1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// See LICENSE file in the project root for full license information.
//
#include "NeonWorkloadFactory.hpp"
#include "armnn/Utils.hpp"
#include "CpuTensorHandle.hpp"
#include "Layer.hpp"
#ifdef ARMCOMPUTENEON_ENABLED
#include "arm_compute/runtime/Allocator.h"
#include "MemCopyWorkload.hpp"
#include "NeonTensorHandle.hpp"
#include "NeonWorkloadUtils.hpp"
#include "NeonWorkloads.hpp"
#endif
#include "MakeWorkloadHelper.hpp"
#include <boost/polymorphic_cast.hpp>
namespace armnn
{
bool NeonWorkloadFactory::IsLayerSupported(const Layer& layer, DataType dataType, std::string& outReasonIfUnsupported)
{
return IWorkloadFactory::IsLayerSupported(Compute::CpuAcc, layer, dataType, outReasonIfUnsupported);
}
#ifdef ARMCOMPUTENEON_ENABLED
NeonWorkloadFactory::NeonWorkloadFactory()
: m_MemoryManager(std::make_unique<arm_compute::Allocator>())
{
}
std::unique_ptr<ITensorHandle> NeonWorkloadFactory::CreateSubTensorHandle(ITensorHandle& parent,
TensorShape const& subTensorShape,
unsigned int const* subTensorOrigin) const
{
BOOST_ASSERT(parent.GetType() == ITensorHandle::Neon);
const arm_compute::TensorShape shape = armcomputetensorutils::BuildArmComputeTensorShape(subTensorShape);
arm_compute::Coordinates coords;
coords.set_num_dimensions(subTensorShape.GetNumDimensions());
for (unsigned int i = 0; i < subTensorShape.GetNumDimensions(); i++)
{
// arm compute indexes tensor coords in reverse order
unsigned int revertedIndex = subTensorShape.GetNumDimensions() - i - 1;
coords.set(i, boost::numeric_cast<int>(subTensorOrigin[revertedIndex]));
}
return std::make_unique<NeonSubTensorHandle>(boost::polymorphic_downcast<INeonTensorHandle*>(&parent)->GetTensor(),
shape, coords);
}
std::unique_ptr<ITensorHandle> NeonWorkloadFactory::CreateTensorHandle(const TensorInfo& tensorInfo) const
{
return std::make_unique<NeonTensorHandle>(tensorInfo);
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateInput(const InputQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<CopyFromCpuToNeonFloat32Workload, CopyFromCpuToNeonUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateOutput(const OutputQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<CopyFromNeonToCpuFloat32Workload, CopyFromNeonToCpuUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateActivation(const ActivationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonActivationFloat32Workload, NeonActivationUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateSoftmax(const SoftmaxQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonSoftmaxFloat32Workload, NeonSoftmaxUint8Workload>(descriptor, info,
m_MemoryManager.Get());
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateSplitter(const SplitterQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonSplitterFloat32Workload, NeonSplitterUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateMerger(const MergerQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonMergerFloat32Workload, NeonMergerUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateFullyConnected(
const FullyConnectedQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<NeonFullyConnectedFloat32Workload, NullWorkload>(descriptor, info, m_MemoryManager.Get());
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreatePermute(const PermuteQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonPermuteFloat32Workload, NeonPermuteUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreatePooling2d(const Pooling2dQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonPooling2dFloat32Workload, NeonPooling2dUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateConvolution2d(
const Convolution2dQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<NeonConvolution2dFloat32Workload, NeonConvolution2dUint8Workload>(descriptor, info,
m_MemoryManager.Get());
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateDepthwiseConvolution2d(
const DepthwiseConvolution2dQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<NeonDepthwiseConvolutionFloat32Workload, NeonDepthwiseConvolutionUint8Workload>(
descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateNormalization(
const NormalizationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<NeonNormalizationFloat32Workload, NullWorkload>(descriptor, info, m_MemoryManager.Get());
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateAddition(const AdditionQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonAdditionFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateMultiplication(
const MultiplicationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<NeonMultiplicationFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateBatchNormalization(
const BatchNormalizationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<NeonBatchNormalizationFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateMemCopy(const MemCopyQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
if (descriptor.m_Inputs.empty() || !descriptor.m_Inputs[0])
{
throw InvalidArgumentException("NeonWorkloadFactory: Invalid null input for MemCopy workload");
}
// Create a workload that will copy tensor data from the inputs, which can have a number of different formats,
// to Neon tensors.
switch (descriptor.m_Inputs[0]->GetType())
{
case ITensorHandle::Cpu:
return MakeWorkload<CopyFromCpuToNeonFloat32Workload, CopyFromCpuToNeonUint8Workload>(descriptor, info);
#if ARMCOMPUTECL_ENABLED
case ITensorHandle::CL:
{
return MakeWorkload<CopyFromClToNeonFloat32Workload, CopyFromClToNeonUint8Workload>(descriptor, info);
}
#endif
default:
throw InvalidArgumentException("NeonWorkloadFactory: Destination type not supported for MemCopy Workload.");
}
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreateResizeBilinear(
const ResizeBilinearQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFakeQuantization(
const FakeQuantizationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateL2Normalization(const L2NormalizationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonL2NormalizationFloat32Workload, NullWorkload>(descriptor, info, m_MemoryManager.Get());
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateConstant(const ConstantQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonConstantFloat32Workload, NeonConstantUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateReshape(const ReshapeQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonReshapeFloat32Workload, NeonReshapeUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFloor(const FloorQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<NeonFloorFloat32Workload, NullWorkload>(descriptor, info);
}
void NeonWorkloadFactory::Finalize()
{
m_MemoryManager.Finalize();
}
#else // Compiled without ArmCompute libs
NeonWorkloadFactory::NeonWorkloadFactory()
{
}
std::unique_ptr<ITensorHandle> NeonWorkloadFactory::CreateSubTensorHandle(ITensorHandle& parent,
TensorShape const& subTensorShape,
unsigned int const* subTensorOrigin) const
{
return nullptr;
}
std::unique_ptr<ITensorHandle> NeonWorkloadFactory::CreateTensorHandle(const TensorInfo& tensorInfo) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateInput(const InputQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateOutput(const OutputQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateActivation(const ActivationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateSoftmax(const SoftmaxQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateSplitter(const SplitterQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateMerger(const MergerQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFullyConnected(const FullyConnectedQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<armnn::IWorkload> NeonWorkloadFactory::CreatePermute(const PermuteQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreatePooling2d(const Pooling2dQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateConvolution2d(const Convolution2dQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateDepthwiseConvolution2d(
const DepthwiseConvolution2dQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateNormalization(const NormalizationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateAddition(const AdditionQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateBatchNormalization(const BatchNormalizationQueueDescriptor& data,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateMultiplication(const MultiplicationQueueDescriptor& data,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateMemCopy(const MemCopyQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateResizeBilinear(const ResizeBilinearQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFakeQuantization(
const FakeQuantizationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateL2Normalization(const L2NormalizationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateConstant(const ConstantQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateReshape(const ReshapeQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
std::unique_ptr<IWorkload> NeonWorkloadFactory::CreateFloor(const FloorQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return nullptr;
}
void NeonWorkloadFactory::Finalize()
{}
#endif
} //namespace armnn
|
