__shared__ float cache[threadPerBlock];//声明共享内存缓冲区，__shared__
__syncthreads();//对线程块中的线程进行同步，只有都完成前面的任务才可以进行后面的
代码：
/*
============================================================================
Name : dot.cu
Author : can
Version :
Copyright : Your copyright notice
Description : CUDA compute reciprocals
============================================================================
*/
#include <iostream>
using namespace std;
static void CheckCudaErrorAux (const char *, unsigned, const char *, cudaError_t);
#define CUDA_CHECK_RETURN(value) CheckCudaErrorAux(__FILE__,__LINE__, #value, value)
#define imin(a,b) (a<b?a:b)
const int N=33*1024;
const int threadPerBlock=256;
const int blockPerGrid=imin(32,(N+threadPerBlock-1)/threadPerBlock);
__global__ void dot(float *a,float *b,float *c)
{
__shared__ float cache[threadPerBlock];//声明共享内存缓冲区，__shared__，
int tid = threadIdx.x + blockIdx.x*blockDim.x;
int cacheIndex = threadIdx.x;
float temp = 0;
while(tid < N)
{
temp += a[tid] * b[tid];
tid += blockDim.x*gridDim.x;
}
cache[cacheIndex] = temp;
__syncthreads();//对线程块中的线程进行同步，只有都完成前面的任务才可以进行后面的
int i = blockDim.x/2;//归约运算
while(i != 0)
{
if(cacheIndex < i)
{
cache[cacheIndex] += cache[cacheIndex + i];
}
__syncthreads();
i /=2;
}
if(cacheIndex == 0)
{
c[blockIdx.x] = cache[0];
}
}
int main()
{
float *a,*b,c,*partial_c;
float *dev_a,*dev_b,*dev_partial_c;
a = (float *)malloc(N*sizeof(float));
b = (float *)malloc(N*sizeof(float));
partial_c = (float *)malloc(blockPerGrid*sizeof(float));
CUDA_CHECK_RETURN(cudaMalloc((void **)&dev_a,N*sizeof(float)));
CUDA_CHECK_RETURN(cudaMalloc((void **)&dev_b,N*sizeof(float)));
CUDA_CHECK_RETURN(cudaMalloc((void **)&dev_partial_c,N*sizeof(float)));
for(int i=0;i<N;i++)
{
a[i] = i;
b[i] = i*2;
}
CUDA_CHECK_RETURN(cudaMemcpy(dev_a,a,N*sizeof(float),cudaMemcpyHostToDevice));
CUDA_CHECK_RETURN(cudaMemcpy(dev_b,b,N*sizeof(float),cudaMemcpyHostToDevice));
dot<<<blockPerGrid,threadPerBlock>>>(dev_a,dev_b,dev_partial_c);
CUDA_CHECK_RETURN(cudaMemcpy(partial_c,dev_partial_c,blockPerGrid*sizeof(float),cudaMemcpyDeviceToHost));
c=0;
for(int i=0;i<blockPerGrid;i++)
{
c += partial_c[i];
}
#define sum_squares(x) (x*(x+1)*(2*x+1)/6)
cout<<"Does GPU value "<<c<<" = "<<2*sum_squares((float)(N-1))<<endl;
cudaFree(dev_a);
cudaFree(dev_b);
cudaFree(dev_partial_c);
free(a);
free(b);
free(partial_c);
return 0;
}
static void CheckCudaErrorAux (const char *file, unsigned line, const char *statement, cudaError_t err)
{
if (err == cudaSuccess)
return;
std::cerr << statement<<" returned " << cudaGetErrorString(err) << "("<<err<< ") at "<<file<<":"<<line << std::endl;
exit (1);
}