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
|
//
// Copyright © 2017 Arm Ltd and Contributors. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "FullyConnectedLayer.hpp"
#include "LayerCloneBase.hpp"
#include <armnn/TypesUtils.hpp>
#include <backendsCommon/CpuTensorHandle.hpp>
#include <backendsCommon/WorkloadData.hpp>
#include <backendsCommon/WorkloadFactory.hpp>
namespace armnn
{
FullyConnectedLayer::FullyConnectedLayer(const FullyConnectedDescriptor& param, const char* name)
: LayerWithParameters(1, 1, LayerType::FullyConnected, param, name)
{
}
std::unique_ptr<IWorkload> FullyConnectedLayer::CreateWorkload(const IWorkloadFactory& factory) const
{
// on this level constant data should not be released..
ARMNN_ASSERT_MSG(m_Weight != nullptr, "FullyConnectedLayer: Weights data should not be null.");
FullyConnectedQueueDescriptor descriptor;
descriptor.m_Weight = m_Weight.get();
if (m_Param.m_BiasEnabled)
{
ARMNN_ASSERT_MSG(m_Bias != nullptr, "FullyConnectedLayer: Bias data should not be null.");
descriptor.m_Bias = m_Bias.get();
}
SetAdditionalInfo(descriptor);
return factory.CreateFullyConnected(descriptor, PrepInfoAndDesc(descriptor));
}
FullyConnectedLayer* FullyConnectedLayer::Clone(Graph& graph) const
{
auto layer = CloneBase<FullyConnectedLayer>(graph, m_Param, GetName());
layer->m_Weight = m_Weight ? std::make_unique<ScopedCpuTensorHandle>(*m_Weight) : nullptr;
if (layer->m_Param.m_BiasEnabled)
{
layer->m_Bias = m_Bias ? std::make_unique<ScopedCpuTensorHandle>(*m_Bias) : nullptr;
}
return std::move(layer);
}
std::vector<TensorShape> FullyConnectedLayer::InferOutputShapes(const std::vector<TensorShape>& inputShapes) const
{
ARMNN_ASSERT(inputShapes.size() == 2);
const TensorShape& inputShape = inputShapes[0];
const TensorShape weightShape = inputShapes[1];
// Output for FC is [1, w[1]].
unsigned int batches = inputShape[0];
unsigned int dimIdx = m_Param.m_TransposeWeightMatrix ? 0 : 1;
return std::vector<TensorShape>({ TensorShape({batches, weightShape[dimIdx]})});
}
void FullyConnectedLayer::ValidateTensorShapesFromInputs()
{
const TensorShape& outputShape = GetOutputSlot(0).GetTensorInfo().GetShape();
VerifyShapeInferenceType(outputShape, m_ShapeInferenceMethod);
// check if we m_Weight data is not nullptr
ARMNN_ASSERT_MSG(m_Weight != nullptr, "FullyConnectedLayer: Weights data should not be null.");
auto inferredShapes = InferOutputShapes({GetInputSlot(0).GetConnection()->GetTensorInfo().GetShape(),
m_Weight->GetTensorInfo().GetShape() });
ARMNN_ASSERT(inferredShapes.size() == 1);
ARMNN_ASSERT(inferredShapes[0].GetDimensionality() == Dimensionality::Specified);
ValidateAndCopyShape(outputShape, inferredShapes[0], m_ShapeInferenceMethod, "FullyConnectedLayer");
}
Layer::ConstantTensors FullyConnectedLayer::GetConstantTensorsByRef()
{
return {m_Weight, m_Bias};
}
void FullyConnectedLayer::Accept(ILayerVisitor& visitor) const
{
ConstTensor weightsTensor(m_Weight->GetTensorInfo(), m_Weight->Map(true));
Optional<ConstTensor> optionalBiasTensor = EmptyOptional();
if (GetParameters().m_BiasEnabled)
{
ConstTensor biasTensor(m_Bias->GetTensorInfo(), m_Bias->GetConstTensor<void>());
optionalBiasTensor = Optional<ConstTensor>(biasTensor);
}
visitor.VisitFullyConnectedLayer(this, GetParameters(), weightsTensor, optionalBiasTensor, GetName());
}
void FullyConnectedLayer::ExecuteStrategy(IStrategy& strategy) const
{
std::vector<armnn::ConstTensor> constTensors { {m_Weight->GetTensorInfo(), m_Weight->Map(true)} };
if (GetParameters().m_BiasEnabled)
{
constTensors.emplace_back(ConstTensor(m_Bias->GetTensorInfo(), m_Bias->Map(true)));
}
strategy.ExecuteStrategy(this, GetParameters(), constTensors, GetName());
}
} // namespace armnn
|
