Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 1 | /* |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 2 | * Copyright (c) 2017-2018 ARM Limited. |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [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 | */ |
| 24 | #ifndef __ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYEROUTPUTSTAGEKERNEL_H__ |
| 25 | #define __ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYEROUTPUTSTAGEKERNEL_H__ |
| 26 | |
| 27 | #include "arm_compute/core/NEON/INEKernel.h" |
| 28 | |
| 29 | namespace arm_compute |
| 30 | { |
| 31 | class ITensor; |
| 32 | /** NEON kernel to accumulate the biases, if provided, or downscale in case of quantized input. |
| 33 | * |
| 34 | * @note We assume bias to be shared |
| 35 | */ |
| 36 | class NEDirectConvolutionLayerOutputStageKernel : public INEKernel |
| 37 | { |
| 38 | public: |
Anthony Barbier | e8a4983 | 2018-01-18 10:04:05 +0000 | [diff] [blame] | 39 | const char *name() const override |
| 40 | { |
| 41 | return "NEDirectConvolutionLayerOutputStageKernel"; |
| 42 | } |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 43 | /** Default constructor */ |
| 44 | NEDirectConvolutionLayerOutputStageKernel(); |
| 45 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 46 | NEDirectConvolutionLayerOutputStageKernel(const NEDirectConvolutionLayerOutputStageKernel &) = delete; |
| 47 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 48 | NEDirectConvolutionLayerOutputStageKernel &operator=(const NEDirectConvolutionLayerOutputStageKernel &) = delete; |
| 49 | /** Allow instances of this class to be moved */ |
| 50 | NEDirectConvolutionLayerOutputStageKernel(NEDirectConvolutionLayerOutputStageKernel &&) = default; |
| 51 | /** Allow instances of this class to be moved */ |
| 52 | NEDirectConvolutionLayerOutputStageKernel &operator=(NEDirectConvolutionLayerOutputStageKernel &&) = default; |
| 53 | /** Default destructor */ |
| 54 | ~NEDirectConvolutionLayerOutputStageKernel() = default; |
| 55 | /** Set the accumulate buffer and the biases of the kernel. |
| 56 | * |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 57 | * @param[in, out] input Input to add the bias to. If @p output is not specified then accumulation is done in-place. |
Vidhya Sudhan Loganathan | f4cb81b | 2018-07-04 15:13:14 +0100 | [diff] [blame] | 58 | * Data type supported: F16/F32 |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 59 | * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p input |
| 60 | * @param[out] output (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) |
Vidhya Sudhan Loganathan | 7485d5a | 2018-07-04 09:34:00 +0100 | [diff] [blame] | 61 | * Data type supported: F16/F32 |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 62 | * @param[in] result_fixedpoint_multiplier (Optional)Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add |
| 63 | * @param[in] result_shift (Optional)Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication |
| 64 | * @param[in] result_offset_after_shift (Optional)Offset to be applied to result before converting it back to QASYMM8 |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 65 | */ |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 66 | void configure(ITensor *input, const ITensor *bias = nullptr, ITensor *output = nullptr, |
| 67 | int result_fixedpoint_multiplier = 0, int result_shift = 0, int result_offset_after_shift = 0); |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 68 | /** Static function to check if given info will lead to a valid configuration of @ref NEDirectConvolutionLayerOutputStageKernel |
| 69 | * |
| 70 | * @param[in] input Input to add the bias to. If @p output is not specified then accumulation is done in-place. |
Vidhya Sudhan Loganathan | f4cb81b | 2018-07-04 15:13:14 +0100 | [diff] [blame] | 71 | * Data type supported: F16/F32 |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 72 | * @param[in] bias (Optional) The shared bias tensor to add. It must be 1D Tensor. Data type supported: Same as @p input |
| 73 | * @param[in] output (Optional) If the output tensor is specified the accumulation is done out-of-place. (Defaults to nullptr) |
Vidhya Sudhan Loganathan | 7485d5a | 2018-07-04 09:34:00 +0100 | [diff] [blame] | 74 | * Data type supported: F16/F32 |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 75 | * @return a status |
| 76 | */ |
| 77 | static Status validate(const ITensorInfo *input, const ITensorInfo *bias = nullptr, const ITensorInfo *output = nullptr); |
| 78 | |
| 79 | // Inherited methods overridden: |
| 80 | void run(const Window &window, const ThreadInfo &info) override; |
| 81 | |
| 82 | private: |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 83 | using OutputStageKernel = void(ITensor *input, const ITensor *bias, const Window &window, ITensor *output, |
| 84 | int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift); |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 85 | |
| 86 | private: |
| 87 | OutputStageKernel *_func; |
| 88 | ITensor *_input; |
| 89 | const ITensor *_bias; |
| 90 | ITensor *_output; |
Georgios Pinitas | f72f936 | 2018-01-12 16:29:45 +0000 | [diff] [blame] | 91 | int _result_fixedpoint_multiplier; |
| 92 | int _result_shift; |
| 93 | int _result_offset_after_shift; |
Michalis Spyrou | b91e34c | 2017-12-20 15:50:55 +0000 | [diff] [blame] | 94 | }; |
| 95 | } // namespace arm_compute |
| 96 | #endif /*__ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYEROUTPUTSTAGEKERNEL_H__ */ |