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
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "ClBatchNormalizationFloatWorkload.hpp"
#include <backends/cl/ClTensorHandle.hpp>
#include <backends/CpuTensorHandle.hpp>
#include <backends/aclCommon/ArmComputeTensorUtils.hpp>
#include <backends/cl/ClLayerSupport.hpp>
#include "ClWorkloadUtils.hpp"
namespace armnn
{
using namespace armcomputetensorutils;
arm_compute::Status ClBatchNormalizationValidate(const TensorInfo& input,
const TensorInfo& output,
const TensorInfo& mean,
const TensorInfo& var,
const TensorInfo& beta,
const TensorInfo& gamma,
const BatchNormalizationDescriptor &desc)
{
const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input);
const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
const arm_compute::TensorInfo aclMeanInfo = BuildArmComputeTensorInfo(mean);
const arm_compute::TensorInfo aclVarInfo = BuildArmComputeTensorInfo(var);
const arm_compute::TensorInfo aclBetaInfo = BuildArmComputeTensorInfo(beta);
const arm_compute::TensorInfo aclGammaInfo = BuildArmComputeTensorInfo(gamma);
return arm_compute::CLBatchNormalizationLayer::validate(&aclInputInfo,
&aclOutputInfo,
&aclMeanInfo,
&aclVarInfo,
&aclBetaInfo,
&aclGammaInfo,
desc.m_Eps);
}
ClBatchNormalizationFloatWorkload::ClBatchNormalizationFloatWorkload(
const BatchNormalizationQueueDescriptor& descriptor, const WorkloadInfo& info)
: FloatWorkload<BatchNormalizationQueueDescriptor>(descriptor, info)
{
m_Mean = std::make_unique<arm_compute::CLTensor>();
BuildArmComputeTensor(*m_Mean, m_Data.m_Mean->GetTensorInfo());
m_Variance = std::make_unique<arm_compute::CLTensor>();
BuildArmComputeTensor(*m_Variance, m_Data.m_Variance->GetTensorInfo());
m_Gamma = std::make_unique<arm_compute::CLTensor>();
BuildArmComputeTensor(*m_Gamma, m_Data.m_Gamma->GetTensorInfo());
m_Beta = std::make_unique<arm_compute::CLTensor>();
BuildArmComputeTensor(*m_Beta, m_Data.m_Beta->GetTensorInfo());
m_Data.ValidateInputsOutputs("ClBatchNormalizationFloatWorkload", 1, 1);
arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
m_Layer.configure(&input,
&output,
m_Mean.get(),
m_Variance.get(),
m_Beta.get(),
m_Gamma.get(),
m_Data.m_Parameters.m_Eps);
InitializeArmComputeClTensorData(*m_Mean, m_Data.m_Mean);
InitializeArmComputeClTensorData(*m_Variance, m_Data.m_Variance);
InitializeArmComputeClTensorData(*m_Beta, m_Data.m_Beta);
InitializeArmComputeClTensorData(*m_Gamma, m_Data.m_Gamma);
// Force Compute Library to perform the necessary copying and reshaping, after which
// delete all the input tensors that will no longer be needed
m_Layer.prepare();
FreeUnusedTensors();
}
void ClBatchNormalizationFloatWorkload::Execute() const
{
ARMNN_SCOPED_PROFILING_EVENT_CL("ClBatchNormalizationFloatWorkload_Execute");
m_Layer.run();
}
void ClBatchNormalizationFloatWorkload::FreeUnusedTensors()
{
FreeTensorIfUnused(m_Mean);
FreeTensorIfUnused(m_Variance);
FreeTensorIfUnused(m_Gamma);
FreeTensorIfUnused(m_Beta);
}
} //namespace armnn
|
