From dbdea0d1c025b18d4d82c278c87454427918f5b4 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Wed, 16 Oct 2019 19:21:40 +0100 Subject: COMPMID-2308: NEConvolutionLayer: support QUANT8_SYMM_PER_CHANNEL filters Change-Id: Ic1bf5f0d21ccd525f84213a360f7e199d7f50577 Signed-off-by: Georgios Pinitas Reviewed-on: https://review.mlplatform.org/c/2177 Tested-by: Arm Jenkins Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- .../NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | 51 +++++++++++++--------- 1 file changed, 30 insertions(+), 21 deletions(-) (limited to 'arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h') diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h index 12c120934e..aa2c23c97c 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h @@ -26,6 +26,8 @@ #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/NEGEMMLowpOffsetContributionKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionOutputStageKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h" @@ -76,24 +78,24 @@ public: * -# Convert b values from QASYMM8 to int32 add b_offset to each of them. * -# Compute the matrix product of the resulting a * b in int32. * - * @note The @p output type is S32 if @p gemm_info.type == GEMMLowpOutputStageType::NONE. It is QASYMM8 otherwise + * @note The @p output type is S32 if @p gemm_info.type == GEMMLowpOutputStageType::NONE. It is QASYMM8/QASYMM8_SIGNED otherwise * - * @param[in] a First input tensor (Matrix A). Data type supported: QASYMM8. + * @param[in] a First input tensor (Matrix A). Data type supported: QASYMM8/QASYMM8_SIGNED. * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a * @param[in] c Third input tensor (Matrix C). It can be a nullptr. Data type supported: S32 - * @param[out] output Output tensor. Data type supported: Data type supported: S32/QASYMM8 + * @param[out] output Output tensor. Data type supported: Data type supported: S32/QASYMM8/QASYMM8_SIGNED * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and * if the reshape of matrix B should be executed only for the first run */ void configure(const ITensor *a, const ITensor *b, const ITensor *c, ITensor *output, const GEMMInfo &gemm_info = GEMMInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixMultiplyCore * - * @note The @p output type is S32 if @p gemm_info.type == GEMMLowpOutputStageType::NONE. It is QASYMM8 otherwise + * @note The @p output type is S32 if @p gemm_info.type == GEMMLowpOutputStageType::NONE. It is QASYMM8/QASYMM8_SIGNED otherwise * - * @param[in] a First input tensor info (Matrix A). Data type supported: QASYMM8. + * @param[in] a First input tensor info (Matrix A). Data type supported: QASYMM8/QASYMM8_SIGNED. * @param[in] b Second input tensor info (Matrix B). Data type supported: same as @p a * @param[in] c Third input tensor info (Matrix C). It can be a nullptr. Data type supported: S32 - * @param[in] output Output tensor info. Data type supported: Data type supported: S32/QASYMM8 + * @param[in] output Output tensor info. Data type supported: Data type supported: S32/QASYMM8/QASYMM8_SIGNED * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and * if the reshape of matrix B should be executed only for the first run * @@ -116,21 +118,28 @@ private: NEGEMMLowpOffsetContributionKernel _offset_contribution_kernel; NEGEMMLowpOffsetContributionOutputStageKernel _offset_contribution_output_stage_kernel; NEActivationLayer _activation_func; - Tensor _vector_sum_col; - Tensor _vector_sum_row; - Tensor _tmp_a; - Tensor _tmp_b; - Tensor _mm_result_s32; - const ITensor *_original_b; - int32_t _a_offset; - int32_t _b_offset; - bool _run_vector_matrix_multiplication; - bool _assembly_path; - bool _fused_assembly_path; - bool _reshape_b_only_on_first_run; - bool _is_prepared; - bool _fuse_output_stage; - bool _run_activation; + NEConvertQuantizedSignednessKernel _convert_to_signed_asymm; + NEConvertQuantizedSignednessKernel _convert_from_signed_asymm; + + Tensor _vector_sum_col; + Tensor _vector_sum_row; + Tensor _tmp_a; + Tensor _tmp_b; + Tensor _mm_result_s32; + Tensor _signed_a; + Tensor _signed_output; + const ITensor *_original_b; + int32_t _a_offset; + int32_t _b_offset; + + bool _run_vector_matrix_multiplication; + bool _assembly_path; + bool _fused_assembly_path; + bool _reshape_b_only_on_first_run; + bool _is_prepared; + bool _fuse_output_stage; + bool _run_activation; + bool _flip_signedness; }; } // namespace arm_compute #endif /*__ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H__ */ -- cgit v1.2.1