1 /* 2 * Copyright (c) 2016-2020 Arm Limited. 3 * 4 * SPDX-License-Identifier: MIT 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to 8 * deal in the Software without restriction, including without limitation the 9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or 10 * sell copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in all 14 * copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 * SOFTWARE. 23 */ 24 #ifndef ARM_COMPUTE_NEPIXELWISEMULTIPLICATION_H 25 #define ARM_COMPUTE_NEPIXELWISEMULTIPLICATION_H 26 27 #include "arm_compute/core/Types.h" 28 #include "arm_compute/runtime/IFunction.h" 29 #include "arm_compute/runtime/NEON/INEOperator.h" 30 31 namespace arm_compute 32 { 33 class ITensor; 34 class ITensorInfo; 35 36 namespace experimental 37 { 38 /** Basic function to run @ref NEPixelWiseMultiplicationKernel */ 39 class NEPixelWiseMultiplication : public INEOperator 40 { 41 public: 42 /** Initialise the kernel's inputs, output and convertion policy. 43 * 44 * Valid configurations (Input1,Input2) -> Output : 45 * 46 * Support: Broadcast? Scale=1/255? 47 * - (U8,U8) -> U8, S16 N Y 48 * - (U8,S16) -> S16 N Y 49 * - (S16,U8) -> S16 N Y 50 * - (S16,S16) -> S16 N Y 51 * - (S32,S32) -> S32 Y N 52 * - (F16,F16) -> F16 N Y 53 * - (F32,F32) -> F32 Y Y 54 * - (QASYMM8,QASYMM8) -> QASYMM8 Y Y 55 * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED Y Y 56 * - (QSYMM16,QSYMM16) -> QSYMM16, S32 N Y 57 * 58 * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. 59 * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. 60 * 61 * @param[in, out] input1 First input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32 62 * This input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 63 * @param[in, out] input2 Second input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32 64 * This input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 65 * @param[out] output Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32/S32 66 * @param[in] scale Scale to apply after multiplication. 67 * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. 68 * If both @p input1, @p input2 and @p output are of datatype S32, scale cannot be 1/255 69 * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if any of the inputs is of quantized datatype 70 * @param[in] rounding_policy Rounding policy. 71 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 72 */ 73 void configure(ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, 74 const ActivationLayerInfo &act_info = ActivationLayerInfo()); 75 /** Static function to check if given info will lead to a valid configuration of @ref NEPixelWiseMultiplication 76 * 77 * Valid configurations (Input1,Input2) -> Output : 78 * 79 * Support: Broadcast? Scale=1/255? 80 * - (U8,U8) -> U8, S16 N Y 81 * - (U8,S16) -> S16 N Y 82 * - (S16,U8) -> S16 N Y 83 * - (S16,S16) -> S16 N Y 84 * - (S32,S32) -> S32 Y N 85 * - (F16,F16) -> F16 N Y 86 * - (F32,F32) -> F32 Y Y 87 * - (QASYMM8,QASYMM8) -> QASYMM8 Y Y 88 * - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED Y Y 89 * - (QSYMM16,QSYMM16) -> QSYMM16, S32 N Y 90 * 91 * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. 92 * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. 93 * 94 * @param[in] input1 First input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32 95 * @param[in] input2 Second input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32 96 * @param[in] output Output tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32/S32 97 * @param[in] scale Scale to apply after multiplication. 98 * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. 99 * If both @p input1, @p input2 and @p output are of datatype S32, scale cannot be 1/255 100 * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if any of the inputs is of quantized datatype 101 * @param[in] rounding_policy Rounding policy. 102 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 103 * 104 * @return a status 105 */ 106 static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, 107 const ActivationLayerInfo &act_info = ActivationLayerInfo()); 108 }; 109 110 /** Basic function to run @ref NEComplexPixelWiseMultiplicationKernel. */ 111 class NEComplexPixelWiseMultiplication : public INEOperator 112 { 113 public: 114 /** Initialise the kernel's inputs, output. 115 * 116 * @param[in, out] input1 An input tensor. Data types supported: F32. Number of channels supported: 2 (complex tensor). 117 * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 118 * @param[in, out] input2 An input tensor. Data types supported: same as @p input1. Number of channels supported: same as @p input1. 119 * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 120 * @param[out] output The output tensor. Data types supported: same as @p input1. Number of channels: same as @p input1. 121 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 122 */ 123 void configure(ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); 124 /** Static function to check if given info will lead to a valid configuration of @ref NEComplexPixelWiseMultiplication 125 * 126 * @param[in] input1 An input tensor info. Data types supported: F32. Number of channels supported: 2 (complex tensor). 127 * @param[in] input2 An input tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. 128 * @param[in] output The output tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. 129 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 130 */ 131 static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); 132 }; 133 } // namespace experimental 134 135 /** Basic function to run @ref NEPixelWiseMultiplicationKernel */ 136 class NEPixelWiseMultiplication : public IFunction 137 { 138 public: 139 /** Default Constructor */ 140 NEPixelWiseMultiplication(); 141 /** Default Destructor */ 142 ~NEPixelWiseMultiplication(); 143 /** Prevent instances of this class from being copied (As this class contains pointers) */ 144 NEPixelWiseMultiplication(const NEPixelWiseMultiplication &) = delete; 145 /** Default move constructor */ 146 NEPixelWiseMultiplication(NEPixelWiseMultiplication &&); 147 /** Prevent instances of this class from being copied (As this class contains pointers) */ 148 NEPixelWiseMultiplication &operator=(const NEPixelWiseMultiplication &) = delete; 149 /** Default move assignment operator */ 150 NEPixelWiseMultiplication &operator=(NEPixelWiseMultiplication &&); 151 /** Initialise the kernel's inputs, output and convertion policy. 152 * 153 * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. 154 * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. 155 * 156 * @param[in, out] input1 An input tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32 157 * This input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 158 * @param[in, out] input2 An input tensor. Data types supported: U8, QASYMM8 (only if @p input1 is QASYMM8), QASYMM8_SIGNED (only if @p input1 is QASYMM8_SIGNED), S16, S32, QSYMM16 (only if @p input1 is QSYMM16), F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). 159 * This input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 160 * @param[out] output Output tensor. Data types supported: 161 * - U8, only if both inputs are U8. 162 * - QASYMM8, only if both inputs are QASYMM8. 163 * - QASYMM8_SIGNED, only if @p input1 is QASYMM8_SIGNED. 164 * - S16. 165 * - QSYMM16, only if both inputs are QSYMM16. 166 * - S32, only if both inputs are S32 or both are QSYMM16. 167 * - F16, only if @p input1 is F16. 168 * - F32, only if both inputs are F32. 169 * @param[in] scale Scale to apply after multiplication. 170 * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. 171 * If both @p input1, @p input2 and @p output are of datatype S32, scale cannot be 1/255 172 * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if any of the inputs is of quantized datatype 173 * @param[in] rounding_policy Rounding policy. 174 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 175 */ 176 void configure(const ITensor *input1, const ITensor *input2, ITensor *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, 177 const ActivationLayerInfo &act_info = ActivationLayerInfo()); 178 /** Static function to check if given info will lead to a valid configuration of @ref NEPixelWiseMultiplication 179 * 180 * @note For @p scale equal to 1/255 only round to nearest even (implemented as round half up) is supported. 181 * For all other scale values only round to zero (implemented as round towards minus infinity) is supported. 182 * 183 * @param[in] input1 An input tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/S32/QSYMM16/F16/F32 184 * @param[in] input2 An input tensor info. Data types supported: U8, QASYMM8 (only if @p input1 is QASYMM8), QASYMM8_SIGNED (only if @p input1 is QASYMM8_SIGNED), S16, S32, QSYMM16 (only if both inputs are QSYMM16), F16 (only if @p input1 is F16), F32 (only if @p input1 is F32). 185 * @param[in] output Output tensor info. Data types supported: 186 * - U8, only if both inputs are U8. 187 * - QASYMM8, only if both inputs are QASYMM8. 188 * - QASYMM8_SIGNED, only if @p input1 is QASYMM8_SIGNED. 189 * - S16. 190 * - QSYMM16, only if both inputs are QSYMM16. 191 * - S32, only if both inputs are S32 or both are QSYMM16. 192 * - F16, only if @p input1 is F16. 193 * - F32, only if both inputs are F32. 194 * @param[in] scale Scale to apply after multiplication. 195 * Scale must be positive and its value must be either 1/255 or 1/2^n where n is between 0 and 15. 196 * If both @p input1, @p input2 and @p output are of datatype S32, scale cannot be 1/255 197 * @param[in] overflow_policy Overflow policy. ConvertPolicy cannot be WRAP if any of the inputs is of quantized datatype 198 * @param[in] rounding_policy Rounding policy. 199 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 200 * 201 * @return a status 202 */ 203 static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale, ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, 204 const ActivationLayerInfo &act_info = ActivationLayerInfo()); 205 206 // Inherited methods overridden: 207 void run() override; 208 209 private: 210 struct Impl; 211 std::unique_ptr<Impl> _impl; 212 }; 213 214 /** Basic function to run @ref NEComplexPixelWiseMultiplicationKernel. */ 215 class NEComplexPixelWiseMultiplication : public IFunction 216 { 217 public: 218 /** Default Constructor */ 219 NEComplexPixelWiseMultiplication(); 220 /** Default Destructor */ 221 ~NEComplexPixelWiseMultiplication(); 222 /** Prevent instances of this class from being copied (As this class contains pointers) */ 223 NEComplexPixelWiseMultiplication(const NEComplexPixelWiseMultiplication &) = delete; 224 /** Default move constructor */ 225 NEComplexPixelWiseMultiplication(NEComplexPixelWiseMultiplication &&); 226 /** Prevent instances of this class from being copied (As this class contains pointers) */ 227 NEComplexPixelWiseMultiplication &operator=(const NEComplexPixelWiseMultiplication &) = delete; 228 /** Default move assignment operator */ 229 NEComplexPixelWiseMultiplication &operator=(NEComplexPixelWiseMultiplication &&); 230 /** Initialise the kernel's inputs, output. 231 * 232 * @param[in, out] input1 An input tensor. Data types supported: F32. Number of channels supported: 2 (complex tensor). 233 * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 234 * @param[in, out] input2 An input tensor. Data types supported: same as @p input1. Number of channels supported: same as @p input1. 235 * The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0. 236 * @param[out] output The output tensor. Data types supported: same as @p input1. Number of channels: same as @p input1. 237 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 238 */ 239 void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); 240 /** Static function to check if given info will lead to a valid configuration of @ref NEComplexPixelWiseMultiplication 241 * 242 * @param[in] input1 An input tensor info. Data types supported: F32. Number of channels supported: 2 (complex tensor). 243 * @param[in] input2 An input tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. 244 * @param[in] output The output tensor info. Data types supported: same as @p input1. Number of channels supported: same as @p input1. 245 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported. 246 */ 247 static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo()); 248 249 // Inherited methods overridden: 250 void run() override; 251 252 private: 253 struct Impl; 254 std::unique_ptr<Impl> _impl; 255 }; 256 } 257 #endif /*ARM_COMPUTE_NEPIXELWISEMULTIPLICATION_H */ 258