unexpected deadlock in MPI - c

Question

I hope someone can help me. My code:
void process(int myid, int numprocs)
{
int i,j, anzahl, rest;
MPI_Status stat;
meta = (int *)malloc(3 * sizeof(int));
if(myid == 0)
{
meta[0] = ASpalten;
meta[1] = AZeilen;
meta[2] = BSpalten;
for (i = 0; i < numprocs; i++) //masternode distributes matrix A to every single core
{
MPI_Send(&meta[0], 3, MPI_INT, i, TAG, MPI_COMM_WORLD);
printf("%d: debug04\n", myid);
MPI_Send(&MA[0], ASpalten*AZeilen, MPI_DOUBLE, i, TAG, MPI_COMM_WORLD);
printf("%d: debug05\n", myid);
MPI_Send(&MB[0], ASpalten*BSpalten, MPI_DOUBLE, i, TAG, MPI_COMM_WORLD);
printf("%d: debug06\n", myid);
}
}
else
{
MPI_Recv(meta, 3, MPI_INT, 0, TAG, MPI_COMM_WORLD, &stat);
printf("%d: debug01\n", myid);
ASpalten = meta[0];
AZeilen = meta[1];
BSpalten=meta[2];
printf("%d: debug02\n", myid);
MA = (double*)malloc(ASpalten*AZeilen*sizeof(double));
MB = (double*)malloc(ASpalten*BSpalten*sizeof(double));
MR = (double*)malloc(AZeilen*BSpalten*sizeof(double));
MPI_Recv(MA, ASpalten*AZeilen, MPI_DOUBLE, 0, TAG, MPI_COMM_WORLD, &stat);
MPI_Recv(MB, ASpalten*BSpalten, MPI_DOUBLE, 0, TAG, MPI_COMM_WORLD, &stat);
printf("%d: debug03\n", myid);
// printf("%d: %f\n", myid, *(MA + _index(1, 1, ASpalten))); //funktioniert
}
The Datatypes:
int ASpalten;
int AZeilen;
int BSpalten;
int *meta; //used to transfer meta data in 1 send
double *MA; //Matrix A
double *MB; //Matrix B
The program is supposed to multiply 2 matrices using MPI. My sample matrix proves that the code is likely valid and I also get this running for up to 130 * 90 matrices (maybe more maybe less), but anyway, as soon, as the number increases, I am getting a possible deadlock: the console prints out "debug4" and that's it. I would be very grateful, if anyone has a clue what is going wrong in my program. I already tried to use MPI_INTEGER instead of MPI_INT, but there is no difference. Any help would be appreciated. The output of the console when using very tiny matrices (PS, I already tried to execute my testcases in different orders too and modified existing ones):
Testcase1 MPI:
0: debug04
0: debug05
0: debug06
0: debug04
1: debug01
1: debug02
0: debug05
1: debug03
1: debugx1
0: debug06
0: debug04......

Answer 1

It seems that process 0 sends messages to proc 0 and proc 0 does receive them.
I modified to
for(i=1;i<numprocs;i++)
to remove the deadlock.
#include <stdio.h>
#include <time.h>
#include <stdlib.h>
#include <string.h>
#include "mpi.h"
int ASpalten;
int AZeilen;
int BSpalten;
int *meta; //used to transfer meta data in 1 send
double *MA; //Matrix A
double *MB; //Matrix B
double *MR; //Matrix B
void process(int myid, int numprocs){
int i,j, anzahl, rest;
int TAG=0;
MPI_Status stat;
meta=(int*)malloc(3*sizeof(int));
if(myid == 0)
{meta[0]=ASpalten;
meta[1]=AZeilen;
meta[2]=BSpalten;
for (i=1; i<numprocs; i++)//masternode distributes matrix A to every single core
{
MPI_Send(&meta[0], 3, MPI_INT, i, TAG, MPI_COMM_WORLD);
printf("%d: debug04\n", myid);
MPI_Send(&MA[0], ASpalten*AZeilen, MPI_DOUBLE, i, TAG, MPI_COMM_WORLD);
printf("%d: debug05\n", myid);
MPI_Send(&MB[0], ASpalten*BSpalten, MPI_DOUBLE, i, TAG, MPI_COMM_WORLD);
printf("%d: debug06\n", myid);
}
}
else
{
MPI_Recv(meta, 3, MPI_INT, 0, TAG, MPI_COMM_WORLD, &stat);
printf("%d: debug01\n", myid);
ASpalten=meta[0];
AZeilen=meta[1];
BSpalten=meta[2];
printf("%d: debug02\n", myid);
MA = (double*)malloc(ASpalten*AZeilen*sizeof(double));
MB = (double*)malloc(ASpalten*BSpalten*sizeof(double));
MR = (double*)malloc(AZeilen*BSpalten*sizeof(double));
MPI_Recv(MA, ASpalten*AZeilen, MPI_DOUBLE, 0, TAG, MPI_COMM_WORLD, &stat);
MPI_Recv(MB, ASpalten*BSpalten, MPI_DOUBLE, 0, TAG, MPI_COMM_WORLD, &stat);
printf("%d: debug03\n", myid);
// printf("%d: %f\n", myid, *(MA + _index(1, 1, ASpalten))); //funktioniert
}
}
int main(int argc,char *argv[])
{
int rank, size;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
ASpalten=130;
AZeilen=90;
BSpalten=200;
if(rank==0){
}
MA = (double*)malloc(ASpalten*AZeilen*sizeof(double));
MB = (double*)malloc(ASpalten*BSpalten*sizeof(double));
MR = (double*)malloc(AZeilen*BSpalten*sizeof(double));
process(rank,size);
MPI_Finalize();
return 0;
}
Bye,
Francis