1# Using MindSpore Lite for Model Inference (C/C++)
2
3## When to Use
4
5MindSpore Lite is an AI engine that provides AI model inference for different hardware devices. It has been used in a wide range of fields, such as image classification, target recognition, facial recognition, and character recognition.
6
7This document describes the general development process for MindSpore Lite model inference.
8
9## Basic Concepts
10
11Before getting started, you need to understand the following basic concepts:
12
13**Tensor**: a special data structure that is similar to arrays and matrices. It is basic data structure used in MindSpore Lite network operations.
14
15**Float16 inference mode**: an inference mode in half-precision format, where a number is represented with 16 bits.
16
17
18
19## Available APIs
20
21APIs involved in MindSpore Lite model inference are categorized into context APIs, model APIs, and tensor APIs.
22
23### Context APIs
24
25| API       | Description       |
26| ------------------ | ----------------- |
27|OH_AI_ContextHandle OH_AI_ContextCreate()|Creates a context object.|
28|void OH_AI_ContextSetThreadNum(OH_AI_ContextHandle context, int32_t thread_num)|Sets the number of runtime threads.|
29| void OH_AI_ContextSetThreadAffinityMode(OH_AI_ContextHandle context, int mode)|Sets the affinity mode for binding runtime threads to CPU cores, which are classified into large, medium, and small cores based on the CPU frequency. You only need to bind the large or medium cores, but not small cores.
30|OH_AI_DeviceInfoHandle OH_AI_DeviceInfoCreate(OH_AI_DeviceType device_type)|Creates a runtime device information object.|
31|void OH_AI_ContextDestroy(OH_AI_ContextHandle *context)|Destroys a context object.|
32|void OH_AI_DeviceInfoSetEnableFP16(OH_AI_DeviceInfoHandle device_info, bool is_fp16)|Sets whether to enable float16 inference. This function is available only for CPU and GPU devices.|
33|void OH_AI_ContextAddDeviceInfo(OH_AI_ContextHandle context, OH_AI_DeviceInfoHandle device_info)|Adds a runtime device information object.|
34
35### Model APIs
36
37| API       | Description       |
38| ------------------ | ----------------- |
39|OH_AI_ModelHandle OH_AI_ModelCreate()|Creates a model object.|
40|OH_AI_Status OH_AI_ModelBuildFromFile(OH_AI_ModelHandle model, const char *model_path,OH_AI_ModelType odel_type, const OH_AI_ContextHandle model_context)|Loads and builds a MindSpore model from a model file.|
41|void OH_AI_ModelDestroy(OH_AI_ModelHandle *model)|Destroys a model object.|
42
43### Tensor APIs
44
45| API       | Description       |
46| ------------------ | ----------------- |
47|OH_AI_TensorHandleArray OH_AI_ModelGetInputs(const OH_AI_ModelHandle model)|Obtains the input tensor array structure of a model.|
48|int64_t OH_AI_TensorGetElementNum(const OH_AI_TensorHandle tensor)|Obtains the number of tensor elements.|
49|const char *OH_AI_TensorGetName(const OH_AI_TensorHandle tensor)|Obtains the name of a tensor.|
50|OH_AI_DataType OH_AI_TensorGetDataType(const OH_AI_TensorHandle tensor)|Obtains the tensor data type.|
51|void *OH_AI_TensorGetMutableData(const OH_AI_TensorHandle tensor)|Obtains the pointer to mutable tensor data.|
52
53## How to Develop
54
55The following figure shows the development process for MindSpore Lite model inference.
56
57**Figure 1** Development process for MindSpore Lite model inference
58
59![how-to-use-mindspore-lite](figures/01.png)
60
61Before moving to the development process, you need to reference related header files and compile functions to generate random input. The sample code is as follows:
62
63```c
64#include <stdlib.h>
65#include <stdio.h>
66#include "mindspore/model.h"
67
68// Generate random input.
69int GenerateInputDataWithRandom(OH_AI_TensorHandleArray inputs) {
70  for (size_t i = 0; i < inputs.handle_num; ++i) {
71    float *input_data = (float *)OH_AI_TensorGetMutableData(inputs.handle_list[i]);
72    if (input_data == NULL) {
73      printf("MSTensorGetMutableData failed.\n");
74      return OH_AI_STATUS_LITE_ERROR;
75    }
76    int64_t num = OH_AI_TensorGetElementNum(inputs.handle_list[i]);
77    const int divisor = 10;
78    for (size_t j = 0; j < num; j++) {
79      input_data[j] = (float)(rand() % divisor) / divisor;  // 0--0.9f
80    }
81  }
82  return OH_AI_STATUS_SUCCESS;
83}
84```
85
86The development process consists of the following main steps:
87
881. Prepare the required model.
89
90    The required model can be downloaded directly or obtained using the model conversion tool.
91  
92     - If the downloaded model is in the `.ms` format, you can use it directly for inference. The following uses the **mobilenetv2.ms** model as an example.
93     - If the downloaded model uses a third-party framework, such as TensorFlow, TensorFlow Lite, Caffe, or ONNX, you can use the [model conversion tool](https://www.mindspore.cn/lite/docs/en/master/use/downloads.html#1-8-1) to convert it to the `.ms` format.
94
952. Create a context, and set parameters such as the number of runtime threads and device type.
96
97    The following describes two typical scenarios:
98
99    Scenario 1: Only the CPU inference context is created.
100  
101    ```c
102    // Create a context, and set the number of runtime threads to 2 and the thread affinity mode to 1 (big cores first).
103    OH_AI_ContextHandle context = OH_AI_ContextCreate();
104    if (context == NULL) {
105      printf("OH_AI_ContextCreate failed.\n");
106      return OH_AI_STATUS_LITE_ERROR;
107    }
108    const int thread_num = 2;
109    OH_AI_ContextSetThreadNum(context, thread_num);
110    OH_AI_ContextSetThreadAffinityMode(context, 1);
111    // Set the device type to CPU, and disable Float16 inference.
112    OH_AI_DeviceInfoHandle cpu_device_info = OH_AI_DeviceInfoCreate(OH_AI_DEVICETYPE_CPU);
113    if (cpu_device_info == NULL) {
114      printf("OH_AI_DeviceInfoCreate failed.\n");
115      OH_AI_ContextDestroy(&context);
116      return OH_AI_STATUS_LITE_ERROR;
117    }
118    OH_AI_DeviceInfoSetEnableFP16(cpu_device_info, false);
119    OH_AI_ContextAddDeviceInfo(context, cpu_device_info);
120    ```
121
122    Scenario 2: The neural network runtime (NNRT) and CPU heterogeneous inference contexts are created.
123
124    NNRT is the runtime for cross-chip inference computing in the AI field. Generally, the acceleration hardware connected to NNRT, such as the NPU, has strong inference capabilities but supports only a limited number of operators, whereas the general-purpose CPU has weak inference capabilities but supports a wide range of operators. MindSpore Lite supports NNRT and CPU heterogeneous inference. Model operators are preferentially scheduled to NNRT for inference. If certain operators are not supported by NNRT, then they are scheduled to the CPU for inference. The following is the sample code for configuring NNRT/CPU heterogeneous inference:
125   <!--Del-->
126   > **NOTE**
127   >
128   > NNRT/CPU heterogeneous inference requires access of NNRT hardware. For details, see [OpenHarmony/ai_neural_network_runtime](https://gitee.com/openharmony/ai_neural_network_runtime).
129   <!--DelEnd-->
130    ```c
131    // Create a context, and set the number of runtime threads to 2 and the thread affinity mode to 1 (big cores first).
132    OH_AI_ContextHandle context = OH_AI_ContextCreate();
133    if (context == NULL) {
134      printf("OH_AI_ContextCreate failed.\n");
135      return OH_AI_STATUS_LITE_ERROR;
136    }
137    // Preferentially use NNRT inference.
138    // Use the NNRT hardware of the first ACCELERATORS class to create the NNRT device information and configure the high-performance inference mode for the NNRT hardware. You can also use OH_AI_GetAllNNRTDeviceDescs() to obtain the list of NNRT devices in the current environment, search for a specific device by device name or type, and use the device as the NNRT inference hardware.
139    OH_AI_DeviceInfoHandle nnrt_device_info = OH_AI_CreateNNRTDeviceInfoByType(OH_AI_NNRTDEVICE_ACCELERATOR);
140    if (nnrt_device_info == NULL) {
141      printf("OH_AI_DeviceInfoCreate failed.\n");
142      OH_AI_ContextDestroy(&context);
143      return OH_AI_STATUS_LITE_ERROR;
144    }
145    OH_AI_DeviceInfoSetPerformanceMode(nnrt_device_info, OH_AI_PERFORMANCE_HIGH);
146    OH_AI_ContextAddDeviceInfo(context, nnrt_device_info);
147
148    // Configure CPU inference.
149    OH_AI_DeviceInfoHandle cpu_device_info = OH_AI_DeviceInfoCreate(OH_AI_DEVICETYPE_CPU);
150    if (cpu_device_info == NULL) {
151      printf("OH_AI_DeviceInfoCreate failed.\n");
152      OH_AI_ContextDestroy(&context);
153      return OH_AI_STATUS_LITE_ERROR;
154    }
155    OH_AI_ContextAddDeviceInfo(context, cpu_device_info);
156    ```
157
158    
159
1603. Create, load, and build the model.
161
162    Call **OH_AI_ModelBuildFromFile** to load and build the model.
163
164    In this example, the **argv[1]** parameter passed to **OH_AI_ModelBuildFromFile** indicates the specified model file path.
165
166    ```c
167    // Create a model.
168    OH_AI_ModelHandle model = OH_AI_ModelCreate();
169    if (model == NULL) {
170      printf("OH_AI_ModelCreate failed.\n");
171      OH_AI_ContextDestroy(&context);
172      return OH_AI_STATUS_LITE_ERROR;
173    }
174
175    // Load and build the inference model. The model type is OH_AI_MODELTYPE_MINDIR.
176    int ret = OH_AI_ModelBuildFromFile(model, argv[1], OH_AI_MODELTYPE_MINDIR, context);
177    if (ret != OH_AI_STATUS_SUCCESS) {
178      printf("OH_AI_ModelBuildFromFile failed, ret: %d.\n", ret);
179      OH_AI_ModelDestroy(&model);
180      return ret;
181    }
182    ```
183
1844. Input data.
185 
186    Before executing model inference, you need to populate data to the input tensor. In this example, random data is used to populate the model.
187
188    ```c
189    // Obtain the input tensor.
190    OH_AI_TensorHandleArray inputs = OH_AI_ModelGetInputs(model);
191    if (inputs.handle_list == NULL) {
192      printf("OH_AI_ModelGetInputs failed, ret: %d.\n", ret);
193      OH_AI_ModelDestroy(&model);
194      return ret;
195    }
196    // Use random data to populate the tensor.
197    ret = GenerateInputDataWithRandom(inputs);
198    if (ret != OH_AI_STATUS_SUCCESS) {
199      printf("GenerateInputDataWithRandom failed, ret: %d.\n", ret);
200      OH_AI_ModelDestroy(&model);
201      return ret;
202    }
203   ```
204
2055. Execute model inference.
206
207    Call **OH_AI_ModelPredict** to perform model inference.
208
209    ```c
210    // Execute model inference.
211    OH_AI_TensorHandleArray outputs;
212    ret = OH_AI_ModelPredict(model, inputs, &outputs, NULL, NULL);
213    if (ret != OH_AI_STATUS_SUCCESS) {
214      printf("OH_AI_ModelPredict failed, ret: %d.\n", ret);
215      OH_AI_ModelDestroy(&model);
216      return ret;
217    }
218    ```
219
2206. Obtain the output.
221
222    After model inference is complete, you can obtain the inference result through the output tensor.
223
224    ```c
225    // Obtain the output tensor and print the information.
226    for (size_t i = 0; i < outputs.handle_num; ++i) {
227      OH_AI_TensorHandle tensor = outputs.handle_list[i];
228      int64_t element_num = OH_AI_TensorGetElementNum(tensor);
229      printf("Tensor name: %s, tensor size is %zu ,elements num: %lld.\n", OH_AI_TensorGetName(tensor),
230            OH_AI_TensorGetDataSize(tensor), element_num);
231      const float *data = (const float *)OH_AI_TensorGetData(tensor);
232      printf("output data is:\n");
233      const int max_print_num = 50;
234      for (int j = 0; j < element_num && j <= max_print_num; ++j) {
235        printf("%f ", data[j]);
236      }
237      printf("\n");
238    }
239    ```
240
2417. Destroy the model.
242
243    If the MindSpore Lite inference framework is no longer needed, you need to destroy the created model.
244
245    ```c
246    // Destroy the model.
247    OH_AI_ModelDestroy(&model);
248    ```
249
250## Verification
251
2521. Write **CMakeLists.txt**.
253
254    ```cmake
255    cmake_minimum_required(VERSION 3.14)
256    project(Demo)
257
258    add_executable(demo main.c)
259
260    target_link_libraries(
261            demo
262            mindspore_lite_ndk
263            pthread
264            dl
265    )
266    ```
267   - To use ohos-sdk for cross compilation, you need to set the native toolchain path for the CMake tool as follows: `-DCMAKE_TOOLCHAIN_FILE="/xxx/native/build/cmake/ohos.toolchain.cmake"`.
268    
269   - The toolchain builds a 64-bit application by default. To build a 32-bit application, add the following configuration: `-DOHOS_ARCH="armeabi-v7a"`.
270
2712. Run the CMake tool.
272
273    - Use hdc_std to connect to the device and put **demo** and **mobilenetv2.ms** to the same directory on the device.
274    - Run the hdc_std shell command to access the device, go to the directory where **demo** is located, and run the following command:
275
276    ```shell
277    ./demo mobilenetv2.ms
278    ```
279
280    The inference is successful if the output is similar to the following:
281
282    ```shell
283    # ./QuickStart ./mobilenetv2.ms                                            
284    Tensor name: Softmax-65, tensor size is 4004 ,elements num: 1001.
285    output data is:
286    0.000018 0.000012 0.000026 0.000194 0.000156 0.001501 0.000240 0.000825 0.000016 0.000006 0.000007 0.000004 0.000004 0.000004 0.000015 0.000099 0.000011 0.000013 0.000005 0.000023 0.000004 0.000008 0.000003 0.000003 0.000008 0.000014 0.000012 0.000006 0.000019 0.000006 0.000018 0.000024 0.000010 0.000002 0.000028 0.000372 0.000010 0.000017 0.000008 0.000004 0.000007 0.000010 0.000007 0.000012 0.000005 0.000015 0.000007 0.000040 0.000004 0.000085 0.000023 
287    ```
288