arovir01 | 9e53a35 | 2018-08-31 15:26:35 +0100 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2017 Arm Ltd. All rights reserved. |
David Beck | ecb56cd | 2018-09-05 12:52:57 +0100 | [diff] [blame] | 3 | // SPDX-License-Identifier: MIT |
arovir01 | 9e53a35 | 2018-08-31 15:26:35 +0100 | [diff] [blame] | 4 | // |
| 5 | |
| 6 | #include "NeonLstmFloatWorkload.hpp" |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 7 | #include "NeonWorkloadUtils.hpp" |
| 8 | |
Aron Virginas-Tar | c9cc804 | 2018-11-01 16:15:57 +0000 | [diff] [blame] | 9 | #include "backendsCommon/CpuTensorHandle.hpp" |
| 10 | #include "aclCommon/ArmComputeTensorUtils.hpp" |
| 11 | #include "neon/NeonTensorHandle.hpp" |
arovir01 | 9e53a35 | 2018-08-31 15:26:35 +0100 | [diff] [blame] | 12 | |
| 13 | namespace armnn |
| 14 | { |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 15 | using namespace armcomputetensorutils; |
| 16 | |
| 17 | NeonLstmFloatWorkload::NeonLstmFloatWorkload(const LstmQueueDescriptor &descriptor, const WorkloadInfo &info) |
arovir01 | 9e53a35 | 2018-08-31 15:26:35 +0100 | [diff] [blame] | 18 | : FloatWorkload<LstmQueueDescriptor>(descriptor, info) |
| 19 | { |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 20 | arm_compute::LSTMParams<arm_compute::ITensor> lstm_param; |
| 21 | |
| 22 | // Basic parameters |
| 23 | m_InputToForgetWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 24 | BuildArmComputeTensor(*m_InputToForgetWeightsTensor, m_Data.m_InputToForgetWeights->GetTensorInfo()); |
| 25 | |
| 26 | m_InputToCellWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 27 | BuildArmComputeTensor(*m_InputToCellWeightsTensor, m_Data.m_InputToCellWeights->GetTensorInfo()); |
| 28 | |
| 29 | m_InputToOutputWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 30 | BuildArmComputeTensor(*m_InputToOutputWeightsTensor, m_Data.m_InputToOutputWeights->GetTensorInfo()); |
| 31 | |
| 32 | m_RecurrentToForgetWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 33 | BuildArmComputeTensor(*m_RecurrentToForgetWeightsTensor, m_Data.m_RecurrentToForgetWeights->GetTensorInfo()); |
| 34 | |
| 35 | m_RecurrentToCellWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 36 | BuildArmComputeTensor(*m_RecurrentToCellWeightsTensor, m_Data.m_RecurrentToCellWeights->GetTensorInfo()); |
| 37 | |
| 38 | m_RecurrentToOutputWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 39 | BuildArmComputeTensor(*m_RecurrentToOutputWeightsTensor, m_Data.m_RecurrentToOutputWeights->GetTensorInfo()); |
| 40 | |
| 41 | m_ForgetGateBiasTensor = std::make_unique<arm_compute::Tensor>(); |
| 42 | BuildArmComputeTensor(*m_ForgetGateBiasTensor, m_Data.m_ForgetGateBias->GetTensorInfo()); |
| 43 | |
| 44 | m_CellBiasTensor = std::make_unique<arm_compute::Tensor>(); |
| 45 | BuildArmComputeTensor(*m_CellBiasTensor, m_Data.m_CellBias->GetTensorInfo()); |
| 46 | |
| 47 | m_OutputGateBiasTensor = std::make_unique<arm_compute::Tensor>(); |
| 48 | BuildArmComputeTensor(*m_OutputGateBiasTensor, m_Data.m_OutputGateBias->GetTensorInfo()); |
| 49 | |
| 50 | // for future reference: check the AndroidNN API for the logic here |
| 51 | if (!m_Data.m_Parameters.m_CifgEnabled) |
| 52 | { |
| 53 | m_InputToInputWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 54 | BuildArmComputeTensor(*m_InputToInputWeightsTensor, m_Data.m_InputToInputWeights->GetTensorInfo()); |
| 55 | |
| 56 | m_RecurrentToInputWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 57 | BuildArmComputeTensor(*m_RecurrentToInputWeightsTensor, m_Data.m_RecurrentToInputWeights->GetTensorInfo()); |
| 58 | |
| 59 | m_CellToInputWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 60 | if (m_Data.m_CellToInputWeights != nullptr) |
| 61 | { |
| 62 | BuildArmComputeTensor(*m_CellToInputWeightsTensor, m_Data.m_CellToInputWeights->GetTensorInfo()); |
| 63 | } |
| 64 | |
| 65 | m_InputGateBiasTensor = std::make_unique<arm_compute::Tensor>(); |
| 66 | BuildArmComputeTensor(*m_InputGateBiasTensor, m_Data.m_InputGateBias->GetTensorInfo()); |
| 67 | |
| 68 | lstm_param.set_cifg_params(m_InputToInputWeightsTensor.get(), |
| 69 | m_RecurrentToInputWeightsTensor.get(), |
| 70 | m_Data.m_CellToInputWeights != nullptr ? m_CellToInputWeightsTensor.get() : nullptr, |
| 71 | m_InputGateBiasTensor.get()); |
| 72 | } |
| 73 | |
| 74 | if (m_Data.m_Parameters.m_ProjectionEnabled) |
| 75 | { |
| 76 | m_ProjectionWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 77 | BuildArmComputeTensor(*m_ProjectionWeightsTensor, m_Data.m_ProjectionWeights->GetTensorInfo()); |
| 78 | |
| 79 | m_ProjectionBiasTensor = std::make_unique<arm_compute::Tensor>(); |
| 80 | if (m_Data.m_ProjectionBias != nullptr) |
| 81 | { |
| 82 | BuildArmComputeTensor(*m_ProjectionBiasTensor, m_Data.m_ProjectionBias->GetTensorInfo()); |
| 83 | } |
| 84 | |
| 85 | lstm_param.set_projection_params(m_ProjectionWeightsTensor.get(), |
| 86 | m_Data.m_ProjectionBias != nullptr ? m_ProjectionBiasTensor.get() : nullptr); |
| 87 | } |
| 88 | |
| 89 | if (m_Data.m_Parameters.m_PeepholeEnabled) |
| 90 | { |
| 91 | m_CellToForgetWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 92 | BuildArmComputeTensor(*m_CellToForgetWeightsTensor, m_Data.m_CellToForgetWeights->GetTensorInfo()); |
| 93 | |
| 94 | m_CellToOutputWeightsTensor = std::make_unique<arm_compute::Tensor>(); |
| 95 | BuildArmComputeTensor(*m_CellToOutputWeightsTensor, m_Data.m_CellToOutputWeights->GetTensorInfo()); |
| 96 | |
| 97 | lstm_param.set_peephole_params(m_CellToForgetWeightsTensor.get(), m_CellToOutputWeightsTensor.get()); |
| 98 | } |
| 99 | |
Derek Lamberti | c81855f | 2019-06-13 17:34:19 +0100 | [diff] [blame] | 100 | const arm_compute::ITensor& input = static_cast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor(); |
| 101 | const arm_compute::ITensor& output_state_in = static_cast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor(); |
| 102 | const arm_compute::ITensor& cell_state_in = static_cast<IAclTensorHandle*>(m_Data.m_Inputs[2])->GetTensor(); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 103 | |
Derek Lamberti | c81855f | 2019-06-13 17:34:19 +0100 | [diff] [blame] | 104 | arm_compute::ITensor& output_state_out = static_cast<IAclTensorHandle*>(m_Data.m_Outputs[1])->GetTensor(); |
| 105 | arm_compute::ITensor& cell_state_out = static_cast<IAclTensorHandle*>(m_Data.m_Outputs[2])->GetTensor(); |
| 106 | arm_compute::ITensor& output = static_cast<IAclTensorHandle*>(m_Data.m_Outputs[3])->GetTensor(); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 107 | |
| 108 | // Get the batch_size and the num_units from the cellStateIn dimensions |
| 109 | const TensorInfo& inputTensorInfo = info.m_InputTensorInfos[2]; |
| 110 | const unsigned int batch_size = boost::numeric_cast<unsigned int>(inputTensorInfo.GetShape()[0]); |
| 111 | const unsigned int num_units = boost::numeric_cast<unsigned int>(inputTensorInfo.GetShape()[1]); |
| 112 | |
| 113 | m_ScratchBuffer = std::make_unique<arm_compute::Tensor>(); |
| 114 | if (m_Data.m_Parameters.m_CifgEnabled) |
| 115 | { |
| 116 | // 2D tensor with dimensions [num_units * 4, batch_size] with CIFG |
Matteo Martincigh | a65b7ae | 2018-11-14 12:39:55 +0000 | [diff] [blame] | 117 | armnn::TensorInfo scratchBuffer1({ batch_size, num_units * 3 }, DataType::Float32); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 118 | BuildArmComputeTensor(*m_ScratchBuffer, scratchBuffer1); |
| 119 | } |
| 120 | else |
| 121 | { |
| 122 | // scratch_buffer [num_units * 3, batch_size] without CIFG |
Matteo Martincigh | a65b7ae | 2018-11-14 12:39:55 +0000 | [diff] [blame] | 123 | armnn::TensorInfo scratchBuffer2({ batch_size, num_units * 4 }, DataType::Float32); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 124 | BuildArmComputeTensor(*m_ScratchBuffer, scratchBuffer2); |
| 125 | } |
| 126 | |
| 127 | float cell_threshold = m_Data.m_Parameters.m_ClippingThresCell; |
| 128 | float projection_threshold = m_Data.m_Parameters.m_ClippingThresProj; |
| 129 | |
| 130 | // for preparing the object for the class ActivationLayerInfo, we need to consider 5 situations |
| 131 | arm_compute::ActivationLayerInfo activationLayerInfo; |
| 132 | if (m_Data.m_Parameters.m_ActivationFunc == 0) |
| 133 | { |
| 134 | // no activation, do nothing |
| 135 | } |
| 136 | else if (m_Data.m_Parameters.m_ActivationFunc == 1) |
| 137 | { |
| 138 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 139 | arm_compute::ActivationLayerInfo::ActivationFunction::RELU); |
| 140 | } |
| 141 | else if (m_Data.m_Parameters.m_ActivationFunc == 3) |
| 142 | { |
| 143 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 144 | arm_compute::ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.0); |
| 145 | } |
| 146 | else if (m_Data.m_Parameters.m_ActivationFunc == 4) |
| 147 | { |
| 148 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 149 | arm_compute::ActivationLayerInfo::ActivationFunction::TANH, 1.0, 1.0); |
| 150 | } |
| 151 | else if (m_Data.m_Parameters.m_ActivationFunc == 6) |
| 152 | { |
| 153 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 154 | arm_compute::ActivationLayerInfo::ActivationFunction::LOGISTIC); |
| 155 | } |
| 156 | else |
| 157 | { |
| 158 | throw armnn::Exception("Wrong Type of Activation Function!"); |
| 159 | } |
| 160 | |
| 161 | |
| 162 | m_LstmLayer.configure(&input, m_InputToForgetWeightsTensor.get(), m_InputToCellWeightsTensor.get(), |
| 163 | m_InputToOutputWeightsTensor.get(), m_RecurrentToForgetWeightsTensor.get(), |
| 164 | m_RecurrentToCellWeightsTensor.get(), m_RecurrentToOutputWeightsTensor.get(), |
| 165 | m_ForgetGateBiasTensor.get(), m_CellBiasTensor.get(), m_OutputGateBiasTensor.get(), |
| 166 | &output_state_in, &cell_state_in, m_ScratchBuffer.get(), &output_state_out, |
| 167 | &cell_state_out, &output, lstm_param, activationLayerInfo, |
| 168 | cell_threshold, projection_threshold); |
| 169 | |
| 170 | armcomputetensorutils::InitialiseArmComputeTensorEmpty(*m_ScratchBuffer); |
| 171 | |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 172 | InitializeArmComputeTensorData(*m_InputToForgetWeightsTensor, |
| 173 | m_Data.m_InputToForgetWeights); |
| 174 | InitializeArmComputeTensorData(*m_InputToCellWeightsTensor, |
| 175 | m_Data.m_InputToCellWeights); |
| 176 | InitializeArmComputeTensorData(*m_InputToOutputWeightsTensor, |
| 177 | m_Data.m_InputToOutputWeights); |
| 178 | InitializeArmComputeTensorData(*m_RecurrentToForgetWeightsTensor, |
| 179 | m_Data.m_RecurrentToForgetWeights); |
| 180 | InitializeArmComputeTensorData(*m_RecurrentToCellWeightsTensor, |
| 181 | m_Data.m_RecurrentToCellWeights); |
| 182 | InitializeArmComputeTensorData(*m_RecurrentToOutputWeightsTensor, |
| 183 | m_Data.m_RecurrentToOutputWeights); |
| 184 | InitializeArmComputeTensorData(*m_ForgetGateBiasTensor, |
| 185 | m_Data.m_ForgetGateBias); |
| 186 | InitializeArmComputeTensorData(*m_CellBiasTensor, |
| 187 | m_Data.m_CellBias); |
| 188 | InitializeArmComputeTensorData(*m_OutputGateBiasTensor, |
| 189 | m_Data.m_OutputGateBias); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 190 | |
| 191 | if (!m_Data.m_Parameters.m_CifgEnabled) |
| 192 | { |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 193 | InitializeArmComputeTensorData(*m_InputToInputWeightsTensor, |
| 194 | m_Data.m_InputToInputWeights); |
| 195 | InitializeArmComputeTensorData(*m_RecurrentToInputWeightsTensor, |
| 196 | m_Data.m_RecurrentToInputWeights); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 197 | if (m_Data.m_CellToInputWeights != nullptr) |
| 198 | { |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 199 | InitializeArmComputeTensorData(*m_CellToInputWeightsTensor, |
| 200 | m_Data.m_CellToInputWeights); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 201 | } |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 202 | InitializeArmComputeTensorData(*m_InputGateBiasTensor, |
| 203 | m_Data.m_InputGateBias); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 204 | } |
| 205 | |
| 206 | if (m_Data.m_Parameters.m_ProjectionEnabled) |
| 207 | { |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 208 | InitializeArmComputeTensorData(*m_ProjectionWeightsTensor, |
| 209 | m_Data.m_ProjectionWeights); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 210 | if (m_Data.m_ProjectionBias != nullptr) |
| 211 | { |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 212 | InitializeArmComputeTensorData(*m_ProjectionBiasTensor, |
| 213 | m_Data.m_ProjectionBias); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 214 | } |
| 215 | } |
| 216 | |
| 217 | if (m_Data.m_Parameters.m_PeepholeEnabled) |
| 218 | { |
Nattapat Chaimanowong | 177d8d2 | 2018-10-16 13:21:27 +0100 | [diff] [blame] | 219 | InitializeArmComputeTensorData(*m_CellToForgetWeightsTensor, |
| 220 | m_Data.m_CellToForgetWeights); |
| 221 | InitializeArmComputeTensorData(*m_CellToOutputWeightsTensor, |
| 222 | m_Data.m_CellToOutputWeights); |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 223 | } |
| 224 | |
| 225 | // Force Compute Library to perform the necessary copying and reshaping, after which |
| 226 | // delete all the input tensors that will no longer be needed |
| 227 | m_LstmLayer.prepare(); |
| 228 | FreeUnusedTensors(); |
arovir01 | 9e53a35 | 2018-08-31 15:26:35 +0100 | [diff] [blame] | 229 | } |
| 230 | |
| 231 | void NeonLstmFloatWorkload::Execute() const |
| 232 | { |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 233 | m_LstmLayer.run(); |
arovir01 | 9e53a35 | 2018-08-31 15:26:35 +0100 | [diff] [blame] | 234 | } |
| 235 | |
Les Bell | de9011b | 2018-10-03 10:37:52 +0100 | [diff] [blame] | 236 | arm_compute::Status NeonLstmFloatWorkloadValidate(const TensorInfo& input, |
| 237 | const TensorInfo& outputStateIn, |
| 238 | const TensorInfo& cellStateIn, |
| 239 | const TensorInfo& scratchBuffer, |
| 240 | const TensorInfo& outputStateOut, |
| 241 | const TensorInfo& cellStateOut, |
| 242 | const TensorInfo& output, |
| 243 | const LstmDescriptor& descriptor, |
| 244 | const TensorInfo& inputToForgetWeights, |
| 245 | const TensorInfo& inputToCellWeights, |
| 246 | const TensorInfo& inputToOutputWeights, |
| 247 | const TensorInfo& recurrentToForgetWeights, |
| 248 | const TensorInfo& recurrentToCellWeights, |
| 249 | const TensorInfo& recurrentToOutputWeights, |
| 250 | const TensorInfo& forgetGateBias, |
| 251 | const TensorInfo& cellBias, |
| 252 | const TensorInfo& outputGateBias, |
| 253 | const TensorInfo* inputToInputWeights, |
| 254 | const TensorInfo* recurrentToInputWeights, |
| 255 | const TensorInfo* cellToInputWeights, |
| 256 | const TensorInfo* inputGateBias, |
| 257 | const TensorInfo* projectionWeights, |
| 258 | const TensorInfo* projectionBias, |
| 259 | const TensorInfo* cellToForgetWeights, |
| 260 | const TensorInfo* cellToOutputWeights) |
| 261 | { |
| 262 | arm_compute::LSTMParams<arm_compute::ITensorInfo> lstm_params_info; |
| 263 | |
| 264 | // The inputs and the outputs |
| 265 | const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input); |
| 266 | const arm_compute::TensorInfo aclOutputStateInInfo = BuildArmComputeTensorInfo(outputStateIn); |
| 267 | const arm_compute::TensorInfo aclCellStateInInfo = BuildArmComputeTensorInfo(cellStateIn); |
| 268 | const arm_compute::TensorInfo aclScratchBufferInfo = BuildArmComputeTensorInfo(scratchBuffer); |
| 269 | const arm_compute::TensorInfo aclOutputStateOutInfo = BuildArmComputeTensorInfo(outputStateOut); |
| 270 | const arm_compute::TensorInfo aclCellStateOutInfo = BuildArmComputeTensorInfo(cellStateOut); |
| 271 | const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output); |
| 272 | |
| 273 | // Basic parameters |
| 274 | const arm_compute::TensorInfo aclInputToForgetWeightsInfo = BuildArmComputeTensorInfo(inputToForgetWeights); |
| 275 | const arm_compute::TensorInfo aclInputToCellWeightsInfo = BuildArmComputeTensorInfo(inputToCellWeights); |
| 276 | const arm_compute::TensorInfo aclInputToOutputWeightsInfo = BuildArmComputeTensorInfo(inputToOutputWeights); |
| 277 | const arm_compute::TensorInfo aclRecurrentToForgetWeightsInfo |
| 278 | = BuildArmComputeTensorInfo(recurrentToForgetWeights); |
| 279 | const arm_compute::TensorInfo aclRecurrentToCellWeightsInfo |
| 280 | = BuildArmComputeTensorInfo(recurrentToCellWeights); |
| 281 | const arm_compute::TensorInfo aclRecurrentToOutputWeightsInfo |
| 282 | = BuildArmComputeTensorInfo(recurrentToOutputWeights); |
| 283 | const arm_compute::TensorInfo aclForgetGateBiasInfo = BuildArmComputeTensorInfo(forgetGateBias); |
| 284 | const arm_compute::TensorInfo aclCellBiasInfo = BuildArmComputeTensorInfo(cellBias); |
| 285 | const arm_compute::TensorInfo aclOutputGateBiasInfo = BuildArmComputeTensorInfo(outputGateBias); |
| 286 | |
| 287 | arm_compute::TensorInfo aclInputToInputWeightsInfo; |
| 288 | arm_compute::TensorInfo aclRecurrentToInputWeightsInfo; |
| 289 | arm_compute::TensorInfo aclCellToInputWeightsInfo; |
| 290 | arm_compute::TensorInfo aclInputGateBiasInfo; |
| 291 | arm_compute::TensorInfo aclProjectionWeightsInfo; |
| 292 | arm_compute::TensorInfo aclProjectionBiasInfo; |
| 293 | arm_compute::TensorInfo aclCellToForgetWeightsInfo; |
| 294 | arm_compute::TensorInfo aclCellToOutputWeightsInfo; |
| 295 | |
| 296 | if (!descriptor.m_CifgEnabled) |
| 297 | { |
| 298 | armnn::TensorInfo inputToInputWInfo = *inputToInputWeights; |
| 299 | aclInputToInputWeightsInfo = BuildArmComputeTensorInfo(inputToInputWInfo); |
| 300 | armnn::TensorInfo recurrentToInputWInfo = *recurrentToInputWeights; |
| 301 | aclRecurrentToInputWeightsInfo = BuildArmComputeTensorInfo(recurrentToInputWInfo); |
| 302 | |
| 303 | if (cellToInputWeights != nullptr) |
| 304 | { |
| 305 | armnn::TensorInfo cellToInputWInfo = *cellToInputWeights; |
| 306 | aclCellToInputWeightsInfo = BuildArmComputeTensorInfo(cellToInputWInfo); |
| 307 | } |
| 308 | armnn::TensorInfo inputGateBiasInfo = *inputGateBias; |
| 309 | aclInputGateBiasInfo = BuildArmComputeTensorInfo(inputGateBiasInfo); |
| 310 | lstm_params_info.set_cifg_params(&aclInputToInputWeightsInfo, &aclRecurrentToInputWeightsInfo, |
| 311 | cellToInputWeights != nullptr ? &aclCellToInputWeightsInfo: nullptr, |
| 312 | &aclInputGateBiasInfo); |
| 313 | } |
| 314 | |
| 315 | if (descriptor.m_ProjectionEnabled) |
| 316 | { |
| 317 | const armnn::TensorInfo& projectionWInfo = *projectionWeights; |
| 318 | aclProjectionWeightsInfo = BuildArmComputeTensorInfo(projectionWInfo); |
| 319 | |
| 320 | if (projectionBias != nullptr) |
| 321 | { |
| 322 | const armnn::TensorInfo& projectionBiasInfo = *projectionBias; |
| 323 | aclProjectionBiasInfo = BuildArmComputeTensorInfo(projectionBiasInfo); |
| 324 | } |
| 325 | lstm_params_info.set_projection_params(&aclProjectionWeightsInfo, |
| 326 | projectionBias != nullptr ? &aclProjectionBiasInfo: nullptr); |
| 327 | } |
| 328 | |
| 329 | if (descriptor.m_PeepholeEnabled) |
| 330 | { |
| 331 | const armnn::TensorInfo& cellToForgetWInfo = *cellToForgetWeights; |
| 332 | aclCellToForgetWeightsInfo = BuildArmComputeTensorInfo(cellToForgetWInfo); |
| 333 | const armnn::TensorInfo& cellToOutputWInfo = *cellToOutputWeights; |
| 334 | aclCellToOutputWeightsInfo = BuildArmComputeTensorInfo(cellToOutputWInfo); |
| 335 | lstm_params_info.set_peephole_params(&aclCellToForgetWeightsInfo, &aclCellToOutputWeightsInfo); |
| 336 | } |
| 337 | |
| 338 | float cell_threshold = descriptor.m_ClippingThresCell; |
| 339 | float projection_threshold = descriptor.m_ClippingThresProj; |
| 340 | |
| 341 | // for preparing the object for the class ActivationLayerInfo, we need to consider 5 situations |
| 342 | arm_compute::ActivationLayerInfo activationLayerInfo; |
| 343 | switch (descriptor.m_ActivationFunc) |
| 344 | { |
| 345 | case 0: |
| 346 | // no activation, do nothing |
| 347 | break; |
| 348 | case 1: |
| 349 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 350 | arm_compute::ActivationLayerInfo::ActivationFunction::RELU); |
| 351 | break; |
| 352 | case 3: |
| 353 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 354 | arm_compute::ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.0); |
| 355 | break; |
| 356 | case 4: |
| 357 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 358 | arm_compute::ActivationLayerInfo::ActivationFunction::TANH, 1.0, 1.0); |
| 359 | break; |
| 360 | case 6: |
| 361 | activationLayerInfo = arm_compute::ActivationLayerInfo( |
| 362 | arm_compute::ActivationLayerInfo::ActivationFunction::LOGISTIC); |
| 363 | break; |
| 364 | default: |
| 365 | throw armnn::Exception("Wrong Type of Activation Function!"); |
| 366 | } |
| 367 | |
| 368 | return arm_compute::NELSTMLayer::validate(&aclInputInfo, |
| 369 | &aclInputToForgetWeightsInfo, |
| 370 | &aclInputToCellWeightsInfo, |
| 371 | &aclInputToOutputWeightsInfo, |
| 372 | &aclRecurrentToForgetWeightsInfo, |
| 373 | &aclRecurrentToCellWeightsInfo, |
| 374 | &aclRecurrentToOutputWeightsInfo, |
| 375 | &aclForgetGateBiasInfo, |
| 376 | &aclCellBiasInfo, |
| 377 | &aclOutputGateBiasInfo, |
| 378 | &aclOutputStateInInfo, |
| 379 | &aclCellStateInInfo, |
| 380 | &aclScratchBufferInfo, |
| 381 | &aclOutputStateOutInfo, |
| 382 | &aclCellStateOutInfo, |
| 383 | &aclOutputInfo, |
| 384 | lstm_params_info, |
| 385 | activationLayerInfo, |
| 386 | cell_threshold, |
| 387 | projection_threshold); |
| 388 | } |
| 389 | |
| 390 | void NeonLstmFloatWorkload::FreeUnusedTensors() |
| 391 | { |
| 392 | FreeTensorIfUnused(m_InputToInputWeightsTensor); |
| 393 | FreeTensorIfUnused(m_InputToForgetWeightsTensor); |
| 394 | FreeTensorIfUnused(m_InputToCellWeightsTensor); |
| 395 | FreeTensorIfUnused(m_InputToOutputWeightsTensor); |
| 396 | FreeTensorIfUnused(m_RecurrentToInputWeightsTensor); |
| 397 | FreeTensorIfUnused(m_RecurrentToForgetWeightsTensor); |
| 398 | FreeTensorIfUnused(m_RecurrentToCellWeightsTensor); |
| 399 | FreeTensorIfUnused(m_RecurrentToOutputWeightsTensor); |
| 400 | FreeTensorIfUnused(m_CellToInputWeightsTensor); |
| 401 | FreeTensorIfUnused(m_CellToForgetWeightsTensor); |
| 402 | FreeTensorIfUnused(m_CellToOutputWeightsTensor); |
| 403 | FreeTensorIfUnused(m_InputGateBiasTensor); |
| 404 | FreeTensorIfUnused(m_ForgetGateBiasTensor); |
| 405 | FreeTensorIfUnused(m_CellBiasTensor); |
| 406 | FreeTensorIfUnused(m_OutputGateBiasTensor); |
| 407 | FreeTensorIfUnused(m_ProjectionWeightsTensor); |
| 408 | FreeTensorIfUnused(m_ProjectionBiasTensor); |
| 409 | FreeTensorIfUnused(m_ScratchBuffer); |
| 410 | } |
| 411 | |
| 412 | } //namespace armnn |