diff options
Diffstat (limited to 'arm_compute/runtime/CL/functions/CLElementwiseOperations.h')
| -rw-r--r-- | arm_compute/runtime/CL/functions/CLElementwiseOperations.h | 586 |
1 files changed, 208 insertions, 378 deletions
diff --git a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h index 55c5fb3455..13844c98a1 100644 --- a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h +++ b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2020 Arm Limited. + * Copyright (c) 2018-2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -24,6 +24,7 @@ #ifndef ARM_COMPUTE_CLELEMENTWISEOPERATIONS_H #define ARM_COMPUTE_CLELEMENTWISEOPERATIONS_H +#include "arm_compute/function_info/ActivationLayerInfo.h" #include "arm_compute/runtime/CL/ICLOperator.h" #include "arm_compute/runtime/IFunction.h" @@ -33,324 +34,7 @@ class ICLTensor; class CLCompileContext; class ITensorInfo; -namespace experimental -{ -/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for addition - * - * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. - * @note The function performs an arithmetic addition between two tensors. - */ -class CLArithmeticAddition : public ICLOperator -{ -public: - /** Default Constructor */ - CLArithmeticAddition(); - /** Initialise the kernel's inputs, output and conversion policy. - * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, ConvertPolicy policy, - const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for addition - * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. - * @param[in] input2 Second tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. - * @param[in] output Output tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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. - * - * @return a status - */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; - -/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for subtraction - * - * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/S32/F16/F32. - * @note The function performs an arithmetic subtraction between two tensors. - */ -class CLArithmeticSubtraction : public ICLOperator -{ -public: - /** Default Constructor */ - CLArithmeticSubtraction(); - /** Initialise the kernel's inputs, output and conversion policy. - * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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 Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, ConvertPolicy policy, - const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for subtraction - * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 - * - * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. - * @param[in] input2 Second tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. - * @param[in] output Output tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/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. - * - * @return a status - */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; - -/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for division - * - * @note The tensor data type for the inputs must be F16/F32. - * @note The function performs an arithmetic division between two tensors. - */ -class CLArithmeticDivision : public ICLOperator -{ -public: - /** Default Constructor */ - CLArithmeticDivision(); - /** Initialise the kernel's inputs, output. - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: 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 Second tensor input. 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: Same as @p input1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticDivision - * - * @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. - * - * @return a status - */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; - -/** Basic function to run @ref CLArithmeticOperationKernel for max - * - * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32. - * @note The function performs a max operation between two tensors. - */ -class CLElementwiseMax : public ICLOperator -{ -public: - /** Default Constructor */ - CLElementwiseMax(); - /** Initialise the kernel's inputs, output and conversion policy. - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/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 Second tensor input. 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: same as @p input1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for max - * - * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/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. - * - * @return a status - */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; - -/** Basic function to run @ref CLArithmeticOperationKernel for min - * - * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32. - * @note The function performs a max operation between two tensors. - */ -class CLElementwiseMin : public ICLOperator -{ -public: - /** Default Constructor */ - CLElementwiseMin(); - /** Initialise the kernel's inputs, output and conversion policy. - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/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 Second tensor input. 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: same as @p input1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for min - * - * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/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. - * - * @return a status - */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; - -/** Basic function to run @ref CLArithmeticOperationKernel for squared difference - * - * @note The tensor data type for the inputs must be QASYMM8/U8/S16/QSYMM16/F16/F32. - * @note The function performs a squared different operation between two tensors (i.e., out[i] = (in1[i] - in2[i])^2 - */ -class CLElementwiseSquaredDiff : public ICLOperator -{ -public: - /** Default Constructor */ - CLElementwiseSquaredDiff(); - /** Initialise the kernel's inputs, output and conversion policy. - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/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 Second tensor input. 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: same as @p input1. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for squared difference - * - * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/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. - * - * @return a status - */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; - -/** Basic function to run @ref CLArithmeticOperationKernel for power - * - * @note The tensor data type for the inputs must be F16/F32. - * @note The function performs an elementwise power of in1 to in2 (i.e., out[i] = in1[i] ^ in2[i]) - */ -class CLElementwisePower : public ICLOperator -{ -public: - /** Default Constructor */ - CLElementwisePower(); - /** Initialise the kernel's inputs, output and conversion policy. - * - * @param[in] compile_context The compile context to be used. - * @param[in, out] input1 First tensor input. Data types supported: 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 Second tensor input. Data types supported: 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[out] output Output tensor. Data types supported:F16/F32. - * @param[in] act_info (Optional) Activation layer information in case of a fused activation. - */ - void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for power - * - * @param[in] input1 First tensor input info. Data types supported: F16/F32. - * @param[in] input2 Second tensor input info. Data types supported: F16/F32. - * @param[in] output Output tensor info. Data types supported: F16/F32. - * @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, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - - // Inherited methods overridden: - void run(ITensorPack &tensors) override; -}; -} // namespace experimental - -/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for addition +/** Basic function to run @ref opencl::kernels::ClSaturatedArithmeticKernel for addition * * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. * @note The function performs an arithmetic addition between two tensors. @@ -372,19 +56,23 @@ public: CLArithmeticAddition &operator=(CLArithmeticAddition &&); /** Initialise the kernel's inputs, output and conversion policy. * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |U8 |U8 |U8 | + * |U8 |U8 |S16 | + * |U8 |S16 |S16 | + * |S16 |U8 |S16 | + * |S16 |S16 |S16 | + * |S32 |S32 |S32 | + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | * * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. @@ -394,7 +82,11 @@ public: * @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(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + ConvertPolicy policy, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and conversion policy. * * Valid configurations (Input1,Input2) -> Output : @@ -420,9 +112,13 @@ public: * @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(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ConvertPolicy policy, + void configure(const CLCompileContext &compile_context, + const ICLTensor *input1, + const ICLTensor *input2, + ICLTensor *output, + ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for addition + /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClSaturatedArithmeticKernel for addition * * Valid configurations (Input1,Input2) -> Output : * @@ -446,7 +142,11 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + ConvertPolicy policy, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -456,7 +156,7 @@ private: std::unique_ptr<Impl> _impl; }; -/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for subtraction +/** Basic function to run @ref opencl::kernels::ClSaturatedArithmeticKernel for subtraction * * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/S32/F16/F32. * @note The function performs an arithmetic subtraction between two tensors. @@ -478,19 +178,23 @@ public: CLArithmeticSubtraction &operator=(CLArithmeticSubtraction &&); /** Initialise the kernel's inputs, output and conversion policy. * - * Valid configurations (Input1,Input2) -> Output : - * - * - (U8,U8) -> U8 - * - (U8,U8) -> S16 - * - (S16,U8) -> S16 - * - (U8,S16) -> S16 - * - (S16,S16) -> S16 - * - (S32,S32) -> S32 - * - (F16,F16) -> F16 - * - (F32,F32) -> F32 - * - (QASYMM8,QASYMM8) -> QASYMM8 - * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED - * - (QSYMM16,QSYMM16) -> QSYMM16 + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |U8 |U8 |U8 | + * |U8 |U8 |S16 | + * |U8 |S16 |S16 | + * |S16 |U8 |S16 | + * |S16 |S16 |S16 | + * |S32 |S32 |S32 | + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | * * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32. * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. @@ -500,7 +204,11 @@ public: * @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(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(const ICLTensor *input1, + const ICLTensor *input2, + ICLTensor *output, + ConvertPolicy policy, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and conversion policy. * * Valid configurations (Input1,Input2) -> Output : @@ -526,9 +234,13 @@ public: * @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(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ConvertPolicy policy, + void configure(const CLCompileContext &compile_context, + const ICLTensor *input1, + const ICLTensor *input2, + ICLTensor *output, + ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); - /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for subtraction + /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClSaturatedArithmeticKernel for subtraction * * Valid configurations (Input1,Input2) -> Output : * @@ -552,7 +264,11 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + ConvertPolicy policy, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -562,7 +278,7 @@ private: std::unique_ptr<Impl> _impl; }; -/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for division +/** Basic function to run @ref opencl::kernels::ClSaturatedArithmeticKernel for division * * @note The tensor data type for the inputs must be F16/F32. * @note The function performs an arithmetic division between two tensors. @@ -584,6 +300,15 @@ public: CLArithmeticDivision &operator=(CLArithmeticDivision &&); /** Initialise the kernel's inputs, output. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | + * * @param[in, out] input1 First tensor input. Data types supported: 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 Second tensor input. Same as @p input1. @@ -591,7 +316,10 @@ public: * @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. */ - void configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output. * * @param[in] compile_context The compile context to be used. @@ -602,7 +330,11 @@ public: * @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. */ - void configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(const CLCompileContext &compile_context, + 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 CLArithmeticDivision * * @param[in] input1 First tensor input info. Data types supported: F16/F32. @@ -612,7 +344,10 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -622,7 +357,7 @@ private: std::unique_ptr<Impl> _impl; }; -/** Basic function to run @ref CLArithmeticOperationKernel for max +/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for max * * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32. * @note The function performs a max operation between two tensors. @@ -644,6 +379,22 @@ public: CLElementwiseMax &operator=(CLElementwiseMax &&); /** Initialise the kernel's inputs, output and conversion policy. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |U8 |U8 |U8 | + * |S16 |S16 |S16 | + * |S32 |S32 |S32 | + * |U32 |U32 |U32 | + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | + * * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/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 Second tensor input. Data types supported: same as @p input1. @@ -651,7 +402,10 @@ public: * @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. */ - void configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and conversion policy. * * @param[in] compile_context The compile context to be used. @@ -662,8 +416,12 @@ public: * @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. */ - void configure(const CLCompileContext &compile_context, ICLTensor *input1, 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 for max + void configure(const CLCompileContext &compile_context, + ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); + /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for max * * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/F16/F32. * @param[in] input2 Second tensor input info. Data types supported: same as @p input1. @@ -672,7 +430,10 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -682,7 +443,7 @@ private: std::unique_ptr<Impl> _impl; }; -/** Basic function to run @ref CLArithmeticOperationKernel for min +/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for min * * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32. * @note The function performs a max operation between two tensors. @@ -704,6 +465,22 @@ public: CLElementwiseMin &operator=(CLElementwiseMin &&); /** Initialise the kernel's inputs, output and conversion policy. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |U8 |U8 |U8 | + * |S16 |S16 |S16 | + * |S32 |S32 |S32 | + * |U32 |U32 |U32 | + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | + * * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/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 Second tensor input. Data types supported: same as @p input1. @@ -711,7 +488,10 @@ public: * @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. */ - void configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and conversion policy. * * @param[in] compile_context The compile context to be used. @@ -722,8 +502,12 @@ public: * @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. */ - void configure(const CLCompileContext &compile_context, ICLTensor *input1, 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 for min + void configure(const CLCompileContext &compile_context, + ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); + /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for min * * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/F16/F32. * @param[in] input2 Second tensor input info. Data types supported: same as @p input1. @@ -732,7 +516,10 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -742,7 +529,7 @@ private: std::unique_ptr<Impl> _impl; }; -/** Basic function to run @ref CLArithmeticOperationKernel for squared difference +/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for squared difference * * @note The tensor data type for the inputs must be QASYMM8/U8/S16/QSYMM16/F16/F32. * @note The function performs a squared different operation between two tensors (i.e., out[i] = (in1[i] - in2[i])^2 @@ -764,6 +551,20 @@ public: CLElementwiseSquaredDiff &operator=(CLElementwiseSquaredDiff &&); /** Initialise the kernel's inputs, output and conversion policy. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |QASYMM8 |QASYMM8 |QASYMM8 | + * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED | + * |QSYMM16 |QSYMM16 |QASYMM16 | + * |U8 |U8 |U8 | + * |S16 |S16 |S16 | + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | + * * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/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 Second tensor input. Data types supported: same as @p input1. @@ -771,7 +572,10 @@ public: * @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. */ - void configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and conversion policy. * * @param[in] compile_context The compile context to be used. @@ -782,8 +586,12 @@ public: * @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. */ - void configure(const CLCompileContext &compile_context, ICLTensor *input1, 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 for squared difference + void configure(const CLCompileContext &compile_context, + ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); + /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for squared difference * * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32. * @param[in] input2 Second tensor input info. Data types supported: same as @p input1. @@ -792,7 +600,10 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; @@ -802,7 +613,7 @@ private: std::unique_ptr<Impl> _impl; }; -/** Basic function to run @ref CLArithmeticOperationKernel for power +/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for power * * @note The tensor data type for the inputs must be F16/F32. * @note The function performs an elementwise power of in1 to in2 (i.e., out[i] = in1[i] ^ in2[i]) @@ -824,6 +635,15 @@ public: CLElementwisePower &operator=(CLElementwisePower &&); /** Initialise the kernel's inputs, output and conversion policy. * + * Valid data layouts: + * - All + * + * Valid data type configurations: + * |src0 |src1 |dst | + * |:--------------|:--------------|:--------------| + * |F16 |F16 |F16 | + * |F32 |F32 |F32 | + * * @param[in, out] input1 First tensor input. Data types supported: 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 Second tensor input. Data types supported: F16/F32. @@ -831,7 +651,10 @@ public: * @param[out] output Output tensor. Data types supported:F16/F32. * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ - void configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + void configure(ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); /** Initialise the kernel's inputs, output and conversion policy. * * @param[in] compile_context The compile context to be used. @@ -842,8 +665,12 @@ public: * @param[out] output Output tensor. Data types supported:F16/F32. * @param[in] act_info (Optional) Activation layer information in case of a fused activation. */ - void configure(const CLCompileContext &compile_context, ICLTensor *input1, 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 for power + void configure(const CLCompileContext &compile_context, + ICLTensor *input1, + ICLTensor *input2, + ICLTensor *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); + /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for power * * @param[in] input1 First tensor input info. Data types supported: F16/F32. * @param[in] input2 Second tensor input info. Data types supported: F16/F32. @@ -852,7 +679,10 @@ public: * * @return a status */ - static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); + static Status validate(const ITensorInfo *input1, + const ITensorInfo *input2, + const ITensorInfo *output, + const ActivationLayerInfo &act_info = ActivationLayerInfo()); // Inherited methods overridden: void run() override; |