• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //
2 // Copyright © 2017 Arm Ltd. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 #pragma once
6 
7 #include <ResolveType.hpp>
8 
9 #include <armnn/INetwork.hpp>
10 
11 #include <CommonTestUtils.hpp>
12 
13 #include <doctest/doctest.h>
14 
15 namespace
16 {
17 template<typename armnn::DataType DataType>
CreatePreluNetwork(const armnn::TensorInfo & inputInfo,const armnn::TensorInfo & alphaInfo,const armnn::TensorInfo & outputInfo)18 INetworkPtr CreatePreluNetwork(const armnn::TensorInfo& inputInfo,
19                                const armnn::TensorInfo& alphaInfo,
20                                const armnn::TensorInfo& outputInfo)
21 {
22     using namespace armnn;
23 
24     INetworkPtr net(INetwork::Create());
25 
26     IConnectableLayer* input = net->AddInputLayer(0, "input");
27     IConnectableLayer* alpha = net->AddInputLayer(1, "alpha");
28     IConnectableLayer* prelu = net->AddPreluLayer("Prelu");
29     IConnectableLayer* output = net->AddOutputLayer(0, "output");
30 
31     Connect(input, prelu, inputInfo, 0, 0);
32     Connect(alpha, prelu, alphaInfo, 0, 1);
33     Connect(prelu, output, outputInfo, 0, 0);
34 
35     return net;
36 }
37 
38 template<armnn::DataType ArmnnType, typename T = armnn::ResolveType<ArmnnType>>
PreluEndToEnd(const std::vector<BackendId> & backends,const std::vector<T> & inputData,const std::vector<T> & alphaData,const std::vector<T> & expectedOutputData,const float qScale,const int32_t qOffset)39 void PreluEndToEnd(const std::vector<BackendId>& backends,
40                    const std::vector<T>& inputData,
41                    const std::vector<T>& alphaData,
42                    const std::vector<T>& expectedOutputData,
43                    const float qScale ,
44                    const int32_t qOffset)
45 {
46     using namespace armnn;
47 
48     armnn::TensorInfo inputInfo({ 2, 2, 2, 1 }, ArmnnType);
49     armnn::TensorInfo alphaInfo({ 1, 2, 2, 1 }, ArmnnType);
50     armnn::TensorInfo outputInfo({ 2, 2, 2, 1 }, ArmnnType);
51 
52     inputInfo.SetQuantizationOffset(qOffset);
53     inputInfo.SetQuantizationScale(qScale);
54     inputInfo.SetConstant(true);
55     alphaInfo.SetQuantizationOffset(qOffset);
56     alphaInfo.SetQuantizationScale(qScale);
57     alphaInfo.SetConstant(true);
58     outputInfo.SetQuantizationOffset(qOffset);
59     outputInfo.SetQuantizationScale(qScale);
60 
61     INetworkPtr net = CreatePreluNetwork<ArmnnType>(inputInfo, alphaInfo, outputInfo);
62 
63     CHECK(net);
64 
65     std::map<int, std::vector<T>> inputTensorData          = { { 0, inputData }, { 1, alphaData} };
66     std::map<int, std::vector<T>> expectedOutputTensorData = { { 0, expectedOutputData } };
67 
68     EndToEndLayerTestImpl<ArmnnType, ArmnnType>(move(net),
69                                                 inputTensorData,
70                                                 expectedOutputTensorData,
71                                                 backends);
72 }
73 
74 template<armnn::DataType ArmnnType, typename T = armnn::ResolveType<ArmnnType>>
PreluEndToEndPositiveTest(const std::vector<BackendId> & backends,const float qScale=1.0f,const int32_t qOffset=2)75 void PreluEndToEndPositiveTest(const std::vector<BackendId>& backends, const float qScale = 1.0f,
76                                const int32_t qOffset = 2)
77 {
78     std::vector<T> inputData{ 1, 2, 3, 4, 5, 6, 7, 8 };
79     std::vector<T> alphaData{ 2, 1, 1, 1 };
80 
81     std::vector<T> expectedOutputData{ 2, 2, 3, 4, 5, 6, 7, 8 };
82 
83     PreluEndToEnd<ArmnnType>(backends, inputData, alphaData, expectedOutputData, qScale, qOffset);
84 }
85 
86 template<armnn::DataType ArmnnType, typename T = armnn::ResolveType<ArmnnType>>
PreluEndToEndNegativeTest(const std::vector<BackendId> & backends,const float qScale=1.0f,const int32_t qOffset=0)87 void PreluEndToEndNegativeTest(const std::vector<BackendId>& backends, const float qScale = 1.0f,
88                                const int32_t qOffset = 0)
89 {
90     std::vector<T> inputData{ 1, -2, 3, 4, 5, 6, 7, 8 };
91     std::vector<T> alphaData{ 1, 2, 1, 1 };
92 
93     std::vector<T> expectedOutputData{ 1, -4, 3, 4, 5, 6, 7, 8 };
94 
95     PreluEndToEnd<ArmnnType>(backends, inputData, alphaData, expectedOutputData, qScale, qOffset);
96 }
97 
98 } // anonymous namespace