[OMPI users] OpenMPI 1.8.8: Segfault when using non-blocking reduce operations with a user-defined operator
Hi, OpenMPI 1.8.8 segfaults when I try to perform a non-blocking reduce operation with a user-defined reduce operator. The problem has already been addressed.here: http://www.open-mpi.org/community/lists/devel/2014/04/14586.php Please fix this issue since it is critical for me! At least you should abort with some more user-friendly output. Best Georg
Re: [OMPI users] OpenMPI 1.8.8: Segfault when using non-blocking reduce operations with a user-defined operator
Nathan, thanks for your rapid response. Do you consider to release 1.8.9? Actually, there is a bug tracking category for that version number. If so, please backport the fix. Best Georg Am 02.10.2015 um 17:59 schrieb Nathan Hjelm: Working on a fix now. Will be in master today then will move it to 1.10.1 and 2.x. -Nathan On Fri, Oct 02, 2015 at 02:24:29PM +0200, Georg Geiser wrote: Hi, OpenMPI 1.8.8 segfaults when I try to perform a non-blocking reduce operation with a user-defined reduce operator. The problem has already been addressed.here: http://www.open-mpi.org/community/lists/devel/2014/04/14586.php Please fix this issue since it is critical for me! At least you should abort with some more user-friendly output. Best Georg ___ users mailing list us...@open-mpi.org Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users Link to this post: http://www.open-mpi.org/community/lists/users/2015/10/27769.php ___ users mailing list us...@open-mpi.org Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users Link to this post: http://www.open-mpi.org/community/lists/users/2015/10/27772.php -- Dr.-Ing. Georg Geiser Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR) Institut für Antriebstechnik | Numerische Methoden | Linder Höhe | 51147 Köln Telefon +49 2203 601-3294 | E-Mail georg.gei...@dlr.de
Re: [OMPI users] MPI_GATHERV error
Diego, try to decrease your displacements by one. Best Georg Am 14.10.2015 um 15:51 schrieb Diego Avesani: dear all, I have some problem with MPI_GATHERV. In my code I generate a complex number DO ij=iNS,iNE X11(ij) = cmplx(1.,0.) ENDDO where iNS,INE change according to the CPU rank, in may case cpu 0 1 10050 cpu 1 10051 20100 cpu 2 20101 30150 cpu 3 30151 40401 according to that I have nNodeGLOBAL = 10050 10050 10050 10251 IDNodeStartGLOBAL =1 10051 20101 30151 After that I apply a MPI_GATHERV as CALL MPI_GATHERV(X11(*iNS:iNE*),MPIdata%nNodes, mpi_double_cpmplex, *PHIH*, nNodeGLOBAL,IDNodeStartGLOBAL, mpi_double_cpmplex, 0, MPI_COMM_WORLD, Err) but when I plot REAL(PHIH(1)), I get 0 I use fortran, do you think that there is an error in nNodeGLOBAL? do I have to allocate the vector nNodeGLOBAL and IDNodeStartGLOBAL satrting from 0 according to the name of the rank? really really thanks a lot Diego ___ users mailing list us...@open-mpi.org Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users Link to this post: http://www.open-mpi.org/community/lists/users/2015/10/27862.php
Re: [OMPI users] MPI_GATHERV error
Hi Diego, displacements start at 0, so 0 means no displacement, i.e., the corresponding data starts at the first entry (by default 1 in Fortran) of the receive field. A displacement of 1 would point to entry number 2. Btw., MPI does not care how you allocate your fields in Fortran. E.g., if you do an allocate(A(-1:10)) the lower and upper bounds are only known to Fortran. You just pass an pointer to this field to MPI and declare that it contains size(A)=12 entries. All displacements are relative to the first entry of that field, so a displacement of 0 points to A(-1), a displacement of 1 to A(0) and so on. Best Georg Am 14.10.2015 um 16:50 schrieb Diego Avesani: dear George, dear all, thanks for the suggestions, it works now. I have just put: IDNodeStartGLOBAL = IDNodeStartGLOBAL -1 but I do not understand why. Can you explain it to me? Thanks Diego On 14 October 2015 at 16:02, Georg Geiser <mailto:georg.gei...@dlr.de>> wrote: Diego, try to decrease your displacements by one. Best Georg Am 14.10.2015 um 15:51 schrieb Diego Avesani: dear all, I have some problem with MPI_GATHERV. In my code I generate a complex number DO ij=iNS,iNE X11(ij) = cmplx(1.,0.) ENDDO where iNS,INE change according to the CPU rank, in may case cpu 0 1 10050 cpu 1 10051 20100 cpu 2 20101 30150 cpu 3 30151 40401 according to that I have nNodeGLOBAL = 10050 10050 10050 10251 IDNodeStartGLOBAL =1 10051 20101 30151 After that I apply a MPI_GATHERV as CALL MPI_GATHERV(X11(*iNS:iNE*),MPIdata%nNodes, mpi_double_cpmplex, *PHIH*, nNodeGLOBAL,IDNodeStartGLOBAL, mpi_double_cpmplex, 0, MPI_COMM_WORLD, Err) but when I plot REAL(PHIH(1)), I get 0 I use fortran, do you think that there is an error in nNodeGLOBAL? do I have to allocate the vector nNodeGLOBAL and IDNodeStartGLOBAL satrting from 0 according to the name of the rank? really really thanks a lot Diego ___ users mailing list us...@open-mpi.org <mailto:us...@open-mpi.org> Subscription:http://www.open-mpi.org/mailman/listinfo.cgi/users Link to this post:http://www.open-mpi.org/community/lists/users/2015/10/27862.php ___ users mailing list us...@open-mpi.org <mailto:us...@open-mpi.org> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users Link to this post: http://www.open-mpi.org/community/lists/users/2015/10/27863.php ___ users mailing list us...@open-mpi.org Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users Link to this post: http://www.open-mpi.org/community/lists/users/2015/10/27864.php
