mpi矩阵乘法c语言代码,MPI中的加农矩阵乘法

使用单个节点执行时它可以正常工作，但不止于此而且它还没有完成 .

我试图在4个节点上使用随机4 * 4矩阵 . 从我可以收集的内容绝对没有调试工具，除了printf进程，等级0完成执行该功能，但其他进程卡在第68行的MPI_Wait上 . 我是C和并行编程的新手，无法弄清楚是什么错误 .

编辑：在调试过程中，我发现了一个令人难以置信的令人困惑的事情 - 调用MPI_Barrier似乎挂起了整个程序 .

在乘法功能完成后，我将主要调用MPI_Barrier . 为了使调试更容易，我决定将功能代码分成几部分并用障碍物分开 . 将屏障放置在主计算循环的末尾(以便每个进程必须完成一次迭代以便任何进程启动下一个进程)导致循环甚至没有达到第二次迭代 .

似乎当非阻塞发送/接收和随后调用MPI_Wait后单个进程到达障碍时，它会导致其他进程无限期地等待并且永远不会到达障碍 .

这是一个示例代码：

void generateArray(int * vector, int size){

int i;

for (i = 0; i < size; i ++){

vector[i] = rand()%10 + 1;

}

}

int test_size = 4;

int main( int argc, char **argv )

{

MPI_Init (&argc, &argv);

int comm_size, myrank;

int my2drank, mycoords[2];

MPI_Comm_size(MPI_COMM_WORLD, &comm_size);

MPI_Comm_rank(MPI_COMM_WORLD, &myrank);

MPI_Request requests[4];

MPI_Status status[4];

MPI_Comm comm_2d;

int dims[2], periods[2];

dims[0] = dims[1] = sqrt(comm_size);

periods[0] = periods[1] = 1;

MPI_Cart_create(MPI_COMM_WORLD, 2, dims, periods, 1, &comm_2d);

int sizelocal = test_size/dims[0];

srand(time(NULL));

int a[sizelocal*sizelocal], b[sizelocal*sizelocal];

generateArray(a, sizelocal*sizelocal);

generateArray(b, sizelocal*sizelocal);

MPI_Comm_rank(comm_2d, &my2drank);

MPI_Cart_coords(comm_2d, my2drank, 2, mycoords);

int uprank, downrank, leftrank, rightrank;

MPI_Cart_shift(comm_2d, 0, -1, &rightrank, &leftrank);

MPI_Cart_shift(comm_2d, 1, -1, &downrank, &uprank);

int *buffersA[2], *buffersB[2];

buffersA[0] = a;

buffersA[1] = (int *)malloc(sizelocal*sizelocal*sizeof(int));

buffersB[0] = b;

buffersB[1] = (int *)malloc(sizelocal*sizelocal*sizeof(int));

int shiftsource, shiftdest;

MPI_Cart_shift(comm_2d, 0, -mycoords[0], &shiftsource, &shiftdest);

MPI_Sendrecv_replace(buffersA[0], sizelocal*sizelocal, MPI_INT,shiftdest, 1, shiftsource, 1, comm_2d, &status[0]);

MPI_Cart_shift(comm_2d, 1, -mycoords[1], &shiftsource, &shiftdest);

MPI_Sendrecv_replace(buffersB[0], sizelocal*sizelocal, MPI_INT,shiftdest, 1, shiftsource, 1, comm_2d, &status[0]);

printf("Rank %d at point 1\n", my2drank);

int i;

for (i = 0; i < 2; i ++){

printf("Rank %d started iteration %d\n", my2drank, i);

MPI_Isend(buffersA[i%2], sizelocal*sizelocal, MPI_INT,leftrank, 1, comm_2d, &requests[0]);

MPI_Isend(buffersB[i%2], sizelocal*sizelocal, MPI_INT,uprank, 1, comm_2d, &requests[1]);

MPI_Irecv(buffersA[(i+1)%2], sizelocal*sizelocal, MPI_INT,rightrank, 1, comm_2d, &requests[2]);

MPI_Irecv(buffersB[(i+1)%2], sizelocal*sizelocal, MPI_INT,downrank, 1, comm_2d, &requests[3]);

MPI_Waitall(4, requests, status);

printf("Rank %d stopped waiting in iteration %d\n", my2drank, i);

MPI_Barrier(comm_2d);

}

printf("Rank %d at point 2\n", my2drank);

MPI_Barrier(MPI_COMM_WORLD);

MPI_Finalize();

return 0;

}

这是输出：

Rank 0 at point 1

Rank 0 started iteration 0

Rank 1 at point 1

Rank 1 started iteration 0

Rank 2 at point 1

Rank 2 started iteration 0

Rank 3 at point 1

Rank 3 started iteration 0

Rank 0 stopped waiting in iteration 0