::unique_ptr<NEGEMMLowpOffsetContributionOutputStageKernel> _offset_contribution_output_stage_kernel; + NEActivationLayer _activation_func; + std::unique_ptr<NEConvertQuantizedSignednessKernel> _convert_to_signed_asymm; + std::unique_ptr<NEConvertQuantizedSignednessKernel> _convert_from_signed_asymm; Tensor _vector_sum_col; Tensor _vector_sum_row; |