1 /* 2 * Copyright (c) 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_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFLOATKERNEL_H 25 #define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFLOATKERNEL_H 26 27 #include "src/core/CL/ICLKernel.h" 28 29 namespace arm_compute 30 { 31 // Forward declarations 32 class ICLTensor; 33 34 /** OpenCL kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8/QASYMM8_SIGNED 35 * 36 * This kernel takes a final int32 accumulator value (the output of the matrix multiplication), and processes it to obtain the final QASYMM8/QASYMM8_SIGNED value. 37 * The following computations will be performed by the kernel: 38 * 39 * -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier 40 * -# Add bias to final result if bias tensor is not a nullptr 41 * -# Requantize 42 * -# Add offset to each result 43 * -# Clamp the value between the specified min and max bounds 44 * -# Clamp the resulting int32 values to 45 * - to the [0..255] range and cast to QASYMM8. 46 * - to the [-128..127] range and cast to QASYMM8_SIGNED. 47 */ 48 class CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel : public ICLKernel 49 { 50 public: 51 /** Constructor */ 52 CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel(); 53 /** Prevent instances of this class from being copied (As this class contains pointers)*/ 54 CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel(const CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel &) = delete; 55 /** Prevent instances of this class from being copied (As this class contains pointers)*/ 56 CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel &operator=(const CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel &) = delete; 57 /** Allow instances of this class to be moved */ 58 CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel(CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel &&) = default; 59 /** Allow instances of this class to be moved */ 60 CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel &operator=(CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel &&) = default; 61 /** Initialise the kernel's input and output. 62 * 63 * @param[in] input Input tensor. Data type supported: S32 64 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. 65 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 66 * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED 67 * @param[in] info Output stage info. Used to pass the quantized output data type 68 */ 69 void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *info); 70 /** Initialise the kernel's input and output. 71 * 72 * @param[in] compile_context The compile context to be used. 73 * @param[in] input Input tensor. Data type supported: S32 74 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. 75 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 76 * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED 77 * @param[in] info Output stage info. Used to pass the quantized output data type 78 */ 79 void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *info); 80 /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel 81 * 82 * @param[in] input Input tensor. Data type supported: S32 83 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. 84 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 85 * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED 86 * @param[in] info Output stage info. Used to pass the quantized output data type 87 * 88 * @return a status 89 */ 90 static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *info); 91 92 // Inherited methods overridden: 93 void run(const Window &window, cl::CommandQueue &queue) override; 94 95 private: 96 const ICLTensor *_input; 97 const ICLTensor *_bias; 98 ICLTensor *_output; 99 }; 100 } // namespace arm_compute 101 #endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFLOATKERNEL_H */ 102