并行Pi的计算

来源：互联网发布：理想国英剧知乎编辑：程序博客网时间：2024/05/17 08:47

主要考虑两种PI的计算方法，一种为积分法的用共享内存试的GPU实现，另外一种为在面积估算MPI伪代码实现。

1.积分法求pi

// 计算PI // 算法：在[0,1]范围内积分 1/(1+x*x)// 做积分，把每一个block中的和放入sum对应的下标中__global__ void kernel1(float* sum, int num){int gid = blockIdx.x*blockDim.x+threadIdx.x;//线程索引float temp;extern float __shared__ pi[]; // 每一个block中的共享内存pi[threadIdx.x] = 0.0; // 初始化__syncthreads();// 计算积分 ， 放入对应线程里while(gid < num){temp = (gid + 0.5f) / num;pi[threadIdx.x] += 4.0/(1 + temp*temp);gid += blockDim.x * gridDim.x}// 对每一个block归约求和for(int i=(blockDim.x>>1);i>0;i>>=1){       if(threadIdx.x<i){          pi[threadIdx.x] += pi[threadIdx.x+i];       }       __syncthreads();    }    if(threadIdx.x==0)       sum[blockIdx.x]=s_pi[0];    }}// 对sum在执行一次归约求和__global__ void  kernel2(float *sum,int num,float *pi){   int id=threadIdx.x;   extern float __shared__ s_sum[];   s_sum[id] = sum[id];    __syncthreads();    for(int i=(blockDim.x>>1);i>0;i>>=1){       if(id<i)       s_sum[id]+=s_sum[id+i];       __syncthreads();}if(id==0){    *d_pi=s_sum[0]/num;} }

2.面积法求pi

/*  算法1.Inscribe a circle in a square2.Randomly generate points in the square3.Determine the number of points in the square that are also in the circleLet r be the number of points in the circle divided by the number of points in the square4.PI ~ 4*r   */npoints = 10000circle_count = 0p = number of tasksnum = npoints/pfind out if I am MASTER or WORKER do j = 1,num   generate 2 random numbers between 0 and 1  xcoordinate = random1  ycoordinate = random2  if (xcoordinate, ycoordinate) inside circle  then circle_count = circle_count + 1end doif I am MASTER  receive from WORKERS their circle_counts  compute PI (use MASTER and WORKER calculations)else if I am WORKER  send to MASTER circle_countendif

两种计算模式，但其实这两种计算方法都比较适合在GPU上实现

详细的过程可以参考：风晨的CUDA入门和 Introduction to Parallel Computing

0 0