ｃｕｄａ基础－－－异步并行执行

实例：ａｓｙｎｃＡＰＩ实例

#include <stdio.h>
#include <cutil_inline.h>

__global__ void 
increment_kernel(int* g_data,int inc_value)
{
int idx = blockIdx.x*blockDm.x+threadIdx.x;
g_data[idx] = g_data[idx]+nc_value;
}

int corect_output(int* data,const int n,const int x)
{
for(int i = 0;i < n;i++)
{
if(data[i] != x)
return 0;
return 1;
}
}

int min(int arg,char** argv)
{
if(cutCheckCmdLineFlag(argc,(const char**)argv,"device"));
cutilDeviceInit(argc,argv);
else
cudaSetDevice(cutGetMaxGflopsDeviceId());

int n = 16*1024*1024;
int nbytes = n*sizeof(int);
int value = 26;
//内存分配
int* a = 0;
cutilSafeCall(cudaMalloc((void**)&a,nbytes));
memset(a,0,nbytes);
//显存分配
int* d_a = 0;
cutilSafeCall(cudaMallocHost((void**)&d_a,nbytes));
cutilSafeCall(cudaMemset(d_a,255,nbytes));

//设置执行参数
dim3 threads = dim3(512,1);
dim3 blocks = dim3(n/threads.x,1);

//创建事件
cudaEvent_t start,stop;
cudaEventCreate(&start);
cudaEventCreate(&stop);


unsigned int timer;
cutCreateCreate(&timer);
cutResetTimer(timer);
cudaThreadSynchronize();
float gpu_time = 0.0f;

／／异步向ＧＰＵ发送任务
cutStartTimer(timer);
cudaEventRecord(start,0);
cudaMemcpyAsync(d_a,a,nbytes,cudaMemcpyHostToDevice,0);
increment_kernel<<<blocks,threads,0,0>>>(d_a,value);
cudaMemcpyAsync(a,d_a,nbytes,cudaMemcpyDeviceToHost,0);
cudaEventRecord(stop,0);
cutStopTimer(timer);

//以上代码异步喜爱那个ＧＰＵ发送要执行的任务，同时ＣＰＵ端执行简单的ｃｏｕｎｔｅｒ＋＋操作－－－ＧＰＵ　ＣＰＵ的执行可以覆盖。在等待上一步操作完成的时间内，ＣＰＵ可以做一部分工作。
unsigned long int counter = 0;
while(cudaEventQury(stop) == cudaErrorNotReady)
{
counter++;
}
cudaEventElapsedTimer(&gpu_tim,start,stop);
printf("timer/gpu_time/counter");

//正确性检测
if(correct_output(a,n,value))
print("passed");

//释放资源
cudaEventDestroy(start);
cudaEventDectroy(stop);
cudaFreeHost(a);
cudaFree(d_a);

cudaThreadExit();
}