josh minor | 4a3c610 | 2020-01-06 16:40:46 -0600 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2019 Arm Ltd. All rights reserved. |
| 3 | // SPDX-License-Identifier: MIT |
| 4 | // |
| 5 | #pragma once |
| 6 | |
| 7 | #include "CommonTestUtils.hpp" |
| 8 | |
| 9 | #include <ResolveType.hpp> |
| 10 | |
| 11 | #include <armnn/INetwork.hpp> |
| 12 | |
Matthew Sloyan | 171214c | 2020-09-09 09:07:37 +0100 | [diff] [blame] | 13 | #include <armnn/utility/NumericCast.hpp> |
| 14 | |
josh minor | 4a3c610 | 2020-01-06 16:40:46 -0600 | [diff] [blame] | 15 | #include <boost/test/unit_test.hpp> |
| 16 | |
| 17 | #include <vector> |
| 18 | |
| 19 | namespace |
| 20 | { |
| 21 | |
| 22 | template<armnn::DataType ArmnnTypeInput> |
| 23 | INetworkPtr CreateElementwiseUnaryNetwork(const TensorShape& inputShape, |
| 24 | const TensorShape& outputShape, |
| 25 | UnaryOperation operation, |
| 26 | const float qScale = 1.0f, |
| 27 | const int32_t qOffset = 0) |
| 28 | { |
| 29 | using namespace armnn; |
| 30 | |
| 31 | INetworkPtr net(INetwork::Create()); |
| 32 | |
| 33 | ElementwiseUnaryDescriptor descriptor(operation); |
| 34 | IConnectableLayer* elementwiseUnaryLayer = net->AddElementwiseUnaryLayer(descriptor, "elementwiseUnary"); |
| 35 | |
| 36 | TensorInfo inputTensorInfo(inputShape, ArmnnTypeInput, qScale, qOffset); |
Matthew Sloyan | 171214c | 2020-09-09 09:07:37 +0100 | [diff] [blame] | 37 | IConnectableLayer* input = net->AddInputLayer(armnn::numeric_cast<LayerBindingId>(0)); |
josh minor | 4a3c610 | 2020-01-06 16:40:46 -0600 | [diff] [blame] | 38 | Connect(input, elementwiseUnaryLayer, inputTensorInfo, 0, 0); |
| 39 | |
| 40 | TensorInfo outputTensorInfo(outputShape, ArmnnTypeInput, qScale, qOffset); |
| 41 | IConnectableLayer* output = net->AddOutputLayer(0, "output"); |
| 42 | Connect(elementwiseUnaryLayer, output, outputTensorInfo, 0, 0); |
| 43 | |
| 44 | return net; |
| 45 | } |
| 46 | |
| 47 | template<armnn::DataType ArmnnInType, |
| 48 | typename TInput = armnn::ResolveType<ArmnnInType>> |
| 49 | void ElementwiseUnarySimpleEndToEnd(const std::vector<BackendId>& backends, |
| 50 | UnaryOperation operation, |
| 51 | const std::vector<float> expectedOutput) |
| 52 | { |
| 53 | using namespace armnn; |
| 54 | |
| 55 | const float qScale = IsQuantizedType<TInput>() ? 0.25f : 1.0f; |
| 56 | const int32_t qOffset = IsQuantizedType<TInput>() ? 50 : 0; |
| 57 | |
| 58 | const TensorShape& inputShape = { 2, 2, 2, 2 }; |
| 59 | const TensorShape& outputShape = { 2, 2, 2, 2 }; |
| 60 | |
| 61 | // Builds up the structure of the network |
| 62 | INetworkPtr net = CreateElementwiseUnaryNetwork<ArmnnInType>(inputShape, outputShape, operation, qScale, qOffset); |
| 63 | |
| 64 | BOOST_TEST_CHECKPOINT("create a network"); |
| 65 | |
| 66 | const std::vector<float> input({ 1, -1, 1, 1, 5, -5, 5, 5, |
| 67 | -3, 3, 3, 3, 4, 4, -4, 4 }); |
| 68 | |
| 69 | // quantize data |
| 70 | std::vector<TInput> qInputData = armnnUtils::QuantizedVector<TInput>(input, qScale, qOffset); |
| 71 | std::vector<TInput> qExpectedOutput = armnnUtils::QuantizedVector<TInput>(expectedOutput, qScale, qOffset); |
| 72 | |
| 73 | std::map<int, std::vector<TInput>> inputTensorData = {{ 0, qInputData }}; |
| 74 | std::map<int, std::vector<TInput>> expectedOutputData = {{ 0, qExpectedOutput }}; |
| 75 | |
| 76 | EndToEndLayerTestImpl<ArmnnInType, ArmnnInType>(move(net), inputTensorData, expectedOutputData, backends); |
| 77 | } |
| 78 | |
| 79 | } // anonymous namespace |