blob: f82e24453a914b571c34dac4a75e5e0c86b5d0c1 (
plain)
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
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "NeonMergerWorkload.hpp"
#include <armnn/ArmNN.hpp>
#include <aclCommon/ArmComputeTensorUtils.hpp>
#include <backendsCommon/CpuTensorHandle.hpp>
#include <neon/NeonTensorHandle.hpp>
namespace armnn
{
using namespace armcomputetensorutils;
arm_compute::Status NeonMergerWorkloadValidate(const std::vector<const TensorInfo*>& inputs,
const TensorInfo& output,
const MergerDescriptor& descriptor)
{
std::vector<arm_compute::TensorInfo> aclInputs;
for (const TensorInfo* input : inputs)
{
arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(*input, armnn::DataLayout::NCHW);
aclInputs.emplace_back(aclInputInfo);
}
const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
arm_compute::DataLayoutDimension aclAxis = arm_compute::DataLayoutDimension::WIDTH;
std::vector<arm_compute::ITensorInfo*> aclInputPtrs;
for (arm_compute::ITensorInfo& input : aclInputs)
{
aclInputPtrs.emplace_back(&input);
}
return arm_compute::NEConcatenateLayer::validate(aclInputPtrs, &aclOutputInfo, aclAxis);
}
NeonMergerWorkload::NeonMergerWorkload(
const MergerQueueDescriptor& descriptor, const WorkloadInfo& info)
: BaseWorkload<MergerQueueDescriptor>(descriptor, info)
{
m_Execute = true;
unsigned int innerAxisOrder = descriptor.m_Parameters.GetNumDimensions() - descriptor.m_Parameters.GetConcatAxis();
if (innerAxisOrder != 1)
{
m_Execute = false;
return;
}
std::vector<arm_compute::ITensor *> aclInputs;
arm_compute::DataLayout aclDataLayout = ConvertDataLayout(armnn::DataLayout::NCHW);
for (auto input : m_Data.m_Inputs)
{
arm_compute::ITensor& aclInput = boost::polymorphic_pointer_downcast<INeonTensorHandle>(input)->GetTensor();
aclInput.info()->set_data_layout(aclDataLayout);
aclInputs.emplace_back(&aclInput);
}
arm_compute::ITensor& output = boost::polymorphic_pointer_downcast<INeonTensorHandle>(
m_Data.m_Outputs[0])->GetTensor();
output.info()->set_data_layout(aclDataLayout);
arm_compute::DataLayoutDimension aclAxis = arm_compute::DataLayoutDimension::WIDTH;
m_Layer.configure(aclInputs, &output, aclAxis);
m_Layer.prepare();
}
void NeonMergerWorkload::Execute() const
{
if (m_Execute)
{
ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonMergerWorkload_Execute");
m_Layer.run();
}
}
} //namespace armnn
|
