1 //
2 // Copyright © 2021, 2023 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5
6 #include "../Serializer.hpp"
7 #include "SerializerTestUtils.hpp"
8
9 #include <armnn/Descriptors.hpp>
10 #include <armnn/INetwork.hpp>
11 #include <armnn/IRuntime.hpp>
12 #include <armnnDeserializer/IDeserializer.hpp>
13 #include <armnn/utility/IgnoreUnused.hpp>
14 #include <armnn/LstmParams.hpp>
15 #include <armnn/QuantizedLstmParams.hpp>
16
17 #include <doctest/doctest.h>
18 #include <fmt/format.h>
19
20
21 TEST_SUITE("SerializerTests")
22 {
23 template<typename Descriptor>
ConstantVector2LstmInputParams(const std::vector<armnn::ConstTensor> & constants,Descriptor & 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:
VerifyLstmLayer(const std::string & layerName,const std::vector<armnn::TensorInfo> & inputInfos,const std::vector<armnn::TensorInfo> & outputInfos,const Descriptor & descriptor,const armnn::LstmInputParams & inputParams)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
ExecuteStrategy(const armnn::IConnectableLayer * layer,const armnn::BaseDescriptor & descriptor,const std::vector<armnn::ConstTensor> & constants,const char * name,const armnn::LayerBindingId id=0)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:
VerifyInputParameters(const armnn::LstmInputParams & params)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, 0.0f , 0);
1322 armnn::TensorInfo cellStateTensorInfo({ batchSize, numUnits}, armnn::DataType::Float32, 0.0f , 0);
1323 armnn::TensorInfo outputStateTensorInfo({ batchSize, outputSize }, armnn::DataType::Float32, 0.0f , 0);
1324 armnn::TensorInfo lstmTensorInfoScratchBuff({ batchSize, numUnits * 4 }, armnn::DataType::Float32, 0.0f , 0);
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
ConstantsVector2QuantizedLstmInputParams(const std::vector<armnn::ConstTensor> & constants)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:
VerifyQuantizedLstmLayer(const std::string & layerName,const std::vector<armnn::TensorInfo> & inputInfos,const std::vector<armnn::TensorInfo> & outputInfos,const armnn::QuantizedLstmInputParams & inputParams)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
ExecuteStrategy(const armnn::IConnectableLayer * layer,const armnn::BaseDescriptor & descriptor,const std::vector<armnn::ConstTensor> & constants,const char * name,const armnn::LayerBindingId id=0)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:
VerifyInputParameters(const armnn::QuantizedLstmInputParams & params)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 }
2743