https://developer.nvidia.com/blog/easy-introduction-cuda-c-and-c/
host 指的是CPU及其内存,device指的是GPU及其内存。
在主机上运行的代码可以管理host 和device上的内存,还可以启动kernels ,这些kernels 是在device上执行的functions。这些kernels 由许多GPU线程并行执行。
- 声明和分配主host 和device内存。
- 初始化host 数据。
- 将数据从host 传输到device。
- 执行一个或多个kernels 。
- 将结果从device传输到host 。
__global__
:声明device kernel functions;
device variables blockDim
, blockIdx
, and threadIdx
获取全局索引
#include <stdio.h>
__global__
void saxpy(int n, float a, float* x, float* y)//kernel that runs in parallel on the GPU,
{
//Device Code
int i = blockIdx.x * blockDim.x + threadIdx.x;//generates a global index
if (i < n) y[i] = a * x[i] + y[i];//n, to ensure there are no out-of-bounds memory accesses.
}
int main(void)//host code
{
//声明两组数组
int N = 1 << 20;
float* x, * y, * d_x, * d_y;
x = (float*)malloc(N * sizeof(float));// The pointers x and y point to the host arrays, allocated with malloc in the typical fashion,
y = (float*)malloc(N * sizeof(float));
cudaMalloc(&d_x, N * sizeof(float));//d_x and d_y arrays point to device arrays allocated with the cudaMalloc function from the CUDA runtime API.
cudaMalloc(&d_y, N * sizeof(float));
for (int i = 0; i < N; i++) {//initializes the host arrays
x[i] = 1.0f;
y[i] = 2.0f;
}
cudaMemcpy(d_x, x, N * sizeof(float), cudaMemcpyHostToDevice);//initialize the device arrays
cudaMemcpy(d_y, y, N * sizeof(float), cudaMemcpyHostToDevice);
// Perform SAXPY on 1M elements
saxpy << <(N + 255) / 256, 256 >> > (N, 2.0f, d_x, d_y);//Launching a Kernel;
// the number of thread blocks in the grid;the number of threads in a thread block.
//Thread blocks and grids can be made one-, two- or three-dimensional by passing dim3
cudaMemcpy(y, d_y, N * sizeof(float), cudaMemcpyDeviceToHost);//get the results back to the host
float maxError = 0.0f;
for (int i = 0; i < N; i++)
maxError = max(maxError, abs(y[i] - 4.0f));
printf("Max error: %f\n", maxError);
cudaFree(d_x);//Cleaning Up
cudaFree(d_y);
free(x);
free(y);
}