CUDA：一维、二维的grid、block的核函数线程分配

最新推荐文章于 2024-07-29 15:05:56 发布

rtoax

最新推荐文章于 2024-07-29 15:05:56 发布

阅读量3.2k

点赞数

分类专栏：【语言与编译】【高性能计算】

本文链接：https://blog.csdn.net/Rong_Toa/article/details/78656908

版权

【语言与编译】同时被 2 个专栏收录

456 篇文章 19 订阅

订阅专栏

【高性能计算】

48 篇文章 11 订阅

订阅专栏

1.一维grid，一维block

    int nx = 1 << 14;
    int ny = 1 << 14;
    int dimx = 32;
    dim3 block(dimx, 1);
    dim3 grid((nx + block.x - 1) / block.x, 1);

核函数

__global__ void kernel_function()
{
    unsigned int ix = threadIdx.x + blockIdx.x * blockDim.x;

    if (ix < nx )
    {
        do_something();
    }
}

2.二维grid，一维block

    int nx = 1 << 14;
    int ny = 1 << 14;
    int dimx = 32;
    dim3 block(dimx, 1);
    dim3 grid((nx + block.x - 1) / block.x, ny);

核函数

__global__ void kernel_function()
{
    unsigned int ix = threadIdx.x + blockIdx.x * blockDim.x;
    unsigned int iy = blockIdx.y;
    unsigned int idx = iy * nx + ix;

    if (ix < nx && iy < ny)
    {
        do_something();
    }
}

3.二维grid，二维block

    int nx = 1 << 14;
    int ny = 1 << 14;
    int dimx = 32;
    int dimy = 32;
    dim3 block(dimx, dimy);
    dim3 grid((nx + block.x - 1) / block.x, (ny + block.y - 1) / block.y);

核函数

__global__ void kernel_function()
{
    unsigned int ix = threadIdx.x + blockIdx.x * blockDim.x;
    unsigned int iy = threadIdx.y + blockIdx.y * blockDim.y;
    unsigned int idx = iy * nx + ix;

    if (ix < nx && iy < ny)
    {
        do_something();
    }
}