Matrices multiply, Cannon algorithm implementation using MPI - process

Question

First of all, of course I saw similar questions and solutions, but my implementation is a little bit different.
The main problem is that, the my code works only for one process, but it doesn't work for more processes.
I don't know what is the cause of this... Probably in communication between processes but I can't figure it out ;/
#include <mpi.h>
#include <stdio.h>
#include <math.h>
#include <iostream>
using namespace std;
int main(int argc, char **argv)
{
int x = 0;
double kk;
int proces;
int numprocs;
int right_neigh, left_neigh, up_neigh, down_neigh;
int tag = 99;
static const int n = 6; //size of matrices
int psa[n][n]; //nxn
int psb[n][n];
int pra[n][n];
int prb[n][n];
int c[n][n];
for (int i = 0; i < n; i++) { //let's make fist matrix
for (int j = 0; j < n; j++) {
psa[i][j] = (int)rand() % 100 + 1;
psb[i][j] = (int)rand() % 100 + 1;
c[i][j] = 0;
}
}
for (int i = 0; i < n; i++) { //an the 2nd one
for (int j = 0; j < n; j++) {
pra[i][j] = psa[i][j];
prb[i][j] = psb[i][j];
}
}
MPI_Status statRecv[2];
MPI_Request reqSend[2], reqRecv[2];
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &proces);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
int PP = numprocs;
double np = numprocs;
kk = sqrt(np);
int k = (int)kk;
if (proces < k) // below neighbour set
{
left_neigh = (proces + k - 1) % k;
right_neigh = (proces + k + 1) % k;
up_neigh = ((k - 1)*k) + proces;
}
if (proces == k)
{
left_neigh = ((proces + k - 1) % k) + k;
right_neigh = ((proces + k + 1) % k) + k;
up_neigh = proces - k;
}
if (proces > k)
{
x = proces / k;
left_neigh = ((proces + k - 1) % k) + x * k;
right_neigh = ((proces + k + 1) % k) + x * k;
up_neigh = proces - k;
}
if (proces == 0 || (proces / k) < (k - 1))
{
down_neigh = proces + k;
}
if ((proces / k) == (k - 1))
{
down_neigh = proces - ((k - 1)*k);
}
x = 0;
for(int kk = 0; kk < PP; kk++) //algorithm
{
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
for (int k = 0; k < n / PP; k++)
{
c[i][j] += psa[i][k] * psb[k][j];
}
}
}
MPI_Irecv(pra, n*n / PP / PP,MPI_FLOAT,left_neigh, tag,MPI_COMM_WORLD, reqRecv);
MPI_Irecv(prb, n*n / PP / PP,MPI_FLOAT,down_neigh,tag,MPI_COMM_WORLD,&reqRecv[1]);
MPI_Isend(psa, n*n / PP / PP,MPI_FLOAT,right_neigh,tag,MPI_COMM_WORLD, reqSend);
MPI_Isend(psb, n*n / PP / PP,MPI_FLOAT,up_neigh,tag,MPI_COMM_WORLD,&reqSend[1]);
MPI_Wait(reqRecv, statRecv);
}
cout << "A" << endl; //show result
for (int i = 0; i < n; i++) {
for (int j = 0; j < n; j++) {
cout << pra[i][j] << " ";
}
cout << endl;
}
cout << "B" << endl;
for (int i = 0; i < n; i++) {
for (int j = 0; j < n; j++) {
cout << prb[i][j] << " ";
}
cout << endl;
}
cout << "C" << endl;
for (int i = 0; i < n; i++) {
for (int j = 0; j < n; j++) {
cout << c[i][j] << " ";
}
cout << endl;
}
MPI_Finalize();
return 0;
}

Answer 1

Ok I made it. Now everything is cool, my friend helped me out. But Admin please do not remove it, it can be helpful for someone.
#include <mpi.h>
#include <stdio.h>
#include <math.h>
#include <iostream> using namespace std; int main(int argc, char **argv) { int x = 0; double kk; int proces; int numprocs; int prawy_sasiad, lewy_sasiad, gorny_sasiad, dolny_sasiad; int tag = 99;
static const int n = 4; //rozmiar tablic
const int PP = 2; // pierwiastek z liczby procesow
int A[n][n] = {}, B[n][n] = {};
for (int i = 0; i < n; i++) {//inicjalizacja macierzy glownych
for (int j = 0; j < n; j++) {
A[i][j] = (int)rand() % 100 + 1;
B[i][j] = (int)rand() % 100 + 1;
}
}
/*
int val = 1;
for (int i = 0; i < n; i++) { //inicjalizacja macierzy glownych
for (int j = 0; j < n; j++) {
A[i][j] = val;
B[i][j] = val;
val++;
}
}
*/
MPI_Status statRecv2;
MPI_Request reqSend2, reqRecv2;
MPI_Status statRecv[2];
MPI_Request reqSend[2], reqRecv[2];
MPI_Init(0, 0);
MPI_Comm_rank(MPI_COMM_WORLD, &proces);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
int pra[n / PP][n / PP] = {}, psa[n / PP][n / PP] = {};// podmacierze
int prb[n / PP][n / PP] = {}, psb[n / PP][n / PP] = {};
//int C[n / PP][n / PP] = {};//wynikowa
int C[n][n] = {};//wynikowa
//cout << proces << endl;
for (int i = 0; i < n / PP; i++)//podzielenie macierzy glownej na podmacierze, kazdy proces otrzymuje inna podmacierz
{
for (int j = 0; j < n / PP; j++)
{
psa[i][j] = A[proces / PP*(n / PP) + i][proces%PP*(n / PP) + j];
psb[i][j] = B[proces / PP*(n / PP) + i][proces%PP*(n / PP) + j];
//cout << A[proces / PP*(n / PP) + i][proces%PP*(n / PP) + j] << " ";
}
//cout << endl;
}
double np = numprocs;
kk = sqrt(np);
int k = (int)kk;
if (proces < k) // ustawienie sasiadow
{
lewy_sasiad = (proces + k - 1) % k;
prawy_sasiad = (proces + k + 1) % k;
gorny_sasiad = ((k - 1)*k) + proces;
}
if (proces == k)
{
lewy_sasiad = ((proces + k - 1) % k) + k;
prawy_sasiad = ((proces + k + 1) % k) + k;
gorny_sasiad = proces - k;
}
if (proces > k)
{
x = proces / k;
lewy_sasiad = ((proces + k - 1) % k) + x * k;
prawy_sasiad = ((proces + k + 1) % k) + x * k;
gorny_sasiad = proces - k;
}
if (proces == 0 || (proces / k) < (k - 1))
{
dolny_sasiad = proces + k;
}
if ((proces / k) == (k - 1))
{
dolny_sasiad = proces - ((k - 1)*k);
}
x = 0;
int p = 0;
do{ //przesuniecia
if (p < proces / PP)// w wierszu
{
MPI_Irecv(pra, n*n / PP / PP, MPI_FLOAT, prawy_sasiad, tag, MPI_COMM_WORLD, &reqRecv2);
MPI_Isend(psa, n*n / PP / PP, MPI_FLOAT, lewy_sasiad, tag, MPI_COMM_WORLD, &reqSend2);
MPI_Wait(&reqRecv2, &statRecv2);
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
psa[i][j] = pra[i][j];
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
if (p < proces % PP)// i w kolumnie
{
MPI_Irecv(prb, n*n / PP / PP, MPI_FLOAT, dolny_sasiad, tag, MPI_COMM_WORLD, &reqRecv2);
MPI_Isend(psb, n*n / PP / PP, MPI_FLOAT, gorny_sasiad, tag, MPI_COMM_WORLD, &reqSend2);
MPI_Wait(&reqRecv2, &statRecv2);
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
psb[i][j] = prb[i][j];
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
p++;
} while (p < n);
//MPI_Barrier(MPI_COMM_WORLD);
for (int kkk = 0; kkk < PP; kkk++) //algorytm
{
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
for (int k = 0; k < n / PP; k++)
{
C[i][j] += psa[i][k] * psb[k][j];
}
}
}
MPI_Irecv(pra, n*n / PP / PP, MPI_FLOAT, prawy_sasiad, tag, MPI_COMM_WORLD, reqRecv);
MPI_Irecv(prb, n*n / PP / PP, MPI_FLOAT, dolny_sasiad, tag, MPI_COMM_WORLD, &reqRecv[1]);
MPI_Isend(psa, n*n / PP / PP, MPI_FLOAT, lewy_sasiad, tag, MPI_COMM_WORLD, reqSend);
MPI_Isend(psb, n*n / PP / PP, MPI_FLOAT, gorny_sasiad, tag, MPI_COMM_WORLD, &reqSend[1]);
MPI_Wait(reqRecv, statRecv);
MPI_Barrier(MPI_COMM_WORLD);
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
psa[i][j] = pra[i][j];
}
}
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
psb[i][j] = prb[i][j];
}
}
}
cout << "Proces: " << proces << " ";
for (int i = 0; i < n / PP; i++)
{
for (int j = 0; j < n / PP; j++)
{
cout << C[i][j] << " ";
}
}
MPI_Finalize();
return 0;
}