Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 1 | /* |
Giorgio Arena | 5ae8d80 | 2021-11-18 18:02:13 +0000 | [diff] [blame] | 2 | * Copyright (c) 2019-2022 Arm Limited. |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 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 | */ |
Georgios Pinitas | 2eb5d16 | 2021-07-02 09:01:49 +0100 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_CPU_GEMMLOWP_QUANTIZEDOWN_INT32TOINT8_SCALEBYFIXEDPOINT_KERNEL_H |
| 25 | #define ARM_COMPUTE_CPU_GEMMLOWP_QUANTIZEDOWN_INT32TOINT8_SCALEBYFIXEDPOINT_KERNEL_H |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 26 | |
| 27 | #include "arm_compute/core/KernelDescriptors.h" |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 28 | |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 29 | #include "src/core/common/Macros.h" |
Georgios Pinitas | 7891a73 | 2021-08-20 21:39:25 +0100 | [diff] [blame] | 30 | #include "src/cpu/ICpuKernel.h" |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 31 | |
| 32 | namespace arm_compute |
| 33 | { |
Georgios Pinitas | 2eb5d16 | 2021-07-02 09:01:49 +0100 | [diff] [blame] | 34 | // Forward declaration |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 35 | class ITensor; |
| 36 | namespace cpu |
| 37 | { |
| 38 | namespace kernels |
| 39 | { |
| 40 | /** Kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8_SIGNED |
| 41 | * |
Manuel Bottini | cfac51c | 2021-06-18 15:47:28 +0100 | [diff] [blame] | 42 | * This kernel takes a final int32 accumulator value (the output of @ref CpuGemmLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8_SIGNED value. |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 43 | * The following computations will be performed by the kernel: |
| 44 | * |
| 45 | * -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier |
| 46 | * -# Add bias to final result if bias tensor is not a nullptr |
| 47 | * -# Round to nearest division by a power-of-two using result_shift |
| 48 | * -# Add offset to each result |
| 49 | * -# Clamp the value between the specified min and max bounds |
| 50 | * -# Clamp the resulting int32 values to the [-128..127] range and cast to QASYMM8_SIGNED. |
| 51 | * |
| 52 | */ |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 53 | class CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel |
| 54 | : public ICpuKernel<CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel> |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 55 | { |
| 56 | public: |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 57 | CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel() = default; |
| 58 | ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel); |
| 59 | /** Initialise the kernel's input and output. |
| 60 | * |
| 61 | * @param[in] src Input tensor info. Data type supported: S32 |
| 62 | * @param[in] bias Biases tensor info. 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] dst Output tensor info. Data type supported: Data type supported: QASYMM8_SIGNED |
| 65 | * @param[in] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add |
| 66 | * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication |
| 67 | * @param[in] result_offset_after_shift Offset to be applied to result before converting it back to QASYMM8_SIGNED |
| 68 | * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED |
| 69 | * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED, |
| 70 | * Along with @p min, this value can be used to implement "rectified linear unit" activation functions |
| 71 | */ |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 72 | void configure(ITensorInfo *src, |
| 73 | ITensorInfo *bias, |
| 74 | ITensorInfo *dst, |
| 75 | int result_fixedpoint_multiplier, |
| 76 | int result_shift, |
| 77 | int result_offset_after_shift, |
| 78 | int min = 0, |
| 79 | int max = 0); |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 80 | /** Static function to check if given info will lead to a valid configuration |
| 81 | * |
| 82 | * Similar to CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::configure() |
| 83 | * |
| 84 | * @return a status |
| 85 | */ |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 86 | static Status |
| 87 | validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, int min = 0, int max = 0); |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 88 | |
| 89 | // Inherited methods overridden: |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 90 | void run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info) override; |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 91 | const char *name() const override; |
| 92 | |
| 93 | private: |
| 94 | /** Template function to run the CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel |
| 95 | * |
| 96 | * @param[in] src Input tensor info |
| 97 | * @param[in] bias Bias tensor info |
| 98 | * @param[out] dst Output tensor info |
| 99 | * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()). |
| 100 | */ |
| 101 | template <bool is_bounded_relu> |
| 102 | void run_internal(const ITensor *src, const ITensor *bias, ITensor *dst, const Window &window); |
| 103 | |
| 104 | /** Common signature for all the specialised CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel functions |
| 105 | * |
| 106 | * @param[in] src Input tensor info |
| 107 | * @param[in] bias Bias tensor info |
| 108 | * @param[out] dst Output tensor info |
| 109 | * @param[in] window Region on which to execute the kernel. |
| 110 | */ |
| 111 | using QuantizeDownFunctionPtr = void (CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::*)( |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 112 | const ITensor *src, const ITensor *bias, ITensor *dst, const Window &window); |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 113 | |
Felix Thomasmathibalan | afd38f0 | 2023-09-27 17:46:17 +0100 | [diff] [blame] | 114 | QuantizeDownFunctionPtr _func{nullptr}; |
| 115 | int _result_fixedpoint_multiplier{0}; |
| 116 | int _result_shift{0}; |
| 117 | int _result_offset_after_shift{0}; |
| 118 | int _min{0}; |
| 119 | int _max{0}; |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 120 | }; |
| 121 | } // namespace kernels |
| 122 | } // namespace cpu |
| 123 | } // namespace arm_compute |
Georgios Pinitas | 2eb5d16 | 2021-07-02 09:01:49 +0100 | [diff] [blame] | 124 | #endif /* ARM_COMPUTE_CPU_GEMMLOWP_QUANTIZEDOWN_INT32TOINT8_SCALEBYFIXEDPOINT_KERNEL_H */ |