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
|
//
// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
// SPDX-License-Identifier: MIT
//
#pragma once
#include "Optimization.hpp"
#include <armnn/backends/TensorHandle.hpp>
#include <armnn/utility/IgnoreUnused.hpp>
#include <armnn/utility/PolymorphicDowncast.hpp>
namespace armnn
{
namespace optimizations
{
static const std::set<armnn::LayerType> broadcastOps{ LayerType::Addition, LayerType::Division,
LayerType::Maximum, LayerType::Minimum,
LayerType::Multiplication, LayerType::Prelu,
LayerType::Subtraction };
class AddBroadcastReshapeLayerImpl
{
public:
/// Run for every ElementwiseBaseLayer. Add Broadcast reshape layer if the inputs shape are different.
void Run(Graph& graph, Layer& layer) const
{
if (std::find(broadcastOps.begin(), broadcastOps.end(), layer.GetType()) != broadcastOps.end())
{
layer.GetInputSlot(0).GetConnectedOutputSlot()->IsTensorInfoSet();
layer.GetInputSlot(1).GetConnectedOutputSlot()->IsTensorInfoSet();
const TensorInfo& inputInfo0 = layer.GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo();
const TensorInfo& inputInfo1 = layer.GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo();
if (inputInfo0.GetNumDimensions() == inputInfo1.GetNumDimensions())
{
return;
}
unsigned int reshapeSlot = 1;
TensorInfo reshapeInfo = inputInfo1;
TensorInfo inputInfo = inputInfo0;
if (inputInfo0.GetNumDimensions() < inputInfo1.GetNumDimensions())
{
reshapeSlot = 0;
reshapeInfo = inputInfo0;
inputInfo = inputInfo1;
}
uint32_t numDimensions = inputInfo.GetNumDimensions();
std::vector<unsigned> reshapedDim;
for (unsigned int i = 0; i < reshapeInfo.GetNumDimensions(); ++i)
{
reshapedDim.push_back(reshapeInfo.GetShape()[i]);
}
std::vector<unsigned int> reshapedDimensions(numDimensions, 1);
std::copy_backward(reshapedDim.begin(), reshapedDim.end(), reshapedDimensions.end());
reshapeInfo.SetShape(armnn::TensorShape{ numDimensions, reshapedDimensions.data() });
// If the parent layer is a Constant layer and it is only used once we can short circuit by just
// changing the tensor info rather than adding a reshape layer.
Layer& parentLayer = layer.GetInputSlot(reshapeSlot).GetConnectedOutputSlot()->GetOwningLayer();
if ((parentLayer.GetType() == armnn::LayerType::Constant) &&
(parentLayer.GetOutputSlot(0).GetNumConnections() == 1))
{
ConstantLayer& constantLayer = static_cast<ConstantLayer&>(parentLayer);
constantLayer.m_LayerOutput = std::make_unique<ScopedTensorHandle>(
ConstTensor(reshapeInfo, constantLayer.m_LayerOutput.get()->GetConstTensor<void>()));
constantLayer.GetOutputSlot().SetTensorInfo(reshapeInfo);
}
else
{
const std::string layerName = "Reshape_for:" + layer.GetNameStr() + "-" + std::to_string(reshapeSlot);
const ReshapeDescriptor descriptor{ reshapeInfo.GetShape() };
ReshapeLayer* reshapeLayer =
graph.InsertNewLayer<ReshapeLayer>(layer.GetInputSlot(reshapeSlot), descriptor, layerName.c_str());
reshapeLayer->GetOutputSlot().SetTensorInfo(reshapeInfo);
}
}
}
protected:
AddBroadcastReshapeLayerImpl() = default;
~AddBroadcastReshapeLayerImpl() = default;
};
using AddBroadcastReshapeLayer = OptimizeForType<Layer, AddBroadcastReshapeLayerImpl>;
} // namespace optimizations
} // namespace armnn
|