CUDA实战

1.第一个程序，输出hello world，1个Block块中含有5个线程

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

__global__ void hello(void)
{
  printf("hello world from GPU!
");
}
int main()
{
    printf("hello world from CPU!
");
    hello<<<1,5>>>();
    //重置CUDA设置释放程序占用的资源
    cudaDeviceReset();
    return 0;
}

2.参数的传入，

#include <stdio.h>
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
__global__ void add(int i,int j)
{
   int count;
   count = i + j;
   printf("
 Sum is %d
",count);
}

int main()
{
    add<<<1,1>>>(10,20);
    cudaDeviceReset();
    return 0;
}

 3.数据的传入与传出，我们的数据要从内存copy到显存上面，然后现在又要从显存上面copy回来

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

__global__ void decrease(int a, int b, int *c)
{
    *c = a + b;
}
int main()
{
    int *c=0;
    int *dev_c=0;
    //初始化CPU上的内存空间
    c = (int*)malloc(sizeof(int));
    //初始化GPU上的内存空间
    cudaMalloc((void**)&dev_c,sizeof(int));
    //调用内核函数
    decrease <<<1,1>>>(15,20,dev_c);
    //等待设备所有线程任务执行完毕
    cudaDeviceSynchronize();
    //将数据从device中复制到hist中
    cudaMemcpy(c,dev_c,sizeof(int),cudaMemcpyDeviceToHost);
    //输出
    printf(" c = %d
",*c);
    //释放内存
    cudaFree(dev_c);
    free(c);
    return 0;
}

 4.传入的值全改为指针类型

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

__global__ void addCuda(int* a, int* b, int* c)
{
    *c = *a - *b;
}

void addWithCuda(int *c,int *a,int *b)
{
    int *dev_c = 0;
    int *dev_a = 0;
    int *dev_b = 0;
    
    //初始化CUDA内存
    cudaMalloc((void**)&dev_c,sizeof(int));
    cudaMalloc((void**)&dev_a,sizeof(int));
    cudaMalloc((void**)&dev_b,sizeof(int));
    
    //从主机复制数据复制到device上
    cudaMemcpy(dev_a,a,sizeof(int),cudaMemcpyHostToDevice);
    cudaMemcpy(dev_b,b,sizeof(int),cudaMemcpyHostToDevice);
    
    //调用内核函数
    addCuda<<<1,1>>>(dev_a,dev_b,dev_c);
    cudaDeviceSynchronize();
    
    //数据复制到host
    cudaMemcpy(c,dev_c,sizeof(int),cudaMemcpyDeviceToHost);
    
    cudaFree(dev_c);
    cudaFree(dev_a);
    cudaFree(dev_b);

}

int main()
{
   int a, b, c;
   a = 30;
   b = 15;
   c = 10;
   //传入参数变量（地址）
   addWithCuda(&c,&a,&b);
   //重置CUDA设备释放程序占用的程序
   cudaDeviceReset();
   printf("Value is %d
", c);

    return 0;
}

5.传入的值全改为指针类型

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

__global__ void deCuda(int* a, int* b, int* c)
{
    *c = *a - *b;
}
int main()
{
   int *a, *b, *c;
   a = (int*)malloc(sizeof(int));
   b = (int*)malloc(sizeof(int));
   c = (int*)malloc(sizeof(int));
   *a=10;
   *b=5;
   *c=0;
   int *dev_c = 0;
   int *dev_a = 0;
   int *dev_b = 0;
    //3.请求CUDA设备的内存（显存），执行CUDA函数
    cudaMalloc((void**)&dev_c, sizeof(int));
    cudaMalloc((void**)&dev_a, sizeof(int));
    cudaMalloc((void**)&dev_b, sizeof(int));
    
    cudaMemcpy(dev_a, a, sizeof(int), cudaMemcpyHostToDevice);
    cudaMemcpy(dev_b, b, sizeof(int), cudaMemcpyHostToDevice);
    
    deCuda<<<1,1>>>(dev_a,dev_b,dev_c);
    cudaMemcpy(c, dev_c, sizeof(int), cudaMemcpyDeviceToHost);
    printf("Value is %d
", *c);
   
    cudaFree(dev_c);
    cudaFree(dev_a);
    cudaFree(dev_b);
    free(a);
    free(b);
    free(c);
   //重置CUDA设备释放程序占用的程序
   cudaDeviceReset();
   return 0;
}

 6. 程序实现向量的加法操作，一个block中含有512个线程

#include <stdio.h>
#include <cuda_runtime.h>
__global__ void add( int *dev_a, int *dev_b, int *dev_c)
{
    int i=threadIdx.x;
    dev_c[i] = dev_a[i] + dev_b[i];
}
int main()
{
    int host_a[512], host_b[512], host_c[512];
    for(int i = 0; i < 512; i++)
    {
        host_a[i] = i;
        host_b[i] = i<<1;
    }
    //定义cudaError,默认为cudaSuccess
    cudaError_t err = cudaSuccess;
    int *dev_a, *dev_b, *dev_c;
    err = cudaMalloc((void**)&dev_a,sizeof(int)*512);
    err = cudaMalloc((void**)&dev_b,sizeof(int)*512);
    err = cudaMalloc((void**)&dev_c,sizeof(int)*512);
    
    if(err!=cudaSuccess)
    {
        printf("the cuadaMalloc on GPU is failed");
        return 1;
    }
    
    printf("SUCCESS
");
    //从host到device
    cudaMemcpy(dev_a,host_a,sizeof(host_a),cudaMemcpyHostToDevice);
    cudaMemcpy(dev_b,host_b,sizeof(host_b),cudaMemcpyHostToDevice);
    
    //调用核函数
    add<<<1,512>>>(dev_a,dev_b,dev_c);
    cudaMemcpy(&host_c,dev_c,sizeof(host_c),cudaMemcpyDeviceToHost);
    for(int i=0; i<512; i++)
    {
        printf("host_a[%d] + host_b[%d] = %d + %d = %d
",i,i,host_a[i],host_b[i],host_c[i]);
    }
    
    //释放内存
    cudaFree(dev_c);
    cudaFree(dev_b);
    cudaFree(dev_a);
    
    return 0;
    
}