ArmNN
 22.05
Functions
LstmSerializationTests.cpp File Reference
#include "../Serializer.hpp"
#include "SerializerTestUtils.hpp"
#include <armnn/Descriptors.hpp>
#include <armnn/INetwork.hpp>
#include <armnn/IRuntime.hpp>
#include <armnnDeserializer/IDeserializer.hpp>
#include <armnn/utility/IgnoreUnused.hpp>
#include <armnn/LstmParams.hpp>
#include <armnn/QuantizedLstmParams.hpp>
#include <doctest/doctest.h>
#include <fmt/format.h>

Go to the source code of this file.

Functions

 TEST_SUITE ("SerializerTests")
 

Function Documentation

◆ TEST_SUITE()

TEST_SUITE ( "SerializerTests"  )

Definition at line 21 of file LstmSerializationTests.cpp.

References IOutputSlot::Connect(), INetwork::Create(), DeserializeNetwork(), LayerVerifierBaseWithDescriptor< Descriptor >::ExecuteStrategy(), armnn::Float32, IConnectableLayer::GetInputSlot(), IConnectableLayer::GetName(), IConnectableLayer::GetOutputSlot(), IConnectableLayer::GetType(), armnn::IgnoreUnused(), armnn::Input, armnn::Lstm, LstmDescriptor::m_ActivationFunc, QuantizedLstmInputParams::m_CellBias, LstmInputParams::m_CellBias, QLstmDescriptor::m_CellClip, QLstmDescriptor::m_CellIntermediateScale, LstmInputParams::m_CellLayerNormWeights, LstmInputParams::m_CellToForgetWeights, LstmInputParams::m_CellToInputWeights, LstmInputParams::m_CellToOutputWeights, LstmDescriptor::m_CifgEnabled, QLstmDescriptor::m_CifgEnabled, LstmDescriptor::m_ClippingThresCell, LstmDescriptor::m_ClippingThresProj, QuantizedLstmInputParams::m_ForgetGateBias, LstmInputParams::m_ForgetGateBias, QLstmDescriptor::m_ForgetIntermediateScale, LstmInputParams::m_ForgetLayerNormWeights, QLstmDescriptor::m_HiddenStateScale, QLstmDescriptor::m_HiddenStateZeroPoint, QuantizedLstmInputParams::m_InputGateBias, LstmInputParams::m_InputGateBias, QLstmDescriptor::m_InputIntermediateScale, LstmInputParams::m_InputLayerNormWeights, QuantizedLstmInputParams::m_InputToCellWeights, LstmInputParams::m_InputToCellWeights, QuantizedLstmInputParams::m_InputToForgetWeights, LstmInputParams::m_InputToForgetWeights, QuantizedLstmInputParams::m_InputToInputWeights, LstmInputParams::m_InputToInputWeights, QuantizedLstmInputParams::m_InputToOutputWeights, LstmInputParams::m_InputToOutputWeights, LstmDescriptor::m_LayerNormEnabled, QLstmDescriptor::m_LayerNormEnabled, QuantizedLstmInputParams::m_OutputGateBias, LstmInputParams::m_OutputGateBias, QLstmDescriptor::m_OutputIntermediateScale, LstmInputParams::m_OutputLayerNormWeights, LstmDescriptor::m_PeepholeEnabled, QLstmDescriptor::m_PeepholeEnabled, LstmInputParams::m_ProjectionBias, QLstmDescriptor::m_ProjectionClip, LstmDescriptor::m_ProjectionEnabled, QLstmDescriptor::m_ProjectionEnabled, LstmInputParams::m_ProjectionWeights, QuantizedLstmInputParams::m_RecurrentToCellWeights, LstmInputParams::m_RecurrentToCellWeights, QuantizedLstmInputParams::m_RecurrentToForgetWeights, LstmInputParams::m_RecurrentToForgetWeights, QuantizedLstmInputParams::m_RecurrentToInputWeights, LstmInputParams::m_RecurrentToInputWeights, QuantizedLstmInputParams::m_RecurrentToOutputWeights, LstmInputParams::m_RecurrentToOutputWeights, LstmDescriptor::m_TimeMajor, armnn::Output, armnn::QAsymmS8, armnn::QAsymmU8, armnn::QLstm, armnn::QSymmS16, armnn::QSymmS8, armnn::QuantizedLstm, SerializeNetwork(), IOutputSlot::SetTensorInfo(), armnn::Signed32, armnn::UnidirectionalSequenceLstm, LayerVerifierBase::VerifyConstTensors(), LayerVerifierBaseWithDescriptor< Descriptor >::VerifyDescriptor(), and LayerVerifierBase::VerifyNameAndConnections().

22 {
23 template<typename Descriptor>
24 armnn::LstmInputParams ConstantVector2LstmInputParams(const std::vector<armnn::ConstTensor>& constants,
25  Descriptor& descriptor)
26 {
27  armnn::LstmInputParams lstmInputParams;
28  size_t i = 0;
29 
30  // Inserting basic paramters
31  lstmInputParams.m_InputToForgetWeights = &constants[i++];
32  lstmInputParams.m_InputToCellWeights = &constants[i++];
33  lstmInputParams.m_InputToOutputWeights = &constants[i++];
34  lstmInputParams.m_RecurrentToForgetWeights = &constants[i++];
35  lstmInputParams.m_RecurrentToCellWeights = &constants[i++];
36  lstmInputParams.m_RecurrentToOutputWeights = &constants[i++];
37  lstmInputParams.m_ForgetGateBias = &constants[i++];
38  lstmInputParams.m_CellBias = &constants[i++];
39  lstmInputParams.m_OutputGateBias = &constants[i++];
40  if (!descriptor.m_CifgEnabled)
41  {
42  lstmInputParams.m_InputToInputWeights = &constants[i++];
43  lstmInputParams.m_RecurrentToInputWeights = &constants[i++];
44  lstmInputParams.m_InputGateBias = &constants[i++];
45  }
46 
47  if (descriptor.m_PeepholeEnabled)
48  {
49  if (!descriptor.m_CifgEnabled)
50  {
51  lstmInputParams.m_CellToInputWeights = &constants[i++];
52  }
53  lstmInputParams.m_CellToForgetWeights = &constants[i++];
54  lstmInputParams.m_CellToOutputWeights = &constants[i++];
55  }
56 
57  if (descriptor.m_ProjectionEnabled)
58  {
59  lstmInputParams.m_ProjectionWeights = &constants[i++];
60  lstmInputParams.m_ProjectionBias = &constants[i++];
61  }
62 
63  if (descriptor.m_LayerNormEnabled)
64  {
65  if (!descriptor.m_CifgEnabled)
66  {
67  lstmInputParams.m_InputLayerNormWeights = &constants[i++];
68  }
69  lstmInputParams.m_ForgetLayerNormWeights = &constants[i++];
70  lstmInputParams.m_CellLayerNormWeights = &constants[i++];
71  lstmInputParams.m_OutputLayerNormWeights = &constants[i++];
72  }
73 
74  return lstmInputParams;
75 }
76 
77 // Works for Lstm, UnidirectionalSequenceLstm and QLstm (QuantizedLstm uses different parameters)
78 template<typename Descriptor>
79 class VerifyLstmLayer : public LayerVerifierBaseWithDescriptor<Descriptor>
80 {
81 public:
82  VerifyLstmLayer(const std::string& layerName,
83  const std::vector<armnn::TensorInfo>& inputInfos,
84  const std::vector<armnn::TensorInfo>& outputInfos,
85  const Descriptor& descriptor,
86  const armnn::LstmInputParams& inputParams)
87  : LayerVerifierBaseWithDescriptor<Descriptor>(layerName, inputInfos, outputInfos, descriptor)
88  , m_InputParams(inputParams) {}
89 
90  void ExecuteStrategy(const armnn::IConnectableLayer* layer,
91  const armnn::BaseDescriptor& descriptor,
92  const std::vector<armnn::ConstTensor>& constants,
93  const char* name,
94  const armnn::LayerBindingId id = 0) override
95  {
96  armnn::IgnoreUnused(constants, id);
97  switch (layer->GetType())
98  {
99  case armnn::LayerType::Input: break;
100  case armnn::LayerType::Output: break;
101  case armnn::LayerType::Lstm:
102  case armnn::LayerType::UnidirectionalSequenceLstm:
103  {
104  this->VerifyNameAndConnections(layer, name);
105  const Descriptor& internalDescriptor = static_cast<const Descriptor&>(descriptor);
106  this->VerifyDescriptor(internalDescriptor);
107  armnn::LstmInputParams lstmParams = ConstantVector2LstmInputParams(constants, internalDescriptor);
108  VerifyInputParameters(lstmParams);
109  break;
110  }
111  case armnn::LayerType::QLstm:
112  {
113  this->VerifyNameAndConnections(layer, name);
114  const Descriptor& internalDescriptor = static_cast<const Descriptor&>(descriptor);
115  this->VerifyDescriptor(internalDescriptor);
116  armnn::LstmInputParams lstmParams = ConstantVector2LstmInputParams(constants, internalDescriptor);
117  VerifyInputParameters(lstmParams);
118  break;
119  }
120  default:
121  {
122  throw armnn::Exception("Unexpected layer type in Lstm test model");
123  }
124  }
125  }
126 
127 protected:
128  void VerifyInputParameters(const armnn::LstmInputParams& params)
129  {
130  this->VerifyConstTensors(
131  "m_InputToInputWeights", m_InputParams.m_InputToInputWeights, params.m_InputToInputWeights);
132  this->VerifyConstTensors(
133  "m_InputToForgetWeights", m_InputParams.m_InputToForgetWeights, params.m_InputToForgetWeights);
134  this->VerifyConstTensors(
135  "m_InputToCellWeights", m_InputParams.m_InputToCellWeights, params.m_InputToCellWeights);
136  this->VerifyConstTensors(
137  "m_InputToOutputWeights", m_InputParams.m_InputToOutputWeights, params.m_InputToOutputWeights);
138  this->VerifyConstTensors(
139  "m_RecurrentToInputWeights", m_InputParams.m_RecurrentToInputWeights, params.m_RecurrentToInputWeights);
140  this->VerifyConstTensors(
141  "m_RecurrentToForgetWeights", m_InputParams.m_RecurrentToForgetWeights, params.m_RecurrentToForgetWeights);
142  this->VerifyConstTensors(
143  "m_RecurrentToCellWeights", m_InputParams.m_RecurrentToCellWeights, params.m_RecurrentToCellWeights);
144  this->VerifyConstTensors(
145  "m_RecurrentToOutputWeights", m_InputParams.m_RecurrentToOutputWeights, params.m_RecurrentToOutputWeights);
146  this->VerifyConstTensors(
147  "m_CellToInputWeights", m_InputParams.m_CellToInputWeights, params.m_CellToInputWeights);
148  this->VerifyConstTensors(
149  "m_CellToForgetWeights", m_InputParams.m_CellToForgetWeights, params.m_CellToForgetWeights);
150  this->VerifyConstTensors(
151  "m_CellToOutputWeights", m_InputParams.m_CellToOutputWeights, params.m_CellToOutputWeights);
152  this->VerifyConstTensors(
153  "m_InputGateBias", m_InputParams.m_InputGateBias, params.m_InputGateBias);
154  this->VerifyConstTensors(
155  "m_ForgetGateBias", m_InputParams.m_ForgetGateBias, params.m_ForgetGateBias);
156  this->VerifyConstTensors(
157  "m_CellBias", m_InputParams.m_CellBias, params.m_CellBias);
158  this->VerifyConstTensors(
159  "m_OutputGateBias", m_InputParams.m_OutputGateBias, params.m_OutputGateBias);
160  this->VerifyConstTensors(
161  "m_ProjectionWeights", m_InputParams.m_ProjectionWeights, params.m_ProjectionWeights);
162  this->VerifyConstTensors(
163  "m_ProjectionBias", m_InputParams.m_ProjectionBias, params.m_ProjectionBias);
164  this->VerifyConstTensors(
165  "m_InputLayerNormWeights", m_InputParams.m_InputLayerNormWeights, params.m_InputLayerNormWeights);
166  this->VerifyConstTensors(
167  "m_ForgetLayerNormWeights", m_InputParams.m_ForgetLayerNormWeights, params.m_ForgetLayerNormWeights);
168  this->VerifyConstTensors(
169  "m_CellLayerNormWeights", m_InputParams.m_CellLayerNormWeights, params.m_CellLayerNormWeights);
170  this->VerifyConstTensors(
171  "m_OutputLayerNormWeights", m_InputParams.m_OutputLayerNormWeights, params.m_OutputLayerNormWeights);
172  }
173 
174 private:
175  armnn::LstmInputParams m_InputParams;
176 };
177 
178 TEST_CASE("SerializeDeserializeLstmCifgPeepholeNoProjection")
179 {
180  armnn::LstmDescriptor descriptor;
181  descriptor.m_ActivationFunc = 4;
182  descriptor.m_ClippingThresProj = 0.0f;
183  descriptor.m_ClippingThresCell = 0.0f;
184  descriptor.m_CifgEnabled = true; // if this is true then we DON'T need to set the OptCifgParams
185  descriptor.m_ProjectionEnabled = false;
186  descriptor.m_PeepholeEnabled = true;
187 
188  const uint32_t batchSize = 1;
189  const uint32_t inputSize = 2;
190  const uint32_t numUnits = 4;
191  const uint32_t outputSize = numUnits;
192 
193  armnn::TensorInfo inputWeightsInfo1({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
194  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
195  armnn::ConstTensor inputToForgetWeights(inputWeightsInfo1, inputToForgetWeightsData);
196 
197  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
198  armnn::ConstTensor inputToCellWeights(inputWeightsInfo1, inputToCellWeightsData);
199 
200  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
201  armnn::ConstTensor inputToOutputWeights(inputWeightsInfo1, inputToOutputWeightsData);
202 
203  armnn::TensorInfo inputWeightsInfo2({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
204  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
205  armnn::ConstTensor recurrentToForgetWeights(inputWeightsInfo2, recurrentToForgetWeightsData);
206 
207  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
208  armnn::ConstTensor recurrentToCellWeights(inputWeightsInfo2, recurrentToCellWeightsData);
209 
210  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
211  armnn::ConstTensor recurrentToOutputWeights(inputWeightsInfo2, recurrentToOutputWeightsData);
212 
213  armnn::TensorInfo inputWeightsInfo3({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
214  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo3.GetNumElements());
215  armnn::ConstTensor cellToForgetWeights(inputWeightsInfo3, cellToForgetWeightsData);
216 
217  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo3.GetNumElements());
218  armnn::ConstTensor cellToOutputWeights(inputWeightsInfo3, cellToOutputWeightsData);
219 
220  std::vector<float> forgetGateBiasData(numUnits, 1.0f);
221  armnn::ConstTensor forgetGateBias(inputWeightsInfo3, forgetGateBiasData);
222 
223  std::vector<float> cellBiasData(numUnits, 0.0f);
224  armnn::ConstTensor cellBias(inputWeightsInfo3, cellBiasData);
225 
226  std::vector<float> outputGateBiasData(numUnits, 0.0f);
227  armnn::ConstTensor outputGateBias(inputWeightsInfo3, outputGateBiasData);
228 
229  armnn::LstmInputParams params;
230  params.m_InputToForgetWeights = &inputToForgetWeights;
231  params.m_InputToCellWeights = &inputToCellWeights;
232  params.m_InputToOutputWeights = &inputToOutputWeights;
233  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
234  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
235  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
236  params.m_ForgetGateBias = &forgetGateBias;
237  params.m_CellBias = &cellBias;
238  params.m_OutputGateBias = &outputGateBias;
239  params.m_CellToForgetWeights = &cellToForgetWeights;
240  params.m_CellToOutputWeights = &cellToOutputWeights;
241 
242  armnn::INetworkPtr network = armnn::INetwork::Create();
243  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
244  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
245  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
246  const std::string layerName("lstm");
247  armnn::IConnectableLayer* const lstmLayer = network->AddLstmLayer(descriptor, params, layerName.c_str());
248  armnn::IConnectableLayer* const scratchBuffer = network->AddOutputLayer(0);
249  armnn::IConnectableLayer* const outputStateOut = network->AddOutputLayer(1);
250  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(2);
251  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(3);
252 
253  // connect up
254  armnn::TensorInfo inputTensorInfo({ batchSize, inputSize }, armnn::DataType::Float32);
255  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
256  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
257  armnn::TensorInfo lstmTensorInfoScratchBuff({ batchSize, numUnits * 3 }, armnn::DataType::Float32);
258 
259  inputLayer->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(0));
260  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
261 
262  outputStateIn->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(1));
263  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
264 
265  cellStateIn->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(2));
266  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
267 
268  lstmLayer->GetOutputSlot(0).Connect(scratchBuffer->GetInputSlot(0));
269  lstmLayer->GetOutputSlot(0).SetTensorInfo(lstmTensorInfoScratchBuff);
270 
271  lstmLayer->GetOutputSlot(1).Connect(outputStateOut->GetInputSlot(0));
272  lstmLayer->GetOutputSlot(1).SetTensorInfo(outputStateTensorInfo);
273 
274  lstmLayer->GetOutputSlot(2).Connect(cellStateOut->GetInputSlot(0));
275  lstmLayer->GetOutputSlot(2).SetTensorInfo(cellStateTensorInfo);
276 
277  lstmLayer->GetOutputSlot(3).Connect(outputLayer->GetInputSlot(0));
278  lstmLayer->GetOutputSlot(3).SetTensorInfo(outputStateTensorInfo);
279 
280  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
281  CHECK(deserializedNetwork);
282 
283  VerifyLstmLayer<armnn::LstmDescriptor> checker(
284  layerName,
285  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
286  {lstmTensorInfoScratchBuff, outputStateTensorInfo, cellStateTensorInfo, outputStateTensorInfo},
287  descriptor,
288  params);
289  deserializedNetwork->ExecuteStrategy(checker);
290 }
291 
292 TEST_CASE("SerializeDeserializeLstmNoCifgWithPeepholeAndProjection")
293 {
294  armnn::LstmDescriptor descriptor;
295  descriptor.m_ActivationFunc = 4;
296  descriptor.m_ClippingThresProj = 0.0f;
297  descriptor.m_ClippingThresCell = 0.0f;
298  descriptor.m_CifgEnabled = false; // if this is true then we DON'T need to set the OptCifgParams
299  descriptor.m_ProjectionEnabled = true;
300  descriptor.m_PeepholeEnabled = true;
301 
302  const uint32_t batchSize = 2;
303  const uint32_t inputSize = 5;
304  const uint32_t numUnits = 20;
305  const uint32_t outputSize = 16;
306 
307  armnn::TensorInfo tensorInfo20x5({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
308  std::vector<float> inputToInputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
309  armnn::ConstTensor inputToInputWeights(tensorInfo20x5, inputToInputWeightsData);
310 
311  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
312  armnn::ConstTensor inputToForgetWeights(tensorInfo20x5, inputToForgetWeightsData);
313 
314  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
315  armnn::ConstTensor inputToCellWeights(tensorInfo20x5, inputToCellWeightsData);
316 
317  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
318  armnn::ConstTensor inputToOutputWeights(tensorInfo20x5, inputToOutputWeightsData);
319 
320  armnn::TensorInfo tensorInfo20({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
321  std::vector<float> inputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
322  armnn::ConstTensor inputGateBias(tensorInfo20, inputGateBiasData);
323 
324  std::vector<float> forgetGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
325  armnn::ConstTensor forgetGateBias(tensorInfo20, forgetGateBiasData);
326 
327  std::vector<float> cellBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
328  armnn::ConstTensor cellBias(tensorInfo20, cellBiasData);
329 
330  std::vector<float> outputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
331  armnn::ConstTensor outputGateBias(tensorInfo20, outputGateBiasData);
332 
333  armnn::TensorInfo tensorInfo20x16({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
334  std::vector<float> recurrentToInputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
335  armnn::ConstTensor recurrentToInputWeights(tensorInfo20x16, recurrentToInputWeightsData);
336 
337  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
338  armnn::ConstTensor recurrentToForgetWeights(tensorInfo20x16, recurrentToForgetWeightsData);
339 
340  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
341  armnn::ConstTensor recurrentToCellWeights(tensorInfo20x16, recurrentToCellWeightsData);
342 
343  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
344  armnn::ConstTensor recurrentToOutputWeights(tensorInfo20x16, recurrentToOutputWeightsData);
345 
346  std::vector<float> cellToInputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
347  armnn::ConstTensor cellToInputWeights(tensorInfo20, cellToInputWeightsData);
348 
349  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
350  armnn::ConstTensor cellToForgetWeights(tensorInfo20, cellToForgetWeightsData);
351 
352  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
353  armnn::ConstTensor cellToOutputWeights(tensorInfo20, cellToOutputWeightsData);
354 
355  armnn::TensorInfo tensorInfo16x20({outputSize, numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
356  std::vector<float> projectionWeightsData = GenerateRandomData<float>(tensorInfo16x20.GetNumElements());
357  armnn::ConstTensor projectionWeights(tensorInfo16x20, projectionWeightsData);
358 
359  armnn::TensorInfo tensorInfo16({outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
360  std::vector<float> projectionBiasData(outputSize, 0.f);
361  armnn::ConstTensor projectionBias(tensorInfo16, projectionBiasData);
362 
363  armnn::LstmInputParams params;
364  params.m_InputToForgetWeights = &inputToForgetWeights;
365  params.m_InputToCellWeights = &inputToCellWeights;
366  params.m_InputToOutputWeights = &inputToOutputWeights;
367  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
368  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
369  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
370  params.m_ForgetGateBias = &forgetGateBias;
371  params.m_CellBias = &cellBias;
372  params.m_OutputGateBias = &outputGateBias;
373 
374  // additional params because: descriptor.m_CifgEnabled = false
375  params.m_InputToInputWeights = &inputToInputWeights;
376  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
377  params.m_CellToInputWeights = &cellToInputWeights;
378  params.m_InputGateBias = &inputGateBias;
379 
380  // additional params because: descriptor.m_ProjectionEnabled = true
381  params.m_ProjectionWeights = &projectionWeights;
382  params.m_ProjectionBias = &projectionBias;
383 
384  // additional params because: descriptor.m_PeepholeEnabled = true
385  params.m_CellToForgetWeights = &cellToForgetWeights;
386  params.m_CellToOutputWeights = &cellToOutputWeights;
387 
388  armnn::INetworkPtr network = armnn::INetwork::Create();
389  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
390  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
391  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
392  const std::string layerName("lstm");
393  armnn::IConnectableLayer* const lstmLayer = network->AddLstmLayer(descriptor, params, layerName.c_str());
394  armnn::IConnectableLayer* const scratchBuffer = network->AddOutputLayer(0);
395  armnn::IConnectableLayer* const outputStateOut = network->AddOutputLayer(1);
396  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(2);
397  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(3);
398 
399  // connect up
400  armnn::TensorInfo inputTensorInfo({ batchSize, inputSize }, armnn::DataType::Float32);
401  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
402  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
403  armnn::TensorInfo lstmTensorInfoScratchBuff({ batchSize, numUnits * 4 }, armnn::DataType::Float32);
404 
405  inputLayer->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(0));
406  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
407 
408  outputStateIn->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(1));
409  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
410 
411  cellStateIn->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(2));
412  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
413 
414  lstmLayer->GetOutputSlot(0).Connect(scratchBuffer->GetInputSlot(0));
415  lstmLayer->GetOutputSlot(0).SetTensorInfo(lstmTensorInfoScratchBuff);
416 
417  lstmLayer->GetOutputSlot(1).Connect(outputStateOut->GetInputSlot(0));
418  lstmLayer->GetOutputSlot(1).SetTensorInfo(outputStateTensorInfo);
419 
420  lstmLayer->GetOutputSlot(2).Connect(cellStateOut->GetInputSlot(0));
421  lstmLayer->GetOutputSlot(2).SetTensorInfo(cellStateTensorInfo);
422 
423  lstmLayer->GetOutputSlot(3).Connect(outputLayer->GetInputSlot(0));
424  lstmLayer->GetOutputSlot(3).SetTensorInfo(outputStateTensorInfo);
425 
426  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
427  CHECK(deserializedNetwork);
428 
429  VerifyLstmLayer<armnn::LstmDescriptor> checker(
430  layerName,
431  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
432  {lstmTensorInfoScratchBuff, outputStateTensorInfo, cellStateTensorInfo, outputStateTensorInfo},
433  descriptor,
434  params);
435  deserializedNetwork->ExecuteStrategy(checker);
436 }
437 
438 TEST_CASE("SerializeDeserializeLstmNoCifgWithPeepholeWithProjectionWithLayerNorm")
439 {
440  armnn::LstmDescriptor descriptor;
441  descriptor.m_ActivationFunc = 4;
442  descriptor.m_ClippingThresProj = 0.0f;
443  descriptor.m_ClippingThresCell = 0.0f;
444  descriptor.m_CifgEnabled = false; // if this is true then we DON'T need to set the OptCifgParams
445  descriptor.m_ProjectionEnabled = true;
446  descriptor.m_PeepholeEnabled = true;
447  descriptor.m_LayerNormEnabled = true;
448 
449  const uint32_t batchSize = 2;
450  const uint32_t inputSize = 5;
451  const uint32_t numUnits = 20;
452  const uint32_t outputSize = 16;
453 
454  armnn::TensorInfo tensorInfo20x5({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
455  std::vector<float> inputToInputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
456  armnn::ConstTensor inputToInputWeights(tensorInfo20x5, inputToInputWeightsData);
457 
458  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
459  armnn::ConstTensor inputToForgetWeights(tensorInfo20x5, inputToForgetWeightsData);
460 
461  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
462  armnn::ConstTensor inputToCellWeights(tensorInfo20x5, inputToCellWeightsData);
463 
464  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
465  armnn::ConstTensor inputToOutputWeights(tensorInfo20x5, inputToOutputWeightsData);
466 
467  armnn::TensorInfo tensorInfo20({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
468  std::vector<float> inputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
469  armnn::ConstTensor inputGateBias(tensorInfo20, inputGateBiasData);
470 
471  std::vector<float> forgetGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
472  armnn::ConstTensor forgetGateBias(tensorInfo20, forgetGateBiasData);
473 
474  std::vector<float> cellBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
475  armnn::ConstTensor cellBias(tensorInfo20, cellBiasData);
476 
477  std::vector<float> outputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
478  armnn::ConstTensor outputGateBias(tensorInfo20, outputGateBiasData);
479 
480  armnn::TensorInfo tensorInfo20x16({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
481  std::vector<float> recurrentToInputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
482  armnn::ConstTensor recurrentToInputWeights(tensorInfo20x16, recurrentToInputWeightsData);
483 
484  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
485  armnn::ConstTensor recurrentToForgetWeights(tensorInfo20x16, recurrentToForgetWeightsData);
486 
487  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
488  armnn::ConstTensor recurrentToCellWeights(tensorInfo20x16, recurrentToCellWeightsData);
489 
490  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
491  armnn::ConstTensor recurrentToOutputWeights(tensorInfo20x16, recurrentToOutputWeightsData);
492 
493  std::vector<float> cellToInputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
494  armnn::ConstTensor cellToInputWeights(tensorInfo20, cellToInputWeightsData);
495 
496  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
497  armnn::ConstTensor cellToForgetWeights(tensorInfo20, cellToForgetWeightsData);
498 
499  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
500  armnn::ConstTensor cellToOutputWeights(tensorInfo20, cellToOutputWeightsData);
501 
502  armnn::TensorInfo tensorInfo16x20({outputSize, numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
503  std::vector<float> projectionWeightsData = GenerateRandomData<float>(tensorInfo16x20.GetNumElements());
504  armnn::ConstTensor projectionWeights(tensorInfo16x20, projectionWeightsData);
505 
506  armnn::TensorInfo tensorInfo16({outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
507  std::vector<float> projectionBiasData(outputSize, 0.f);
508  armnn::ConstTensor projectionBias(tensorInfo16, projectionBiasData);
509 
510  std::vector<float> inputLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
511  armnn::ConstTensor inputLayerNormWeights(tensorInfo20, forgetGateBiasData);
512 
513  std::vector<float> forgetLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
514  armnn::ConstTensor forgetLayerNormWeights(tensorInfo20, forgetGateBiasData);
515 
516  std::vector<float> cellLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
517  armnn::ConstTensor cellLayerNormWeights(tensorInfo20, forgetGateBiasData);
518 
519  std::vector<float> outLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
520  armnn::ConstTensor outLayerNormWeights(tensorInfo20, forgetGateBiasData);
521 
522  armnn::LstmInputParams params;
523  params.m_InputToForgetWeights = &inputToForgetWeights;
524  params.m_InputToCellWeights = &inputToCellWeights;
525  params.m_InputToOutputWeights = &inputToOutputWeights;
526  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
527  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
528  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
529  params.m_ForgetGateBias = &forgetGateBias;
530  params.m_CellBias = &cellBias;
531  params.m_OutputGateBias = &outputGateBias;
532 
533  // additional params because: descriptor.m_CifgEnabled = false
534  params.m_InputToInputWeights = &inputToInputWeights;
535  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
536  params.m_CellToInputWeights = &cellToInputWeights;
537  params.m_InputGateBias = &inputGateBias;
538 
539  // additional params because: descriptor.m_ProjectionEnabled = true
540  params.m_ProjectionWeights = &projectionWeights;
541  params.m_ProjectionBias = &projectionBias;
542 
543  // additional params because: descriptor.m_PeepholeEnabled = true
544  params.m_CellToForgetWeights = &cellToForgetWeights;
545  params.m_CellToOutputWeights = &cellToOutputWeights;
546 
547  // additional params because: despriptor.m_LayerNormEnabled = true
548  params.m_InputLayerNormWeights = &inputLayerNormWeights;
549  params.m_ForgetLayerNormWeights = &forgetLayerNormWeights;
550  params.m_CellLayerNormWeights = &cellLayerNormWeights;
551  params.m_OutputLayerNormWeights = &outLayerNormWeights;
552 
553  armnn::INetworkPtr network = armnn::INetwork::Create();
554  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
555  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
556  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
557  const std::string layerName("lstm");
558  armnn::IConnectableLayer* const lstmLayer = network->AddLstmLayer(descriptor, params, layerName.c_str());
559  armnn::IConnectableLayer* const scratchBuffer = network->AddOutputLayer(0);
560  armnn::IConnectableLayer* const outputStateOut = network->AddOutputLayer(1);
561  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(2);
562  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(3);
563 
564  // connect up
565  armnn::TensorInfo inputTensorInfo({ batchSize, inputSize }, armnn::DataType::Float32);
566  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
567  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
568  armnn::TensorInfo lstmTensorInfoScratchBuff({ batchSize, numUnits * 4 }, armnn::DataType::Float32);
569 
570  inputLayer->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(0));
571  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
572 
573  outputStateIn->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(1));
574  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
575 
576  cellStateIn->GetOutputSlot(0).Connect(lstmLayer->GetInputSlot(2));
577  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
578 
579  lstmLayer->GetOutputSlot(0).Connect(scratchBuffer->GetInputSlot(0));
580  lstmLayer->GetOutputSlot(0).SetTensorInfo(lstmTensorInfoScratchBuff);
581 
582  lstmLayer->GetOutputSlot(1).Connect(outputStateOut->GetInputSlot(0));
583  lstmLayer->GetOutputSlot(1).SetTensorInfo(outputStateTensorInfo);
584 
585  lstmLayer->GetOutputSlot(2).Connect(cellStateOut->GetInputSlot(0));
586  lstmLayer->GetOutputSlot(2).SetTensorInfo(cellStateTensorInfo);
587 
588  lstmLayer->GetOutputSlot(3).Connect(outputLayer->GetInputSlot(0));
589  lstmLayer->GetOutputSlot(3).SetTensorInfo(outputStateTensorInfo);
590 
591  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
592  CHECK(deserializedNetwork);
593 
594  VerifyLstmLayer<armnn::LstmDescriptor> checker(
595  layerName,
596  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
597  {lstmTensorInfoScratchBuff, outputStateTensorInfo, cellStateTensorInfo, outputStateTensorInfo},
598  descriptor,
599  params);
600  deserializedNetwork->ExecuteStrategy(checker);
601 }
602 
603 TEST_CASE("EnsureLstmLayersBackwardCompatibility")
604 {
605  // The hex data below is a flat buffer containing a lstm layer with no Cifg, with peephole and projection
606  // enabled. That data was obtained before additional layer normalization parameters where added to the
607  // lstm serializer. That way it can be tested if a lstm model with the old parameter configuration can
608  // still be loaded
609  const std::vector<uint8_t> lstmNoCifgWithPeepholeAndProjectionModel =
610  {
611  0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x10, 0x00, 0x04, 0x00, 0x08, 0x00, 0x0C, 0x00, 0x0A, 0x00,
612  0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00,
613  0xDC, 0x29, 0x00, 0x00, 0x38, 0x29, 0x00, 0x00, 0xB4, 0x28, 0x00, 0x00, 0x94, 0x01, 0x00, 0x00, 0x3C, 0x01,
614  0x00, 0x00, 0xE0, 0x00, 0x00, 0x00, 0x84, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
615  0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x04, 0x00,
616  0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x70, 0xD6, 0xFF, 0xFF,
617  0x00, 0x00, 0x00, 0x0B, 0x04, 0x00, 0x00, 0x00, 0x06, 0xD7, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x88, 0xD7,
618  0xFF, 0xFF, 0x08, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xF6, 0xD6, 0xFF, 0xFF, 0x07, 0x00, 0x00, 0x00,
619  0x10, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00,
620  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
621  0xE8, 0xD7, 0xFF, 0xFF, 0x03, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xC8, 0xD6, 0xFF, 0xFF, 0x00, 0x00,
622  0x00, 0x0B, 0x04, 0x00, 0x00, 0x00, 0x5E, 0xD7, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0xE0, 0xD7, 0xFF, 0xFF,
623  0x08, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x4E, 0xD7, 0xFF, 0xFF, 0x06, 0x00, 0x00, 0x00, 0x10, 0x00,
624  0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
625  0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x40, 0xD8,
626  0xFF, 0xFF, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0xD7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x0B,
627  0x04, 0x00, 0x00, 0x00, 0xB6, 0xD7, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x38, 0xD8, 0xFF, 0xFF, 0x08, 0x00,
628  0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0xA6, 0xD7, 0xFF, 0xFF, 0x05, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00,
629  0x03, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
630  0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x98, 0xD8, 0xFF, 0xFF,
631  0x03, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x78, 0xD7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x0B, 0x04, 0x00,
632  0x00, 0x00, 0x0E, 0xD8, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x16, 0xD8, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00,
633  0xFA, 0xD7, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x10, 0x00,
634  0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
635  0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xEC, 0xD8, 0xFF, 0xFF, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00,
636  0x00, 0x00, 0x6C, 0xD8, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x23, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A, 0x00,
637  0x12, 0x00, 0x04, 0x00, 0x08, 0x00, 0x0C, 0x00, 0x0A, 0x00, 0x00, 0x00, 0xE0, 0x25, 0x00, 0x00, 0xD0, 0x25,
638  0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x26, 0x00, 0x48, 0x00, 0x04, 0x00, 0x08, 0x00, 0x0C, 0x00,
639  0x10, 0x00, 0x14, 0x00, 0x18, 0x00, 0x1C, 0x00, 0x20, 0x00, 0x24, 0x00, 0x28, 0x00, 0x2C, 0x00, 0x30, 0x00,
640  0x34, 0x00, 0x38, 0x00, 0x3C, 0x00, 0x40, 0x00, 0x44, 0x00, 0x26, 0x00, 0x00, 0x00, 0xC4, 0x23, 0x00, 0x00,
641  0xF8, 0x21, 0x00, 0x00, 0x2C, 0x20, 0x00, 0x00, 0xF0, 0x1A, 0x00, 0x00, 0xB4, 0x15, 0x00, 0x00, 0x78, 0x10,
642  0x00, 0x00, 0xF0, 0x0F, 0x00, 0x00, 0x68, 0x0F, 0x00, 0x00, 0xE0, 0x0E, 0x00, 0x00, 0x14, 0x0D, 0x00, 0x00,
643  0xD8, 0x07, 0x00, 0x00, 0x50, 0x07, 0x00, 0x00, 0xC8, 0x06, 0x00, 0x00, 0x8C, 0x01, 0x00, 0x00, 0x14, 0x01,
644  0x00, 0x00, 0x8C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xEE, 0xD7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03,
645  0x64, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xFE, 0xD8, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x14, 0x00,
646  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
647  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
648  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
649  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
650  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x5A, 0xD8, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01,
651  0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x72, 0xD8,
652  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x64, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x82, 0xD9, 0xFF, 0xFF,
653  0x04, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
654  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
655  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
656  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
657  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xDE, 0xD8,
658  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
659  0x14, 0x00, 0x00, 0x00, 0xF6, 0xD8, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x54, 0x00, 0x00, 0x00, 0x04, 0x00,
660  0x00, 0x00, 0x06, 0xDA, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
661  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
662  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
663  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
664  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x52, 0xD9, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00,
665  0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x6A, 0xD9, 0xFF, 0xFF, 0x00, 0x00,
666  0x00, 0x03, 0x14, 0x05, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x7A, 0xDA, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00,
667  0x40, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
668  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
669  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
670  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
671  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
672  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
673  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
674  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
675  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
676  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
677  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
678  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
679  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
680  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
681  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
682  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
683  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
684  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
685  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
686  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
687  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
688  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
689  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
690  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
691  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
692  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
693  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
694  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
695  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
696  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
697  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
698  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
699  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
700  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
701  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
702  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
703  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
704  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
705  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
706  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
707  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
708  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
709  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
710  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
711  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
712  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
713  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
714  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
715  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
716  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
717  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
718  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
719  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
720  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
721  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
722  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
723  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
724  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
725  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
726  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
727  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
728  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
729  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
730  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
731  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
732  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
733  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
734  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
735  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
736  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
737  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
738  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x86, 0xDE, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00,
739  0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0xA2, 0xDE,
740  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x64, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xB2, 0xDF, 0xFF, 0xFF,
741  0x04, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
742  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
743  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
744  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
745  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0E, 0xDF,
746  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
747  0x14, 0x00, 0x00, 0x00, 0x26, 0xDF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x64, 0x00, 0x00, 0x00, 0x04, 0x00,
748  0x00, 0x00, 0x36, 0xE0, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
749  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
750  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
751  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
752  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
753  0x00, 0x00, 0x00, 0x00, 0x92, 0xDF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
754  0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0xAA, 0xDF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03,
755  0x14, 0x05, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xBA, 0xE0, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x40, 0x01,
756  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
757  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
758  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
759  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
760  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
761  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
762  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
763  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
764  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
765  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
766  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
767  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
768  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
769  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
770  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
771  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
772  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
773  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
774  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
775  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
776  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
777  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
778  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
779  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
780  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
781  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
782  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
783  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
784  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
785  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
786  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
787  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
788  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
789  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
790  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
791  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
792  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
793  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
794  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
795  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
796  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
797  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
798  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
799  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
800  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
801  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
802  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
803  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
804  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
805  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
806  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
807  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
808  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
809  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
810  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
811  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
812  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
813  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
814  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
815  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
816  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
817  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
818  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
819  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
820  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
821  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
822  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
823  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
824  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
825  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
826  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
827  0x00, 0x00, 0x00, 0x00, 0xC6, 0xE4, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
828  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0xE2, 0xE4, 0xFF, 0xFF,
829  0x00, 0x00, 0x00, 0x03, 0xA4, 0x01, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xF2, 0xE5, 0xFF, 0xFF, 0x04, 0x00,
830  0x00, 0x00, 0x64, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
831  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
832  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
833  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
834  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
835  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
836  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
837  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
838  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
839  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
840  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
841  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
842  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
843  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
844  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
845  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
846  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
847  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
848  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
849  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
850  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
851  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
852  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x8E, 0xE6, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01,
853  0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x05, 0x00,
854  0x00, 0x00, 0xAA, 0xE6, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x64, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
855  0xBA, 0xE7, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
856  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
857  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
858  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
859  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
860  0x00, 0x00, 0x16, 0xE7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
861  0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x2E, 0xE7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x64, 0x00,
862  0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x3E, 0xE8, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00,
863  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
864  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
865  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
866  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
867  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x9A, 0xE7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00,
868  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0xB2, 0xE7, 0xFF, 0xFF,
869  0x00, 0x00, 0x00, 0x03, 0x64, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xC2, 0xE8, 0xFF, 0xFF, 0x04, 0x00,
870  0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
871  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
872  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
873  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
874  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1E, 0xE8, 0xFF, 0xFF,
875  0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00,
876  0x00, 0x00, 0x36, 0xE8, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x14, 0x05, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
877  0x46, 0xE9, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x40, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
878  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
879  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
880  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
881  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
882  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
883  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
884  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
885  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
886  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
887  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
888  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
889  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
890  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
891  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
892  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
893  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
894  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
895  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
896  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
897  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
898  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
899  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
900  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
901  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
902  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
903  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
904  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
905  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
906  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
907  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
908  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
909  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
910  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
911  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
912  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
913  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
914  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
915  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
916  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
917  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
918  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
919  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
920  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
921  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
922  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
923  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
924  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
925  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
926  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
927  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
928  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
929  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
930  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
931  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
932  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
933  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
934  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
935  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
936  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
937  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
938  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
939  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
940  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
941  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
942  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
943  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
944  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
945  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
946  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
947  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
948  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x52, 0xED, 0xFF, 0xFF,
949  0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00,
950  0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x6E, 0xED, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0x14, 0x05, 0x00, 0x00,
951  0x04, 0x00, 0x00, 0x00, 0x7E, 0xEE, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x40, 0x01, 0x00, 0x00, 0x00, 0x00,
952  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
953  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
954  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
955  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
956  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
957  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
958  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
959  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
960  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
961  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
962  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
963  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
964  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
965  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
966  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
967  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
968  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
969  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
970  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
971  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
972  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
973  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
974  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
975  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
976  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
977  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
978  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
979  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
980  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
981  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
982  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
983  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
984  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
985  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
986  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
987  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
988  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
989  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
990  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
991  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
992  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
993  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
994  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
995  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
996  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
997  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
998  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
999  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1000  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1001  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1002  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1003  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1004  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1005  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1006  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1007  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1008  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1009  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1010  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1011  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1012  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1013  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1014  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1015  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1016  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1017  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1018  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1019  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1020  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1021  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1022  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1023  0x8A, 0xF2, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00,
1024  0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0xA6, 0xF2, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03,
1025  0x14, 0x05, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xB6, 0xF3, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x40, 0x01,
1026  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1027  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1028  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1029  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1030  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1031  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1032  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1033  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1034  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1035  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1036  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1037  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1038  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1039  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1040  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1041  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1042  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1043  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1044  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1045  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1046  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1047  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1048  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1049  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1050  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1051  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1052  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1053  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1054  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1055  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1056  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1057  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1058  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1059  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1060  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1061  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1062  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1063  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1064  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1065  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1066  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1067  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1068  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1069  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1070  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1071  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1072  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1073  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1074  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1075  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1076  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1077  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1078  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1079  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1080  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1081  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1082  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1083  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1084  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1085  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1086  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1087  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1088  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1089  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1090  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1091  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1092  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1093  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1094  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1095  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1096  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1097  0x00, 0x00, 0x00, 0x00, 0xC2, 0xF7, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
1098  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0xDE, 0xF7, 0xFF, 0xFF,
1099  0x00, 0x00, 0x00, 0x03, 0xA4, 0x01, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xEE, 0xF8, 0xFF, 0xFF, 0x04, 0x00,
1100  0x00, 0x00, 0x64, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1101  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1102  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1103  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1104  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1105  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1106  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1107  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1108  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1109  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1110  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1111  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1112  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1113  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1114  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1115  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1116  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1117  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1118  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1119  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1120  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1121  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1122  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x8A, 0xF9, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01,
1123  0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x05, 0x00,
1124  0x00, 0x00, 0xA6, 0xF9, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0xA4, 0x01, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
1125  0xB6, 0xFA, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x64, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1126  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1127  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1128  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1129  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1130  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1131  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1132  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1133  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1134  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1135  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1136  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1137  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1138  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1139  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1140  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1141  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1142  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1143  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1144  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1145  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1146  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1147  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x52, 0xFB,
1148  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
1149  0x14, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x6E, 0xFB, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0xA4, 0x01,
1150  0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x7E, 0xFC, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x64, 0x00, 0x00, 0x00,
1151  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1152  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1153  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1154  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1155  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1156  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1157  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1158  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1159  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1160  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1161  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1162  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1163  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1164  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1165  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1166  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1167  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1168  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1169  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1170  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1171  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1172  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1173  0x00, 0x00, 0x00, 0x00, 0x1A, 0xFD, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
1174  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x10, 0x00, 0x0C, 0x00,
1175  0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x06, 0x00, 0x07, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
1176  0x01, 0x01, 0x04, 0x00, 0x00, 0x00, 0x2E, 0xFE, 0xFF, 0xFF, 0x03, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00,
1177  0x22, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6C, 0x73,
1178  0x74, 0x6D, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xEC, 0x00, 0x00, 0x00, 0xD0, 0x00, 0x00, 0x00,
1179  0xB4, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x88, 0x00, 0x00, 0x00, 0x5C, 0x00, 0x00, 0x00, 0x30, 0x00,
1180  0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x14, 0xFF, 0xFF, 0xFF, 0x03, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
1181  0xA6, 0xFD, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00,
1182  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x3C, 0xFF, 0xFF, 0xFF, 0x02, 0x00, 0x00, 0x00,
1183  0x04, 0x00, 0x00, 0x00, 0xCE, 0xFD, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
1184  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x64, 0xFF, 0xFF, 0xFF,
1185  0x01, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xF6, 0xFD, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00,
1186  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00,
1187  0xB4, 0xFE, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0x1A, 0xFE, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00,
1188  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x50, 0x00, 0x00, 0x00,
1189  0xF0, 0xFF, 0xFF, 0xFF, 0x02, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00,
1190  0x10, 0x00, 0x04, 0x00, 0x08, 0x00, 0x08, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
1191  0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00,
1192  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xE8, 0xFE, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x09, 0x04, 0x00, 0x00, 0x00,
1193  0x7E, 0xFF, 0xFF, 0xFF, 0x0C, 0x00, 0x00, 0x00, 0x08, 0x00, 0x0C, 0x00, 0x04, 0x00, 0x08, 0x00, 0x08, 0x00,
1194  0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x76, 0xFF, 0xFF, 0xFF, 0x02, 0x00, 0x00, 0x00,
1195  0x10, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
1196  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
1197  0x68, 0xFF, 0xFF, 0xFF, 0x04, 0x00, 0x00, 0x00, 0xCE, 0xFE, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00,
1198  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00,
1199  0x08, 0x00, 0x0E, 0x00, 0x07, 0x00, 0x08, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x0C, 0x00,
1200  0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x08, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
1201  0x08, 0x00, 0x0E, 0x00, 0x04, 0x00, 0x08, 0x00, 0x08, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x01, 0x00,
1202  0x00, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x18, 0x00, 0x04, 0x00, 0x08, 0x00, 0x0C, 0x00, 0x10, 0x00, 0x14, 0x00,
1203  0x0E, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x10, 0x00,
1204  0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1205  0x01, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x08, 0x00, 0x08, 0x00, 0x00, 0x00, 0x04, 0x00, 0x08, 0x00,
1206  0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6E, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00,
1207  0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x08, 0x00,
1208  0x0C, 0x00, 0x07, 0x00, 0x08, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x09, 0x04, 0x00, 0x00, 0x00,
1209  0xF6, 0xFF, 0xFF, 0xFF, 0x0C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x0A, 0x00, 0x04, 0x00, 0x06, 0x00,
1210  0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0E, 0x00, 0x14, 0x00, 0x00, 0x00, 0x04, 0x00, 0x08, 0x00,
1211  0x0C, 0x00, 0x10, 0x00, 0x0E, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x10, 0x00,
1212  0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
1213  0x01, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00, 0x08, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x04, 0x00, 0x08, 0x00,
1214  0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x10, 0x00, 0x08, 0x00, 0x07, 0x00, 0x0C, 0x00,
1215  0x0A, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00,
1216  0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x00
1217  };
1218 
1219  armnn::INetworkPtr deserializedNetwork =
1220  DeserializeNetwork(std::string(lstmNoCifgWithPeepholeAndProjectionModel.begin(),
1221  lstmNoCifgWithPeepholeAndProjectionModel.end()));
1222 
1223  CHECK(deserializedNetwork);
1224 
1225  // generating the same model parameters which where used to serialize the model (Layer norm is not specified)
1226  armnn::LstmDescriptor descriptor;
1227  descriptor.m_ActivationFunc = 4;
1228  descriptor.m_ClippingThresProj = 0.0f;
1229  descriptor.m_ClippingThresCell = 0.0f;
1230  descriptor.m_CifgEnabled = false;
1231  descriptor.m_ProjectionEnabled = true;
1232  descriptor.m_PeepholeEnabled = true;
1233 
1234  const uint32_t batchSize = 2u;
1235  const uint32_t inputSize = 5u;
1236  const uint32_t numUnits = 20u;
1237  const uint32_t outputSize = 16u;
1238 
1239  armnn::TensorInfo tensorInfo20x5({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
1240  std::vector<float> inputToInputWeightsData(tensorInfo20x5.GetNumElements(), 0.0f);
1241  armnn::ConstTensor inputToInputWeights(tensorInfo20x5, inputToInputWeightsData);
1242 
1243  std::vector<float> inputToForgetWeightsData(tensorInfo20x5.GetNumElements(), 0.0f);
1244  armnn::ConstTensor inputToForgetWeights(tensorInfo20x5, inputToForgetWeightsData);
1245 
1246  std::vector<float> inputToCellWeightsData(tensorInfo20x5.GetNumElements(), 0.0f);
1247  armnn::ConstTensor inputToCellWeights(tensorInfo20x5, inputToCellWeightsData);
1248 
1249  std::vector<float> inputToOutputWeightsData(tensorInfo20x5.GetNumElements(), 0.0f);
1250  armnn::ConstTensor inputToOutputWeights(tensorInfo20x5, inputToOutputWeightsData);
1251 
1252  armnn::TensorInfo tensorInfo20({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
1253  std::vector<float> inputGateBiasData(tensorInfo20.GetNumElements(), 0.0f);
1254  armnn::ConstTensor inputGateBias(tensorInfo20, inputGateBiasData);
1255 
1256  std::vector<float> forgetGateBiasData(tensorInfo20.GetNumElements(), 0.0f);
1257  armnn::ConstTensor forgetGateBias(tensorInfo20, forgetGateBiasData);
1258 
1259  std::vector<float> cellBiasData(tensorInfo20.GetNumElements(), 0.0f);
1260  armnn::ConstTensor cellBias(tensorInfo20, cellBiasData);
1261 
1262  std::vector<float> outputGateBiasData(tensorInfo20.GetNumElements(), 0.0f);
1263  armnn::ConstTensor outputGateBias(tensorInfo20, outputGateBiasData);
1264 
1265  armnn::TensorInfo tensorInfo20x16({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
1266  std::vector<float> recurrentToInputWeightsData(tensorInfo20x16.GetNumElements(), 0.0f);
1267  armnn::ConstTensor recurrentToInputWeights(tensorInfo20x16, recurrentToInputWeightsData);
1268 
1269  std::vector<float> recurrentToForgetWeightsData(tensorInfo20x16.GetNumElements(), 0.0f);
1270  armnn::ConstTensor recurrentToForgetWeights(tensorInfo20x16, recurrentToForgetWeightsData);
1271 
1272  std::vector<float> recurrentToCellWeightsData(tensorInfo20x16.GetNumElements(), 0.0f);
1273  armnn::ConstTensor recurrentToCellWeights(tensorInfo20x16, recurrentToCellWeightsData);
1274 
1275  std::vector<float> recurrentToOutputWeightsData(tensorInfo20x16.GetNumElements(), 0.0f);
1276  armnn::ConstTensor recurrentToOutputWeights(tensorInfo20x16, recurrentToOutputWeightsData);
1277 
1278  std::vector<float> cellToInputWeightsData(tensorInfo20.GetNumElements(), 0.0f);
1279  armnn::ConstTensor cellToInputWeights(tensorInfo20, cellToInputWeightsData);
1280 
1281  std::vector<float> cellToForgetWeightsData(tensorInfo20.GetNumElements(), 0.0f);
1282  armnn::ConstTensor cellToForgetWeights(tensorInfo20, cellToForgetWeightsData);
1283 
1284  std::vector<float> cellToOutputWeightsData(tensorInfo20.GetNumElements(), 0.0f);
1285  armnn::ConstTensor cellToOutputWeights(tensorInfo20, cellToOutputWeightsData);
1286 
1287  armnn::TensorInfo tensorInfo16x20({outputSize, numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
1288  std::vector<float> projectionWeightsData(tensorInfo16x20.GetNumElements(), 0.0f);
1289  armnn::ConstTensor projectionWeights(tensorInfo16x20, projectionWeightsData);
1290 
1291  armnn::TensorInfo tensorInfo16({outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
1292  std::vector<float> projectionBiasData(outputSize, 0.0f);
1293  armnn::ConstTensor projectionBias(tensorInfo16, projectionBiasData);
1294 
1295  armnn::LstmInputParams params;
1296  params.m_InputToForgetWeights = &inputToForgetWeights;
1297  params.m_InputToCellWeights = &inputToCellWeights;
1298  params.m_InputToOutputWeights = &inputToOutputWeights;
1299  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
1300  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
1301  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
1302  params.m_ForgetGateBias = &forgetGateBias;
1303  params.m_CellBias = &cellBias;
1304  params.m_OutputGateBias = &outputGateBias;
1305 
1306  // additional params because: descriptor.m_CifgEnabled = false
1307  params.m_InputToInputWeights = &inputToInputWeights;
1308  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
1309  params.m_CellToInputWeights = &cellToInputWeights;
1310  params.m_InputGateBias = &inputGateBias;
1311 
1312  // additional params because: descriptor.m_ProjectionEnabled = true
1313  params.m_ProjectionWeights = &projectionWeights;
1314  params.m_ProjectionBias = &projectionBias;
1315 
1316  // additional params because: descriptor.m_PeepholeEnabled = true
1317  params.m_CellToForgetWeights = &cellToForgetWeights;
1318  params.m_CellToOutputWeights = &cellToOutputWeights;
1319 
1320  const std::string layerName("lstm");
1321  armnn::TensorInfo inputTensorInfo({ batchSize, inputSize }, armnn::DataType::Float32);
1322  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
1323  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
1324  armnn::TensorInfo lstmTensorInfoScratchBuff({ batchSize, numUnits * 4 }, armnn::DataType::Float32);
1325 
1326  VerifyLstmLayer<armnn::LstmDescriptor> checker(
1327  layerName,
1328  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
1329  {lstmTensorInfoScratchBuff, outputStateTensorInfo, cellStateTensorInfo, outputStateTensorInfo},
1330  descriptor,
1331  params);
1332  deserializedNetwork->ExecuteStrategy(checker);
1333 }
1334 
1335 armnn::QuantizedLstmInputParams ConstantsVector2QuantizedLstmInputParams(
1336  const std::vector<armnn::ConstTensor>& constants)
1337 {
1338  armnn::QuantizedLstmInputParams params;
1339 
1340  // index for constants vector
1341  size_t i = 0;
1342 
1343  // Get input parameters
1344  params.m_InputToInputWeights = &constants[i++];
1345  params.m_InputToForgetWeights = &constants[i++];
1346  params.m_InputToCellWeights = &constants[i++];
1347  params.m_InputToOutputWeights = &constants[i++];
1348 
1349  params.m_RecurrentToInputWeights = &constants[i++];
1350  params.m_RecurrentToForgetWeights = &constants[i++];
1351  params.m_RecurrentToCellWeights = &constants[i++];
1352  params.m_RecurrentToOutputWeights = &constants[i++];
1353 
1354  params.m_InputGateBias = &constants[i++];
1355  params.m_ForgetGateBias = &constants[i++];
1356  params.m_CellBias = &constants[i++];
1357  params.m_OutputGateBias = &constants[i++];
1358 
1359  return params;
1360 }
1361 
1362 class VerifyQuantizedLstmLayer : public LayerVerifierBase
1363 {
1364 
1365 public:
1366  VerifyQuantizedLstmLayer(const std::string& layerName,
1367  const std::vector<armnn::TensorInfo>& inputInfos,
1368  const std::vector<armnn::TensorInfo>& outputInfos,
1369  const armnn::QuantizedLstmInputParams& inputParams)
1370  : LayerVerifierBase(layerName, inputInfos, outputInfos), m_InputParams(inputParams) {}
1371 
1372  void ExecuteStrategy(const armnn::IConnectableLayer* layer,
1373  const armnn::BaseDescriptor& descriptor,
1374  const std::vector<armnn::ConstTensor>& constants,
1375  const char* name,
1376  const armnn::LayerBindingId id = 0) override
1377  {
1378  armnn::IgnoreUnused(descriptor, constants, id);
1379  switch (layer->GetType())
1380  {
1381  case armnn::LayerType::Input: break;
1382  case armnn::LayerType::Output: break;
1383  case armnn::LayerType::QuantizedLstm:
1384  {
1385  VerifyNameAndConnections(layer, name);
1386  armnn::QuantizedLstmInputParams params = ConstantsVector2QuantizedLstmInputParams(constants);
1387  VerifyInputParameters(params);
1388  break;
1389  }
1390  default:
1391  {
1392  throw armnn::Exception(fmt::format("Unexpected layer type in QuantizedLstm test model:",
1393  layer->GetName()));
1394  }
1395  }
1396  }
1397 
1398 protected:
1399  void VerifyInputParameters(const armnn::QuantizedLstmInputParams& params)
1400  {
1401  VerifyConstTensors("m_InputToInputWeights",
1402  m_InputParams.m_InputToInputWeights, params.m_InputToInputWeights);
1403  VerifyConstTensors("m_InputToForgetWeights",
1404  m_InputParams.m_InputToForgetWeights, params.m_InputToForgetWeights);
1405  VerifyConstTensors("m_InputToCellWeights",
1406  m_InputParams.m_InputToCellWeights, params.m_InputToCellWeights);
1407  VerifyConstTensors("m_InputToOutputWeights",
1408  m_InputParams.m_InputToOutputWeights, params.m_InputToOutputWeights);
1409  VerifyConstTensors("m_RecurrentToInputWeights",
1410  m_InputParams.m_RecurrentToInputWeights, params.m_RecurrentToInputWeights);
1411  VerifyConstTensors("m_RecurrentToForgetWeights",
1412  m_InputParams.m_RecurrentToForgetWeights, params.m_RecurrentToForgetWeights);
1413  VerifyConstTensors("m_RecurrentToCellWeights",
1414  m_InputParams.m_RecurrentToCellWeights, params.m_RecurrentToCellWeights);
1415  VerifyConstTensors("m_RecurrentToOutputWeights",
1416  m_InputParams.m_RecurrentToOutputWeights, params.m_RecurrentToOutputWeights);
1417  VerifyConstTensors("m_InputGateBias",
1418  m_InputParams.m_InputGateBias, params.m_InputGateBias);
1419  VerifyConstTensors("m_ForgetGateBias",
1420  m_InputParams.m_ForgetGateBias, params.m_ForgetGateBias);
1421  VerifyConstTensors("m_CellBias",
1422  m_InputParams.m_CellBias, params.m_CellBias);
1423  VerifyConstTensors("m_OutputGateBias",
1424  m_InputParams.m_OutputGateBias, params.m_OutputGateBias);
1425  }
1426 
1427 private:
1428  armnn::QuantizedLstmInputParams m_InputParams;
1429 };
1430 
1431 TEST_CASE("SerializeDeserializeQuantizedLstm")
1432 {
1433  const uint32_t batchSize = 1;
1434  const uint32_t inputSize = 2;
1435  const uint32_t numUnits = 4;
1436  const uint32_t outputSize = numUnits;
1437 
1438  // Scale/Offset for input/output, cellState In/Out, weights, bias
1439  float inputOutputScale = 0.0078125f;
1440  int32_t inputOutputOffset = 128;
1441 
1442  float cellStateScale = 0.00048828125f;
1443  int32_t cellStateOffset = 0;
1444 
1445  float weightsScale = 0.00408021f;
1446  int32_t weightsOffset = 100;
1447 
1448  float biasScale = 3.1876640625e-05f;
1449  int32_t biasOffset = 0;
1450 
1451  // The shape of weight data is {outputSize, inputSize} = {4, 2}
1452  armnn::TensorShape inputToInputWeightsShape = {4, 2};
1453  std::vector<uint8_t> inputToInputWeightsData = {1, 2, 3, 4, 5, 6, 7, 8};
1454  armnn::TensorInfo inputToInputWeightsInfo(inputToInputWeightsShape,
1455  armnn::DataType::QAsymmU8,
1456  weightsScale,
1457  weightsOffset,
1458  true);
1459  armnn::ConstTensor inputToInputWeights(inputToInputWeightsInfo, inputToInputWeightsData);
1460 
1461  armnn::TensorShape inputToForgetWeightsShape = {4, 2};
1462  std::vector<uint8_t> inputToForgetWeightsData = {1, 2, 3, 4, 5, 6, 7, 8};
1463  armnn::TensorInfo inputToForgetWeightsInfo(inputToForgetWeightsShape,
1464  armnn::DataType::QAsymmU8,
1465  weightsScale,
1466  weightsOffset,
1467  true);
1468  armnn::ConstTensor inputToForgetWeights(inputToForgetWeightsInfo, inputToForgetWeightsData);
1469 
1470  armnn::TensorShape inputToCellWeightsShape = {4, 2};
1471  std::vector<uint8_t> inputToCellWeightsData = {1, 2, 3, 4, 5, 6, 7, 8};
1472  armnn::TensorInfo inputToCellWeightsInfo(inputToCellWeightsShape,
1473  armnn::DataType::QAsymmU8,
1474  weightsScale,
1475  weightsOffset,
1476  true);
1477  armnn::ConstTensor inputToCellWeights(inputToCellWeightsInfo, inputToCellWeightsData);
1478 
1479  armnn::TensorShape inputToOutputWeightsShape = {4, 2};
1480  std::vector<uint8_t> inputToOutputWeightsData = {1, 2, 3, 4, 5, 6, 7, 8};
1481  armnn::TensorInfo inputToOutputWeightsInfo(inputToOutputWeightsShape,
1482  armnn::DataType::QAsymmU8,
1483  weightsScale,
1484  weightsOffset,
1485  true);
1486  armnn::ConstTensor inputToOutputWeights(inputToOutputWeightsInfo, inputToOutputWeightsData);
1487 
1488  // The shape of recurrent weight data is {outputSize, outputSize} = {4, 4}
1489  armnn::TensorShape recurrentToInputWeightsShape = {4, 4};
1490  std::vector<uint8_t> recurrentToInputWeightsData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
1491  armnn::TensorInfo recurrentToInputWeightsInfo(recurrentToInputWeightsShape,
1492  armnn::DataType::QAsymmU8,
1493  weightsScale,
1494  weightsOffset,
1495  true);
1496  armnn::ConstTensor recurrentToInputWeights(recurrentToInputWeightsInfo, recurrentToInputWeightsData);
1497 
1498  armnn::TensorShape recurrentToForgetWeightsShape = {4, 4};
1499  std::vector<uint8_t> recurrentToForgetWeightsData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
1500  armnn::TensorInfo recurrentToForgetWeightsInfo(recurrentToForgetWeightsShape,
1501  armnn::DataType::QAsymmU8,
1502  weightsScale,
1503  weightsOffset,
1504  true);
1505  armnn::ConstTensor recurrentToForgetWeights(recurrentToForgetWeightsInfo, recurrentToForgetWeightsData);
1506 
1507  armnn::TensorShape recurrentToCellWeightsShape = {4, 4};
1508  std::vector<uint8_t> recurrentToCellWeightsData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
1509  armnn::TensorInfo recurrentToCellWeightsInfo(recurrentToCellWeightsShape,
1510  armnn::DataType::QAsymmU8,
1511  weightsScale,
1512  weightsOffset,
1513  true);
1514  armnn::ConstTensor recurrentToCellWeights(recurrentToCellWeightsInfo, recurrentToCellWeightsData);
1515 
1516  armnn::TensorShape recurrentToOutputWeightsShape = {4, 4};
1517  std::vector<uint8_t> recurrentToOutputWeightsData = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
1518  armnn::TensorInfo recurrentToOutputWeightsInfo(recurrentToOutputWeightsShape,
1519  armnn::DataType::QAsymmU8,
1520  weightsScale,
1521  weightsOffset,
1522  true);
1523  armnn::ConstTensor recurrentToOutputWeights(recurrentToOutputWeightsInfo, recurrentToOutputWeightsData);
1524 
1525  // The shape of bias data is {outputSize} = {4}
1526  armnn::TensorShape inputGateBiasShape = {4};
1527  std::vector<int32_t> inputGateBiasData = {1, 2, 3, 4};
1528  armnn::TensorInfo inputGateBiasInfo(inputGateBiasShape,
1529  armnn::DataType::Signed32,
1530  biasScale,
1531  biasOffset,
1532  true);
1533  armnn::ConstTensor inputGateBias(inputGateBiasInfo, inputGateBiasData);
1534 
1535  armnn::TensorShape forgetGateBiasShape = {4};
1536  std::vector<int32_t> forgetGateBiasData = {1, 2, 3, 4};
1537  armnn::TensorInfo forgetGateBiasInfo(forgetGateBiasShape,
1538  armnn::DataType::Signed32,
1539  biasScale,
1540  biasOffset,
1541  true);
1542  armnn::ConstTensor forgetGateBias(forgetGateBiasInfo, forgetGateBiasData);
1543 
1544  armnn::TensorShape cellBiasShape = {4};
1545  std::vector<int32_t> cellBiasData = {1, 2, 3, 4};
1546  armnn::TensorInfo cellBiasInfo(cellBiasShape,
1547  armnn::DataType::Signed32,
1548  biasScale,
1549  biasOffset,
1550  true);
1551  armnn::ConstTensor cellBias(cellBiasInfo, cellBiasData);
1552 
1553  armnn::TensorShape outputGateBiasShape = {4};
1554  std::vector<int32_t> outputGateBiasData = {1, 2, 3, 4};
1555  armnn::TensorInfo outputGateBiasInfo(outputGateBiasShape,
1556  armnn::DataType::Signed32,
1557  biasScale,
1558  biasOffset,
1559  true);
1560  armnn::ConstTensor outputGateBias(outputGateBiasInfo, outputGateBiasData);
1561 
1562  armnn::QuantizedLstmInputParams params;
1563  params.m_InputToInputWeights = &inputToInputWeights;
1564  params.m_InputToForgetWeights = &inputToForgetWeights;
1565  params.m_InputToCellWeights = &inputToCellWeights;
1566  params.m_InputToOutputWeights = &inputToOutputWeights;
1567  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
1568  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
1569  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
1570  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
1571  params.m_InputGateBias = &inputGateBias;
1572  params.m_ForgetGateBias = &forgetGateBias;
1573  params.m_CellBias = &cellBias;
1574  params.m_OutputGateBias = &outputGateBias;
1575 
1576  armnn::INetworkPtr network = armnn::INetwork::Create();
1577  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
1578  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
1579  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
1580  const std::string layerName("QuantizedLstm");
1581  armnn::IConnectableLayer* const quantizedLstmLayer = network->AddQuantizedLstmLayer(params, layerName.c_str());
1582  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(0);
1583  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(1);
1584 
1585  // Connect up
1586  armnn::TensorInfo inputTensorInfo({ batchSize, inputSize },
1587  armnn::DataType::QAsymmU8,
1588  inputOutputScale,
1589  inputOutputOffset);
1590  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits },
1591  armnn::DataType::QSymmS16,
1592  cellStateScale,
1593  cellStateOffset);
1594  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize },
1595  armnn::DataType::QAsymmU8,
1596  inputOutputScale,
1597  inputOutputOffset);
1598 
1599  inputLayer->GetOutputSlot(0).Connect(quantizedLstmLayer->GetInputSlot(0));
1600  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
1601 
1602  cellStateIn->GetOutputSlot(0).Connect(quantizedLstmLayer->GetInputSlot(1));
1603  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
1604 
1605  outputStateIn->GetOutputSlot(0).Connect(quantizedLstmLayer->GetInputSlot(2));
1606  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
1607 
1608  quantizedLstmLayer->GetOutputSlot(0).Connect(cellStateOut->GetInputSlot(0));
1609  quantizedLstmLayer->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
1610 
1611  quantizedLstmLayer->GetOutputSlot(1).Connect(outputLayer->GetInputSlot(0));
1612  quantizedLstmLayer->GetOutputSlot(1).SetTensorInfo(outputStateTensorInfo);
1613 
1614  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
1615  CHECK(deserializedNetwork);
1616 
1617  VerifyQuantizedLstmLayer checker(layerName,
1618  {inputTensorInfo, cellStateTensorInfo, outputStateTensorInfo},
1619  {cellStateTensorInfo, outputStateTensorInfo},
1620  params);
1621 
1622  deserializedNetwork->ExecuteStrategy(checker);
1623 }
1624 
1625 TEST_CASE("SerializeDeserializeQLstmBasic")
1626 {
1627  armnn::QLstmDescriptor descriptor;
1628 
1629  descriptor.m_CifgEnabled = true;
1630  descriptor.m_ProjectionEnabled = false;
1631  descriptor.m_PeepholeEnabled = false;
1632  descriptor.m_LayerNormEnabled = false;
1633 
1634  descriptor.m_CellClip = 0.0f;
1635  descriptor.m_ProjectionClip = 0.0f;
1636 
1637  descriptor.m_InputIntermediateScale = 0.00001f;
1638  descriptor.m_ForgetIntermediateScale = 0.00001f;
1639  descriptor.m_CellIntermediateScale = 0.00001f;
1640  descriptor.m_OutputIntermediateScale = 0.00001f;
1641 
1642  descriptor.m_HiddenStateScale = 0.07f;
1643  descriptor.m_HiddenStateZeroPoint = 0;
1644 
1645  const unsigned int numBatches = 2;
1646  const unsigned int inputSize = 5;
1647  const unsigned int outputSize = 4;
1648  const unsigned int numUnits = 4;
1649 
1650  // Scale/Offset quantization info
1651  float inputScale = 0.0078f;
1652  int32_t inputOffset = 0;
1653 
1654  float outputScale = 0.0078f;
1655  int32_t outputOffset = 0;
1656 
1657  float cellStateScale = 3.5002e-05f;
1658  int32_t cellStateOffset = 0;
1659 
1660  float weightsScale = 0.007f;
1661  int32_t weightsOffset = 0;
1662 
1663  float biasScale = 3.5002e-05f / 1024;
1664  int32_t biasOffset = 0;
1665 
1666  // Weights and bias tensor and quantization info
1667  armnn::TensorInfo inputWeightsInfo({numUnits, inputSize},
1668  armnn::DataType::QSymmS8,
1669  weightsScale,
1670  weightsOffset,
1671  true);
1672 
1673  armnn::TensorInfo recurrentWeightsInfo({numUnits, outputSize},
1674  armnn::DataType::QSymmS8,
1675  weightsScale,
1676  weightsOffset,
1677  true);
1678 
1679  armnn::TensorInfo biasInfo({numUnits}, armnn::DataType::Signed32, biasScale, biasOffset, true);
1680 
1681  std::vector<int8_t> inputToForgetWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
1682  std::vector<int8_t> inputToCellWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
1683  std::vector<int8_t> inputToOutputWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
1684 
1685  armnn::ConstTensor inputToForgetWeights(inputWeightsInfo, inputToForgetWeightsData);
1686  armnn::ConstTensor inputToCellWeights(inputWeightsInfo, inputToCellWeightsData);
1687  armnn::ConstTensor inputToOutputWeights(inputWeightsInfo, inputToOutputWeightsData);
1688 
1689  std::vector<int8_t> recurrentToForgetWeightsData =
1690  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
1691  std::vector<int8_t> recurrentToCellWeightsData =
1692  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
1693  std::vector<int8_t> recurrentToOutputWeightsData =
1694  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
1695 
1696  armnn::ConstTensor recurrentToForgetWeights(recurrentWeightsInfo, recurrentToForgetWeightsData);
1697  armnn::ConstTensor recurrentToCellWeights(recurrentWeightsInfo, recurrentToCellWeightsData);
1698  armnn::ConstTensor recurrentToOutputWeights(recurrentWeightsInfo, recurrentToOutputWeightsData);
1699 
1700  std::vector<int32_t> forgetGateBiasData(numUnits, 1);
1701  std::vector<int32_t> cellBiasData(numUnits, 0);
1702  std::vector<int32_t> outputGateBiasData(numUnits, 0);
1703 
1704  armnn::ConstTensor forgetGateBias(biasInfo, forgetGateBiasData);
1705  armnn::ConstTensor cellBias(biasInfo, cellBiasData);
1706  armnn::ConstTensor outputGateBias(biasInfo, outputGateBiasData);
1707 
1708  // Set up params
1709  armnn::LstmInputParams params;
1710  params.m_InputToForgetWeights = &inputToForgetWeights;
1711  params.m_InputToCellWeights = &inputToCellWeights;
1712  params.m_InputToOutputWeights = &inputToOutputWeights;
1713 
1714  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
1715  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
1716  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
1717 
1718  params.m_ForgetGateBias = &forgetGateBias;
1719  params.m_CellBias = &cellBias;
1720  params.m_OutputGateBias = &outputGateBias;
1721 
1722  // Create network
1723  armnn::INetworkPtr network = armnn::INetwork::Create();
1724  const std::string layerName("qLstm");
1725 
1726  armnn::IConnectableLayer* const input = network->AddInputLayer(0);
1727  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(1);
1728  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(2);
1729 
1730  armnn::IConnectableLayer* const qLstmLayer = network->AddQLstmLayer(descriptor, params, layerName.c_str());
1731 
1732  armnn::IConnectableLayer* const outputStateOut = network->AddOutputLayer(0);
1733  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(1);
1734  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(2);
1735 
1736  // Input/Output tensor info
1737  armnn::TensorInfo inputInfo({numBatches , inputSize},
1738  armnn::DataType::QAsymmS8,
1739  inputScale,
1740  inputOffset);
1741 
1742  armnn::TensorInfo cellStateInfo({numBatches , numUnits},
1743  armnn::DataType::QSymmS16,
1744  cellStateScale,
1745  cellStateOffset);
1746 
1747  armnn::TensorInfo outputStateInfo({numBatches , outputSize},
1748  armnn::DataType::QAsymmS8,
1749  outputScale,
1750  outputOffset);
1751 
1752  // Connect input/output slots
1753  input->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(0));
1754  input->GetOutputSlot(0).SetTensorInfo(inputInfo);
1755 
1756  outputStateIn->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(1));
1757  outputStateIn->GetOutputSlot(0).SetTensorInfo(cellStateInfo);
1758 
1759  cellStateIn->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(2));
1760  cellStateIn->GetOutputSlot(0).SetTensorInfo(outputStateInfo);
1761 
1762  qLstmLayer->GetOutputSlot(0).Connect(outputStateOut->GetInputSlot(0));
1763  qLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateInfo);
1764 
1765  qLstmLayer->GetOutputSlot(1).Connect(cellStateOut->GetInputSlot(0));
1766  qLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateInfo);
1767 
1768  qLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
1769  qLstmLayer->GetOutputSlot(2).SetTensorInfo(outputStateInfo);
1770 
1771  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
1772  CHECK(deserializedNetwork);
1773 
1774  VerifyLstmLayer<armnn::QLstmDescriptor> checker(
1775  layerName,
1776  {inputInfo, cellStateInfo, outputStateInfo},
1777  {outputStateInfo, cellStateInfo, outputStateInfo},
1778  descriptor,
1779  params);
1780 
1781  deserializedNetwork->ExecuteStrategy(checker);
1782 }
1783 
1784 TEST_CASE("SerializeDeserializeQLstmCifgLayerNorm")
1785 {
1786  armnn::QLstmDescriptor descriptor;
1787 
1788  // CIFG params are used when CIFG is disabled
1789  descriptor.m_CifgEnabled = true;
1790  descriptor.m_ProjectionEnabled = false;
1791  descriptor.m_PeepholeEnabled = false;
1792  descriptor.m_LayerNormEnabled = true;
1793 
1794  descriptor.m_CellClip = 0.0f;
1795  descriptor.m_ProjectionClip = 0.0f;
1796 
1797  descriptor.m_InputIntermediateScale = 0.00001f;
1798  descriptor.m_ForgetIntermediateScale = 0.00001f;
1799  descriptor.m_CellIntermediateScale = 0.00001f;
1800  descriptor.m_OutputIntermediateScale = 0.00001f;
1801 
1802  descriptor.m_HiddenStateScale = 0.07f;
1803  descriptor.m_HiddenStateZeroPoint = 0;
1804 
1805  const unsigned int numBatches = 2;
1806  const unsigned int inputSize = 5;
1807  const unsigned int outputSize = 4;
1808  const unsigned int numUnits = 4;
1809 
1810  // Scale/Offset quantization info
1811  float inputScale = 0.0078f;
1812  int32_t inputOffset = 0;
1813 
1814  float outputScale = 0.0078f;
1815  int32_t outputOffset = 0;
1816 
1817  float cellStateScale = 3.5002e-05f;
1818  int32_t cellStateOffset = 0;
1819 
1820  float weightsScale = 0.007f;
1821  int32_t weightsOffset = 0;
1822 
1823  float layerNormScale = 3.5002e-05f;
1824  int32_t layerNormOffset = 0;
1825 
1826  float biasScale = layerNormScale / 1024;
1827  int32_t biasOffset = 0;
1828 
1829  // Weights and bias tensor and quantization info
1830  armnn::TensorInfo inputWeightsInfo({numUnits, inputSize},
1831  armnn::DataType::QSymmS8,
1832  weightsScale,
1833  weightsOffset,
1834  true);
1835 
1836  armnn::TensorInfo recurrentWeightsInfo({numUnits, outputSize},
1837  armnn::DataType::QSymmS8,
1838  weightsScale,
1839  weightsOffset,
1840  true);
1841 
1842  armnn::TensorInfo biasInfo({numUnits},
1843  armnn::DataType::Signed32,
1844  biasScale,
1845  biasOffset,
1846  true);
1847 
1848  armnn::TensorInfo layerNormWeightsInfo({numUnits},
1849  armnn::DataType::QSymmS16,
1850  layerNormScale,
1851  layerNormOffset,
1852  true);
1853 
1854  // Mandatory params
1855  std::vector<int8_t> inputToForgetWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
1856  std::vector<int8_t> inputToCellWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
1857  std::vector<int8_t> inputToOutputWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
1858 
1859  armnn::ConstTensor inputToForgetWeights(inputWeightsInfo, inputToForgetWeightsData);
1860  armnn::ConstTensor inputToCellWeights(inputWeightsInfo, inputToCellWeightsData);
1861  armnn::ConstTensor inputToOutputWeights(inputWeightsInfo, inputToOutputWeightsData);
1862 
1863  std::vector<int8_t> recurrentToForgetWeightsData =
1864  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
1865  std::vector<int8_t> recurrentToCellWeightsData =
1866  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
1867  std::vector<int8_t> recurrentToOutputWeightsData =
1868  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
1869 
1870  armnn::ConstTensor recurrentToForgetWeights(recurrentWeightsInfo, recurrentToForgetWeightsData);
1871  armnn::ConstTensor recurrentToCellWeights(recurrentWeightsInfo, recurrentToCellWeightsData);
1872  armnn::ConstTensor recurrentToOutputWeights(recurrentWeightsInfo, recurrentToOutputWeightsData);
1873 
1874  std::vector<int32_t> forgetGateBiasData(numUnits, 1);
1875  std::vector<int32_t> cellBiasData(numUnits, 0);
1876  std::vector<int32_t> outputGateBiasData(numUnits, 0);
1877 
1878  armnn::ConstTensor forgetGateBias(biasInfo, forgetGateBiasData);
1879  armnn::ConstTensor cellBias(biasInfo, cellBiasData);
1880  armnn::ConstTensor outputGateBias(biasInfo, outputGateBiasData);
1881 
1882  // Layer Norm
1883  std::vector<int16_t> forgetLayerNormWeightsData =
1884  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
1885  std::vector<int16_t> cellLayerNormWeightsData =
1886  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
1887  std::vector<int16_t> outputLayerNormWeightsData =
1888  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
1889 
1890  armnn::ConstTensor forgetLayerNormWeights(layerNormWeightsInfo, forgetLayerNormWeightsData);
1891  armnn::ConstTensor cellLayerNormWeights(layerNormWeightsInfo, cellLayerNormWeightsData);
1892  armnn::ConstTensor outputLayerNormWeights(layerNormWeightsInfo, outputLayerNormWeightsData);
1893 
1894  // Set up params
1895  armnn::LstmInputParams params;
1896 
1897  // Mandatory params
1898  params.m_InputToForgetWeights = &inputToForgetWeights;
1899  params.m_InputToCellWeights = &inputToCellWeights;
1900  params.m_InputToOutputWeights = &inputToOutputWeights;
1901 
1902  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
1903  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
1904  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
1905 
1906  params.m_ForgetGateBias = &forgetGateBias;
1907  params.m_CellBias = &cellBias;
1908  params.m_OutputGateBias = &outputGateBias;
1909 
1910  // Layer Norm
1911  params.m_ForgetLayerNormWeights = &forgetLayerNormWeights;
1912  params.m_CellLayerNormWeights = &cellLayerNormWeights;
1913  params.m_OutputLayerNormWeights = &outputLayerNormWeights;
1914 
1915  // Create network
1916  armnn::INetworkPtr network = armnn::INetwork::Create();
1917  const std::string layerName("qLstm");
1918 
1919  armnn::IConnectableLayer* const input = network->AddInputLayer(0);
1920  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(1);
1921  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(2);
1922 
1923  armnn::IConnectableLayer* const qLstmLayer = network->AddQLstmLayer(descriptor, params, layerName.c_str());
1924 
1925  armnn::IConnectableLayer* const outputStateOut = network->AddOutputLayer(0);
1926  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(1);
1927  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(2);
1928 
1929  // Input/Output tensor info
1930  armnn::TensorInfo inputInfo({numBatches , inputSize},
1931  armnn::DataType::QAsymmS8,
1932  inputScale,
1933  inputOffset);
1934 
1935  armnn::TensorInfo cellStateInfo({numBatches , numUnits},
1936  armnn::DataType::QSymmS16,
1937  cellStateScale,
1938  cellStateOffset);
1939 
1940  armnn::TensorInfo outputStateInfo({numBatches , outputSize},
1941  armnn::DataType::QAsymmS8,
1942  outputScale,
1943  outputOffset);
1944 
1945  // Connect input/output slots
1946  input->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(0));
1947  input->GetOutputSlot(0).SetTensorInfo(inputInfo);
1948 
1949  outputStateIn->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(1));
1950  outputStateIn->GetOutputSlot(0).SetTensorInfo(cellStateInfo);
1951 
1952  cellStateIn->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(2));
1953  cellStateIn->GetOutputSlot(0).SetTensorInfo(outputStateInfo);
1954 
1955  qLstmLayer->GetOutputSlot(0).Connect(outputStateOut->GetInputSlot(0));
1956  qLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateInfo);
1957 
1958  qLstmLayer->GetOutputSlot(1).Connect(cellStateOut->GetInputSlot(0));
1959  qLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateInfo);
1960 
1961  qLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
1962  qLstmLayer->GetOutputSlot(2).SetTensorInfo(outputStateInfo);
1963 
1964  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
1965  CHECK(deserializedNetwork);
1966 
1967  VerifyLstmLayer<armnn::QLstmDescriptor> checker(layerName,
1968  {inputInfo, cellStateInfo, outputStateInfo},
1969  {outputStateInfo, cellStateInfo, outputStateInfo},
1970  descriptor,
1971  params);
1972 
1973  deserializedNetwork->ExecuteStrategy(checker);
1974 }
1975 
1976 TEST_CASE("SerializeDeserializeQLstmAdvanced")
1977 {
1978  armnn::QLstmDescriptor descriptor;
1979 
1980  descriptor.m_CifgEnabled = false;
1981  descriptor.m_ProjectionEnabled = true;
1982  descriptor.m_PeepholeEnabled = true;
1983  descriptor.m_LayerNormEnabled = true;
1984 
1985  descriptor.m_CellClip = 0.1f;
1986  descriptor.m_ProjectionClip = 0.1f;
1987 
1988  descriptor.m_InputIntermediateScale = 0.00001f;
1989  descriptor.m_ForgetIntermediateScale = 0.00001f;
1990  descriptor.m_CellIntermediateScale = 0.00001f;
1991  descriptor.m_OutputIntermediateScale = 0.00001f;
1992 
1993  descriptor.m_HiddenStateScale = 0.07f;
1994  descriptor.m_HiddenStateZeroPoint = 0;
1995 
1996  const unsigned int numBatches = 2;
1997  const unsigned int inputSize = 5;
1998  const unsigned int outputSize = 4;
1999  const unsigned int numUnits = 4;
2000 
2001  // Scale/Offset quantization info
2002  float inputScale = 0.0078f;
2003  int32_t inputOffset = 0;
2004 
2005  float outputScale = 0.0078f;
2006  int32_t outputOffset = 0;
2007 
2008  float cellStateScale = 3.5002e-05f;
2009  int32_t cellStateOffset = 0;
2010 
2011  float weightsScale = 0.007f;
2012  int32_t weightsOffset = 0;
2013 
2014  float layerNormScale = 3.5002e-05f;
2015  int32_t layerNormOffset = 0;
2016 
2017  float biasScale = layerNormScale / 1024;
2018  int32_t biasOffset = 0;
2019 
2020  // Weights and bias tensor and quantization info
2021  armnn::TensorInfo inputWeightsInfo({numUnits, inputSize},
2022  armnn::DataType::QSymmS8,
2023  weightsScale,
2024  weightsOffset,
2025  true);
2026 
2027  armnn::TensorInfo recurrentWeightsInfo({numUnits, outputSize},
2028  armnn::DataType::QSymmS8,
2029  weightsScale,
2030  weightsOffset,
2031  true);
2032 
2033  armnn::TensorInfo biasInfo({numUnits},
2034  armnn::DataType::Signed32,
2035  biasScale,
2036  biasOffset,
2037  true);
2038 
2039  armnn::TensorInfo peepholeWeightsInfo({numUnits},
2040  armnn::DataType::QSymmS16,
2041  weightsScale,
2042  weightsOffset,
2043  true);
2044 
2045  armnn::TensorInfo layerNormWeightsInfo({numUnits},
2046  armnn::DataType::QSymmS16,
2047  layerNormScale,
2048  layerNormOffset,
2049  true);
2050 
2051  armnn::TensorInfo projectionWeightsInfo({outputSize, numUnits},
2052  armnn::DataType::QSymmS8,
2053  weightsScale,
2054  weightsOffset,
2055  true);
2056 
2057  // Mandatory params
2058  std::vector<int8_t> inputToForgetWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
2059  std::vector<int8_t> inputToCellWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
2060  std::vector<int8_t> inputToOutputWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
2061 
2062  armnn::ConstTensor inputToForgetWeights(inputWeightsInfo, inputToForgetWeightsData);
2063  armnn::ConstTensor inputToCellWeights(inputWeightsInfo, inputToCellWeightsData);
2064  armnn::ConstTensor inputToOutputWeights(inputWeightsInfo, inputToOutputWeightsData);
2065 
2066  std::vector<int8_t> recurrentToForgetWeightsData =
2067  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
2068  std::vector<int8_t> recurrentToCellWeightsData =
2069  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
2070  std::vector<int8_t> recurrentToOutputWeightsData =
2071  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
2072 
2073  armnn::ConstTensor recurrentToForgetWeights(recurrentWeightsInfo, recurrentToForgetWeightsData);
2074  armnn::ConstTensor recurrentToCellWeights(recurrentWeightsInfo, recurrentToCellWeightsData);
2075  armnn::ConstTensor recurrentToOutputWeights(recurrentWeightsInfo, recurrentToOutputWeightsData);
2076 
2077  std::vector<int32_t> forgetGateBiasData(numUnits, 1);
2078  std::vector<int32_t> cellBiasData(numUnits, 0);
2079  std::vector<int32_t> outputGateBiasData(numUnits, 0);
2080 
2081  armnn::ConstTensor forgetGateBias(biasInfo, forgetGateBiasData);
2082  armnn::ConstTensor cellBias(biasInfo, cellBiasData);
2083  armnn::ConstTensor outputGateBias(biasInfo, outputGateBiasData);
2084 
2085  // CIFG
2086  std::vector<int8_t> inputToInputWeightsData = GenerateRandomData<int8_t>(inputWeightsInfo.GetNumElements());
2087  std::vector<int8_t> recurrentToInputWeightsData =
2088  GenerateRandomData<int8_t>(recurrentWeightsInfo.GetNumElements());
2089  std::vector<int32_t> inputGateBiasData(numUnits, 1);
2090 
2091  armnn::ConstTensor inputToInputWeights(inputWeightsInfo, inputToInputWeightsData);
2092  armnn::ConstTensor recurrentToInputWeights(recurrentWeightsInfo, recurrentToInputWeightsData);
2093  armnn::ConstTensor inputGateBias(biasInfo, inputGateBiasData);
2094 
2095  // Peephole
2096  std::vector<int16_t> cellToInputWeightsData = GenerateRandomData<int16_t>(peepholeWeightsInfo.GetNumElements());
2097  std::vector<int16_t> cellToForgetWeightsData = GenerateRandomData<int16_t>(peepholeWeightsInfo.GetNumElements());
2098  std::vector<int16_t> cellToOutputWeightsData = GenerateRandomData<int16_t>(peepholeWeightsInfo.GetNumElements());
2099 
2100  armnn::ConstTensor cellToInputWeights(peepholeWeightsInfo, cellToInputWeightsData);
2101  armnn::ConstTensor cellToForgetWeights(peepholeWeightsInfo, cellToForgetWeightsData);
2102  armnn::ConstTensor cellToOutputWeights(peepholeWeightsInfo, cellToOutputWeightsData);
2103 
2104  // Projection
2105  std::vector<int8_t> projectionWeightsData = GenerateRandomData<int8_t>(projectionWeightsInfo.GetNumElements());
2106  std::vector<int32_t> projectionBiasData(outputSize, 1);
2107 
2108  armnn::ConstTensor projectionWeights(projectionWeightsInfo, projectionWeightsData);
2109  armnn::ConstTensor projectionBias(biasInfo, projectionBiasData);
2110 
2111  // Layer Norm
2112  std::vector<int16_t> inputLayerNormWeightsData =
2113  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
2114  std::vector<int16_t> forgetLayerNormWeightsData =
2115  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
2116  std::vector<int16_t> cellLayerNormWeightsData =
2117  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
2118  std::vector<int16_t> outputLayerNormWeightsData =
2119  GenerateRandomData<int16_t>(layerNormWeightsInfo.GetNumElements());
2120 
2121  armnn::ConstTensor inputLayerNormWeights(layerNormWeightsInfo, inputLayerNormWeightsData);
2122  armnn::ConstTensor forgetLayerNormWeights(layerNormWeightsInfo, forgetLayerNormWeightsData);
2123  armnn::ConstTensor cellLayerNormWeights(layerNormWeightsInfo, cellLayerNormWeightsData);
2124  armnn::ConstTensor outputLayerNormWeights(layerNormWeightsInfo, outputLayerNormWeightsData);
2125 
2126  // Set up params
2127  armnn::LstmInputParams params;
2128 
2129  // Mandatory params
2130  params.m_InputToForgetWeights = &inputToForgetWeights;
2131  params.m_InputToCellWeights = &inputToCellWeights;
2132  params.m_InputToOutputWeights = &inputToOutputWeights;
2133 
2134  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
2135  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
2136  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
2137 
2138  params.m_ForgetGateBias = &forgetGateBias;
2139  params.m_CellBias = &cellBias;
2140  params.m_OutputGateBias = &outputGateBias;
2141 
2142  // CIFG
2143  params.m_InputToInputWeights = &inputToInputWeights;
2144  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
2145  params.m_InputGateBias = &inputGateBias;
2146 
2147  // Peephole
2148  params.m_CellToInputWeights = &cellToInputWeights;
2149  params.m_CellToForgetWeights = &cellToForgetWeights;
2150  params.m_CellToOutputWeights = &cellToOutputWeights;
2151 
2152  // Projection
2153  params.m_ProjectionWeights = &projectionWeights;
2154  params.m_ProjectionBias = &projectionBias;
2155 
2156  // Layer Norm
2157  params.m_InputLayerNormWeights = &inputLayerNormWeights;
2158  params.m_ForgetLayerNormWeights = &forgetLayerNormWeights;
2159  params.m_CellLayerNormWeights = &cellLayerNormWeights;
2160  params.m_OutputLayerNormWeights = &outputLayerNormWeights;
2161 
2162  // Create network
2163  armnn::INetworkPtr network = armnn::INetwork::Create();
2164  const std::string layerName("qLstm");
2165 
2166  armnn::IConnectableLayer* const input = network->AddInputLayer(0);
2167  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(1);
2168  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(2);
2169 
2170  armnn::IConnectableLayer* const qLstmLayer = network->AddQLstmLayer(descriptor, params, layerName.c_str());
2171 
2172  armnn::IConnectableLayer* const outputStateOut = network->AddOutputLayer(0);
2173  armnn::IConnectableLayer* const cellStateOut = network->AddOutputLayer(1);
2174  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(2);
2175 
2176  // Input/Output tensor info
2177  armnn::TensorInfo inputInfo({numBatches , inputSize},
2178  armnn::DataType::QAsymmS8,
2179  inputScale,
2180  inputOffset);
2181 
2182  armnn::TensorInfo cellStateInfo({numBatches , numUnits},
2183  armnn::DataType::QSymmS16,
2184  cellStateScale,
2185  cellStateOffset);
2186 
2187  armnn::TensorInfo outputStateInfo({numBatches , outputSize},
2188  armnn::DataType::QAsymmS8,
2189  outputScale,
2190  outputOffset);
2191 
2192  // Connect input/output slots
2193  input->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(0));
2194  input->GetOutputSlot(0).SetTensorInfo(inputInfo);
2195 
2196  outputStateIn->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(1));
2197  outputStateIn->GetOutputSlot(0).SetTensorInfo(cellStateInfo);
2198 
2199  cellStateIn->GetOutputSlot(0).Connect(qLstmLayer->GetInputSlot(2));
2200  cellStateIn->GetOutputSlot(0).SetTensorInfo(outputStateInfo);
2201 
2202  qLstmLayer->GetOutputSlot(0).Connect(outputStateOut->GetInputSlot(0));
2203  qLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateInfo);
2204 
2205  qLstmLayer->GetOutputSlot(1).Connect(cellStateOut->GetInputSlot(0));
2206  qLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateInfo);
2207 
2208  qLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
2209  qLstmLayer->GetOutputSlot(2).SetTensorInfo(outputStateInfo);
2210 
2211  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
2212  CHECK(deserializedNetwork);
2213 
2214  VerifyLstmLayer<armnn::QLstmDescriptor> checker(layerName,
2215  {inputInfo, cellStateInfo, outputStateInfo},
2216  {outputStateInfo, cellStateInfo, outputStateInfo},
2217  descriptor,
2218  params);
2219 
2220  deserializedNetwork->ExecuteStrategy(checker);
2221 }
2222 
2223 TEST_CASE("SerializeDeserializeUnidirectionalSequenceLstmCifgPeepholeNoProjection")
2224 {
2225  armnn::UnidirectionalSequenceLstmDescriptor descriptor;
2226  descriptor.m_ActivationFunc = 4;
2227  descriptor.m_ClippingThresProj = 0.0f;
2228  descriptor.m_ClippingThresCell = 0.0f;
2229  descriptor.m_CifgEnabled = true; // if this is true then we DON'T need to set the OptCifgParams
2230  descriptor.m_ProjectionEnabled = false;
2231  descriptor.m_PeepholeEnabled = true;
2232  descriptor.m_TimeMajor = false;
2233 
2234  const uint32_t batchSize = 1;
2235  const uint32_t timeSize = 2;
2236  const uint32_t inputSize = 2;
2237  const uint32_t numUnits = 4;
2238  const uint32_t outputSize = numUnits;
2239 
2240  armnn::TensorInfo inputWeightsInfo1({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2241  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
2242  armnn::ConstTensor inputToForgetWeights(inputWeightsInfo1, inputToForgetWeightsData);
2243 
2244  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
2245  armnn::ConstTensor inputToCellWeights(inputWeightsInfo1, inputToCellWeightsData);
2246 
2247  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
2248  armnn::ConstTensor inputToOutputWeights(inputWeightsInfo1, inputToOutputWeightsData);
2249 
2250  armnn::TensorInfo inputWeightsInfo2({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2251  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
2252  armnn::ConstTensor recurrentToForgetWeights(inputWeightsInfo2, recurrentToForgetWeightsData);
2253 
2254  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
2255  armnn::ConstTensor recurrentToCellWeights(inputWeightsInfo2, recurrentToCellWeightsData);
2256 
2257  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
2258  armnn::ConstTensor recurrentToOutputWeights(inputWeightsInfo2, recurrentToOutputWeightsData);
2259 
2260  armnn::TensorInfo inputWeightsInfo3({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
2261  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo3.GetNumElements());
2262  armnn::ConstTensor cellToForgetWeights(inputWeightsInfo3, cellToForgetWeightsData);
2263 
2264  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo3.GetNumElements());
2265  armnn::ConstTensor cellToOutputWeights(inputWeightsInfo3, cellToOutputWeightsData);
2266 
2267  std::vector<float> forgetGateBiasData(numUnits, 1.0f);
2268  armnn::ConstTensor forgetGateBias(inputWeightsInfo3, forgetGateBiasData);
2269 
2270  std::vector<float> cellBiasData(numUnits, 0.0f);
2271  armnn::ConstTensor cellBias(inputWeightsInfo3, cellBiasData);
2272 
2273  std::vector<float> outputGateBiasData(numUnits, 0.0f);
2274  armnn::ConstTensor outputGateBias(inputWeightsInfo3, outputGateBiasData);
2275 
2276  armnn::LstmInputParams params;
2277  params.m_InputToForgetWeights = &inputToForgetWeights;
2278  params.m_InputToCellWeights = &inputToCellWeights;
2279  params.m_InputToOutputWeights = &inputToOutputWeights;
2280  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
2281  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
2282  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
2283  params.m_ForgetGateBias = &forgetGateBias;
2284  params.m_CellBias = &cellBias;
2285  params.m_OutputGateBias = &outputGateBias;
2286  params.m_CellToForgetWeights = &cellToForgetWeights;
2287  params.m_CellToOutputWeights = &cellToOutputWeights;
2288 
2289  armnn::INetworkPtr network = armnn::INetwork::Create();
2290  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
2291  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
2292  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
2293  const std::string layerName("UnidirectionalSequenceLstm");
2294  armnn::IConnectableLayer* const unidirectionalSequenceLstmLayer =
2295  network->AddUnidirectionalSequenceLstmLayer(descriptor, params, layerName.c_str());
2296  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(0);
2297 
2298  // connect up
2299  armnn::TensorInfo inputTensorInfo({ batchSize, timeSize, inputSize }, armnn::DataType::Float32);
2300  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
2301  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
2302  armnn::TensorInfo outputStateOutTensorInfo({ batchSize, timeSize, outputSize }, armnn::DataType::Float32);
2303  armnn::TensorInfo cellStateOutTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
2304  armnn::TensorInfo outputTensorInfo({ batchSize, timeSize, outputSize }, armnn::DataType::Float32);
2305 
2306  inputLayer->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(0));
2307  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
2308 
2309  outputStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(1));
2310  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
2311 
2312  cellStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(2));
2313  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
2314 
2315  unidirectionalSequenceLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateOutTensorInfo);
2316  unidirectionalSequenceLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateOutTensorInfo);
2317  unidirectionalSequenceLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
2318  unidirectionalSequenceLstmLayer->GetOutputSlot(2).SetTensorInfo(outputTensorInfo);
2319 
2320  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
2321  CHECK(deserializedNetwork);
2322 
2323  VerifyLstmLayer<armnn::UnidirectionalSequenceLstmDescriptor> checker(
2324  layerName,
2325  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
2326  {outputStateOutTensorInfo, cellStateOutTensorInfo, outputTensorInfo},
2327  descriptor,
2328  params);
2329  deserializedNetwork->ExecuteStrategy(checker);
2330 }
2331 
2332 TEST_CASE("SerializeDeserializeUnidirectionalSequenceLstmNoCifgWithPeepholeAndProjection")
2333 {
2334  armnn::UnidirectionalSequenceLstmDescriptor descriptor;
2335  descriptor.m_ActivationFunc = 4;
2336  descriptor.m_ClippingThresProj = 0.0f;
2337  descriptor.m_ClippingThresCell = 0.0f;
2338  descriptor.m_CifgEnabled = false; // if this is true then we DON'T need to set the OptCifgParams
2339  descriptor.m_ProjectionEnabled = true;
2340  descriptor.m_PeepholeEnabled = true;
2341  descriptor.m_TimeMajor = false;
2342 
2343  const uint32_t batchSize = 2;
2344  const uint32_t timeSize = 2;
2345  const uint32_t inputSize = 5;
2346  const uint32_t numUnits = 20;
2347  const uint32_t outputSize = 16;
2348 
2349  armnn::TensorInfo tensorInfo20x5({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2350  std::vector<float> inputToInputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2351  armnn::ConstTensor inputToInputWeights(tensorInfo20x5, inputToInputWeightsData);
2352 
2353  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2354  armnn::ConstTensor inputToForgetWeights(tensorInfo20x5, inputToForgetWeightsData);
2355 
2356  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2357  armnn::ConstTensor inputToCellWeights(tensorInfo20x5, inputToCellWeightsData);
2358 
2359  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2360  armnn::ConstTensor inputToOutputWeights(tensorInfo20x5, inputToOutputWeightsData);
2361 
2362  armnn::TensorInfo tensorInfo20({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
2363  std::vector<float> inputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2364  armnn::ConstTensor inputGateBias(tensorInfo20, inputGateBiasData);
2365 
2366  std::vector<float> forgetGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2367  armnn::ConstTensor forgetGateBias(tensorInfo20, forgetGateBiasData);
2368 
2369  std::vector<float> cellBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2370  armnn::ConstTensor cellBias(tensorInfo20, cellBiasData);
2371 
2372  std::vector<float> outputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2373  armnn::ConstTensor outputGateBias(tensorInfo20, outputGateBiasData);
2374 
2375  armnn::TensorInfo tensorInfo20x16({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2376  std::vector<float> recurrentToInputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2377  armnn::ConstTensor recurrentToInputWeights(tensorInfo20x16, recurrentToInputWeightsData);
2378 
2379  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2380  armnn::ConstTensor recurrentToForgetWeights(tensorInfo20x16, recurrentToForgetWeightsData);
2381 
2382  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2383  armnn::ConstTensor recurrentToCellWeights(tensorInfo20x16, recurrentToCellWeightsData);
2384 
2385  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2386  armnn::ConstTensor recurrentToOutputWeights(tensorInfo20x16, recurrentToOutputWeightsData);
2387 
2388  std::vector<float> cellToInputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2389  armnn::ConstTensor cellToInputWeights(tensorInfo20, cellToInputWeightsData);
2390 
2391  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2392  armnn::ConstTensor cellToForgetWeights(tensorInfo20, cellToForgetWeightsData);
2393 
2394  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2395  armnn::ConstTensor cellToOutputWeights(tensorInfo20, cellToOutputWeightsData);
2396 
2397  armnn::TensorInfo tensorInfo16x20({outputSize, numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
2398  std::vector<float> projectionWeightsData = GenerateRandomData<float>(tensorInfo16x20.GetNumElements());
2399  armnn::ConstTensor projectionWeights(tensorInfo16x20, projectionWeightsData);
2400 
2401  armnn::TensorInfo tensorInfo16({outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2402  std::vector<float> projectionBiasData(outputSize, 0.f);
2403  armnn::ConstTensor projectionBias(tensorInfo16, projectionBiasData);
2404 
2405  armnn::LstmInputParams params;
2406  params.m_InputToForgetWeights = &inputToForgetWeights;
2407  params.m_InputToCellWeights = &inputToCellWeights;
2408  params.m_InputToOutputWeights = &inputToOutputWeights;
2409  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
2410  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
2411  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
2412  params.m_ForgetGateBias = &forgetGateBias;
2413  params.m_CellBias = &cellBias;
2414  params.m_OutputGateBias = &outputGateBias;
2415 
2416  // additional params because: descriptor.m_CifgEnabled = false
2417  params.m_InputToInputWeights = &inputToInputWeights;
2418  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
2419  params.m_CellToInputWeights = &cellToInputWeights;
2420  params.m_InputGateBias = &inputGateBias;
2421 
2422  // additional params because: descriptor.m_ProjectionEnabled = true
2423  params.m_ProjectionWeights = &projectionWeights;
2424  params.m_ProjectionBias = &projectionBias;
2425 
2426  // additional params because: descriptor.m_PeepholeEnabled = true
2427  params.m_CellToForgetWeights = &cellToForgetWeights;
2428  params.m_CellToOutputWeights = &cellToOutputWeights;
2429 
2430  armnn::INetworkPtr network = armnn::INetwork::Create();
2431  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
2432  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
2433  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
2434  const std::string layerName("unidirectionalSequenceLstm");
2435  armnn::IConnectableLayer* const unidirectionalSequenceLstmLayer =
2436  network->AddUnidirectionalSequenceLstmLayer(descriptor, params, layerName.c_str());
2437  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(0);
2438 
2439  // connect up
2440  armnn::TensorInfo inputTensorInfo({ batchSize, timeSize, inputSize }, armnn::DataType::Float32);
2441  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
2442  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
2443  armnn::TensorInfo outputStateOutTensorInfo({ batchSize, timeSize, outputSize }, armnn::DataType::Float32);
2444  armnn::TensorInfo cellStateOutTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
2445  armnn::TensorInfo outputTensorInfo({ batchSize, timeSize, outputSize }, armnn::DataType::Float32);
2446 
2447  inputLayer->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(0));
2448  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
2449 
2450  outputStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(1));
2451  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
2452 
2453  cellStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(2));
2454  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
2455 
2456  unidirectionalSequenceLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateOutTensorInfo);
2457  unidirectionalSequenceLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateOutTensorInfo);
2458  unidirectionalSequenceLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
2459  unidirectionalSequenceLstmLayer->GetOutputSlot(2).SetTensorInfo(outputTensorInfo);
2460 
2461  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
2462  CHECK(deserializedNetwork);
2463 
2464  VerifyLstmLayer<armnn::UnidirectionalSequenceLstmDescriptor> checker(
2465  layerName,
2466  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
2467  {outputStateOutTensorInfo, cellStateOutTensorInfo, outputTensorInfo},
2468  descriptor,
2469  params);
2470  deserializedNetwork->ExecuteStrategy(checker);
2471 }
2472 
2473 TEST_CASE("SerializeDeserializeUnidirectionalSequenceLstmNoCifgWithPeepholeWithProjectionWithLayerNorm")
2474 {
2475  armnn::UnidirectionalSequenceLstmDescriptor descriptor;
2476  descriptor.m_ActivationFunc = 4;
2477  descriptor.m_ClippingThresProj = 0.0f;
2478  descriptor.m_ClippingThresCell = 0.0f;
2479  descriptor.m_CifgEnabled = false; // if this is true then we DON'T need to set the OptCifgParams
2480  descriptor.m_ProjectionEnabled = true;
2481  descriptor.m_PeepholeEnabled = true;
2482  descriptor.m_LayerNormEnabled = true;
2483  descriptor.m_TimeMajor = false;
2484 
2485  const uint32_t batchSize = 2;
2486  const uint32_t timeSize = 2;
2487  const uint32_t inputSize = 5;
2488  const uint32_t numUnits = 20;
2489  const uint32_t outputSize = 16;
2490 
2491  armnn::TensorInfo tensorInfo20x5({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2492  std::vector<float> inputToInputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2493  armnn::ConstTensor inputToInputWeights(tensorInfo20x5, inputToInputWeightsData);
2494 
2495  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2496  armnn::ConstTensor inputToForgetWeights(tensorInfo20x5, inputToForgetWeightsData);
2497 
2498  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2499  armnn::ConstTensor inputToCellWeights(tensorInfo20x5, inputToCellWeightsData);
2500 
2501  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x5.GetNumElements());
2502  armnn::ConstTensor inputToOutputWeights(tensorInfo20x5, inputToOutputWeightsData);
2503 
2504  armnn::TensorInfo tensorInfo20({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
2505  std::vector<float> inputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2506  armnn::ConstTensor inputGateBias(tensorInfo20, inputGateBiasData);
2507 
2508  std::vector<float> forgetGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2509  armnn::ConstTensor forgetGateBias(tensorInfo20, forgetGateBiasData);
2510 
2511  std::vector<float> cellBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2512  armnn::ConstTensor cellBias(tensorInfo20, cellBiasData);
2513 
2514  std::vector<float> outputGateBiasData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2515  armnn::ConstTensor outputGateBias(tensorInfo20, outputGateBiasData);
2516 
2517  armnn::TensorInfo tensorInfo20x16({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2518  std::vector<float> recurrentToInputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2519  armnn::ConstTensor recurrentToInputWeights(tensorInfo20x16, recurrentToInputWeightsData);
2520 
2521  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2522  armnn::ConstTensor recurrentToForgetWeights(tensorInfo20x16, recurrentToForgetWeightsData);
2523 
2524  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2525  armnn::ConstTensor recurrentToCellWeights(tensorInfo20x16, recurrentToCellWeightsData);
2526 
2527  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(tensorInfo20x16.GetNumElements());
2528  armnn::ConstTensor recurrentToOutputWeights(tensorInfo20x16, recurrentToOutputWeightsData);
2529 
2530  std::vector<float> cellToInputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2531  armnn::ConstTensor cellToInputWeights(tensorInfo20, cellToInputWeightsData);
2532 
2533  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2534  armnn::ConstTensor cellToForgetWeights(tensorInfo20, cellToForgetWeightsData);
2535 
2536  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2537  armnn::ConstTensor cellToOutputWeights(tensorInfo20, cellToOutputWeightsData);
2538 
2539  armnn::TensorInfo tensorInfo16x20({outputSize, numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
2540  std::vector<float> projectionWeightsData = GenerateRandomData<float>(tensorInfo16x20.GetNumElements());
2541  armnn::ConstTensor projectionWeights(tensorInfo16x20, projectionWeightsData);
2542 
2543  armnn::TensorInfo tensorInfo16({outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2544  std::vector<float> projectionBiasData(outputSize, 0.f);
2545  armnn::ConstTensor projectionBias(tensorInfo16, projectionBiasData);
2546 
2547  std::vector<float> inputLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2548  armnn::ConstTensor inputLayerNormWeights(tensorInfo20, forgetGateBiasData);
2549 
2550  std::vector<float> forgetLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2551  armnn::ConstTensor forgetLayerNormWeights(tensorInfo20, forgetGateBiasData);
2552 
2553  std::vector<float> cellLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2554  armnn::ConstTensor cellLayerNormWeights(tensorInfo20, forgetGateBiasData);
2555 
2556  std::vector<float> outLayerNormWeightsData = GenerateRandomData<float>(tensorInfo20.GetNumElements());
2557  armnn::ConstTensor outLayerNormWeights(tensorInfo20, forgetGateBiasData);
2558 
2559  armnn::LstmInputParams params;
2560  params.m_InputToForgetWeights = &inputToForgetWeights;
2561  params.m_InputToCellWeights = &inputToCellWeights;
2562  params.m_InputToOutputWeights = &inputToOutputWeights;
2563  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
2564  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
2565  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
2566  params.m_ForgetGateBias = &forgetGateBias;
2567  params.m_CellBias = &cellBias;
2568  params.m_OutputGateBias = &outputGateBias;
2569 
2570  // additional params because: descriptor.m_CifgEnabled = false
2571  params.m_InputToInputWeights = &inputToInputWeights;
2572  params.m_RecurrentToInputWeights = &recurrentToInputWeights;
2573  params.m_CellToInputWeights = &cellToInputWeights;
2574  params.m_InputGateBias = &inputGateBias;
2575 
2576  // additional params because: descriptor.m_ProjectionEnabled = true
2577  params.m_ProjectionWeights = &projectionWeights;
2578  params.m_ProjectionBias = &projectionBias;
2579 
2580  // additional params because: descriptor.m_PeepholeEnabled = true
2581  params.m_CellToForgetWeights = &cellToForgetWeights;
2582  params.m_CellToOutputWeights = &cellToOutputWeights;
2583 
2584  // additional params because: despriptor.m_LayerNormEnabled = true
2585  params.m_InputLayerNormWeights = &inputLayerNormWeights;
2586  params.m_ForgetLayerNormWeights = &forgetLayerNormWeights;
2587  params.m_CellLayerNormWeights = &cellLayerNormWeights;
2588  params.m_OutputLayerNormWeights = &outLayerNormWeights;
2589 
2590  armnn::INetworkPtr network = armnn::INetwork::Create();
2591  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
2592  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
2593  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
2594  const std::string layerName("unidirectionalSequenceLstm");
2595  armnn::IConnectableLayer* const unidirectionalSequenceLstmLayer =
2596  network->AddUnidirectionalSequenceLstmLayer(descriptor, params, layerName.c_str());
2597  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(0);
2598 
2599  // connect up
2600  armnn::TensorInfo inputTensorInfo({ batchSize, timeSize, inputSize }, armnn::DataType::Float32);
2601  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
2602  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
2603  armnn::TensorInfo outputStateOutTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
2604  armnn::TensorInfo cellStateOutTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
2605  armnn::TensorInfo outputTensorInfo({ batchSize, timeSize, outputSize }, armnn::DataType::Float32);
2606 
2607  inputLayer->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(0));
2608  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
2609 
2610  outputStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(1));
2611  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
2612 
2613  cellStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(2));
2614  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
2615 
2616  unidirectionalSequenceLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateOutTensorInfo);
2617  unidirectionalSequenceLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateOutTensorInfo);
2618  unidirectionalSequenceLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
2619  unidirectionalSequenceLstmLayer->GetOutputSlot(2).SetTensorInfo(outputTensorInfo);
2620 
2621  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
2622  CHECK(deserializedNetwork);
2623 
2624  VerifyLstmLayer<armnn::UnidirectionalSequenceLstmDescriptor> checker(
2625  layerName,
2626  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
2627  {outputStateOutTensorInfo, cellStateOutTensorInfo, outputTensorInfo},
2628  descriptor,
2629  params);
2630  deserializedNetwork->ExecuteStrategy(checker);
2631 }
2632 
2633 TEST_CASE("SerializeDeserializeUnidirectionalSequenceLstmCifgPeepholeNoProjectionTimeMajor")
2634 {
2635  armnn::UnidirectionalSequenceLstmDescriptor descriptor;
2636  descriptor.m_ActivationFunc = 4;
2637  descriptor.m_ClippingThresProj = 0.0f;
2638  descriptor.m_ClippingThresCell = 0.0f;
2639  descriptor.m_CifgEnabled = true; // if this is true then we DON'T need to set the OptCifgParams
2640  descriptor.m_ProjectionEnabled = false;
2641  descriptor.m_PeepholeEnabled = true;
2642  descriptor.m_TimeMajor = true;
2643 
2644  const uint32_t batchSize = 1;
2645  const uint32_t timeSize = 2;
2646  const uint32_t inputSize = 2;
2647  const uint32_t numUnits = 4;
2648  const uint32_t outputSize = numUnits;
2649 
2650  armnn::TensorInfo inputWeightsInfo1({numUnits, inputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2651  std::vector<float> inputToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
2652  armnn::ConstTensor inputToForgetWeights(inputWeightsInfo1, inputToForgetWeightsData);
2653 
2654  std::vector<float> inputToCellWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
2655  armnn::ConstTensor inputToCellWeights(inputWeightsInfo1, inputToCellWeightsData);
2656 
2657  std::vector<float> inputToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo1.GetNumElements());
2658  armnn::ConstTensor inputToOutputWeights(inputWeightsInfo1, inputToOutputWeightsData);
2659 
2660  armnn::TensorInfo inputWeightsInfo2({numUnits, outputSize}, armnn::DataType::Float32, 0.0f, 0, true);
2661  std::vector<float> recurrentToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
2662  armnn::ConstTensor recurrentToForgetWeights(inputWeightsInfo2, recurrentToForgetWeightsData);
2663 
2664  std::vector<float> recurrentToCellWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
2665  armnn::ConstTensor recurrentToCellWeights(inputWeightsInfo2, recurrentToCellWeightsData);
2666 
2667  std::vector<float> recurrentToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo2.GetNumElements());
2668  armnn::ConstTensor recurrentToOutputWeights(inputWeightsInfo2, recurrentToOutputWeightsData);
2669 
2670  armnn::TensorInfo inputWeightsInfo3({numUnits}, armnn::DataType::Float32, 0.0f, 0, true);
2671  std::vector<float> cellToForgetWeightsData = GenerateRandomData<float>(inputWeightsInfo3.GetNumElements());
2672  armnn::ConstTensor cellToForgetWeights(inputWeightsInfo3, cellToForgetWeightsData);
2673 
2674  std::vector<float> cellToOutputWeightsData = GenerateRandomData<float>(inputWeightsInfo3.GetNumElements());
2675  armnn::ConstTensor cellToOutputWeights(inputWeightsInfo3, cellToOutputWeightsData);
2676 
2677  std::vector<float> forgetGateBiasData(numUnits, 1.0f);
2678  armnn::ConstTensor forgetGateBias(inputWeightsInfo3, forgetGateBiasData);
2679 
2680  std::vector<float> cellBiasData(numUnits, 0.0f);
2681  armnn::ConstTensor cellBias(inputWeightsInfo3, cellBiasData);
2682 
2683  std::vector<float> outputGateBiasData(numUnits, 0.0f);
2684  armnn::ConstTensor outputGateBias(inputWeightsInfo3, outputGateBiasData);
2685 
2686  armnn::LstmInputParams params;
2687  params.m_InputToForgetWeights = &inputToForgetWeights;
2688  params.m_InputToCellWeights = &inputToCellWeights;
2689  params.m_InputToOutputWeights = &inputToOutputWeights;
2690  params.m_RecurrentToForgetWeights = &recurrentToForgetWeights;
2691  params.m_RecurrentToCellWeights = &recurrentToCellWeights;
2692  params.m_RecurrentToOutputWeights = &recurrentToOutputWeights;
2693  params.m_ForgetGateBias = &forgetGateBias;
2694  params.m_CellBias = &cellBias;
2695  params.m_OutputGateBias = &outputGateBias;
2696  params.m_CellToForgetWeights = &cellToForgetWeights;
2697  params.m_CellToOutputWeights = &cellToOutputWeights;
2698 
2699  armnn::INetworkPtr network = armnn::INetwork::Create();
2700  armnn::IConnectableLayer* const inputLayer = network->AddInputLayer(0);
2701  armnn::IConnectableLayer* const cellStateIn = network->AddInputLayer(1);
2702  armnn::IConnectableLayer* const outputStateIn = network->AddInputLayer(2);
2703  const std::string layerName("UnidirectionalSequenceLstm");
2704  armnn::IConnectableLayer* const unidirectionalSequenceLstmLayer =
2705  network->AddUnidirectionalSequenceLstmLayer(descriptor, params, layerName.c_str());
2706  armnn::IConnectableLayer* const outputLayer = network->AddOutputLayer(0);
2707 
2708  // connect up
2709  armnn::TensorInfo inputTensorInfo({ timeSize, batchSize, inputSize }, armnn::DataType::Float32);
2710  armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32);
2711  armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32);
2712  armnn::TensorInfo outputStateOutTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
2713  armnn::TensorInfo cellStateOutTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
2714  armnn::TensorInfo outputTensorInfo({ timeSize, batchSize, outputSize }, armnn::DataType::Float32);
2715 
2716  inputLayer->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(0));
2717  inputLayer->GetOutputSlot(0).SetTensorInfo(inputTensorInfo);
2718 
2719  outputStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(1));
2720  outputStateIn->GetOutputSlot(0).SetTensorInfo(outputStateTensorInfo);
2721 
2722  cellStateIn->GetOutputSlot(0).Connect(unidirectionalSequenceLstmLayer->GetInputSlot(2));
2723  cellStateIn->GetOutputSlot(0).SetTensorInfo(cellStateTensorInfo);
2724 
2725  unidirectionalSequenceLstmLayer->GetOutputSlot(0).SetTensorInfo(outputStateOutTensorInfo);
2726  unidirectionalSequenceLstmLayer->GetOutputSlot(1).SetTensorInfo(cellStateOutTensorInfo);
2727  unidirectionalSequenceLstmLayer->GetOutputSlot(2).Connect(outputLayer->GetInputSlot(0));
2728  unidirectionalSequenceLstmLayer->GetOutputSlot(2).SetTensorInfo(outputTensorInfo);
2729 
2730  armnn::INetworkPtr deserializedNetwork = DeserializeNetwork(SerializeNetwork(*network));
2731  CHECK(deserializedNetwork);
2732 
2733  VerifyLstmLayer<armnn::UnidirectionalSequenceLstmDescriptor> checker(
2734  layerName,
2735  {inputTensorInfo, outputStateTensorInfo, cellStateTensorInfo},
2736  {outputStateOutTensorInfo, cellStateOutTensorInfo, outputTensorInfo},
2737  descriptor,
2738  params);
2739  deserializedNetwork->ExecuteStrategy(checker);
2740 }
2741 
2742 }
armnn::LstmDescriptor::m_ProjectionEnabled
bool m_ProjectionEnabled
Enable/disable the projection layer.
Definition: Descriptors.hpp:1131
armnn::LstmInputParams::m_ProjectionWeights
const ConstTensor * m_ProjectionWeights
Definition: LstmParams.hpp:55
armnn::IConnectableLayer
Interface for a layer that is connectable to other layers via InputSlots and OutputSlots.
Definition: INetwork.hpp:66
armnn::LstmInputParams::m_CellBias
const ConstTensor * m_CellBias
Definition: LstmParams.hpp:53
armnn::QuantizedLstmInputParams::m_RecurrentToOutputWeights
const ConstTensor * m_RecurrentToOutputWeights
Definition: QuantizedLstmParams.hpp:41
armnn::LstmDescriptor::m_ClippingThresProj
float m_ClippingThresProj
Clipping threshold value for the projection.
Definition: Descriptors.hpp:1125
DeserializeNetwork
armnn::INetworkPtr DeserializeNetwork(const std::string &serializerString)
Definition: SerializerTestUtils.cpp:153
armnn::LstmInputParams::m_CellToOutputWeights
const ConstTensor * m_CellToOutputWeights
Definition: LstmParams.hpp:50
armnn::QuantizedLstmInputParams::m_RecurrentToForgetWeights
const ConstTensor * m_RecurrentToForgetWeights
Definition: QuantizedLstmParams.hpp:39
armnn::QLstmDescriptor::m_PeepholeEnabled
bool m_PeepholeEnabled
Enable/disable peephole.
Definition: Descriptors.hpp:1401
armnn::QLstmDescriptor::m_HiddenStateScale
float m_HiddenStateScale
Hidden State quantization scale.
Definition: Descriptors.hpp:1417
armnn::LstmInputParams::m_CellToInputWeights
const ConstTensor * m_CellToInputWeights
Definition: LstmParams.hpp:48
armnn::QLstmDescriptor::m_OutputIntermediateScale
float m_OutputIntermediateScale
Output intermediate quantization scale.
Definition: Descriptors.hpp:1413
armnn::LstmInputParams::m_InputGateBias
const ConstTensor * m_InputGateBias
Definition: LstmParams.hpp:51
LayerVerifierBaseWithDescriptor::VerifyDescriptor
void VerifyDescriptor(const Descriptor &descriptor)
Definition: SerializerTestUtils.hpp:84
armnn::LstmInputParams::m_RecurrentToCellWeights
const ConstTensor * m_RecurrentToCellWeights
Definition: LstmParams.hpp:46
LayerVerifierBaseWithDescriptor::ExecuteStrategy
void ExecuteStrategy(const armnn::IConnectableLayer *layer, const armnn::BaseDescriptor &descriptor, const std::vector< armnn::ConstTensor > &constants, const char *name, const armnn::LayerBindingId id=0) override
Definition: SerializerTestUtils.hpp:61
armnn::LstmInputParams::m_ForgetLayerNormWeights
const ConstTensor * m_ForgetLayerNormWeights
Definition: LstmParams.hpp:58
armnn::LstmInputParams::m_CellToForgetWeights
const ConstTensor * m_CellToForgetWeights
Definition: LstmParams.hpp:49
armnn::LstmDescriptor::m_TimeMajor
bool m_TimeMajor
Enable/disable time major.
Definition: Descriptors.hpp:1135
armnn::IgnoreUnused
void IgnoreUnused(Ts &&...)
Definition: IgnoreUnused.hpp:14
armnn::LayerBindingId
int LayerBindingId
Type of identifiers for bindable layers (inputs, outputs).
Definition: Types.hpp:290
armnn::LstmInputParams::m_OutputGateBias
const ConstTensor * m_OutputGateBias
Definition: LstmParams.hpp:54
armnn::IOutputSlot::SetTensorInfo
virtual void SetTensorInfo(const TensorInfo &tensorInfo)=0
armnn::BaseDescriptor
Base class for all descriptors.
Definition: Descriptors.hpp:22
armnn::LstmInputParams::m_InputLayerNormWeights
const ConstTensor * m_InputLayerNormWeights
Definition: LstmParams.hpp:57
armnn::QLstmDescriptor::m_LayerNormEnabled
bool m_LayerNormEnabled
Enable/disable layer normalization.
Definition: Descriptors.hpp:1405
armnn::LstmInputParams::m_RecurrentToOutputWeights
const ConstTensor * m_RecurrentToOutputWeights
Definition: LstmParams.hpp:47
armnn::LstmDescriptor
An LstmDescriptor for the LstmLayer.
Definition: Descriptors.hpp:1083
armnn::LstmInputParams::m_ProjectionBias
const ConstTensor * m_ProjectionBias
Definition: LstmParams.hpp:56
LayerVerifierBase::VerifyNameAndConnections
void VerifyNameAndConnections(const armnn::IConnectableLayer *layer, const char *name)
Definition: SerializerTestUtils.cpp:40
armnn::QuantizedLstmInputParams::m_InputToForgetWeights
const ConstTensor * m_InputToForgetWeights
Definition: QuantizedLstmParams.hpp:34
armnn::QLstmDescriptor::m_ProjectionClip
float m_ProjectionClip
Clipping threshold value for the projection.
Definition: Descriptors.hpp:1397
armnn::ConstTensor
A tensor defined by a TensorInfo (shape and data type) and an immutable backing store.
Definition: Tensor.hpp:327
armnn::QLstmDescriptor::m_InputIntermediateScale
float m_InputIntermediateScale
Input intermediate quantization scale.
Definition: Descriptors.hpp:1407
armnn::LstmDescriptor::m_PeepholeEnabled
bool m_PeepholeEnabled
Enable/disable peephole.
Definition: Descriptors.hpp:1129
armnn::QLstmDescriptor
A QLstmDescriptor for the QLstmLayer.
Definition: Descriptors.hpp:1361
armnn::LstmDescriptor::m_ActivationFunc
uint32_t m_ActivationFunc
The activation function to use.
Definition: Descriptors.hpp:1121
LayerVerifierBase::VerifyConstTensors
void VerifyConstTensors(const std::string &tensorName, const armnn::ConstTensor *expectedPtr, const armnn::ConstTensor *actualPtr)
Definition: SerializerTestUtils.cpp:78
armnn::IConnectableLayer::GetType
virtual LayerType GetType() const =0
Returns the armnn::LayerType of this layer.
armnn::QuantizedLstmInputParams::m_RecurrentToInputWeights
const ConstTensor * m_RecurrentToInputWeights
Definition: QuantizedLstmParams.hpp:38
armnn::LstmDescriptor::m_ClippingThresCell
float m_ClippingThresCell
Clipping threshold value for the cell state.
Definition: Descriptors.hpp:1123
armnn::QLstmDescriptor::m_ForgetIntermediateScale
float m_ForgetIntermediateScale
Forget intermediate quantization scale.
Definition: Descriptors.hpp:1409
armnn::LstmInputParams::m_CellLayerNormWeights
const ConstTensor * m_CellLayerNormWeights
Definition: LstmParams.hpp:59
armnn::LstmInputParams::m_ForgetGateBias
const ConstTensor * m_ForgetGateBias
Definition: LstmParams.hpp:52
armnn::LstmInputParams::m_InputToCellWeights
const ConstTensor * m_InputToCellWeights
Definition: LstmParams.hpp:42
armnn::LstmInputParams::m_InputToOutputWeights
const ConstTensor * m_InputToOutputWeights
Definition: LstmParams.hpp:43
armnn::QLstmDescriptor::m_CellClip
float m_CellClip
Clipping threshold value for the cell state.
Definition: Descriptors.hpp:1395
armnn::LstmDescriptor::m_CifgEnabled
bool m_CifgEnabled
Enable/disable cifg (coupled input & forget gate).
Definition: Descriptors.hpp:1127
armnn::LstmInputParams::m_RecurrentToForgetWeights
const ConstTensor * m_RecurrentToForgetWeights
Definition: LstmParams.hpp:45
armnn::Exception
Base class for all ArmNN exceptions so that users can filter to just those.
Definition: Exceptions.hpp:46
armnn::QLstmDescriptor::m_ProjectionEnabled
bool m_ProjectionEnabled
Enable/disable the projection layer.
Definition: Descriptors.hpp:1403
armnn::QuantizedLstmInputParams::m_RecurrentToCellWeights
const ConstTensor * m_RecurrentToCellWeights
Definition: QuantizedLstmParams.hpp:40
armnn::IConnectableLayer::GetInputSlot
virtual const IInputSlot & GetInputSlot(unsigned int index) const =0
Get a const input slot handle by slot index.
armnn::LstmInputParams::m_RecurrentToInputWeights
const ConstTensor * m_RecurrentToInputWeights
Definition: LstmParams.hpp:44
armnn::QuantizedLstmInputParams::m_InputToOutputWeights
const ConstTensor * m_InputToOutputWeights
Definition: QuantizedLstmParams.hpp:36
armnn::LstmDescriptor::m_LayerNormEnabled
bool m_LayerNormEnabled
Enable/disable layer normalization.
Definition: Descriptors.hpp:1133
armnn::IConnectableLayer::GetOutputSlot
virtual const IOutputSlot & GetOutputSlot(unsigned int index) const =0
Get the const output slot handle by slot index.
armnn::IConnectableLayer::GetName
virtual const char * GetName() const =0
Returns the name of the layer.
armnn::INetworkPtr
std::unique_ptr< INetwork, void(*)(INetwork *network)> INetworkPtr
Definition: INetwork.hpp:241
armnn::IOutputSlot::Connect
virtual int Connect(IInputSlot &destination)=0
SerializeNetwork
std::string SerializeNetwork(const armnn::INetwork &network)
Definition: SerializerTestUtils.cpp:159
armnn::LstmInputParams::m_OutputLayerNormWeights
const ConstTensor * m_OutputLayerNormWeights
Definition: LstmParams.hpp:60
armnn::QLstmDescriptor::m_CellIntermediateScale
float m_CellIntermediateScale
Cell intermediate quantization scale.
Definition: Descriptors.hpp:1411
armnn::INetwork::Create
static INetworkPtr Create(NetworkOptions networkOptions={})
Definition: Network.cpp:476
armnn::QLstmDescriptor::m_CifgEnabled
bool m_CifgEnabled
Enable/disable CIFG (coupled input & forget gate).
Definition: Descriptors.hpp:1399
armnn::LstmInputParams::m_InputToForgetWeights
const ConstTensor * m_InputToForgetWeights
Definition: LstmParams.hpp:41
armnn::QLstmDescriptor::m_HiddenStateZeroPoint
int32_t m_HiddenStateZeroPoint
Hidden State zero point.
Definition: Descriptors.hpp:1415
armnn::LstmInputParams::m_InputToInputWeights
const ConstTensor * m_InputToInputWeights
Definition: LstmParams.hpp:40