search

FreeFem / FreeFem-sources

FreeFEM source code
https://freefem.org/
Other
794 stars 192 forks source link

MUMPS error #51

Closed sgarnotel closed 5 years ago

sgarnotel commented 6 years ago

OS: ArchLinux FreeFem++ version: develop

The MUMPS/MUMPS_seq plugin does not work anymore. The compilation process complete correctly, but during the check, LapMUMPS_seq.edp fails

For example with this file mumps.edp.txt, the result is:

FreeFem++ mumps.edp
 Matrix morse type:6
 BuildSolverMUMPSseq<double>
[---:09016] *** An error occurred in MPI_Comm_rank
[---:09016] *** reported by process [3860070401,0]
[---:09016] *** on communicator MPI_COMM_WORLD
[---:09016] *** MPI_ERR_COMM: invalid communicator
[---:09016] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[---:09016] ***    and potentially your MPI job)

Any idea about this bug ?

prj- commented 5 years ago

Do you still have this error if you comment out this line?

sgarnotel commented 5 years ago

Yes I still have this error.

By commenting the line, that works well

prj- commented 5 years ago

Would that be an appropriate fix for you? There is really no need to call MPI_Comm_rank and I'm guessing there is something wrong in the mpiseq library. I'll make the commit once you give me the green light.

sgarnotel commented 5 years ago

That fix the problem when I use load "MUMPS_seq" But when I use load "MUMPS" in a sequential script, MUMPS_seq is not loaded by defaut and the error is still here

prj- commented 5 years ago

Have you made the same fix in MUMPS.cpp?

sgarnotel commented 5 years ago

Ok, that fix all. I have not realized that SolveMUMPS_seq is equally defined in MUMPS.cpp.

Thanks a lot, all work fine now

frederichecht commented 5 years ago

remark in plugin/seq dir MUMPS.cpp and MUMPS_seq.cpp is the same code Best Regards,

Frédéric Hecht.

Laboratoire Jacques-Louis Lions, UPMC Sorbonne Université BC187, 4 Place Jussieu, 75252 PARIS cedex 05, France Campus Jussieu, Barre 15-25, 3 etage Bureau 307 Projet Alpines , Inria de Paris, 2 rue Simone Iff Voie DQ12 75012 Paris tel: +33 1 44274411, mob: +33 6 62198986, fax: +33 1 44277200 frederic.hecht@sorbonne-universite.fr https://www.ljll.math.upmc.fr/hecht software: FreeFem++ web site: http://www.freefem.org/ff++

Le 20 mars 2019 à 16:18, Simon Garnotel notifications@github.com a écrit :

Ok, that fix all. I have not realized that SolveMUMPS_seq is equally defined in MUMPS.cpp.

Thanks a lot, all work fine now

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/FreeFem/FreeFem-sources/issues/51#issuecomment-474879436, or mute the thread https://github.com/notifications/unsubscribe-auth/AFb2jPHiWN2UU8K-HNtSbBUW4HJ4WtNhks5vYlEtgaJpZM4XcZEm.

maximefiore1 commented 7 months ago

Hello, I apologize for sending this message but I also used the command load "MUMPS_seq" in Freefem++ and when I defined a sparsematrix (NSMAT=LNS(Uvvp,Uvvp,solver=sparsesolver);) (I think the error comes at this line), I get the same type of error as the one described previously:

[visu01:44612] An error occurred in MPI_Bcast [visu01:44612] reported by process [3541041153,0] [visu01:44612] on communicator MPI_COMM_WORLD [visu01:44612] MPI_ERR_COMM: invalid communicator [visu01:44612] MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, [visu01:44612] and potentially your MPI job)

I am using a 4.5 version of Freefem++, would you know if I make something wrong with the code.

I thank you,

Maxime Fiore

frederichecht commented 7 months ago

REmark,

MUMPS is a parallel solver work with MPII; if you are on sequential code (no mpi) You say Freefem++ but I think you use FreeFem++-mpi and the plugging "MUMPS_seq"

1) try with FreeFem++
and please send all the output.

maximefiore1 commented 7 months ago

Hello,

I thank you for your answer, I have tried with the different executable available, this is the output obtained:

with FreeFem++ executable:

[pando01:81420] 1 more process has sent help message help-mpi-btl-openib.txt / no device params found [pando01:81420] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages [pando01:81419] An error occurred in MPI_Bcast [pando01:81419] reported by process [2850029569,0] [pando01:81419] on communicator MPI_COMM_WORLD [pando01:81419] MPI_ERR_COMM: invalid communicator [pando01:81419] MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, [pando01:81419] and potentially your MPI job)

with FreeFem++-nw executable:

[pando01:82088] 1 more process has sent help message help-mpi-btl-openib.txt / no device params found [pando01:82088] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages [pando01:82087] An error occurred in MPI_Bcast [pando01:82087] reported by process [3611557889,0] [pando01:82087] on communicator MPI_COMM_WORLD [pando01:82087] MPI_ERR_COMM: invalid communicator [pando01:82087] MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, [pando01:82087] and potentially your MPI job)

with Freefem++-mpi executable:

[pando01:82516] An error occurred in MPI_Init [pando01:82516] reported by process [3585671169,0] [pando01:82516] on a NULL communicator [pando01:82516] Unknown error [pando01:82516] MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, [pando01:82516] and potentially your MPI job) [pando01:82517] 1 more process has sent help message help-mpi-btl-openib.txt / no device params found [pando01:82517] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages

I can share also the code if needed.

I thank you for your help.

Maxime Fiore

maximefiore1 commented 7 months ago

Hello,

I have found a solution normally, the script I was using was for older versions of Freefem++ I think and the beginning of the script started with: load "MUMPS" load "MUMPS_seq"

In the version of Freefem++ I am currently using, by simply commenting these lines, the code works perfectly.

I apologize for having disturbed you.

Best Regards,