From 9c700378f2227cb9d51455ed4a5086daaac5532a Mon Sep 17 00:00:00 2001 From: Michele Di Giorgio Date: Wed, 8 Jan 2020 11:33:44 +0000 Subject: COMPMID-2769: Add support for QASYMM8_SIGNED in NEFullyConnectedLayer Change-Id: I4c35c522375ae5a5de78716e079ebb9ffad15956 Signed-off-by: Michele Di Giorgio Reviewed-on: https://review.mlplatform.org/c/2581 Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins --- .../runtime/NEON/functions/NEFullyConnectedLayer.h | 12 ++++---- .../runtime/NEON/functions/NEGEMMLowpOutputStage.h | 36 +++++++++++++++++++++- 2 files changed, 41 insertions(+), 7 deletions(-) (limited to 'arm_compute') diff --git a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h index 784637a796..78f12daf9c 100644 --- a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h +++ b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2019 ARM Limited. + * Copyright (c) 2017-2020 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -126,12 +126,12 @@ public: NEFullyConnectedLayer &operator=(NEFullyConnectedLayer &&) = default; /** Set the input and output tensors. * - * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32. + * @param[in] input Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[in] weights Weights tensor. The weights must be 2 dimensional. * If this function is called after a Convolution Layer, the (transposed) weights will have as many rows as the product of the first 3 input's dimensions. * If it is called after another FullyConnected Layer, the (transposed) weights will have as many rows as the input's first dimension. * Data type supported: Same as @p input. - * @param[in] biases Bias tensor. Can be nullptr. Data type supported: Same as @p weights, S32 if @p weights is QASYMM8. + * @param[in] biases Bias tensor. Can be nullptr. Data type supported: Same as @p weights, S32 if @p weights is QASYMM8/QASYMM8_SIGNED. * @param[out] output Destination tensor. Its shape should be equal to the output of a matrix multiplication between: * - The output of im2col on the input and the (transposed) 2D weights, if the function is called after a Convolution Layer * - The input tensor and the (transposed) 2D weights, if the function is called after another FullyConnected Layer. @@ -142,12 +142,12 @@ public: FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEFullyConnectedLayer * - * @param[in] input Source tensor info. Data type supported: QASYMM8/F16/F32. + * @param[in] input Source tensor info. Data type supported: QASYMM8/QASYMM8_SIGNED/F16/F32. * @param[in] weights Weights tensor info. The weights must be 2 dimensional. * If this function is called after a Convolution Layer, the (transposed) weights will have as many rows as the product of the first 3 input's dimensions. * If it is called after another FullyConnected Layer, the (transposed) weights will have as many rows as the input's first dimension. * Data type supported: Same as @p input. - * @param[in] biases Bias tensor. Can be nullptr. Data type supported: Same as @p weights, S32 if @p weights is QASYMM8. + * @param[in] biases Bias tensor. Can be nullptr. Data type supported: Same as @p weights, S32 if @p weights is QASYMM8/QASYMM8_SIGNED. * @param[in] output Destination tensor info. Its shape should be equal to the output of a matrix multiplication between: * - The output of im2col on the input and the (transposed) 2D weights, if the function is called after a Convolution Layer * - The input tensor and the (transposed) 2D weights, if the function is called after another FullyConnected Layer. @@ -177,7 +177,7 @@ private: weights_transformations::NEFullyConnectedLayerReshapeWeightsManaged _reshape_weights_managed_function; NEGEMM _mm_gemm; NEGEMMLowpMatrixMultiplyCore _mm_gemmlowp; - NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint _gemmlowp_output_stage; + NEGEMMLowpOutputStage _gemmlowp_output_stage; NEGEMMMatrixAccumulateBiasesKernel _accumulate_biases_kernel; Tensor _flatten_output; Tensor _gemmlowp_output; diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h index b483d03c85..ca2cbbc268 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2019 ARM Limited. + * Copyright (c) 2017-2020 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -264,5 +264,39 @@ public: */ static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0); }; + +/** Basic function to execute GEMMLowpQuantizeDown kernels on NEON. + * + * This function calls the following NEON kernels: + * + * -# @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel + * -# @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel + * -# @ref NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel + * -# @ref NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel +*/ +class NEGEMMLowpOutputStage : public INESimpleFunctionNoBorder +{ +public: + /** Initialise the kernel's inputs, output + * + * @param[in] input Input tensor. Data type supported: S32 + * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. + * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. + * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED/QSYMM16 + * @param[in] info GEMMLowp output stage metadata. + */ + void configure(const ITensor *input, const ITensor *bias, ITensor *output, const GEMMLowpOutputStageInfo &info); + /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpOutputStage + * + * @param[in] input Input tensor info. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32 + * @param[in] bias Biases tensor info. Only shared biases supported and it can be a nullptr if the addition of biases is not required. + * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. + * @param[in] output Output tensor info. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED/QSYMM16 + * @param[in] info GEMMLowp output stage metadata. + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo &info); +}; } // namespace arm_compute #endif /*ARM_COMPUTE_NEGEMMLOWPOUTPUTSTAGE_H */ -- cgit v1.2.1