Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 1 | /* |
Sheri Zhang | ac6499a | 2021-02-10 15:32:38 +0000 | [diff] [blame] | 2 | * Copyright (c) 2019-2021 Arm Limited. |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +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 | */ |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_NELSTMLAYERQUANTIZED_H |
| 25 | #define ARM_COMPUTE_NELSTMLAYERQUANTIZED_H |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 26 | |
| 27 | #include "arm_compute/core/Types.h" |
| 28 | #include "arm_compute/runtime/NEON/functions/NEActivationLayer.h" |
| 29 | #include "arm_compute/runtime/NEON/functions/NEArithmeticAddition.h" |
| 30 | #include "arm_compute/runtime/NEON/functions/NEConcatenateLayer.h" |
| 31 | #include "arm_compute/runtime/NEON/functions/NEDequantizationLayer.h" |
| 32 | #include "arm_compute/runtime/NEON/functions/NEElementwiseOperations.h" |
| 33 | #include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h" |
| 34 | #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h" |
| 35 | #include "arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h" |
| 36 | #include "arm_compute/runtime/NEON/functions/NEPixelWiseMultiplication.h" |
| 37 | #include "arm_compute/runtime/NEON/functions/NEQuantizationLayer.h" |
| 38 | #include "arm_compute/runtime/NEON/functions/NESlice.h" |
| 39 | #include "arm_compute/runtime/NEON/functions/NETranspose.h" |
| 40 | |
| 41 | #include "arm_compute/runtime/common/LSTMParams.h" |
| 42 | |
| 43 | namespace arm_compute |
| 44 | { |
| 45 | // Forward declarations |
| 46 | class ITensor; |
| 47 | |
| 48 | /** Basic function to run @ref NELSTMLayerQuantized |
| 49 | * |
Michele Di Giorgio | 33f41fa | 2021-03-09 14:09:08 +0000 | [diff] [blame] | 50 | * This function calls the following functions/kernels: |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 51 | * |
| 52 | * -# @ref NEGEMMLowpMatrixMultiplyCore Quantized matrix multiplication core. Accumulators are 32-bit integers |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 53 | * -# @ref NEGEMMLowpOutputStage Convert 32-bit integers into QSYMM16 |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 54 | * -# @ref NETranspose Matrix transpose |
| 55 | * -# @ref NEConcatenateLayer Tensor concatenation |
Manuel Bottini | 10c53f1 | 2019-07-17 16:11:53 +0100 | [diff] [blame] | 56 | * -# @ref NEActivationLayer Activation functions (tanh and logistic) |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 57 | * -# @ref NEArithmeticAddition Elementwise addition |
| 58 | * -# @ref NEPixelWiseMultiplication Elementwise multiplication |
| 59 | * -# @ref NESlice Tensor slicing |
| 60 | * -# @ref NEDequantizationLayer Dequantize into float |
| 61 | * -# @ref NEQuantizationLayer Quantize from float |
| 62 | * */ |
| 63 | class NELSTMLayerQuantized : public IFunction |
| 64 | { |
| 65 | public: |
| 66 | /** Default constructor */ |
| 67 | NELSTMLayerQuantized(std::shared_ptr<IMemoryManager> memory_manager = nullptr); |
| 68 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 69 | NELSTMLayerQuantized(const NELSTMLayerQuantized &) = delete; |
Michalis Spyrou | 770dfeb | 2020-11-04 18:55:34 +0000 | [diff] [blame] | 70 | /** Prevent instances of this class from being moved (As this class contains pointers) */ |
| 71 | NELSTMLayerQuantized(NELSTMLayerQuantized &&) = delete; |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 72 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 73 | NELSTMLayerQuantized &operator=(const NELSTMLayerQuantized &) = delete; |
Michalis Spyrou | 770dfeb | 2020-11-04 18:55:34 +0000 | [diff] [blame] | 74 | /** Prevent instances of this class from being moved (As this class contains pointers) */ |
| 75 | NELSTMLayerQuantized &operator=(NELSTMLayerQuantized &&) = delete; |
Michalis Spyrou | ebcebf1 | 2020-10-21 00:04:14 +0100 | [diff] [blame] | 76 | /** Default destructor */ |
| 77 | ~NELSTMLayerQuantized(); |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 78 | /** Initialize function's tensors. |
| 79 | * |
Teresa Charlin | 6268742 | 2021-04-28 10:58:49 +0100 | [diff] [blame] | 80 | * Valid data layouts: |
| 81 | * - All |
| 82 | * |
| 83 | * Valid data type configurations: |
| 84 | * |src0 - src8 |src9 - src12 |src13 |src14 |dst0 |dst1 | |
| 85 | * |:-----------|:------------|:-------|:------|:------|:------| |
| 86 | * |QASYMM8 |S32 |QSYMM16 |QASYMM8|QSYMM16|QASYMM8| |
| 87 | * |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 88 | * @param[in] input Source tensor. Input is a 2D tensor with dimensions [input_size, batch_size]. Data types supported: QASYMM8. |
| 89 | * @param[in] input_to_input_weights 2D weights tensor with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 90 | * @param[in] input_to_forget_weights 2D weights tensor with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 91 | * @param[in] input_to_cell_weights 2D weights tensor with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 92 | * @param[in] input_to_output_weights 2D weights tensor with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 93 | * @param[in] recurrent_to_input_weights 2D weights tensor with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 94 | * @param[in] recurrent_to_forget_weights 2D weights tensor with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 95 | * @param[in] recurrent_to_cell_weights 2D weights tensor with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 96 | * @param[in] recurrent_to_output_weights 2D weights tensor with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 97 | * @param[in] input_gate_bias 1D weights tensor with dimensions [output_size]. Data type supported: S32. |
| 98 | * @param[in] forget_gate_bias 1D weights tensor with dimensions [output_size]. Data type supported: S32. |
| 99 | * @param[in] cell_bias 1D weights tensor with dimensions [output_size]. Data type supported: S32. |
| 100 | * @param[in] output_gate_bias 1D weights tensor with dimensions [output_size]. Data type supported: S32. |
| 101 | * @param[in] cell_state_in 2D tensor with dimensions [output_size, batch_size]. Data type supported: QSYMM16. |
| 102 | * @param[in] output_state_in 2D tensor with dimensions [output_size, batch_size]. Data type supported: Same as @p input. |
| 103 | * @param[out] cell_state_out Destination tensor. Output is a 2D tensor with dimensions [output_size, batch_size]. Data type supported: QSYMM16. |
| 104 | * @param[out] output_state_out Destination tensor. Output is a 2D tensor with dimensions [output_size, batch_size].Data types supported: Same as @p input. |
| 105 | */ |
| 106 | void configure(const ITensor *input, |
| 107 | const ITensor *input_to_input_weights, const ITensor *input_to_forget_weights, const ITensor *input_to_cell_weights, const ITensor *input_to_output_weights, |
| 108 | const ITensor *recurrent_to_input_weights, const ITensor *recurrent_to_forget_weights, const ITensor *recurrent_to_cell_weights, const ITensor *recurrent_to_output_weights, |
| 109 | const ITensor *input_gate_bias, const ITensor *forget_gate_bias, const ITensor *cell_bias, const ITensor *output_gate_bias, |
| 110 | ITensor *cell_state_in, const ITensor *output_state_in, |
| 111 | ITensor *cell_state_out, ITensor *output_state_out); |
| 112 | |
| 113 | /** Static function to check if given info will lead to a valid configuration of @ref NELSTMLayer |
| 114 | * |
| 115 | * @param[in] input Source tensor info. Input is a 2D tensor info with dimensions [input_size, batch_size]. Data types supported: QASYMM8. |
| 116 | * @param[in] input_to_input_weights 2D weights tensor info with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 117 | * @param[in] input_to_forget_weights 2D weights tensor info with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 118 | * @param[in] input_to_cell_weights 2D weights tensor info with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 119 | * @param[in] input_to_output_weights 2D weights tensor info with dimensions [input_size, output_size]. Data type supported: Same as @p input. |
| 120 | * @param[in] recurrent_to_input_weights 2D weights tensor info with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 121 | * @param[in] recurrent_to_forget_weights 2D weights tensor info with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 122 | * @param[in] recurrent_to_cell_weights 2D weights tensor info with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 123 | * @param[in] recurrent_to_output_weights 2D weights tensor info with dimensions [output_size, output_size]. Data type supported: Same as @p input. |
| 124 | * @param[in] input_gate_bias 1D weights tensor info with dimensions [output_size]. Data type supported: S32. |
| 125 | * @param[in] forget_gate_bias 1D weights tensor info with dimensions [output_size]. Data type supported: S32. |
| 126 | * @param[in] cell_bias 1D weights tensor info with dimensions [output_size]. Data type supported: S32. |
| 127 | * @param[in] output_gate_bias 1D weights tensor info with dimensions [output_size]. Data type supported: S32. |
| 128 | * @param[in] cell_state_in 2D tensor info with dimensions [output_size, batch_size]. Data type supported: QSYMM16. |
| 129 | * @param[in] output_state_in 2D tensor info with dimensions [output_size, batch_size]. Data type supported: Same as @p input. |
| 130 | * @param[out] cell_state_out Destination tensor info. Output is a 2D tensor info with dimensions [output_size, batch_size]. Data type supported: QSYMM16. |
| 131 | * @param[out] output_state_out Destination tensor info. Output is a 2D tensor info with dimensions [output_size, batch_size].Data types supported: Same as @p input. |
| 132 | * |
| 133 | * @return a status |
| 134 | */ |
| 135 | static Status validate(const ITensorInfo *input, |
| 136 | const ITensorInfo *input_to_input_weights, const ITensorInfo *input_to_forget_weights, const ITensorInfo *input_to_cell_weights, const ITensorInfo *input_to_output_weights, |
| 137 | const ITensorInfo *recurrent_to_input_weights, const ITensorInfo *recurrent_to_forget_weights, const ITensorInfo *recurrent_to_cell_weights, const ITensorInfo *recurrent_to_output_weights, |
| 138 | const ITensorInfo *input_gate_bias, const ITensorInfo *forget_gate_bias, const ITensorInfo *cell_bias, const ITensorInfo *output_gate_bias, |
| 139 | const ITensorInfo *cell_state_in, const ITensorInfo *output_state_in, |
| 140 | const ITensorInfo *cell_state_out, const ITensorInfo *output_state_out); |
| 141 | |
| 142 | // Inherited methods overridden: |
| 143 | void run() override; |
| 144 | void prepare() override; |
| 145 | |
| 146 | private: |
| 147 | MemoryGroup _memory_group; |
| 148 | |
| 149 | // Functions used |
Manuel Bottini | ae58bdf | 2021-06-17 17:18:45 +0100 | [diff] [blame] | 150 | NEGEMMLowpMatrixMultiplyCore _gemmlowp; |
| 151 | NEGEMMLowpOutputStage _output_stage; |
| 152 | NETranspose _transpose_weights; |
| 153 | NEConcatenateLayer _concat_input_weights; |
| 154 | NEConcatenateLayer _concat_recurrent_weights; |
| 155 | NEConcatenateLayer _concat_weights; |
| 156 | NEConcatenateLayer _concat_inputs; |
| 157 | NEConcatenateLayer _concat_bias; |
| 158 | NEActivationLayer _sigmoid_forget_gate; |
| 159 | NEActivationLayer _sigmoid_input_gate; |
| 160 | NEActivationLayer _sigmoid_output_gate; |
| 161 | NEActivationLayer _tanh_modulation_gate; |
| 162 | NEActivationLayer _tanh_output_state; |
| 163 | NEArithmeticAddition _add1; |
| 164 | NEArithmeticAddition _add2; |
| 165 | NEPixelWiseMultiplication _mul1; |
| 166 | NEPixelWiseMultiplication _mul2; |
| 167 | NEPixelWiseMultiplication _mul3; |
| 168 | NESlice _slice_input_tensor; |
| 169 | NESlice _slice_forget_tensor; |
| 170 | NESlice _slice_cell_tensor; |
| 171 | NESlice _slice_output_tensor; |
| 172 | NEDequantizationLayer _dequantize; |
| 173 | NEQuantizationLayer _quantize; |
Michalis Spyrou | ba27e44 | 2019-05-28 10:04:57 +0100 | [diff] [blame] | 174 | |
| 175 | // Tensor pointers |
| 176 | const ITensor *_input_to_input_weights; |
| 177 | const ITensor *_input_to_forget_weights; |
| 178 | const ITensor *_input_to_cell_weights; |
| 179 | const ITensor *_input_to_output_weights; |
| 180 | const ITensor *_recurrent_to_input_weights; |
| 181 | const ITensor *_recurrent_to_forget_weights; |
| 182 | const ITensor *_recurrent_to_cell_weights; |
| 183 | const ITensor *_recurrent_to_output_weights; |
| 184 | const ITensor *_input_gate_bias; |
| 185 | const ITensor *_forget_gate_bias; |
| 186 | const ITensor *_cell_bias; |
| 187 | const ITensor *_output_gate_bias; |
| 188 | |
| 189 | // Temporary tensors |
| 190 | Tensor _recurrent_weights; |
| 191 | Tensor _input_weights; |
| 192 | Tensor _weights; |
| 193 | Tensor _input; |
| 194 | Tensor _weights_transposed; |
| 195 | Tensor _output_highp; |
| 196 | Tensor _output_lowp; |
| 197 | Tensor _bias; |
| 198 | Tensor _forget_gate_input; |
| 199 | Tensor _input_gate_input; |
| 200 | Tensor _output_gate_input; |
| 201 | Tensor _input_modulation_gate_input; |
| 202 | Tensor _forget_gate_output; |
| 203 | Tensor _input_gate_output; |
| 204 | Tensor _output_gate_output; |
| 205 | Tensor _input_modulation_gate_output; |
| 206 | Tensor _cell_state1; |
| 207 | Tensor _cell_state2; |
| 208 | Tensor _output_state_tmp; |
| 209 | Tensor _output_state_out_symm; |
| 210 | Tensor _output_state_out_f32; |
| 211 | |
| 212 | bool _is_prepared; |
| 213 | }; |
| 214 | } // namespace arm_compute |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 215 | #endif /* ARM_COMPUTE_NELSTMLAYERQUANTIZED_H */ |