Hi All,
Here is my sample program. I am using channel sock of mpich-3.0.4.
I am running it as
> mpiexec -n 1 ./server.out
> mpiexec -n 1 ./client.out
Here my client program (client.c) hangs in MPI_Finalize.
There is an assert in the server.c where server exits.
There is no way to detect that in client.
Even if we detect that using some timeout strategy, the client hangs in
the finalize step.
Could you please suggest what is going wrong here or is this a bug in
sock channel?
Thanks,
Hirak
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
The MPICH team is pleased to announce the availability of a new preview release (mpich-3.2a2). This preview release adds several capabilities including support for the proposed MPI-3.1 standard (contains nonblocking collective I/O), full Fortran 2008 support (enabled by default), support for the Mellanox MXM interface for InfiniBand, support for the Mellanox HCOLL interface for collective communication, support for OFED InfiniBand for Xeon and Xeon Phi architectures, and significant improvements to the MPICH/portals4 implementation. These features represent a subset of those planned for the 3.2.x release series.
Complete list of features planned for mpich-3.2:
https://trac.mpich.org/projects/mpich/roadmap
MPICH website:
http://www.mpich.org
A list of changes in this release is included at the end of this email.
Regards,
The MPICH Team
===============================================================================
Changes in 3.2a2
===============================================================================
# Added support for proposed MPI-3.1 features including nonblocking
collective I/O, address manipulation routines, thread-safety for
MPI initialization and pre-init functionality. Proposed MPI-3.1
features are implemented as MPIX_ functions and the MPI version
number is still set to 3.0.
# Fortran 2008 bindings are enabled by default and fully supported.
# Added support for the Mellanox MXM InfiniBand interface. (thanks
to Mellanox for the code contribution).
# Added support for the Mellanox HCOLL interface for collectives.
(thanks to Mellanox for the code contribution).
# Added support for OFED IB on Xeon and Xeon Phi. (thanks to RIKEN
and University of Tokyo for the contribution).
# Significant stability improvements to the MPICH/portals4
implementation.
# Several other minor bug fixes, memory leak fixes, and code cleanup.
A full list of changes is available at the following link:
http://git.mpich.org/mpich.git/shortlog/v3.1.2..v3.2a2
A full list of bugs that have been fixed is available at the
following link:
https://trac.mpich.org/projects/mpich/query?status=closed&group=resolution&…
--
Pavan Balaji ✉️
http://www.mcs.anl.gov/~balaji
Try init tab_indice and tab_taille to NULL (it does not matter that it is
initialized for non-root, but the compiler likely isn't smart enough to
tell)
2014-11-15 12:38 GMT-08:00 Chafik sanaa <san.chafik(a)gmail.com>:
> when i running the following program, i have this error (error C4703:
> variable de pointeur locale potentiellement non initialisée 'tab_indice'
> utilisée and error C4703: variable de pointeur locale potentiellement non
> initialisée 'tab_taille' utilisée ):
> program:
> /////////////////////////////////////////////
> // MPI_Scatterv.c
> // test de la fonction MPI_Scatterv
> /////////////////////////////////////////////
>
> #include <stdio.h>
> #include <stdlib.h>
> #include <mpi.h>
>
> #define DIMENSION 10
>
> int main(int argc, char** argv)
> {
> int myrank, i, n_procs;
> int * sendbuf; //buffer à disperser
> int * tab_indice; //indice de début de chaque subdivision
> int * tab_taille; //nombre d'éléments à envoyer
> // pour chaque processus
> int * rbuf; //buffer de reception
> int taille; //taille de la partie reçue
>
> //Initialisation de MPI
> MPI_Init(&argc, &argv);
> MPI_Comm_size(MPI_COMM_WORLD, &n_procs);
> MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
>
> if (myrank == 0) {
> //allocation mémoire
> sendbuf = (int *)malloc(n_procs*DIMENSION*sizeof(int));
> tab_indice = (int *)malloc(n_procs*sizeof(int));
> tab_taille = (int *)malloc(n_procs*sizeof(int));
> //remplissage du buffer à disperser
> for (i = 0; i<n_procs*DIMENSION; i++) sendbuf[i] = i;
> //initialisation des subdivisions
> for (i = 0; i<n_procs; i++){
> //nombre d'éléments à envoyer
> tab_taille[i] = DIMENSION - i;
> //indice de début du processus i
> // = celui de i-1 + nombre d'éléments
> // envoyés par i-1
> if (i != 0) tab_indice[i] = tab_indice[i - 1] + tab_taille[i - 1];
> else tab_indice[i] = 0;
> }
> }
>
> //communication de la taille de la partie reçue à chaque processus
> MPI_Scatter(tab_taille, 1, MPI_INT, &taille, 1, MPI_INT, 0,
> MPI_COMM_WORLD);
> //allocation du buffer de reception
> rbuf = (int *)malloc(taille*sizeof(int));
>
> //dispersion
> MPI_Scatterv(&sendbuf, tab_taille, tab_indice, MPI_INT, rbuf,
> DIMENSION, MPI_INT, 0, MPI_COMM_WORLD);
>
> //affichage
> printf("processus %d : [ ", myrank);
> for (i = 0; i<taille; i++) printf("%d ", rbuf[i]);
> printf("]\n");
>
> //desallocation mémoire
> free(rbuf);
> if (myrank == 0) {
> free(sendbuf);
> free(tab_indice);
> free(tab_taille);
> }
>
> MPI_Finalize();
> exit(EXIT_SUCCESS);
> }
>
> _______________________________________________
> discuss mailing list discuss(a)mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
>
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
when i running the following program, i have this error (error C4703:
variable de pointeur locale potentiellement non initialisée 'tab_indice'
utilisée and error C4703: variable de pointeur locale potentiellement non
initialisée 'tab_taille' utilisée ):
program:
/////////////////////////////////////////////
// MPI_Scatterv.c
// test de la fonction MPI_Scatterv
/////////////////////////////////////////////
#include <stdio.h>
#include <stdlib.h>
#include <mpi.h>
#define DIMENSION 10
int main(int argc, char** argv)
{
int myrank, i, n_procs;
int * sendbuf; //buffer à disperser
int * tab_indice; //indice de début de chaque subdivision
int * tab_taille; //nombre d'éléments à envoyer
// pour chaque processus
int * rbuf; //buffer de reception
int taille; //taille de la partie reçue
//Initialisation de MPI
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &n_procs);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
if (myrank == 0) {
//allocation mémoire
sendbuf = (int *)malloc(n_procs*DIMENSION*sizeof(int));
tab_indice = (int *)malloc(n_procs*sizeof(int));
tab_taille = (int *)malloc(n_procs*sizeof(int));
//remplissage du buffer à disperser
for (i = 0; i<n_procs*DIMENSION; i++) sendbuf[i] = i;
//initialisation des subdivisions
for (i = 0; i<n_procs; i++){
//nombre d'éléments à envoyer
tab_taille[i] = DIMENSION - i;
//indice de début du processus i
// = celui de i-1 + nombre d'éléments
// envoyés par i-1
if (i != 0) tab_indice[i] = tab_indice[i - 1] + tab_taille[i - 1];
else tab_indice[i] = 0;
}
}
//communication de la taille de la partie reçue à chaque processus
MPI_Scatter(tab_taille, 1, MPI_INT, &taille, 1, MPI_INT, 0, MPI_COMM_WORLD);
//allocation du buffer de reception
rbuf = (int *)malloc(taille*sizeof(int));
//dispersion
MPI_Scatterv(&sendbuf, tab_taille, tab_indice, MPI_INT, rbuf,
DIMENSION, MPI_INT, 0, MPI_COMM_WORLD);
//affichage
printf("processus %d : [ ", myrank);
for (i = 0; i<taille; i++) printf("%d ", rbuf[i]);
printf("]\n");
//desallocation mémoire
free(rbuf);
if (myrank == 0) {
free(sendbuf);
free(tab_indice);
free(tab_taille);
}
MPI_Finalize();
exit(EXIT_SUCCESS);
}
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
FYI, this is what we heard back from the Mellanox folks:
I think, issue could be because of older MXM (part of MOFED) being used. We can ask him to try latest MXM from HPCX (http://bgate.mellanox.com/products/hpcx)
— Pavan
> On Nov 14, 2014, at 5:35 AM, Balaji, Pavan <balaji(a)anl.gov> wrote:
>
>
>> On Nov 14, 2014, at 2:55 AM, Victor Vysotskiy <victor.vysotskiy(a)teokem.lu.se> wrote:
>> finally I got reply from admins about the status of MXM on our cluster:
>>
>>> MXM is part of the Mellanox OFED stack, which is not used on Triolith.
>>> Please note that NSC recommends using Intel MPI when possible.
>
> You can install MXM on the vanilla OFED, even without Mellanox OFED. I don’t know why it’s failing to detect the network adapter, though. I’ve pinged the Mellanox folks to see if they can help.
>
>> Of course, the Intel MPI is installed and available. However, since Intel MPI is based on MPICH, it suffers from the same problems as MPICH does.
>
> Right. Intel folks occassionally do release updates. For the next update, we’ll ask them to make sure to pick up this patch. But it’s going to be a few months out.
>
> Regards,
>
> — Pavan
>
> --
> Pavan Balaji ✉️
> http://www.mcs.anl.gov/~balaji
>
> _______________________________________________
> discuss mailing list discuss(a)mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
--
Pavan Balaji ✉️
http://www.mcs.anl.gov/~balaji
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
> On Nov 14, 2014, at 2:55 AM, Victor Vysotskiy <victor.vysotskiy(a)teokem.lu.se> wrote:
> finally I got reply from admins about the status of MXM on our cluster:
>
> >MXM is part of the Mellanox OFED stack, which is not used on Triolith.
> >Please note that NSC recommends using Intel MPI when possible.
You can install MXM on the vanilla OFED, even without Mellanox OFED. I don’t know why it’s failing to detect the network adapter, though. I’ve pinged the Mellanox folks to see if they can help.
> Of course, the Intel MPI is installed and available. However, since Intel MPI is based on MPICH, it suffers from the same problems as MPICH does.
Right. Intel folks occassionally do release updates. For the next update, we’ll ask them to make sure to pick up this patch. But it’s going to be a few months out.
Regards,
— Pavan
--
Pavan Balaji ✉️
http://www.mcs.anl.gov/~balaji
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
Hi Pavan,
finally I got reply from admins about the status of MXM on our cluster:
>MXM is part of the Mellanox OFED stack, which is not used on Triolith.
>Please note that NSC recommends using Intel MPI when possible.
Of course, the Intel MPI is installed and available. However, since Intel MPI is based on MPICH, it suffers from the same problems as MPICH does.
Best,
Victor.
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
-enable-x works.
Thanks Pavan
________________________________________
From: Balaji, Pavan [balaji(a)anl.gov]
Sent: Tuesday, November 11, 2014 6:44 PM
To: discuss(a)mpich.org
Subject: Re: [mpich-discuss] mpiexec and DISPLAY for remote hosts
Did you try passing -enable-x to mpiexec?
— Pavan
> On Nov 11, 2014, at 5:00 PM, Isaila, Florin D. <fisaila(a)mcs.anl.gov> wrote:
>
> Hi,
>
> I am having troubles running mpiexec (from MPICH 3.1) on remote hosts with xterm (for debugging purposes) from an Ubuntu box.
>
> The following command works perfectly in local:
> mpiexec -n 1 xterm
>
> However, if I use a machine file:
> mpiexec -f macfile -n 1 xterm
> xterm Xt error: Can't open display:
> xterm: DISPLAY is not set
>
> ===================================================================================
> = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
> = PID 19225 RUNNING AT thwomp.mcs.anl.gov
> = EXIT CODE: 1
> = CLEANING UP REMAINING PROCESSES
> = YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
> ===================================================================================
>
> X forwarding is activated:
> fisaila@howard:bin$ grep X11Forwarding /etc/ssh/sshd_config
> X11Forwarding yes
>
> The following command works:
> fisaila@howard:bin$ ssh thwomp.mcs.anl.gov xterm
>
> Any suggestion how to solve this?
>
> Thanks
> Florin
>
>
> _______________________________________________
> discuss mailing list discuss(a)mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
--
Pavan Balaji ✉️
http://www.mcs.anl.gov/~balaji
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
_______________________________________________
discuss mailing list discuss(a)mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
