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
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include <backends/CpuTensorHandle.hpp>
#include <backends/MemCopyWorkload.hpp>
#include <backends/MakeWorkloadHelper.hpp>
#include "RefWorkloadFactory.hpp"
#include "workloads/RefWorkloads.hpp"
#include "Layer.hpp"
#include <boost/log/trivial.hpp>
namespace armnn
{
template <typename F32Workload, typename U8Workload, typename QueueDescriptorType>
std::unique_ptr<IWorkload> RefWorkloadFactory::MakeWorkload(const QueueDescriptorType& descriptor,
const WorkloadInfo& info) const
{
return armnn::MakeWorkload<NullWorkload, F32Workload, U8Workload>(descriptor, info);
}
RefWorkloadFactory::RefWorkloadFactory()
{
}
bool RefWorkloadFactory::IsLayerSupported(const Layer& layer, boost::optional<DataType> dataType,
std::string& outReasonIfUnsupported)
{
return IWorkloadFactory::IsLayerSupported(Compute::CpuRef, layer, dataType, outReasonIfUnsupported);
}
std::unique_ptr<ITensorHandle> RefWorkloadFactory::CreateTensorHandle(const TensorInfo& tensorInfo) const
{
return std::make_unique<ScopedCpuTensorHandle>(tensorInfo);
}
std::unique_ptr<ITensorHandle> RefWorkloadFactory::CreateTensorHandle(const TensorInfo& tensorInfo,
DataLayout dataLayout) const
{
return std::make_unique<ScopedCpuTensorHandle>(tensorInfo);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateInput(const InputQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
if (info.m_InputTensorInfos.empty() )
{
throw InvalidArgumentException("RefWorkloadFactory::CreateInput: Input cannot be zero length");
}
if (info.m_OutputTensorInfos.empty())
{
throw InvalidArgumentException("RefWorkloadFactory::CreateInput: Output cannot be zero length");
}
if (info.m_InputTensorInfos[0].GetNumBytes() != info.m_OutputTensorInfos[0].GetNumBytes())
{
throw InvalidArgumentException("RefWorkloadFactory::CreateInput: data input and output differ in byte count.");
}
return MakeWorkload<CopyMemGenericWorkload, CopyMemGenericWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateOutput(const OutputQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
if (info.m_InputTensorInfos.empty() )
{
throw InvalidArgumentException("RefWorkloadFactory::CreateOutput: Input cannot be zero length");
}
if (info.m_OutputTensorInfos.empty())
{
throw InvalidArgumentException("RefWorkloadFactory::CreateOutput: Output cannot be zero length");
}
if (info.m_InputTensorInfos[0].GetNumBytes() != info.m_OutputTensorInfos[0].GetNumBytes())
{
throw InvalidArgumentException("RefWorkloadFactory::CreateOutput: data input and output differ in byte count.");
}
return MakeWorkload<CopyMemGenericWorkload, CopyMemGenericWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateActivation(const ActivationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefActivationFloat32Workload, RefActivationUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateSoftmax(const SoftmaxQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefSoftmaxFloat32Workload, RefSoftmaxUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateSplitter(const SplitterQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefSplitterFloat32Workload, RefSplitterUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateMerger(const MergerQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefMergerFloat32Workload, RefMergerUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateFullyConnected(
const FullyConnectedQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefFullyConnectedFloat32Workload, RefFullyConnectedUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreatePermute(const PermuteQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return armnn::MakeWorkload<RefPermuteFloat16Workload, RefPermuteFloat32Workload, RefPermuteUint8Workload>
(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreatePooling2d(const Pooling2dQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefPooling2dFloat32Workload, RefPooling2dUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateConvolution2d(
const Convolution2dQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefConvolution2dFloat32Workload, RefConvolution2dUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateDepthwiseConvolution2d(
const DepthwiseConvolution2dQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefDepthwiseConvolution2dFloat32Workload,
RefDepthwiseConvolution2dUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateNormalization(
const NormalizationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefNormalizationFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateAddition(const AdditionQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefAdditionFloat32Workload, RefAdditionUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateMultiplication(
const MultiplicationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefMultiplicationFloat32Workload, RefMultiplicationUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateBatchNormalization(
const BatchNormalizationQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefBatchNormalizationFloat32Workload, RefBatchNormalizationUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateMemCopy(const MemCopyQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
if (descriptor.m_Inputs.empty())
{
throw InvalidArgumentException("RefWorkloadFactory: CreateMemCopy() expected an input tensor.");
}
return std::make_unique<CopyMemGenericWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateResizeBilinear(const ResizeBilinearQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefResizeBilinearFloat32Workload, RefResizeBilinearUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateFakeQuantization(
const FakeQuantizationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefFakeQuantizationFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateL2Normalization(const L2NormalizationQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefL2NormalizationFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateConstant(const ConstantQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefConstantFloat32Workload, RefConstantUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateReshape(const ReshapeQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefReshapeFloat32Workload, RefReshapeUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateFloor(const FloorQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefFloorFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateLstm(const LstmQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefLstmFloat32Workload, NullWorkload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateConvertFp16ToFp32(
const ConvertFp16ToFp32QueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return std::make_unique<RefConvertFp16ToFp32Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreateConvertFp32ToFp16(
const ConvertFp32ToFp16QueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return std::make_unique<RefConvertFp32ToFp16Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateDivision(
const DivisionQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefDivisionFloat32Workload, RefDivisionUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateSubtraction(
const SubtractionQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefSubtractionFloat32Workload, RefSubtractionUint8Workload>(descriptor, info);
}
std::unique_ptr<armnn::IWorkload> RefWorkloadFactory::CreateMean(
const MeanQueueDescriptor& descriptor, const WorkloadInfo& info) const
{
return MakeWorkload<RefMeanFloat32Workload, RefMeanUint8Workload>(descriptor, info);
}
std::unique_ptr<IWorkload> RefWorkloadFactory::CreatePad(const PadQueueDescriptor& descriptor,
const WorkloadInfo& info) const
{
return MakeWorkload<RefPadFloat32Workload, RefPadUint8Workload>(descriptor, info);
}
} // namespace armnn
|
