最简单的 CUDA 程序，hello world from CUDA

hello_world.cu

#include <stdio.h>

// CUDA 内核函数 - 在GPU上执行
__global__ void helloFromGPU()
{
    // 每个线程都会执行这个函数
    printf("Hello World from GPU! Thread %d in block %d\n", threadIdx.x, blockIdx.x);
}

int main()
{
    // 在主机(CPU)上打印
    printf("Hello World from CPU!\n");

    // 启动内核函数
    // 参数: 块数量, 每个块的线程数量
    helloFromGPU<<<2, 5>>>();

    // 等待GPU完成
    cudaDeviceSynchronize();

    return 0;
}

编译运行：

nvcc hello_world.cu -o hello_world

输出如下：

Hello World from CPU!
Hello World from GPU! Thread 0 in block 1
Hello World from GPU! Thread 1 in block 1
Hello World from GPU! Thread 2 in block 1
Hello World from GPU! Thread 3 in block 1
Hello World from GPU! Thread 4 in block 1
Hello World from GPU! Thread 0 in block 0
Hello World from GPU! Thread 1 in block 0
Hello World from GPU! Thread 2 in block 0
Hello World from GPU! Thread 3 in block 0
Hello World from GPU! Thread 4 in block 0