[C++知识库]CUDA并行化向量相乘Example

该Example使用最基本的cudaMalloc(), cudaMemcpy()等函数以及Kernel函数实现了一个并行化向量相乘的案例 :

#include <iostream>

using namespace std;

static void HandleError(cudaError_t err, const char *file, int line) {

    // CUDA error handling from the "CUDA by example" book
    if (err != cudaSuccess) {
        printf("%s in %s at line %d\n", cudaGetErrorString(err), file, line);
        exit(EXIT_FAILURE);
    }
}

#define HANDLE_ERROR(err) (HandleError( err, __FILE__, __LINE__ ))

__global__ void product(const int *a, const int *b, int *c) {
    unsigned int index = threadIdx.x;
    c[index] = a[index] * b[index];
}

int main() {
    const int n = 10;

    int a[n];
    int b[n];

    // 初始化数据
    for (int i = 0; i < n; i++) {
        a[i] = i;
        b[i] = i;
    }

    int *aDev, *bDev, *cDev;

    // 申请device内存（device指代GPU，host指代CPU）
    HANDLE_ERROR(cudaMalloc((void **) &aDev, sizeof(int) * n));
    HANDLE_ERROR(cudaMalloc((void **) &bDev, sizeof(int) * n));
    HANDLE_ERROR(cudaMalloc((void **) &cDev, sizeof(int) * n));

    // 将host数据拷贝到device
    HANDLE_ERROR(cudaMemcpy((void *) aDev, (void *) a, sizeof(int) * n, cudaMemcpyHostToDevice));
    HANDLE_ERROR(cudaMemcpy((void *) bDev, (void *) b, sizeof(int) * n, cudaMemcpyHostToDevice));

    // 定义kernel的执行配置
    dim3 gridSize(1);
    dim3 blockSize(n);

    // 执行kernel
    product<<<gridSize, blockSize>>>(aDev, bDev, cDev);

    int c[n];

    // 将device得到的结果拷贝到host
    HANDLE_ERROR(cudaMemcpy(c, cDev, sizeof(int) * n, cudaMemcpyDeviceToHost));

    // 检查执行结果
    for (int i: c) {
        cout << i << endl;
    }

    // 释放device内存
    HANDLE_ERROR(cudaFree(aDev));
    HANDLE_ERROR(cudaFree(bDev));
    HANDLE_ERROR(cudaFree(cDev));

    return 0;
}