1 //
2 // Copyright © 2021, 2023 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5
6 #pragma once
7
8 #include "TestUtils.hpp"
9
10 #include <armnn_delegate.hpp>
11 #include <DelegateTestInterpreter.hpp>
12
13 #include <flatbuffers/flatbuffers.h>
14 #include <tensorflow/lite/kernels/register.h>
15 #include <tensorflow/lite/version.h>
16
17 #include <schema_generated.h>
18
19 #include <doctest/doctest.h>
20
21 namespace
22 {
CreateShapeTfLiteModel(tflite::TensorType inputTensorType,tflite::TensorType outputTensorType,const std::vector<int32_t> & inputTensorShape,const std::vector<int32_t> & outputTensorShape,float quantScale=1.0f,int quantOffset=0)23 std::vector<char> CreateShapeTfLiteModel(tflite::TensorType inputTensorType,
24 tflite::TensorType outputTensorType,
25 const std::vector<int32_t>& inputTensorShape,
26 const std::vector<int32_t>& outputTensorShape,
27 float quantScale = 1.0f,
28 int quantOffset = 0)
29 {
30 using namespace tflite;
31 flatbuffers::FlatBufferBuilder flatBufferBuilder;
32
33 std::vector<flatbuffers::Offset<tflite::Buffer>> buffers;
34 buffers.push_back(CreateBuffer(flatBufferBuilder));
35 buffers.push_back(CreateBuffer(flatBufferBuilder));
36 buffers.push_back(CreateBuffer(flatBufferBuilder));
37
38 auto quantizationParameters =
39 CreateQuantizationParameters(flatBufferBuilder,
40 0,
41 0,
42 flatBufferBuilder.CreateVector<float>({ quantScale }),
43 flatBufferBuilder.CreateVector<int64_t>({ quantOffset }));
44
45 std::array<flatbuffers::Offset<Tensor>, 2> tensors;
46 tensors[0] = CreateTensor(flatBufferBuilder,
47 flatBufferBuilder.CreateVector<int32_t>(inputTensorShape.data(),
48 inputTensorShape.size()),
49 inputTensorType,
50 1,
51 flatBufferBuilder.CreateString("input"),
52 quantizationParameters);
53 tensors[1] = CreateTensor(flatBufferBuilder,
54 flatBufferBuilder.CreateVector<int32_t>(outputTensorShape.data(),
55 outputTensorShape.size()),
56 outputTensorType,
57 2,
58 flatBufferBuilder.CreateString("output"),
59 quantizationParameters);
60
61 const std::vector<int32_t> operatorInputs({ 0 });
62 const std::vector<int32_t> operatorOutputs({ 1 });
63
64 flatbuffers::Offset<Operator> shapeOperator =
65 CreateOperator(flatBufferBuilder,
66 0,
67 flatBufferBuilder.CreateVector<int32_t>(operatorInputs.data(),
68 operatorInputs.size()),
69 flatBufferBuilder.CreateVector<int32_t>(operatorOutputs.data(),
70 operatorOutputs.size()),
71 BuiltinOptions_ShapeOptions,
72 CreateShapeOptions(flatBufferBuilder, outputTensorType).Union());
73
74 flatbuffers::Offset<flatbuffers::String> modelDescription =
75 flatBufferBuilder.CreateString("ArmnnDelegate: SHAPE Operator Model");
76
77 flatbuffers::Offset<OperatorCode> operatorCode =
78 CreateOperatorCode(flatBufferBuilder, tflite::BuiltinOperator_SHAPE);
79
80 const std::vector<int32_t> subgraphInputs({ 0 });
81 const std::vector<int32_t> subgraphOutputs({ 1 });
82
83 flatbuffers::Offset<SubGraph> subgraph =
84 CreateSubGraph(flatBufferBuilder,
85 flatBufferBuilder.CreateVector(tensors.data(), tensors.size()),
86 flatBufferBuilder.CreateVector<int32_t>(subgraphInputs.data(),
87 subgraphInputs.size()),
88 flatBufferBuilder.CreateVector<int32_t>(subgraphOutputs.data(),
89 subgraphOutputs.size()),
90 flatBufferBuilder.CreateVector(&shapeOperator, 1));
91
92 flatbuffers::Offset<Model> flatbufferModel =
93 CreateModel(flatBufferBuilder,
94 TFLITE_SCHEMA_VERSION,
95 flatBufferBuilder.CreateVector(&operatorCode, 1),
96 flatBufferBuilder.CreateVector(&subgraph, 1),
97 modelDescription,
98 flatBufferBuilder.CreateVector(buffers.data(), buffers.size()));
99
100 flatBufferBuilder.Finish(flatbufferModel, armnnDelegate::FILE_IDENTIFIER);
101
102 return std::vector<char>(flatBufferBuilder.GetBufferPointer(),
103 flatBufferBuilder.GetBufferPointer() + flatBufferBuilder.GetSize());
104 }
105
106 template<typename T, typename K>
ShapeTest(tflite::TensorType inputTensorType,tflite::TensorType outputTensorType,std::vector<armnn::BackendId> & backends,std::vector<int32_t> & inputShape,std::vector<T> & inputValues,std::vector<K> & expectedOutputValues,std::vector<int32_t> & expectedOutputShape,float quantScale=1.0f,int quantOffset=0)107 void ShapeTest(tflite::TensorType inputTensorType,
108 tflite::TensorType outputTensorType,
109 std::vector<armnn::BackendId>& backends,
110 std::vector<int32_t>& inputShape,
111 std::vector<T>& inputValues,
112 std::vector<K>& expectedOutputValues,
113 std::vector<int32_t>& expectedOutputShape,
114 float quantScale = 1.0f,
115 int quantOffset = 0)
116 {
117 using namespace delegateTestInterpreter;
118 std::vector<char> modelBuffer = CreateShapeTfLiteModel(inputTensorType,
119 outputTensorType,
120 inputShape,
121 expectedOutputShape,
122 quantScale,
123 quantOffset);
124
125 // Setup interpreter with just TFLite Runtime.
126 auto tfLiteInterpreter = DelegateTestInterpreter(modelBuffer);
127 CHECK(tfLiteInterpreter.AllocateTensors() == kTfLiteOk);
128 CHECK(tfLiteInterpreter.Invoke() == kTfLiteOk);
129 std::vector<K> tfLiteOutputValues = tfLiteInterpreter.GetOutputResult<K>(0);
130 std::vector<int32_t> tfLiteOutputShape = tfLiteInterpreter.GetOutputShape(0);
131
132 // Setup interpreter with Arm NN Delegate applied.
133 auto armnnInterpreter = DelegateTestInterpreter(modelBuffer, backends);
134 CHECK(armnnInterpreter.AllocateTensors() == kTfLiteOk);
135 CHECK(armnnInterpreter.Invoke() == kTfLiteOk);
136 std::vector<K> armnnOutputValues = armnnInterpreter.GetOutputResult<K>(0);
137 std::vector<int32_t> armnnOutputShape = armnnInterpreter.GetOutputShape(0);
138
139 armnnDelegate::CompareOutputData<K>(tfLiteOutputValues, armnnOutputValues, expectedOutputValues);
140 armnnDelegate::CompareOutputShape(tfLiteOutputShape, armnnOutputShape, expectedOutputShape);
141
142 tfLiteInterpreter.Cleanup();
143 armnnInterpreter.Cleanup();
144 }
145
146 } // anonymous namespace
147