Write the first CUDA program

1. Environmental preparation

Make sure cuda The driver has been installed correctly , Execute the following command to view ：

nvidia-smi

There are specific graphics card information display , It indicates that the graphics card and its driver are working normally . If it does not work properly, you can refer to course Installation driver .

2. Source code preparation

The following source code output graphics card support CUDA Program information ,cuda_status Used to judge the status of obtaining information ,cuda_status by 0 Show success .

#include <iostream>
#include "cuda_runtime.h"
#include "device_launch_parameters.h"

using namespace std;

int main()
{
    
    int dev = 0;
    cudaDeviceProp dev_prop;
    cudaError_t cuda_status;
    cuda_status = cudaGetDeviceProperties(&dev_prop, dev);
    cout << "cuda_status:" << cuda_status << endl;
    cout << " Use GPU device " << dev << ": " << dev_prop.name << endl;
    cout << "SM The number of ：" << dev_prop.multiProcessorCount << endl;
    cout << " Shared memory size per thread block ：" << dev_prop.sharedMemPerBlock / 1024.0 << " KB" << endl;
    cout << " Maximum number of threads per thread block ：" << dev_prop.maxThreadsPerBlock << endl;
    cout << " Every SM Is the maximum number of threads ：" << dev_prop.maxThreadsPerMultiProcessor << endl;
    cout << " Every SM The maximum number of thread bundles ：" << dev_prop.maxThreadsPerMultiProcessor / 32 << endl;

    return 0;
}

3. Compile and run

The above source code is saved in the file GPU_info.cu in , The compile command is as follows ：

nvcc  GPU_info.cu  -o gpu_info
./gpu_info

As shown in the figure below , It is the result of correct execution .