From 8b2a7d3aa119e7f1d6a03690d05eb27c5d178b9f Mon Sep 17 00:00:00 2001 From: Giorgio Arena Date: Tue, 11 Feb 2020 17:21:31 +0000 Subject: COMPMID-3101 Fuse activation with floating point elementwise operation layers in CL Signed-off-by: Giorgio Arena Change-Id: I1693f8664ba7c0dc8c076bbe7365cef1e667bd25 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2718 Tested-by: Arm Jenkins Reviewed-by: Gian Marco Iodice Comments-Addressed: Arm Jenkins --- .../core/CL/kernels/CLElementwiseOperationKernel.h | 53 ++++++++++++---------- .../CL/kernels/CLPixelWiseMultiplicationKernel.h | 24 ++++++---- 2 files changed, 44 insertions(+), 33 deletions(-) (limited to 'arm_compute/core') diff --git a/arm_compute/core/CL/kernels/CLElementwiseOperationKernel.h b/arm_compute/core/CL/kernels/CLElementwiseOperationKernel.h index 34fafaa3a5..85961f28bc 100644 --- a/arm_compute/core/CL/kernels/CLElementwiseOperationKernel.h +++ b/arm_compute/core/CL/kernels/CLElementwiseOperationKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2019 ARM Limited. + * Copyright (c) 2018-2020 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -97,6 +97,8 @@ protected: */ void configure_common(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output); + ActivationLayerInfo _act_info; + private: const ICLTensor *_input1; /**< Source tensor 1 */ const ICLTensor *_input2; /**< Source tensor 2 */ @@ -114,25 +116,28 @@ public: /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel * - * @param[in] op Arithmetic operation to be executed. - * @param[in] input1 First tensor input. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. - * @param[in] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[in] output Output tensor. Data types supported: Same as @p input1. - * @param[in] policy Policy to use to handle overflow. + * @param[in] op Arithmetic operation to be executed. + * @param[in] input1 First tensor input. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. + * @param[in] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[in] output Output tensor. Data types supported: Same as @p input1. + * @param[in] policy Policy to use to handle overflow. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ - void configure(ArithmeticOperation op, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const ConvertPolicy &policy); + void configure(ArithmeticOperation op, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const ConvertPolicy &policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel * - * @param[in] op Arithmetic operation to be executed. - * @param[in] input1 First tensor input info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. - * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1. - * @param[in] output Output tensor info. Data types supported: Same as @p input1. - * @param[in] policy Policy to use to handle overflow. + * @param[in] op Arithmetic operation to be executed. + * @param[in] input1 First tensor input info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. + * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1. + * @param[in] output Output tensor info. Data types supported: Same as @p input1. + * @param[in] policy Policy to use to handle overflow. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. * * @return a Status */ - static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy); + static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); protected: // Inherited methods overridden: @@ -157,23 +162,25 @@ public: /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel * - * @param[in] op Arithmetic operation to be executed. - * @param[in] input1 First tensor input. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. - * @param[in] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[in] output Output tensor. Data types supported: Same as @p input1. + * @param[in] op Arithmetic operation to be executed. + * @param[in] input1 First tensor input. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. + * @param[in] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[in] output Output tensor. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ - void configure(ArithmeticOperation op, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output); + void configure(ArithmeticOperation op, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel * - * @param[in] op Arithmetic operation to be executed. - * @param[in] input1 First tensor input info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. - * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1. - * @param[in] output Output tensor info. Data types supported: Same as @p input1. + * @param[in] op Arithmetic operation to be executed. + * @param[in] input1 First tensor input info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/QSYMM16/F16/U32/S32/F32. + * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1. + * @param[in] output Output tensor info. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. * * @return a Status */ - static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output); + static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); protected: // Inherited methods overridden: diff --git a/arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h b/arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h index 58471ab299..eacdb44c09 100644 --- a/arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h +++ b/arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h @@ -55,9 +55,10 @@ public: * 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. Supported overflow policies: Wrap, Saturate * @param[in] rounding_policy Rounding policy. Supported rounding modes: to zero, to nearest even. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ void configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, float scale, - ConvertPolicy overflow_policy, RoundingPolicy rounding_policy); + ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLPixelWiseMultiplicationKernel * * @param[in] input1 An input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. @@ -67,11 +68,12 @@ public: * 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. Supported overflow policies: Wrap, Saturate * @param[in] rounding_policy Rounding policy. Supported rounding modes: to zero, to nearest even. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. * * @return a status */ static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, - ConvertPolicy overflow_policy, RoundingPolicy rounding_policy); + ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run(const Window &window, cl::CommandQueue &queue) override; @@ -99,20 +101,22 @@ public: CLComplexPixelWiseMultiplicationKernel &operator=(CLComplexPixelWiseMultiplicationKernel &&) = default; /** Initialise the kernel's input, output and border mode. * - * @param[in] input1 An input tensor. Data types supported: F32. Number of channels supported: 2. - * @param[in] input2 An input tensor. Data types supported: same as @p input1. Number of channels supported: same as @p input1. - * @param[out] output The output tensor, Data types supported: same as @p input1. Number of channels supported: same as @p input1. + * @param[in] input1 An input tensor. Data types supported: F32. Number of channels supported: 2. + * @param[in] input2 An input tensor. Data types supported: same as @p input1. Number of channels supported: same as @p input1. + * @param[out] output The output tensor, Data types supported: same as @p input1. Number of channels supported: same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ - void configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output); + void configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref CLComplexPixelWiseMultiplicationKernel * - * @param[in] input1 An input tensor info. Data types supported: F32. Number of channels supported: 2. - * @param[in] input2 An input tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. - * @param[in] output The output tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. + * @param[in] input1 An input tensor info. Data types supported: F32. Number of channels supported: 2. + * @param[in] input2 An input tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. + * @param[in] output The output tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output); + static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run(const Window &window, cl::CommandQueue &queue) override; -- cgit v1.2.1