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
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "NeonWorkloadFactoryHelper.hpp"
#include <Graph.hpp>
#include <Network.hpp>
#include <neon/NeonWorkloadFactory.hpp>
#include <doctest/doctest.h>
TEST_SUITE("NeonOptimizedNetwork")
{
TEST_CASE("OptimizeValidateCpuAccDeviceSupportLayerNoFallback")
{
// build up the structure of the network
armnn::INetworkPtr net(armnn::INetwork::Create());
armnn::IConnectableLayer* input = net->AddInputLayer(0);
armnn::IConnectableLayer* output = net->AddOutputLayer(0);
input->GetOutputSlot(0).Connect(output->GetInputSlot(0));
input->GetOutputSlot(0).SetTensorInfo(armnn::TensorInfo({ 1, 1, 4, 4 }, armnn::DataType::Float32));
armnn::IRuntime::CreationOptions options;
armnn::IRuntimePtr runtime(armnn::IRuntime::Create(options));
std::vector<armnn::BackendId> backends = { armnn::Compute::CpuAcc };
armnn::IOptimizedNetworkPtr optNet = armnn::Optimize(*net, backends, runtime->GetDeviceSpec());
CHECK(optNet);
// validate workloads
armnn::NeonWorkloadFactory fact =
NeonWorkloadFactoryHelper::GetFactory(NeonWorkloadFactoryHelper::GetMemoryManager());
armnn::Graph& graph = GetGraphForTesting(optNet.get());
for (auto&& layer : graph)
{
CHECK(layer->GetBackendId() == armnn::Compute::CpuAcc);
CHECK_NOTHROW(
layer->CreateWorkload(fact));
}
}
TEST_CASE("OptimizeValidateDeviceNonSupportLayerNoFallback")
{
// build up the structure of the network
armnn::INetworkPtr net(armnn::INetwork::Create());
armnn::IConnectableLayer* input = net->AddInputLayer(0);
// This layer configuration isn't supported by CpuAcc and isn't allowed to fall back, so Optimize will return null.
armnn::NormalizationDescriptor descriptor;
armnn::IConnectableLayer* normalize = net->AddNormalizationLayer(descriptor);
armnn::IConnectableLayer* output = net->AddOutputLayer(0);
input->GetOutputSlot(0).Connect(normalize->GetInputSlot(0));
normalize->GetOutputSlot(0).Connect(output->GetInputSlot(0));
input->GetOutputSlot(0).SetTensorInfo(armnn::TensorInfo({ 1, 1, 4, 4 }, armnn::DataType::Float32));
normalize->GetOutputSlot(0).SetTensorInfo(armnn::TensorInfo({ 1, 1, 4, 4 }, armnn::DataType::Float32));
armnn::IRuntime::CreationOptions options;
armnn::IRuntimePtr runtime(armnn::IRuntime::Create(options));
std::vector<armnn::BackendId> backends = { armnn::Compute::CpuAcc };
std::vector<std::string> errMessages;
try
{
Optimize(*net, backends, runtime->GetDeviceSpec(), armnn::OptimizerOptions(), errMessages);
FAIL("Should have thrown an exception.");
}
catch (const armnn::InvalidArgumentException& e)
{
// Different exceptions are thrown on different backends
}
CHECK(errMessages.size() > 0);
}
TEST_CASE("FastMathEnabledTestOnCpuAcc")
{
armnn::INetworkPtr net(armnn::INetwork::Create());
armnn::IConnectableLayer* input = net->AddInputLayer(0);
armnn::IConnectableLayer* output = net->AddOutputLayer(0);
input->GetOutputSlot(0).Connect(output->GetInputSlot(0));
input->GetOutputSlot(0).SetTensorInfo(armnn::TensorInfo({ 1, 1, 4, 4 }, armnn::DataType::Float32));
armnn::IRuntime::CreationOptions options;
armnn::IRuntimePtr runtime(armnn::IRuntime::Create(options));
std::vector<armnn::BackendId> backends = {armnn::Compute::CpuAcc};
armnn::OptimizerOptions optimizerOptions;
armnn::BackendOptions modelOptions("CpuAcc", {{"FastMathEnabled", true}});
optimizerOptions.m_ModelOptions.push_back(modelOptions);
armnn::IOptimizedNetworkPtr optimizedNet = armnn::Optimize(
*net, backends, runtime->GetDeviceSpec(), optimizerOptions);
CHECK(optimizedNet);
auto modelOptionsOut = GetModelOptionsForTesting(optimizedNet.get());
CHECK(modelOptionsOut.size() == 1);
CHECK(modelOptionsOut[0].GetOption(0).GetName() == "FastMathEnabled");
CHECK(modelOptionsOut[0].GetOption(0).GetValue().AsBool() == true);
}
TEST_CASE("NumberOfThreadsTestOnCpuAcc")
{
armnn::INetworkPtr net(armnn::INetwork::Create());
armnn::IConnectableLayer* input = net->AddInputLayer(0);
armnn::IConnectableLayer* output = net->AddOutputLayer(0);
input->GetOutputSlot(0).Connect(output->GetInputSlot(0));
input->GetOutputSlot(0).SetTensorInfo(armnn::TensorInfo({ 1, 1, 4, 4 }, armnn::DataType::Float32));
armnn::IRuntime::CreationOptions options;
armnn::IRuntimePtr runtime(armnn::IRuntime::Create(options));
unsigned int numberOfThreads = 2;
std::vector<armnn::BackendId> backends = {armnn::Compute::CpuAcc};
armnn::OptimizerOptions optimizerOptions;
armnn::BackendOptions modelOptions("CpuAcc", {{"NumberOfThreads", numberOfThreads}});
optimizerOptions.m_ModelOptions.push_back(modelOptions);
armnn::IOptimizedNetworkPtr optimizedNet = armnn::Optimize(
*net, backends, runtime->GetDeviceSpec(), optimizerOptions);
CHECK(optimizedNet);
std::unique_ptr<armnn::Graph> graphPtr;
armnn::OptimizedNetworkImpl impl(std::move(graphPtr), optimizerOptions.m_ModelOptions);
auto modelOptionsOut = impl.GetModelOptions();
CHECK(modelOptionsOut.size() == 1);
CHECK(modelOptionsOut[0].GetOption(0).GetName() == "NumberOfThreads");
CHECK(modelOptionsOut[0].GetOption(0).GetValue().AsUnsignedInt() == numberOfThreads);
}
}
|
