problem in MPI

[joud]

Hi

I'm beginner in Ubuntu and I have installed Openmpi & Mpich without problem

but when I try to execute some examples, it don't work....why???

just this example works


#include <stdio.h>
#include <mpi.h>

int main(int argc, char** argv) {
int myrank, nprocs;

MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);

printf("Hello from processor %d of %d\n", myrank, nprocs);

MPI_Finalize();
return 0;
}
I want to do a difficult code, for example: using two processors,the first adds two numbers and the second do the multiplication
please, help me

thanks in advance

[talonmies]

You are probably unwittingly mixing openMPI and mpich2 libraries and binaries. Try only having one or the other installed and see what happens.

[joud]

Thanks talonmies

I try to uninstall Mpich and leave only OpenMpi

but the problem does not resolve

[joud]

any reply may help me!!!!

[SevenMachines]

As a test, how are you trying to run your example? For example, here...


$ sudo apt-get install openmpi-bin libopenmpi-dev
$ mpicc -o example example.c
$ mpirun -np 4 ./example
Hello from processor 0 of 4
Hello from processor 1 of 4
Hello from processor 3 of 4
Hello from processor 2 of 4So it certainly runs, whether it 'works' i couldnt say :)

[EDIT] Sorry, ignore that, i didnt read the post correctly.. its friday :) If you post some example you're having trouble with though?

[joud]

ok no problem :)

Yes I try the code that you post and all run , but only with a simple example
and with this example , nooooo


#include<stdio.h>
#include<stdlib.h>
#include"mpi.h"
#define MCW MPI_COMM_WORLD
int * produitMatrice(int * a,int* b,int m,int l,int k){
int i,j,x;int *res; res=(int*)malloc(m*k*sizeof(int));
for(i=0;i<m;i++)
for(j=0;j<m;j++)
for(*(res+(i*k)+j)=0,x=0;x<l;x++){
*(res+(i*k)+j)+=*(a+(i*l)+x) * (*(b+(x*k)+j));
}
return res;
}
int main(int argc,char ** argv){
int me,size,m,l,k,x,*matA,*matB,i,j,*ligne,*colonne,*m atC;
MPI_Status st;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MCW,&me);
MPI_Comm_size(MCW,&size);
if(!me){
printf("Donnez le nombre des lignes de A: ");scanf("%d",&m);
printf("Donnez le nombre des colonnes de A: ");scanf("%d",&l);
printf("Donnez le nombre des colonnes de B: ");scanf("%d",&k);
/// lecture des matrices A et B
matA=(int*)malloc(m*l*sizeof(int));
matB=(int*)malloc(l*k*sizeof(int));
for(i=0;i<m;i++)for(j=0;j<l;j++) *(matA+i*l+j)=i+j+2;
for(i=0;i<l;i++)for(j=0;j<k;j++) *(matB+i*k+j)=i*j+2;
for(printf("\n\n"),i=0;i<m;printf("\n"),i++)for(j=0;j<l;j++)printf("%d\t",*(matA+i*l+j));
for(printf("\n\n"),i=0;i<l;printf("\n"),i++)for(j=0;j<k;j++)printf("%d\t",*(matB+i*k+j));
}
MPI_Bcast(&m,1,MPI_INT,0,MCW);
MPI_Bcast(&l,1,MPI_INT,0,MCW);
MPI_Bcast(&k,1,MPI_INT,0,MCW);
matC = (int*)malloc(m*k*sizeof(int));
ligne = (int*)malloc(m/size*l*sizeof(int));
colonne = (int*)malloc(l*k/size*sizeof(int));
/// envoi de matrice A
if(!me){
for(i=0;i<m/size;i++)for(x=0;x<l;x++) *(ligne+(i*l)+x)=*(matA+(i*l)+x);
printf("matrice A vers processus[%d]\n",me);for(i=0;i<m/size;printf("\n"),i++)for(j=0;j<l;j++) printf("%d\t",*(matA+(i*l)+j));printf("\n");

for(i=0;i<l;i++)for(x=0;x<k/size;x++) {*(colonne+(i*k/size)+x)=*(matB+(i*m)+x);}
printf("matrice B vers processus[%d]\n",me);
for(i=0;i<l;printf("\n"),i++)for(j=0;j<k/size;j++) printf("%d\t",*(colonne+(i*k/size)+j));printf("\n");

for(i=1;i<size;i++){
MPI_Send(matA+(i*l*m/size),l*m/size,MPI_INT,i,i+size,MCW);
printf("matrice A vers processus[%d]\n",i);
for(j=0;j<m/size;printf("\n"),j++)for(x=0;x<l;x++) printf("%d\t",*(matA+(i*m/size*l+(l*j))+x));printf("\n");
// créer portion
int *portion; portion =(int*)malloc(l*k/size*sizeof(int));
for(j=0;j<l;j++)for(x=0;x<k/size;x++) *(portion+(j*k/size)+x)=*(matB+(i*k/size+(j*m))+x);
printf("matrice B vers processus[%d]\n",i);
for(j=0;j<l;printf("\n"),j++)for(x=0;x<k/size;x++) printf("%d\t",*(portion+(j*k/size)+x));printf("\n");
// fin
MPI_Send(portion,l*k/size,MPI_INT,i,i+size,MCW);
}
}
else{
MPI_Recv(ligne,l*m/size,MPI_INT,0,me+size,MCW,&st);
MPI_Recv(colonne,l*k/size,MPI_INT,0,me+size,MCW,&st);
}
printf("mat A processus[%d]\n",me);for(j=0;j<m/size;printf("\n"),j++)for(i=0;i<l;i++) printf("%d\t",*(ligne+(j*l)+i));printf("\n");
printf("mat B processus[%d]\n",me);for(j=0;j<l;printf("\n"),j++)for(i=0;i<k/size;i++) printf("%d\t",*(colonne+(j*k/size)+i));printf("\n");

int * res;
res=(int*)malloc((m/size)*(k/size)*sizeof(int));
res=produitMatrice(ligne,colonne,m/size,l,k/size);printf("produit \n");
/// MAJ matC
for(i=0;i<m/size;i++)for(j=0;j<k/size;j++) *(c+(i*k/size)+j)=*(res)
for(i=1;i<size-1;i++){
MPI_Send(colonne,l*k/size,MPI_INT,(me+1)%size,me+size,MCW);
MPI_Recv(colonne,l*k/size,MPI_INT,(me-1)%size,me+size,MCW,&st);
int * res;
res=(int*)malloc((m/size)*(k/size)*sizeof(int));
res=produitMatrice(ligne,colonne,m/size,l,k/size);printf("produit \n");
}


if(!me){
int * res;res=(int*)malloc((m/size)*(k/size)*sizeof(int));
res=produitMatrice(ligne,colonne,m/size,l,k/size);printf("produit \n");
for(i=0;i<m/size;printf("\n"),i++)for(j=0;j<k/size;j++) printf("%d\t",*(res+(i*k/size)+j));printf("\n");
}
MPI_Finalize();
}

the error is:

.........@ubuntu:~$ mpicc mat.c -o mat
mat.c: In function ‘main’:
mat.c:71: error: ‘c’ undeclared (first use in this function)
mat.c:71: error: (Each undeclared identifier is reported only once
mat.c:71: error: for each function it appears in.)
mat.c:72: error: expected ‘;’ before ‘for’


I don't understand

[joud]

I think that the problem is in library stdlib.h , it doesn't work with me or always in ubuntu?????

[talonmies]

So despite what you said earlier in the thread, the problem is not that you can't run MPI programs at all. It is that you can't compile a specific example? How did you possibly imagine someone could help you before you had actually posted the code?

The compiler is telling you there is are two errors with this line:



for(i=0;i<m/size;i++)for(j=0;j<k/size;j++) *(c+(i*k/size)+j)=*(res)

and there are. Neither has anything to do with MPI. If you cannot understand why that line is wrong, even after the compiler gives an error which says exactly what is wrong, perhaps you should spend some time learning basic C syntax before you go worrying about parallel programming APIs like MPI.

[joud]

Sorry for being late
Thanks talonmies for your reply... I try to correct the error and compile it without problem

but I don't know how to use MPI
I compile the code and no problem, what can I do to execute the code using more processus in the same machine??

[joud]

............@ubuntu:~$ mpicc example.c -o example

I'm trying to compile the example by using this command but how can I execute the example ?

[joud]

Sorry , I'm late, I had problem in my OS Ubuntu

but i'm return now :)

Please .... help me

[Arndt]

............@ubuntu:~$ mpicc example.c -o example

I'm trying to compile the example by using this command but how can I execute the example ?

I don't know mpicc, but if it works like gcc, it will produce an executable file "example", which you can run by issuing the command "./example".

[SevenMachines]

you need to run in the openmpi environment rather than from the directly. As a simple example, as mentioned in post #5, could be run with something like

$ mpirun -np 4 ./exampleThis says, 'Run the example binary in openmpi using 4 processes'.

[joud]

Thanks ,
I know that it must use this command mpirun -np 4 ./example

but.. how can I make process n°1 do some instructions, process n°2 do auther instructions...etc

for exmaple, when I want to calculate the multiplication of two matrices A*B=C

what can I do???

[SevenMachines]

theres an example of matrix multiplication here,
http://www.pdc.kth.se/training/Tutor/MPI/Templates/mm/mm.c
but obviously you would need to alter it to make it useful. More importantly you need to get stuck into a good tutorial/book on mpi, how to split a job onto different processes is quite fundamental to mpi.

[joud]

Thanks SevenMachines for your aid

mmmmmm can you give me tutorials on mpi, in english or french language as you like

I have some tutorials but may your choice is more comprehensin and clear :)

speacially how to split a job onto different processes

[SevenMachines]

I doubt my choices would be any more comprehensive or clear than yours to be honest, i had a need to learn some mpi a while ago but nothing too advanced. I think I read 'A Users Guide to MPI (http://www.wellesley.edu/CS/courses/CS331/notes/mpi.guide.pdf)', which is free and seemed ok. The author has a new book which isnt free, but the source code for the examples is, and is on his homepage http://www.cs.usfca.edu/~peter/ppmpi/ (http://www.cs.usfca.edu/%7Epeter/ppmpi/) so it might be worth looking through some of the examples. I'll try to think of more, its been a while though.

[joud]

Thank you very much SevenMachines for all your efforts to help me

are you have or know some examples in mpi which for sorting arrays of course in mpi?

just a simple example..