From 79fa9a22022824735986f74557bf38095eb2284d Mon Sep 17 00:00:00 2001 From: Manuel Bottini Date: Fri, 22 Feb 2019 17:54:22 +0000 Subject: COMPMID-2009: Add support for QASYMM8 in NEPixelWiseMultiplicationKernel Change-Id: I58536e945d069c96a065b82cc14960f54afc6e1a Signed-off-by: Manuel Bottini Reviewed-on: https://review.mlplatform.org/c/781 Tested-by: Arm Jenkins Reviewed-by: Pablo Marquez --- .../NEON/kernels/NEPixelWiseMultiplicationKernel.h | 38 +++++++++++++++------- .../NEON/functions/NEPixelWiseMultiplication.h | 31 +++++++++++------- 2 files changed, 46 insertions(+), 23 deletions(-) (limited to 'arm_compute') diff --git a/arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h b/arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h index 41ea91495f..2a8e36b1de 100644 --- a/arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h +++ b/arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2018 ARM Limited. + * Copyright (c) 2016-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -56,12 +56,12 @@ public: * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. * - * @param[in] input1 An input tensor. Data types supported: U8/S16/F16/F32 - * @param[in] input2 An input tensor. Data types supported: U8, S16/F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). - * @param[out] output The output tensor. Data types supported: U8 (Only if both inputs are U8), S16/F16 (only if @p input1 is F16), F32 (only if both inputs are F32). + * @param[in] input1 An input tensor. Data types supported: U8/QASYMM8/S16/F16/F32 + * @param[in] input2 An input tensor. Data types supported: U8, QASYMM8 (only if @p input1 is QASYMM8), S16, F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). + * @param[out] output Output tensor. Data types supported: U8 (Only if both inputs are U8), QASYMM8 (only if both inputs are QASYMM8), S16/F16 (only if @p input1 is F16), F32 (only if both inputs are F32). * @param[in] scale Scale to apply after multiplication. * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. - * @param[in] overflow_policy Overflow policy. + * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if datatype is QASYMM8. * @param[in] rounding_policy Rounding policy. */ void configure(const ITensor *input1, const ITensor *input2, ITensor *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy); @@ -70,12 +70,12 @@ public: * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. * - * @param[in] input1 An input tensor. Data types supported: U8/S16/F16/F32 - * @param[in] input2 An input tensor. Data types supported: U8, S16/F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). - * @param[in] output The output tensor. Data types supported: U8 (Only if both inputs are U8), S16/F16 (only if @p input1 is F16), F32 (only if both inputs are F32). + * @param[in] input1 An input tensor info. Data types supported: U8/QASYMM8/S16/F16/F32 + * @param[in] input2 An input tensor info. Data types supported: U8, QASYMM8 (only if @p input1 is QASYMM8), S16, F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). + * @param[in] output Output tensor info. Data types supported: U8 (Only if both inputs are U8), QASYMM8 (only if both inputs are QASYMM8), S16/F16 (only if @p input1 is F16), F32 (only if both inputs are F32). * @param[in] scale Scale to apply after multiplication. * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. - * @param[in] overflow_policy Overflow policy. + * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if datatype is QASYMM8. * @param[in] rounding_policy Rounding policy. * * @return a status @@ -92,6 +92,7 @@ private: * @param[in] input1_ptr Pointer to the first input tensor. * @param[in] input2_ptr Pointer to the second input tensor. * @param[out] output_ptr Pointer to the output tensor. + * @param[in] scale Integer scale factor. */ using MulFunctionInt = void(const void *__restrict input1_ptr, const void *__restrict input2_ptr, void *__restrict output_ptr, int scale); /** Common signature for all the specialised multiplication functions with float scaling factor @@ -99,11 +100,26 @@ private: * @param[in] input1_ptr Pointer to the first input tensor. * @param[in] input2_ptr Pointer to the second input tensor. * @param[out] output_ptr Pointer to the output tensor. + * @param[in] scale Float scale factor. */ using MulFunctionFloat = void(const void *__restrict input1_ptr, const void *__restrict input2_ptr, void *__restrict output_ptr, float scale); + /** Common signature for all the specialised QASYMM8 multiplication functions with float scaling factor + * + * @param[in] input1_ptr Pointer to the first input tensor. + * @param[in] input2_ptr Pointer to the second input tensor. + * @param[out] output_ptr Pointer to the output tensor. + * @param[in] scale Float scale factor. + * @param[in] input1_qua_info Quantization Info of tensor input1. + * @param[in] input2_qua_info Quantization Info of tensor input2. + * @param[in] output_qua_info Quantization Info of tensor output. + * + */ + using MulFunctionQASYMM8 = void(const void *__restrict input1_ptr, const void *__restrict input2_ptr, void *__restrict output_ptr, float scale, + const QuantizationInfo &input1_qua_info, const QuantizationInfo &input2_qua_info, const QuantizationInfo &output_qua_info); - MulFunctionFloat *_func_float; - MulFunctionInt *_func_int; + MulFunctionFloat *_func_float; + MulFunctionInt *_func_int; + MulFunctionQASYMM8 *_func_qasymm8; private: const ITensor *_input1; diff --git a/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h b/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h index 371bb2e13e..869dd4e1d5 100644 --- a/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h +++ b/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2018 ARM Limited. + * Copyright (c) 2016-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -37,24 +37,31 @@ class NEPixelWiseMultiplication : public INESimpleFunction public: /** Initialise the kernel's inputs, output and convertion policy. * - * @param[in, out] input1 An input tensor. Data types supported: U8/S16/F16/F32. - * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. - * @param[in, out] input2 An input tensor. Data types supported: same as @p input1. - * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. - * @param[out] output Output tensor. Data types supported: U8/S16/F16/F32. + * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. + * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. + * + * @param[in, out] input1 An input tensor. Data types supported: U8/QASYMM8/S16/F16/F32 + * This input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. + * @param[in, out] input2 An input tensor. Data types supported: U8, QASYMM8 (only if @p input1 is QASYMM8), S16, F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). + * This input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. + * @param[out] output Output tensor. Data types supported: U8 (Only if both inputs are U8), QASYMM8 (only if both inputs are QASYMM8), S16/F16 (only if @p input1 is F16), F32 (only if both inputs are F32). * @param[in] scale Scale to apply after multiplication. * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. - * @param[in] overflow_policy Overflow policy. + * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if datatype is QASYMM8. * @param[in] rounding_policy Rounding policy. */ void configure(ITensor *input1, ITensor *input2, ITensor *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy); /** Static function to check if given info will lead to a valid configuration of @ref NEPixelWiseMultiplication * - * @param[in] input1 First tensor info input. Data types supported: U8/S16/F16/F32. - * @param[in] input2 Second tensor info input. Data types supported: U8/S16/F16/F32. - * @param[in] output Output tensor info. Data types supported: U8/S16/F16/F32. - * @param[in] scale Scale to apply after multiplication. Must be positive. - * @param[in] overflow_policy Overflow policy. + * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. + * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. + * + * @param[in] input1 An input tensor info. Data types supported: U8/QASYMM8/S16/F16/F32 + * @param[in] input2 An input tensor info. Data types supported: U8, QASYMM8 (only if @p input1 is QASYMM8), S16, F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). + * @param[in] output Output tensor info. Data types supported: U8 (Only if both inputs are U8), QASYMM8 (only if both inputs are QASYMM8), S16/F16 (only if @p input1 is F16), F32 (only if both inputs are F32). + * @param[in] scale Scale to apply after multiplication. + * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. + * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if datatype is QASYMM8. * @param[in] rounding_policy Rounding policy. * * @return a status -- cgit v1.2.1