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
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "RefDepthwiseConvolution2dWorkload.hpp"
#include "ConvImpl.hpp"
#include "RefWorkloadUtils.hpp"
#include "Decoders.hpp"
#include "Encoders.hpp"
#include "Profiling.hpp"
#include <ResolveType.hpp>
namespace armnn
{
RefDepthwiseConvolution2dWorkload::RefDepthwiseConvolution2dWorkload(
const DepthwiseConvolution2dQueueDescriptor& descriptor, const WorkloadInfo& info)
: RefBaseWorkload<DepthwiseConvolution2dQueueDescriptor>(descriptor, info)
{
WorkloadInfo detailsInfo;
detailsInfo.m_InputTensorInfos = info.m_InputTensorInfos;
detailsInfo.m_OutputTensorInfos = info.m_OutputTensorInfos;
detailsInfo.m_WeightsTensorInfo = armnn::Optional<armnn::TensorInfo>(info.m_InputTensorInfos[1]);
if (descriptor.m_Parameters.m_BiasEnabled)
{
detailsInfo.m_BiasTensorInfo = armnn::Optional<armnn::TensorInfo>(info.m_InputTensorInfos[2]);
}
// Report Profiling Details
ARMNN_REPORT_PROFILING_WORKLOAD_DESC("RefDepthwiseConvolution2dWorkload_Construct",
descriptor.m_Parameters,
detailsInfo,
this->GetGuid());
}
void RefDepthwiseConvolution2dWorkload::PostAllocationConfigure()
{
PostAllocationConfigure(m_Data.m_Inputs, m_Data.m_Outputs);
}
void RefDepthwiseConvolution2dWorkload::PostAllocationConfigure(std::vector<ITensorHandle*> inputs,
std::vector<ITensorHandle*> outputs)
{
IgnoreUnused(outputs);
const TensorInfo& rFilterInfo = GetTensorInfo(inputs[1]);
m_FilterShape = rFilterInfo.GetShape();
m_FilterDecoder = MakeDecoder<float>(rFilterInfo);
if (m_Data.m_Parameters.m_BiasEnabled)
{
const TensorInfo& biasInfo = GetTensorInfo(inputs[2]);
m_BiasDecoder = MakeDecoder<float>(biasInfo);
}
}
void RefDepthwiseConvolution2dWorkload::Execute() const
{
Execute(m_Data.m_Inputs, m_Data.m_Outputs);
}
void RefDepthwiseConvolution2dWorkload::ExecuteAsync(WorkingMemDescriptor &workingMemDescriptor)
{
PostAllocationConfigure(workingMemDescriptor.m_Inputs, workingMemDescriptor.m_Outputs);
Execute(workingMemDescriptor.m_Inputs, workingMemDescriptor.m_Outputs);
}
void RefDepthwiseConvolution2dWorkload::Execute(std::vector<ITensorHandle*> inputs,
std::vector<ITensorHandle*> outputs) const
{
ARMNN_SCOPED_PROFILING_EVENT(Compute::CpuRef, "RefDepthwiseConvolution2dWorkload_Execute");
std::unique_ptr<Decoder<float>> pBiasDecoder{};
std::unique_ptr<Decoder<float>> inputDecoder = MakeDecoder<float>(GetTensorInfo(inputs[0]), inputs[0]->Map());
std::unique_ptr<Encoder<float>> OutputEncoder = MakeEncoder<float>(GetTensorInfo(outputs[0]), outputs[0]->Map());
const TensorShape& inputShape = GetTensorInfo(inputs[0]).GetShape();
const TensorShape& outputShape = GetTensorInfo(outputs[0]).GetShape();
m_FilterDecoder->Reset(inputs[1]->Map());
if (m_Data.m_Parameters.m_BiasEnabled)
{
m_BiasDecoder->Reset(inputs[2]->Map());
}
Convolve(inputShape, *inputDecoder, outputShape, *OutputEncoder,
m_FilterShape, *m_FilterDecoder, m_Data.m_Parameters.m_BiasEnabled, m_BiasDecoder.get(),
m_Data.m_Parameters.m_DataLayout, m_Data.m_Parameters.m_PadTop, m_Data.m_Parameters.m_PadLeft,
m_Data.m_Parameters.m_StrideX, m_Data.m_Parameters.m_StrideY,
m_Data.m_Parameters.m_DilationX,
m_Data.m_Parameters.m_DilationY, true);
}
} //namespace armnn
|
