narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 1 | // |
Teresa Charlin | 5266473 | 2020-06-29 16:27:03 +0100 | [diff] [blame] | 2 | // Copyright © 2017 Arm Ltd and Contributors. All rights reserved. |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 3 | // SPDX-License-Identifier: MIT |
| 4 | // |
| 5 | |
| 6 | #pragma once |
| 7 | |
Matteo Martincigh | f02e6cd | 2019-05-17 12:15:30 +0100 | [diff] [blame] | 8 | #include "CommonTestUtils.hpp" |
| 9 | |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 10 | #include <armnn/INetwork.hpp> |
Aron Virginas-Tar | d4f0fea | 2019-04-09 14:08:06 +0100 | [diff] [blame] | 11 | #include <ResolveType.hpp> |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 12 | |
| 13 | namespace{ |
| 14 | |
| 15 | armnn::INetworkPtr CreateGatherNetwork(const armnn::TensorInfo& paramsInfo, |
| 16 | const armnn::TensorInfo& indicesInfo, |
| 17 | const armnn::TensorInfo& outputInfo, |
| 18 | const std::vector<int32_t>& indicesData) |
| 19 | { |
| 20 | armnn::INetworkPtr net(armnn::INetwork::Create()); |
| 21 | |
Teresa Charlin | 5266473 | 2020-06-29 16:27:03 +0100 | [diff] [blame] | 22 | armnn::GatherDescriptor descriptor; |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 23 | armnn::IConnectableLayer* paramsLayer = net->AddInputLayer(0); |
| 24 | armnn::IConnectableLayer* indicesLayer = net->AddConstantLayer(armnn::ConstTensor(indicesInfo, indicesData)); |
Teresa Charlin | 5266473 | 2020-06-29 16:27:03 +0100 | [diff] [blame] | 25 | armnn::IConnectableLayer* gatherLayer = net->AddGatherLayer(descriptor, "gather"); |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 26 | armnn::IConnectableLayer* outputLayer = net->AddOutputLayer(0, "output"); |
| 27 | Connect(paramsLayer, gatherLayer, paramsInfo, 0, 0); |
| 28 | Connect(indicesLayer, gatherLayer, indicesInfo, 0, 1); |
| 29 | Connect(gatherLayer, outputLayer, outputInfo, 0, 0); |
| 30 | |
| 31 | return net; |
| 32 | } |
| 33 | |
| 34 | template<armnn::DataType ArmnnType, typename T = armnn::ResolveType<ArmnnType>> |
| 35 | void GatherEndToEnd(const std::vector<BackendId>& backends) |
| 36 | { |
| 37 | armnn::TensorInfo paramsInfo({ 8 }, ArmnnType); |
| 38 | armnn::TensorInfo indicesInfo({ 3 }, armnn::DataType::Signed32); |
| 39 | armnn::TensorInfo outputInfo({ 3 }, ArmnnType); |
| 40 | |
| 41 | paramsInfo.SetQuantizationScale(1.0f); |
| 42 | paramsInfo.SetQuantizationOffset(0); |
| 43 | outputInfo.SetQuantizationScale(1.0f); |
| 44 | outputInfo.SetQuantizationOffset(0); |
| 45 | |
| 46 | // Creates structures for input & output. |
| 47 | std::vector<T> paramsData{ |
| 48 | 1, 2, 3, 4, 5, 6, 7, 8 |
| 49 | }; |
| 50 | |
| 51 | std::vector<int32_t> indicesData{ |
| 52 | 7, 6, 5 |
| 53 | }; |
| 54 | |
| 55 | std::vector<T> expectedOutput{ |
| 56 | 8, 7, 6 |
| 57 | }; |
| 58 | |
| 59 | // Builds up the structure of the network |
| 60 | armnn::INetworkPtr net = CreateGatherNetwork(paramsInfo, indicesInfo, outputInfo, indicesData); |
| 61 | |
| 62 | BOOST_TEST_CHECKPOINT("create a network"); |
| 63 | |
| 64 | std::map<int, std::vector<T>> inputTensorData = {{ 0, paramsData }}; |
| 65 | std::map<int, std::vector<T>> expectedOutputData = {{ 0, expectedOutput }}; |
| 66 | |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 67 | EndToEndLayerTestImpl<ArmnnType, ArmnnType>(move(net), inputTensorData, expectedOutputData, backends); |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 68 | } |
| 69 | |
| 70 | template<armnn::DataType ArmnnType, typename T = armnn::ResolveType<ArmnnType>> |
| 71 | void GatherMultiDimEndToEnd(const std::vector<BackendId>& backends) |
| 72 | { |
| 73 | armnn::TensorInfo paramsInfo({ 3, 2, 3}, ArmnnType); |
| 74 | armnn::TensorInfo indicesInfo({ 2, 3 }, armnn::DataType::Signed32); |
| 75 | armnn::TensorInfo outputInfo({ 2, 3, 2, 3 }, ArmnnType); |
| 76 | |
| 77 | paramsInfo.SetQuantizationScale(1.0f); |
| 78 | paramsInfo.SetQuantizationOffset(0); |
| 79 | outputInfo.SetQuantizationScale(1.0f); |
| 80 | outputInfo.SetQuantizationOffset(0); |
| 81 | |
| 82 | // Creates structures for input & output. |
| 83 | std::vector<T> paramsData{ |
| 84 | 1, 2, 3, |
| 85 | 4, 5, 6, |
| 86 | |
| 87 | 7, 8, 9, |
| 88 | 10, 11, 12, |
| 89 | |
| 90 | 13, 14, 15, |
| 91 | 16, 17, 18 |
| 92 | }; |
| 93 | |
| 94 | std::vector<int32_t> indicesData{ |
| 95 | 1, 2, 1, |
| 96 | 2, 1, 0 |
| 97 | }; |
| 98 | |
| 99 | std::vector<T> expectedOutput{ |
| 100 | 7, 8, 9, |
| 101 | 10, 11, 12, |
| 102 | 13, 14, 15, |
| 103 | 16, 17, 18, |
| 104 | 7, 8, 9, |
| 105 | 10, 11, 12, |
| 106 | |
| 107 | 13, 14, 15, |
| 108 | 16, 17, 18, |
| 109 | 7, 8, 9, |
| 110 | 10, 11, 12, |
| 111 | 1, 2, 3, |
| 112 | 4, 5, 6 |
| 113 | }; |
| 114 | |
| 115 | // Builds up the structure of the network |
| 116 | armnn::INetworkPtr net = CreateGatherNetwork(paramsInfo, indicesInfo, outputInfo, indicesData); |
| 117 | |
| 118 | BOOST_TEST_CHECKPOINT("create a network"); |
| 119 | |
| 120 | std::map<int, std::vector<T>> inputTensorData = {{ 0, paramsData }}; |
| 121 | std::map<int, std::vector<T>> expectedOutputData = {{ 0, expectedOutput }}; |
| 122 | |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 123 | EndToEndLayerTestImpl<ArmnnType, ArmnnType>(move(net), inputTensorData, expectedOutputData, backends); |
narpra01 | db2b160 | 2019-01-23 15:23:11 +0000 | [diff] [blame] | 124 | } |
| 125 | |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 126 | } // anonymous namespace |