【MPI高性能计算】用集合通信改进梯形求和积分

ゞ浴缸里的玫瑰 2022-04-10 06:49 186阅读 0赞

## 简述 ##

*  [【MPI高性能计算】梯形面积积分计算][MPI]
 *  里面有梯形积分公式的具体写法

这里采用MPI\_Reduce的集合通信的方式来计算

## API ##

int MPI_Reduce(const void *sendbuf, void *recvbuf, int count,
                   MPI_Datatype datatype, MPI_Op op, int root,
                   MPI_Comm comm)

注意，在send和recv上，放的指针最好别指向同一个空间

#include <mpi.h>
    #include <stdio.h>
    #include <string>
    #include <string.h>
    #pragma warning(disable : 4996)
    const int MAX_STRING = 100;
    
    #define FUN(x) (x * x)
    
    int main(int argc, char **argv) { 
    	int comm_sz;
    	int my_rank;
    	if (argc == 1) return 0;
    	
    	int n = strtol(argv[1], NULL, 10);
    	double a = 0, b = 1;
    	MPI_Init(NULL, NULL);
    	MPI_Comm_size(MPI_COMM_WORLD, &comm_sz);
    	MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
    
    	int left = n % comm_sz;
    	int localn = n / comm_sz + (my_rank < left);
    	double h = (b - a) / n, locala, localb;
    
    	if (my_rank < left) {  locala = a + my_rank * localn * h; }
    	else {  locala = a + left * (localn + 1) * h + (my_rank - left) * localn * h; }
    	localb = locala + localn * h;
    
    	double x = locala; // initial x
    	double localSum = ( FUN(locala) + FUN(localb)) / 2, totalSum=0;
    	for (int i = 1; i < localn; ++i) { 
    		x += h;
    		localSum += FUN(x);
    	}
    	localSum *= h;
    
    	MPI_Reduce(&localSum, &totalSum, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
    
    	if (my_rank == 0) printf("comm_sz: %d, n: %d, ans: %.5f\n", comm_sz, n, totalSum);
    	
    	MPI_Finalize();
    }

*  一样能算的正确，主要是会简化，同时用这样的方式也可以加速计算，毕竟求和的过程也被默认并行了。

PS D:\Code\C++\repo\MPITest\x64\Debug> mpiexec -n 3 ./MPITest.exe 1000
    comm_sz: 3, n: 1000, ans: 0.33333
    PS D:\Code\C++\repo\MPITest\x64\Debug>

[MPI]: https://blog.csdn.net/a19990412/article/details/85320564