CUDA:将含有多级指针的结构体指针传入设备，并在设备中

致远的方

已于 2023-03-23 15:25:22 修改

阅读量294

点赞数

文章标签： c++ c语言

于 2023-03-20 22:08:37 首次发布

本文链接：https://blog.csdn.net/zhiyuandefang/article/details/129677786

版权

示意图：

可直接运行，见代码：

#include <cuda_runtime.h>
#include <stdio.h>

struct PARA
{
    int c;
};

struct ELE
{
    int b;
    PARA *h_para;
    PARA *d_para;
};

struct LINE
{
    int a;
    ELE *h_ele;
    ELE *d_ele;
};
struct TASK
{
    int s;
    LINE *h_line;
    LINE *d_line;
};
__global__ void printtask(TASK *d_task)
{
    printf("g_s: %d\n", d_task->s);
    printf("g_a: %d\n", d_task->d_line->a);
    printf("g_b: %d\n", d_task->d_line->d_ele->b);
    printf("g_c: %d\n", d_task->d_line->d_ele->d_para->c);
    d_task->s = 11;
    d_task->d_line->a = 12;
    d_task->d_line->d_ele->b = 13;
    d_task->d_line->d_ele->d_para->c = 14;
}
// 多级结构体指针的嵌套传入gpu的方法
int main()
{
    // 主机结构体
    TASK *h_task = (TASK *)malloc(sizeof(TASK));
    h_task->s = 1;
    h_task->h_line = (LINE *)malloc(sizeof(LINE));
    h_task->h_line->a = 2;
    h_task->h_line->h_ele = (ELE *)malloc(sizeof(ELE));
    h_task->h_line->h_ele->b = 3;
    h_task->h_line->h_ele->h_para = (PARA *)malloc(sizeof(PARA));
    h_task->h_line->h_ele->h_para->c = 4;
    printf("h_s1 : %d\n", h_task->s);
    printf("h_a1 : %d\n", h_task->h_line->a);
    printf("h_b1 : %d\n", h_task->h_line->h_ele->b);
    printf("h_c1 : %d\n", h_task->h_line->h_ele->h_para->c);

    // 设备结构体
    TASK *d_task;
    // 设备内存的初始化必须通过主机端指针来进行，会在设备端生成同名的指针
    cudaMalloc((void **)&d_task, sizeof(TASK));
    cudaMalloc((void **)&h_task->d_line, sizeof(LINE));
    cudaMalloc((void **)&h_task->h_line->d_ele, sizeof(ELE));
    cudaMalloc((void **)&h_task->h_line->h_ele->d_para, sizeof(PARA));
    // 将主机指针指向的内容传入设备指针指向的内存
    cudaMemcpy(d_task, h_task, sizeof(TASK), cudaMemcpyHostToDevice);
    cudaMemcpy(h_task->d_line, h_task->h_line, sizeof(LINE), cudaMemcpyHostToDevice);
    cudaMemcpy(h_task->h_line->d_ele, h_task->h_line->h_ele, sizeof(ELE), cudaMemcpyHostToDevice);
    cudaMemcpy(h_task->h_line->h_ele->d_para, h_task->h_line->h_ele->h_para, sizeof(PARA), cudaMemcpyHostToDevice);

    printtask<<<1, 1>>>(d_task);
    cudaDeviceSynchronize();
    // 将设备指针指向的内容传入主机指针指向的内存
    cudaMemcpy(h_task, d_task, sizeof(TASK), cudaMemcpyDeviceToHost);
    cudaMemcpy(h_task->h_line, h_task->d_line, sizeof(LINE), cudaMemcpyDeviceToHost);
    cudaMemcpy(h_task->h_line->h_ele, h_task->h_line->d_ele, sizeof(ELE), cudaMemcpyDeviceToHost);
    cudaMemcpy(h_task->h_line->h_ele->h_para, h_task->h_line->h_ele->d_para, sizeof(PARA), cudaMemcpyDeviceToHost);

    printf("h_s2 : %d\n", h_task->s);
    printf("h_a2 : %d\n", h_task->h_line->a);
    printf("h_b2 : %d\n", h_task->h_line->h_ele->b);
    printf("h_c2 : %d\n", h_task->h_line->h_ele->h_para->c);
    return 0;
}