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_NEACCUMULATEKERNEL_H 25 #define ARM_COMPUTE_NEACCUMULATEKERNEL_H 26 27 #include "src/core/NEON/INESimpleKernel.h" 28 29 #include <cstdint> 30 31 namespace arm_compute 32 { 33 class ITensor; 34 35 /** Interface for the accumulate kernel 36 * 37 * Accumulation is computed by: 38 * @f[ accum(x,y) = accum(x,y) + input(x,y) @f] 39 */ 40 class NEAccumulateKernel : public INESimpleKernel 41 { 42 public: name()43 const char *name() const override 44 { 45 return "NEAccumulateKernel"; 46 } 47 /** Default constructor */ 48 NEAccumulateKernel() = default; 49 /** Prevent instances of this class from being copied (As this class contains pointers) */ 50 NEAccumulateKernel(const NEAccumulateKernel &) = delete; 51 /** Prevent instances of this class from being copied (As this class contains pointers) */ 52 NEAccumulateKernel &operator=(const NEAccumulateKernel &) = delete; 53 /** Allow instances of this class to be moved */ 54 NEAccumulateKernel(NEAccumulateKernel &&) = default; 55 /** Allow instances of this class to be moved */ 56 NEAccumulateKernel &operator=(NEAccumulateKernel &&) = default; 57 /** Default destructor */ 58 ~NEAccumulateKernel() = default; 59 /** Set the input and accumulation tensors 60 * 61 * @param[in] input Source tensor. Data type supported: U8. 62 * @param[out] accum Destination tensor. Data type supported: S16. 63 */ 64 void configure(const ITensor *input, ITensor *accum); 65 66 // Inherited methods overridden: 67 void run(const Window &window, const ThreadInfo &info) override; 68 }; 69 70 /** Interface for the accumulate weighted kernel 71 * 72 * Weighted accumulation is computed: 73 * @f[ accum(x,y) = (1 - \alpha)*accum(x,y) + \alpha*input(x,y) @f] 74 * 75 * Where @f$ 0 \le \alpha \le 1 @f$ 76 * Conceptually, the rounding for this is defined as: 77 * @f[ output(x,y)= uint8( (1 - \alpha) * float32( int32( output(x,y) ) ) + \alpha * float32( int32( input(x,y) ) ) ) @f] 78 */ 79 class NEAccumulateWeightedKernel : public INESimpleKernel 80 { 81 public: name()82 const char *name() const override 83 { 84 return "NEAccumulateWeightedKernel"; 85 } 86 /** Default constructor */ 87 NEAccumulateWeightedKernel(); 88 /** Prevent instances of this class from being copied (As this class contains pointers) */ 89 NEAccumulateWeightedKernel(const NEAccumulateWeightedKernel &) = delete; 90 /** Prevent instances of this class from being copied (As this class contains pointers) */ 91 NEAccumulateWeightedKernel &operator=(const NEAccumulateWeightedKernel &) = delete; 92 /** Allow instances of this class to be moved */ 93 NEAccumulateWeightedKernel(NEAccumulateWeightedKernel &&) = default; 94 /** Allow instances of this class to be moved */ 95 NEAccumulateWeightedKernel &operator=(NEAccumulateWeightedKernel &&) = default; 96 /** Default destructor */ 97 ~NEAccumulateWeightedKernel() = default; 98 /** Set the input and accumulation tensors, and the scale value 99 * 100 * @param[in] input Source tensor. Data type supported: U8. 101 * @param[in] alpha Scalar value in the range [0.0f, 1.0f] 102 * @param[in,out] accum Accumulated tensor. Data type supported: U8. 103 */ 104 void configure(const ITensor *input, float alpha, ITensor *accum); 105 106 // Inherited methods overridden: 107 void run(const Window &window, const ThreadInfo &info) override; 108 109 protected: 110 float _alpha; 111 }; 112 113 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC 114 /** Interface for the accumulate weighted kernel using F16 */ 115 class NEAccumulateWeightedFP16Kernel : public NEAccumulateWeightedKernel 116 { 117 public: name()118 const char *name() const override 119 { 120 return "NEAccumulateWeightedFP16Kernel"; 121 } 122 /** Default constructor */ 123 NEAccumulateWeightedFP16Kernel() = default; 124 /** Prevent instances of this class from being copied (As this class contains pointers) */ 125 NEAccumulateWeightedFP16Kernel(const NEAccumulateWeightedFP16Kernel &) = delete; 126 /** Prevent instances of this class from being copied (As this class contains pointers) */ 127 NEAccumulateWeightedFP16Kernel &operator=(const NEAccumulateWeightedFP16Kernel &) = delete; 128 /** Allow instances of this class to be moved */ 129 NEAccumulateWeightedFP16Kernel(NEAccumulateWeightedFP16Kernel &&) = default; 130 /** Allow instances of this class to be moved */ 131 NEAccumulateWeightedFP16Kernel &operator=(NEAccumulateWeightedFP16Kernel &&) = default; 132 /** Default destructor */ 133 ~NEAccumulateWeightedFP16Kernel() = default; 134 // Inherited methods overridden: 135 void run(const Window &window, const ThreadInfo &info) override; 136 }; 137 #else /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ 138 /** Interface for the accumulate weighted kernel using F16 */ 139 using NEAccumulateWeightedFP16Kernel = NEAccumulateWeightedKernel; 140 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ 141 142 /** Interface for the accumulate squared kernel 143 * 144 * The accumulation of squares is computed: 145 * @f[ accum(x,y) = saturate_{int16} ( (uint16) accum(x,y) + (((uint16)(input(x,y)^2)) >> (shift)) ) @f] 146 * 147 * Where @f$ 0 \le shift \le 15 @f$ 148 */ 149 class NEAccumulateSquaredKernel : public INESimpleKernel 150 { 151 public: name()152 const char *name() const override 153 { 154 return "NEAccumulateSquaredKernel"; 155 } 156 /** Default constructor */ 157 NEAccumulateSquaredKernel(); 158 /** Prevent instances of this class from being copied (As this class contains pointers) */ 159 NEAccumulateSquaredKernel(const NEAccumulateSquaredKernel &) = delete; 160 /** Prevent instances of this class from being copied (As this class contains pointers) */ 161 NEAccumulateSquaredKernel &operator=(const NEAccumulateSquaredKernel &) = delete; 162 /** Allow instances of this class to be moved */ 163 NEAccumulateSquaredKernel(NEAccumulateSquaredKernel &&) = default; 164 /** Allow instances of this class to be moved */ 165 NEAccumulateSquaredKernel &operator=(NEAccumulateSquaredKernel &&) = default; 166 /** Default destructor */ 167 ~NEAccumulateSquaredKernel() = default; 168 /** Set the input and accumulation tensors and the shift value. 169 * 170 * @param[in] input Source tensor. Data type supported: U8. 171 * @param[in] shift Shift value in the range of [0, 15] 172 * @param[in,out] accum Accumulated tensor. Data type supported: S16. 173 */ 174 void configure(const ITensor *input, uint32_t shift, ITensor *accum); 175 176 // Inherited methods overridden: 177 void run(const Window &window, const ThreadInfo &info) override; 178 179 private: 180 uint32_t _shift; 181 }; 182 } // namespace arm_compute 183 #endif /*ARM_COMPUTE_NEACCUMULATEKERNEL_H */ 184