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_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFIXEDPOINTKERNEL_H 25 #define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFIXEDPOINTKERNEL_H 26 27 #include "src/core/CL/ICLKernel.h" 28 29 namespace arm_compute 30 { 31 class ICLTensor; 32 33 /** OpenCL kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8/QASYMM8_SIGNED/QSYMM16 34 * 35 * This kernel takes a final int32 accumulator value (the output of the matrix multiplication), and processes it to obtain the final quantized value. 36 * The following computations will be performed by the kernel: 37 * 38 * -# Compute fixed point multiplication between each entry of input by gemmlowp_multiplier 39 * -# Add bias to final result if bias tensor is not a nullptr 40 * -# Round to nearest division by a power-of-two using result_shift 41 * -# Add offset to each result 42 * -# Clamp the value between the specified min and max bounds 43 * -# Clamp the resulting int32 values to the proper quantized range and cast to QASYMM8/QASYMM8_SIGNED/QSYMM16. 44 */ 45 class CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel : public ICLKernel 46 { 47 public: 48 /** Constructor */ 49 CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel(); 50 /** Prevent instances of this class from being copied (As this class contains pointers)*/ 51 CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel(const CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &) = delete; 52 /** Prevent instances of this class from being copied (As this class contains pointers)*/ 53 CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &operator=(const CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &) = delete; 54 /** Allow instances of this class to be moved */ 55 CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel(CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &&) = default; 56 /** Allow instances of this class to be moved */ 57 CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &operator=(CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &&) = default; 58 /** Initialise the kernel's input and output. 59 * 60 * @param[in] compile_context The compile context to be used. 61 * @param[in] input Input tensor. Data type supported: S32 62 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. 63 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 64 * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED/QSYMM16. 65 * @param[in] info Output stage info. Used to pass the quantized output data type 66 */ 67 void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *info); 68 /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel 69 * 70 * @param[in] input Input tensor. Data type supported: S32 71 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. 72 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 73 * @param[in] output Output tensor. Data type supported: Data type supported: QSYMM8/QASYMM8_SIGNED/QSYMM16. 74 * @param[in] info Output stage info. Used to pass the quantized output data type 75 * 76 * @return a status 77 */ 78 static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *info); 79 80 // Inherited methods overridden: 81 void run(const Window &window, cl::CommandQueue &queue) override; 82 83 private: 84 const ICLTensor *_input; 85 const ICLTensor *_bias; 86 ICLTensor *_output; 87 }; 88 } // namespace arm_compute 89 #endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFIXEDPOINTKERNEL_H */ 90