Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2017 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 | */ |
Gian Marco | 58c5794 | 2017-11-28 09:10:03 +0000 | [diff] [blame] | 24 | #ifndef __ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEKERNEL_H__ |
| 25 | #define __ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEKERNEL_H__ |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 26 | |
| 27 | #include "arm_compute/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 |
| 34 | * |
| 35 | * This kernel takes a final int32 accumulator value (the output of @ref CLGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8 value. |
| 36 | * The following computations will be performed by the kernel: |
| 37 | * |
| 38 | * -# Add offset terms to final result |
| 39 | * -# Multiply each entry of result by result_mult_int |
| 40 | * -# Add bias to final result if bias tensor is not a nullptr |
| 41 | * -# Shift the int32 accumulator by result_shift |
| 42 | * -# Clamp the value between the specified min and max bounds |
| 43 | * -# Clamp the resulting int32 values to the [0..255] range and cast to QASYMM8. |
| 44 | * |
| 45 | */ |
| 46 | class CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel : public ICLKernel |
| 47 | { |
| 48 | public: |
| 49 | /** Constructor */ |
| 50 | CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel(); |
| 51 | /** Prevent instances of this class from being copied (As this class contains pointers)*/ |
| 52 | CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &) = delete; |
| 53 | /** Prevent instances of this class from being copied (As this class contains pointers)*/ |
| 54 | CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &operator=(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &) = delete; |
| 55 | /** Allow instances of this class to be moved */ |
| 56 | CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel(CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &&) = default; |
| 57 | /** Allow instances of this class to be moved */ |
| 58 | CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &operator=(CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &&) = default; |
| 59 | /** Initialise the kernel's input and output. |
| 60 | * |
| 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 |
| 65 | * @param[in] result_offset Offset to be added to each element of the input matrix |
| 66 | * @param[in] result_mult_int Value to be multiplied to each element of the input matrix when once the result_offset has been add |
| 67 | * @param[in] result_shift Number of bits to shift right the result before converting back to QASYMM8 |
| 68 | * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8 |
| 69 | * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8, |
| 70 | * Along with @p min, this value can be used to implement "rectified linear unit" activation functions |
| 71 | */ |
| 72 | void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_offset, int result_mult_int, int result_shift, int min = 0, int max = 0); |
Gian Marco | 58c5794 | 2017-11-28 09:10:03 +0000 | [diff] [blame] | 73 | /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel |
| 74 | * |
| 75 | * @param[in] input Input tensor. Data type supported: S32 |
| 76 | * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required. |
| 77 | * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. |
| 78 | * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8 |
| 79 | * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8 |
| 80 | * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8, |
| 81 | * Along with @p min, this value can be used to implement "rectified linear unit" activation functions |
| 82 | * |
| 83 | * @return an error status |
| 84 | */ |
| 85 | static Error validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0); |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 86 | |
| 87 | // Inherited methods overridden: |
| 88 | void run(const Window &window, cl::CommandQueue &queue) override; |
| 89 | |
| 90 | private: |
| 91 | const ICLTensor *_input; |
| 92 | const ICLTensor *_bias; |
| 93 | ICLTensor *_output; |
| 94 | }; |
| 95 | } // namespace arm_compute |
| 96 | |
Gian Marco | 58c5794 | 2017-11-28 09:10:03 +0000 | [diff] [blame] | 97 | #endif /* __ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEKERNEL_H__ */ |