ArmNN
 24.02
TransposeConvolution2d.cpp
Go to the documentation of this file.
1 //
2 // Copyright © 2017 Arm Ltd. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 
7 
9 
10 namespace armnn
11 {
12 
13 using namespace armnnUtils;
14 
16  const TensorShape& inputShape,
17  Decoder<float>& inputDecoder,
18  const TensorShape& outputShape,
19  Encoder<float>& outputEncoder,
20  const TensorShape& weightsShape,
21  Decoder<float>& weightsDecoder,
22  Decoder<float>* biasesDecoder)
23 {
24  if (descriptor.m_BiasEnabled && !biasesDecoder)
25  {
26  throw InvalidArgumentException("Biases enabled but no bias data provided");
27  }
28  const DataLayoutIndexed dataLayoutIndexed(descriptor.m_DataLayout);
29  const unsigned int channelsIndex = dataLayoutIndexed.GetChannelsIndex();
30  const unsigned int heightIndex = dataLayoutIndexed.GetHeightIndex();
31  const unsigned int widthIndex = dataLayoutIndexed.GetWidthIndex();
32 
33  const unsigned int numBatches = inputShape[0];
34 
35  const unsigned int inputWidth = inputShape[widthIndex];
36  const unsigned int inputHeight = inputShape[heightIndex];
37  const unsigned int inputDepth = inputShape[channelsIndex];
38 
39  const unsigned int weightsHeight = weightsShape[heightIndex];
40  const unsigned int weightsWidth = weightsShape[widthIndex];
41  const unsigned int weightsDepth = weightsShape[channelsIndex];
42 
43  const unsigned int outputHeight = outputShape[heightIndex];
44  const unsigned int outputWidth = outputShape[widthIndex];
45  const unsigned int outputDepth = outputShape[channelsIndex];
46 
47  const unsigned int paddingLeft = descriptor.m_PadLeft;
48  const unsigned int paddingTop = descriptor.m_PadTop;
49 
50  const unsigned int strideX = descriptor.m_StrideX;
51  const unsigned int strideY = descriptor.m_StrideY;
52 
53  std::vector<float> outputBuffer(outputShape.GetNumElements(), 0);
54 
55  const std::vector<float> inputVec = inputDecoder.DecodeTensor(inputShape);
56  const std::vector<float> filterVec = weightsDecoder.DecodeTensor(weightsShape);
57 
58  for (unsigned int batch = 0u; batch < numBatches; ++batch)
59  {
60  for (unsigned int yInput = 0u; yInput < inputHeight; ++yInput)
61  {
62  for (unsigned int xInput = 0u; xInput < inputWidth; ++xInput)
63  {
64  unsigned int xOutputOrigin = xInput * strideX - paddingLeft;
65  unsigned int yOutputOrigin = yInput * strideY - paddingTop;
66 
67  for (unsigned int dOutput = 0u; dOutput < outputDepth; ++dOutput)
68  {
69  for (unsigned int yWeights = 0u; yWeights < weightsHeight; ++yWeights)
70  {
71  for (unsigned int xWeights = 0u; xWeights < weightsWidth; ++xWeights)
72  {
73  unsigned int yOutput = yOutputOrigin + yWeights;
74  unsigned int xOutput = xOutputOrigin + xWeights;
75 
76  if (yOutput < outputHeight && xOutput< outputWidth)
77  {
78  for (unsigned int dInput = 0u; dInput < inputDepth; dInput++)
79  {
80  unsigned int inputIndex;
81  unsigned int outputIndex;
82  unsigned int weightsIndex;
83 
84  if(descriptor.m_DataLayout == armnn::DataLayout::NHWC)
85  {
86  inputIndex = batch * inputHeight * inputWidth * inputDepth +
87  yInput * inputWidth * inputDepth +
88  xInput * inputDepth +
89  dInput;
90 
91  weightsIndex = dOutput * weightsHeight * weightsWidth * weightsDepth +
92  yWeights * weightsWidth * weightsDepth +
93  xWeights * weightsDepth +
94  dInput;
95 
96  outputIndex = batch * outputHeight * outputWidth * outputDepth +
97  yOutput * outputWidth * outputDepth +
98  xOutput * outputDepth +
99  dOutput;
100  }
101  else
102  {
103  inputIndex = batch * inputDepth * inputHeight * inputWidth +
104  dInput * inputHeight * inputWidth +
105  yInput * inputWidth +
106  xInput;
107 
108  weightsIndex = dOutput * weightsDepth * weightsHeight * weightsWidth +
109  dInput * weightsHeight * weightsWidth +
110  yWeights * weightsWidth +
111  xWeights;
112 
113  outputIndex = batch * outputDepth * outputHeight * outputWidth +
114  dOutput * outputHeight * outputWidth +
115  yOutput * outputWidth +
116  xOutput;
117  }
118 
119  outputBuffer[outputIndex] += inputVec[inputIndex] * filterVec[weightsIndex];
120  }
121  }
122  }
123  }
124 
125  }
126  }
127  }
128  }
129 
130  // Apply bias (if enabled)
131  if (descriptor.m_BiasEnabled)
132  {
133  outputEncoder[0];
134  Decoder<float>& rBiasesDecoder = *biasesDecoder;
135 
136  for (unsigned int batch = 0u; batch < numBatches; ++batch)
137  {
138  for (unsigned int dOutput = 0u; dOutput < outputDepth; ++dOutput)
139  {
140  rBiasesDecoder[dOutput];
141  for (unsigned int yOutput = 0u; yOutput < outputHeight; ++yOutput)
142  {
143  for (unsigned int xOutput = 0u; xOutput < outputWidth; ++xOutput)
144  {
145  const unsigned int outputIndex =
146  dataLayoutIndexed.GetIndex(outputShape, batch, dOutput, yOutput, xOutput);
147  outputBuffer[outputIndex] += rBiasesDecoder.Get();
148  }
149  }
150  }
151  }
152  }
153  outputEncoder[0];
154  for (float output : outputBuffer)
155  {
156  outputEncoder.Set(output);
157  ++outputEncoder;
158  }
159 }
160 
161 } // namespace armnn
armnn::Decoder< float >
armnn::TransposeConvolution2dDescriptor::m_PadLeft
uint32_t m_PadLeft
Padding left value in the width dimension.
Definition: Descriptors.hpp:1469
armnn::TransposeConvolution2dImpl
void TransposeConvolution2dImpl(const TransposeConvolution2dDescriptor &descriptor, const TensorShape &inputShape, Decoder< float > &inputDecoder, const TensorShape &outputShape, Encoder< float > &outputEncoder, const TensorShape &weightsShape, Decoder< float > &weightsDecoder, Decoder< float > *biasesDecoder)
Definition: TransposeConvolution2d.cpp:15
armnn::TransposeConvolution2dDescriptor::m_StrideX
uint32_t m_StrideX
Stride value when proceeding through input for the width dimension.
Definition: Descriptors.hpp:1477
armnn::Encoder::Set
virtual void Set(IType right)=0
armnn::DataLayout::NHWC
@ NHWC
armnnUtils::DataLayoutIndexed
Provides access to the appropriate indexes for Channels, Height and Width based on DataLayout.
Definition: DataLayoutIndexed.hpp:17
armnnUtils::DataLayoutIndexed::GetHeightIndex
unsigned int GetHeightIndex() const
Definition: DataLayoutIndexed.hpp:24
armnn::TensorShape
Definition: Tensor.hpp:20
armnn::Encoder< float >
armnnUtils
Definition: CompatibleTypes.hpp:10
armnn::InvalidArgumentException
Definition: Exceptions.hpp:80
armnn::Decoder::DecodeTensor
virtual std::vector< float > DecodeTensor(const TensorShape &tensorShape, bool isDepthwise=false)=0
armnnUtils::DataLayoutIndexed::GetWidthIndex
unsigned int GetWidthIndex() const
Definition: DataLayoutIndexed.hpp:25
armnn::TransposeConvolution2dDescriptor::m_StrideY
uint32_t m_StrideY
Stride value when proceeding through input for the height dimension.
Definition: Descriptors.hpp:1479
armnn::Decoder::Get
virtual IType Get() const =0
armnn::TransposeConvolution2dDescriptor::m_PadTop
uint32_t m_PadTop
Padding top value in the height dimension.
Definition: Descriptors.hpp:1473
armnn
Copyright (c) 2021 ARM Limited and Contributors.
Definition: 01_00_quick_start.dox:6
armnn::TransposeConvolution2dDescriptor
A TransposeConvolution2dDescriptor for the TransposeConvolution2dLayer.
Definition: Descriptors.hpp:1440
armnnUtils::DataLayoutIndexed::GetChannelsIndex
unsigned int GetChannelsIndex() const
Definition: DataLayoutIndexed.hpp:23
armnn::TransposeConvolution2dDescriptor::m_BiasEnabled
bool m_BiasEnabled
Enable/disable bias.
Definition: Descriptors.hpp:1481
armnn::TransposeConvolution2dDescriptor::m_DataLayout
DataLayout m_DataLayout
The data layout to be used (NCHW, NHWC).
Definition: Descriptors.hpp:1483
armnnUtils::DataLayoutIndexed::GetIndex
unsigned int GetIndex(const armnn::TensorShape &shape, unsigned int batchIndex, unsigned int channelIndex, unsigned int heightIndex, unsigned int widthIndex) const
Definition: DataLayoutIndexed.hpp:28
DataLayoutIndexed.hpp
armnn::TensorShape::GetNumElements
unsigned int GetNumElements() const
Function that calculates the tensor elements by multiplying all dimension size which are Specified.
Definition: Tensor.cpp:181
TransposeConvolution2d.hpp