Function main

code/Chapter12/12-6-mpi-hip-cannon.cpp:46–135

    hipFree(d_C);
}
int main(int argc, char *argv[]){
    int id,p,part,num,i,
    upRank,downRank,leftRank,rightRank;
    float *a, *b, *c;
    float *A,*B,*C;
    int coord[2],x,y;//本进程坐标
    int position[2] = {0, 0};//确定（0，0）位置上的进程id
    double start_time, end_time;
    MPI_Comm MPI_COMM_CART;
    MPI_Status status;
    MPI_Init(&argc, &argv);
    MPI_Comm_size(MPI_COMM_WORLD, &p);
    int periodic[2];
    int size[2];
    part = sqrt(p);
    //如果进程数不是平方数，则终止进程
    if(part*part!=p){
        printf("总进程数必须是一个平方数!\n");
        MPI_Finalize();
        return 0;
    }
    //如果行数不是进程数的整数倍
    if(DIMS%p!=0){
        printf("总进程数开方之后必须能整除矩阵总行数!\n");
        MPI_Finalize();
        return 0;
    }
    num = DIMS/part;
    //建立笛卡尔通信域
    size[0] = size[1] = part;//虚拟拓扑维数
    periodic[0] = periodic[1] = 1;//虚拟拓扑中进程下标是否循环
    MPI_Cart_create(MPI_COMM_WORLD, 2, size, periodic, 1, &MPI_COMM_CART);
    MPI_Comm_rank(MPI_COMM_CART,&id);
    MPI_Cart_coords(MPI_COMM_CART,id,2,coord);
    x = coord[0];
    y = coord[1];
    //为各个进程分配对应部分块矩阵的空间
    a = (float *)malloc(sizeof(float)*num*num);
    b = (float *)malloc(sizeof(float)*num*num);
    c = (float *)malloc(sizeof(float)*num*num);
    if (id==0){
        A = (float *)malloc(sizeof(float)*DIMS*DIMS);
        B = (float *)malloc(sizeof(float)*DIMS*DIMS);
        C = (float *)malloc(sizeof(float)*DIMS*DIMS);
        Init_Matrix(A,DIMS*DIMS,2);
        Init_Matrix(B,DIMS*DIMS,2);
    }
    start_time = MPI_Wtime();
    //分发矩阵
    Matrix_cannon_scatter(DIMS,id,num,part,A,a,MPI_COMM_CART);
    Matrix_cannon_scatter(DIMS,id,num,part,B,b,MPI_COMM_CART);
    Init_Matrix(c,num*num,1);
    //获得当前矩阵块横坐标方向相聚x的左右方向的邻居进程号
    //分块a循环左移x位
    MPI_Cart_shift(MPI_COMM_CART, 1, x, &leftRank, &rightRank);
    MPI_Sendrecv_replace(a, num*num, MPI_FLOAT, leftRank, 0, rightRank, 0, MPI_COMM_CART, &status);
    //获得当前矩阵块纵坐标方向相聚y的上下邻居进程号
    //分块b循环上移y位