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
99
100
101
102
103
104
105
106
107
108
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#pragma once
#include "CreateWorkload.hpp"
#include <backends/MemCopyWorkload.hpp>
#include <backends/reference/RefWorkloadFactory.hpp>
#if ARMCOMPUTECL_ENABLED
#include <backends/cl/ClTensorHandle.hpp>
#endif
#if ARMCOMPUTENEON_ENABLED
#include <backends/neon/NeonTensorHandle.hpp>
#endif
using namespace armnn;
namespace
{
using namespace std;
template<typename IComputeTensorHandle>
boost::test_tools::predicate_result CompareTensorHandleShape(IComputeTensorHandle* tensorHandle,
std::initializer_list<unsigned int> expectedDimensions)
{
arm_compute::ITensorInfo* info = tensorHandle->GetTensor().info();
auto infoNumDims = info->num_dimensions();
auto numExpectedDims = expectedDimensions.size();
if (infoNumDims != numExpectedDims)
{
boost::test_tools::predicate_result res(false);
res.message() << "Different number of dimensions [" << info->num_dimensions()
<< "!=" << expectedDimensions.size() << "]";
return res;
}
size_t i = info->num_dimensions() - 1;
for (unsigned int expectedDimension : expectedDimensions)
{
if (info->dimension(i) != expectedDimension)
{
boost::test_tools::predicate_result res(false);
res.message() << "Different dimension [" << info->dimension(i) << "!=" << expectedDimension << "]";
return res;
}
i--;
}
return true;
}
template<typename IComputeTensorHandle>
void CreateMemCopyWorkloads(IWorkloadFactory& factory)
{
Graph graph;
RefWorkloadFactory refFactory;
// Creates the layers we're testing.
Layer* const layer1 = graph.AddLayer<MemCopyLayer>("layer1");
Layer* const layer2 = graph.AddLayer<MemCopyLayer>("layer2");
// Creates extra layers.
Layer* const input = graph.AddLayer<InputLayer>(0, "input");
Layer* const output = graph.AddLayer<OutputLayer>(0, "output");
// Connects up.
TensorInfo tensorInfo({2, 3}, DataType::Float32);
Connect(input, layer1, tensorInfo);
Connect(layer1, layer2, tensorInfo);
Connect(layer2, output, tensorInfo);
input->CreateTensorHandles(graph, refFactory);
layer1->CreateTensorHandles(graph, factory);
layer2->CreateTensorHandles(graph, refFactory);
output->CreateTensorHandles(graph, refFactory);
// make the workloads and check them
auto workload1 = MakeAndCheckWorkload<CopyMemGenericWorkload>(*layer1, graph, factory);
auto workload2 = MakeAndCheckWorkload<CopyMemGenericWorkload>(*layer2, graph, refFactory);
MemCopyQueueDescriptor queueDescriptor1 = workload1->GetData();
BOOST_TEST(queueDescriptor1.m_Inputs.size() == 1);
BOOST_TEST(queueDescriptor1.m_Outputs.size() == 1);
auto inputHandle1 = boost::polymorphic_downcast<ConstCpuTensorHandle*>(queueDescriptor1.m_Inputs[0]);
auto outputHandle1 = boost::polymorphic_downcast<IComputeTensorHandle*>(queueDescriptor1.m_Outputs[0]);
BOOST_TEST((inputHandle1->GetTensorInfo() == TensorInfo({2, 3}, DataType::Float32)));
BOOST_TEST(CompareTensorHandleShape<IComputeTensorHandle>(outputHandle1, {2, 3}));
MemCopyQueueDescriptor queueDescriptor2 = workload2->GetData();
BOOST_TEST(queueDescriptor2.m_Inputs.size() == 1);
BOOST_TEST(queueDescriptor2.m_Outputs.size() == 1);
auto inputHandle2 = boost::polymorphic_downcast<IComputeTensorHandle*>(queueDescriptor2.m_Inputs[0]);
auto outputHandle2 = boost::polymorphic_downcast<CpuTensorHandle*>(queueDescriptor2.m_Outputs[0]);
BOOST_TEST(CompareTensorHandleShape<IComputeTensorHandle>(inputHandle2, {2, 3}));
BOOST_TEST((outputHandle2->GetTensorInfo() == TensorInfo({2, 3}, DataType::Float32)));
}
} //namespace
|
