diff options
author | Giorgio Arena <giorgio.arena@arm.com> | 2020-02-11 17:21:31 +0000 |
---|---|---|
committer | SiCong Li <sicong.li@arm.com> | 2020-04-03 08:51:12 +0000 |
commit | 8b2a7d3aa119e7f1d6a03690d05eb27c5d178b9f (patch) | |
tree | 9fb4f4f328f7a17de13bef109834e8ad8a21d2ee /arm_compute/runtime/NEON | |
parent | 15e4d876643c37e1db36ee1190ec52319479ffaf (diff) | |
download | ComputeLibrary-8b2a7d3aa119e7f1d6a03690d05eb27c5d178b9f.tar.gz |
COMPMID-3101 Fuse activation with floating point elementwise operation layers in CL
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: I1693f8664ba7c0dc8c076bbe7365cef1e667bd25
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2718
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON')
4 files changed, 93 insertions, 73 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h b/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h index e17c255d2a..6cab5b3547 100644 --- a/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h +++ b/arm_compute/runtime/NEON/functions/NEArithmeticAddition.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2019 ARM Limited. + * Copyright (c) 2016-2020 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -37,22 +37,24 @@ class NEArithmeticAddition : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 - * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 - * @param[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 - * @param[in] policy Policy to use to handle overflow. + * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 + * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 + * @param[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 + * @param[in] policy Policy to use to handle overflow. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy); + void configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticAddition * - * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 - * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 - * @param[in] output Output tensor. Data types supported: U8/SQASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 - * @param[in] policy Policy to use to handle overflow. + * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 + * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 + * @param[in] output Output tensor. Data types supported: U8/SQASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32 + * @param[in] policy Policy to use to handle overflow + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy); + static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); }; } // namespace arm_compute #endif /*ARM_COMPUTE_NEARITHMETICADDITION_H */ diff --git a/arm_compute/runtime/NEON/functions/NEArithmeticSubtraction.h b/arm_compute/runtime/NEON/functions/NEArithmeticSubtraction.h index c8c3fd3d2f..69d7b4bcfb 100644 --- a/arm_compute/runtime/NEON/functions/NEArithmeticSubtraction.h +++ b/arm_compute/runtime/NEON/functions/NEArithmeticSubtraction.h @@ -45,22 +45,24 @@ class NEArithmeticSubtraction : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/QSYMM16/S16/F16/F32 - * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/QSYMM16/S16/F16/F32 - * @param[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/QSYMM16/S16/F16/F32 - * @param[in] policy Policy to use to handle overflow. Convert policy cannot be WRAP if datatype is quantized. + * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/QSYMM16/S16/F16/F32 + * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/QSYMM16/S16/F16/F32 + * @param[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/QSYMM16/S16/F16/F32 + * @param[in] policy Policy to use to handle overflow. Convert policy cannot be WRAP if datatype is quantized. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy); + void configure(ITensor *input1, ITensor *input2, ITensor *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticSubtraction * - * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32 - * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32 - * @param[in] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32 - * @param[in] policy Policy to use to handle overflow. Convert policy cannot be WRAP if datatype is quantized. + * @param[in] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32 + * @param[in] input2 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32 + * @param[in] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/F32 + * @param[in] policy Policy to use to handle overflow. Convert policy cannot be WRAP if datatype is quantized. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy); + static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); }; } #endif /* ARM_COMPUTE_NEARITHMETICSUBTRACTION_H */ diff --git a/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h b/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h index e5af6bc841..cac105cdb9 100644 --- a/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h +++ b/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h @@ -41,20 +41,22 @@ class NEElementwiseMax : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32. - * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32. + * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output); + void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for max * - * @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/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] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/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. Currently not supported. * * @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()); }; /** Basic function to run @ref NEArithmeticOperationKernel for min @@ -67,20 +69,22 @@ class NEElementwiseMin : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32. - * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32. + * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output); + void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for min * - * @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/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] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/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. Currently not supported. * * @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()); }; /** Basic function to run @ref NEArithmeticOperationKernel for squared difference @@ -93,20 +97,22 @@ class NEElementwiseSquaredDiff : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32. - * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32. + * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output); + void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for squared difference * - * @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/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] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/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. Currently not supported. * * @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()); }; /** Basic function to run @ref NEArithmeticOperationKernel for division @@ -119,20 +125,22 @@ class NEElementwiseDivision : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in, out] input1 First tensor input. Data types supported: F16/F32. - * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in, out] input1 First tensor input. Data types supported: F16/F32. + * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output); + void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for division * - * @param[in] input1 First tensor input info. Data types supported: F16/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] input1 First tensor input info. Data types supported: F16/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. Currently not supported. * * @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()); }; /** Basic function to run @ref NEArithmeticOperationKernel for power @@ -146,20 +154,22 @@ class NEElementwisePower : public INESimpleFunction public: /** Initialise the kernel's inputs, output and conversion policy. * - * @param[in, out] input1 First tensor input. Data types supported: F16/F32. - * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. - * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in, out] input1 First tensor input. Data types supported: F16/F32. + * @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1. + * @param[out] output Output tensor. Data types supported: Same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output); + void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for power * - * @param[in] input1 First tensor input info. Data types supported: F16/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] input1 First tensor input info. Data types supported: F16/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. Currently not supported. * * @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()); }; /** Basic function to run @ref NEComparisonOperationKernel. diff --git a/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h b/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h index ede4327bfb..2b31032931 100644 --- a/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h +++ b/arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h @@ -57,8 +57,10 @@ 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. ConvertPolicy cannot be WRAP if datatype is QASYMM8, QASYMM8_SIGNED or QSYMM16. * @param[in] rounding_policy Rounding policy. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy); + void configure(ITensor *input1, ITensor *input2, ITensor *output, float scale, 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 NEPixelWiseMultiplication * * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. @@ -79,10 +81,12 @@ 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. ConvertPolicy cannot be WRAP if datatype is QASYMM8, QASYMM8_SIGNED or QSYMM16. * @param[in] rounding_policy Rounding policy. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy); + static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); }; /** Basic function to run @ref NEComplexPixelWiseMultiplicationKernel. */ @@ -91,20 +95,22 @@ class NEComplexPixelWiseMultiplication : public INESimpleFunction public: /** Initialise the kernel's inputs, output. * - * @param[in, out] input1 An input tensor. Data types supported: F32. Number of channels supported: 2 (complex tensor). - * 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. Number of channels 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 The output tensor. Data types supported: same as @p input1. Number of channels: same as @p input1. + * @param[in, out] input1 An input tensor. Data types supported: F32. Number of channels supported: 2 (complex tensor). + * 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. Number of channels 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 The output tensor. Data types supported: same as @p input1. Number of channels: same as @p input1. + * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. */ - void configure(ITensor *input1, ITensor *input2, ITensor *output); + void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Static function to check if given info will lead to a valid configuration of @ref NEComplexPixelWiseMultiplication * - * @param[in] input1 An input tensor info. Data types supported: F32. Number of channels supported: 2 (complex tensor). - * @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 (complex tensor). + * @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. Currently not supported. */ - 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()); }; } #endif /*ARM_COMPUTE_NEPIXELWISEMULTIPLICATION_H */ |