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
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "RefFullyConnectedWorkload.hpp"
#include "FullyConnected.hpp"
#include "RefWorkloadUtils.hpp"
#include "Profiling.hpp"
namespace armnn
{
unsigned int GetNumActivations(const TensorInfo& inputInfo)
{
unsigned int numActivations = 1; // Total number of activations in the input.
for (unsigned int i = 1; i < inputInfo.GetNumDimensions(); i++)
{
numActivations *= inputInfo.GetShape()[i];
}
return numActivations;
}
RefFullyConnectedWorkload::RefFullyConnectedWorkload(
const FullyConnectedQueueDescriptor& descriptor, const WorkloadInfo& info)
: RefBaseWorkload<FullyConnectedQueueDescriptor>(descriptor, info)
, m_InputShape(info.m_InputTensorInfos[0].GetShape())
, m_WeightShape(info.m_InputTensorInfos[1].GetShape())
, m_OutputShape(info.m_OutputTensorInfos[0].GetShape())
, m_NumActivations(GetNumActivations(info.m_InputTensorInfos[0]))
{
}
void RefFullyConnectedWorkload::Execute() const
{
Execute(m_Data.m_Inputs, m_Data.m_Outputs);
}
void RefFullyConnectedWorkload::ExecuteAsync(WorkingMemDescriptor &workingMemDescriptor)
{
Execute(workingMemDescriptor.m_Inputs, workingMemDescriptor.m_Outputs);
}
void RefFullyConnectedWorkload::Execute(std::vector<ITensorHandle*> inputs, std::vector<ITensorHandle*> outputs) const
{
ARMNN_SCOPED_PROFILING_EVENT(Compute::CpuRef, "RefFullyConnectedWorkload_Execute");
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());
std::unique_ptr<Decoder<float>> weightsDecoder = MakeDecoder<float>(GetTensorInfo(inputs[1]), inputs[1]->Map());
std::unique_ptr<Decoder<float>> biasDecoder;
if (m_Data.m_Parameters.m_BiasEnabled)
{
biasDecoder = MakeDecoder<float>(GetTensorInfo(inputs[2]), inputs[2]->Map());
}
FullyConnected(m_InputShape,
*inputDecoder,
m_OutputShape,
*OutputEncoder,
m_WeightShape,
*weightsDecoder,
biasDecoder.get(),
m_Data.m_Parameters.m_BiasEnabled,
m_NumActivations,
m_Data.m_Parameters.m_TransposeWeightMatrix);
}
} //namespace armnn
|
