September 2015 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] Open solaris 64bit printf bug?
by Satish Balay 03 Sep '15

03 Sep '15

These compilers are pretty old - we just use solaris for the 'diversity in testing' feature.. $ cc -V cc: Sun C 5.9 SunOS_i386 Patch 124868-10 2009/04/30 There is a slightly newer version of compiler - I don't remember why I don't use it [I suspect it had issues] $ /opt/solarisstudio12.3/bin/cc -V cc: Sun C 5.12 SunOS_i386 2011/11/16 I tried switching to this version to test - but the hang persisted.. Satish On Thu, 3 Sep 2015, Barry Smith wrote: > > > On Sep 3, 2015, at 10:52 PM, Satish Balay <balay(a)mcs.anl.gov> wrote: > > > > I couldn't get this build [with 64-bit-indices] compiled in debug > > mode. The compiler just hangs with a couple of source files. > > > > CC arch-opensolaris-misc/obj/src/mat/impls/aij/seq/inode.o > > CC arch-opensolaris-misc/obj/src/dm/impls/da/fdda.o > > > > I had to manaully compile these sources with '-O' - instead of '-g' to > > get the build going :( > > I noticed that same hang but wasn't bright enough to try removing the -g on them. > > We use to have trouble with compilers with optimization turned on, now we have trouble with optimization turned off! > > Barry > > > > > Satish > > > > > > On Thu, 3 Sep 2015, Barry Smith wrote: > > > >> > >> And just to annoy me I cannot get -start_in_debugger to work on that machine. Valgrind has no complaints on a Mac with the examples. > >> > >> I've found some miss-use of PetscViewerASCIIPushSynchronized() and %d in parts of plex, I don't see how they could cause this problem but I am fixing them and will push that. Maybe we'll get lucky. > >> > >> Barry > >> > >>> On Sep 3, 2015, at 10:06 AM, Tobin Isaac <tisaac(a)ices.utexas.edu> wrote: > >>> > >>> > >>> Here's a partial diff from http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2015/09/03/examples_ne…: > >>> > >>> 232c232 > >>> < field 0 with 2 components > >>> --- > >>>> field 0 with 0 components > >>> 234,242c234,242 > >>> < ( 2) dim 2 offset 0 0.5 0 > >>> < ( 3) dim 2 offset 2 1 0 > >>> < ( 4) dim 2 offset 4 0 0.5 > >>> < ( 5) dim 2 offset 6 0.5 0.5 > >>> < ( 6) dim 2 offset 8 1 0.5 > >>> < ( 7) dim 2 offset 10 0 1 > >>> < ( 8) dim 2 offset 12 0.5 1 > >>> < ( 9) dim 2 offset 14 0.5 0.25 > >>> < ( 10) dim 2 offset 16 0.25 0.5 > >>> --- > >>>> ( 2) dim 0 offset 2 0.5 0 > >>>> ( 3) dim 0 offset 2 1 0 > >>>> ( 4) dim 0 offset 2 0 0.5 > >>>> ( 5) dim 0 offset 2 0.5 0.5 > >>>> ( 6) dim 0 offset 2 1 0.5 > >>>> ( 7) dim 0 offset 2 0 1 > >>>> ( 8) dim 0 offset 2 0.5 1 > >>>> ( 9) dim 0 offset 2 0.5 0.25 > >>>> ( 10) dim 0 offset 2 0.25 0.5 > >>> > >>> The numbers that printf is making 0 (or 2) clearly aren't, because > >>> they are used to iterate over a section and display its contents. But > >>> other PetscInt's are printed fine elsewhere. Weird. > >>> > >>> Toby > >>> > >> > >> > > > >

1 0

Re: [petsc-dev] Open solaris 64bit printf bug?
by Barry Smith 03 Sep '15

03 Sep '15

> On Sep 3, 2015, at 10:52 PM, Satish Balay <balay(a)mcs.anl.gov> wrote: > > I couldn't get this build [with 64-bit-indices] compiled in debug > mode. The compiler just hangs with a couple of source files. > > CC arch-opensolaris-misc/obj/src/mat/impls/aij/seq/inode.o > CC arch-opensolaris-misc/obj/src/dm/impls/da/fdda.o > > I had to manaully compile these sources with '-O' - instead of '-g' to > get the build going :( I noticed that same hang but wasn't bright enough to try removing the -g on them. We use to have trouble with compilers with optimization turned on, now we have trouble with optimization turned off! Barry > > Satish > > > On Thu, 3 Sep 2015, Barry Smith wrote: > >> >> And just to annoy me I cannot get -start_in_debugger to work on that machine. Valgrind has no complaints on a Mac with the examples. >> >> I've found some miss-use of PetscViewerASCIIPushSynchronized() and %d in parts of plex, I don't see how they could cause this problem but I am fixing them and will push that. Maybe we'll get lucky. >> >> Barry >> >>> On Sep 3, 2015, at 10:06 AM, Tobin Isaac <tisaac(a)ices.utexas.edu> wrote: >>> >>> >>> Here's a partial diff from http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2015/09/03/examples_ne…: >>> >>> 232c232 >>> < field 0 with 2 components >>> --- >>>> field 0 with 0 components >>> 234,242c234,242 >>> < ( 2) dim 2 offset 0 0.5 0 >>> < ( 3) dim 2 offset 2 1 0 >>> < ( 4) dim 2 offset 4 0 0.5 >>> < ( 5) dim 2 offset 6 0.5 0.5 >>> < ( 6) dim 2 offset 8 1 0.5 >>> < ( 7) dim 2 offset 10 0 1 >>> < ( 8) dim 2 offset 12 0.5 1 >>> < ( 9) dim 2 offset 14 0.5 0.25 >>> < ( 10) dim 2 offset 16 0.25 0.5 >>> --- >>>> ( 2) dim 0 offset 2 0.5 0 >>>> ( 3) dim 0 offset 2 1 0 >>>> ( 4) dim 0 offset 2 0 0.5 >>>> ( 5) dim 0 offset 2 0.5 0.5 >>>> ( 6) dim 0 offset 2 1 0.5 >>>> ( 7) dim 0 offset 2 0 1 >>>> ( 8) dim 0 offset 2 0.5 1 >>>> ( 9) dim 0 offset 2 0.5 0.25 >>>> ( 10) dim 0 offset 2 0.25 0.5 >>> >>> The numbers that printf is making 0 (or 2) clearly aren't, because >>> they are used to iterate over a section and display its contents. But >>> other PetscInt's are printed fine elsewhere. Weird. >>> >>> Toby >>> >> >> >

1 0

Re: [petsc-dev] Open solaris 64bit printf bug?
by Satish Balay 03 Sep '15

03 Sep '15

I couldn't get this build [with 64-bit-indices] compiled in debug mode. The compiler just hangs with a couple of source files. CC arch-opensolaris-misc/obj/src/mat/impls/aij/seq/inode.o CC arch-opensolaris-misc/obj/src/dm/impls/da/fdda.o I had to manaully compile these sources with '-O' - instead of '-g' to get the build going :( Satish On Thu, 3 Sep 2015, Barry Smith wrote: > > And just to annoy me I cannot get -start_in_debugger to work on that machine. Valgrind has no complaints on a Mac with the examples. > > I've found some miss-use of PetscViewerASCIIPushSynchronized() and %d in parts of plex, I don't see how they could cause this problem but I am fixing them and will push that. Maybe we'll get lucky. > > Barry > > > On Sep 3, 2015, at 10:06 AM, Tobin Isaac <tisaac(a)ices.utexas.edu> wrote: > > > > > > Here's a partial diff from http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2015/09/03/examples_ne…: > > > > 232c232 > > < field 0 with 2 components > > --- > >> field 0 with 0 components > > 234,242c234,242 > > < ( 2) dim 2 offset 0 0.5 0 > > < ( 3) dim 2 offset 2 1 0 > > < ( 4) dim 2 offset 4 0 0.5 > > < ( 5) dim 2 offset 6 0.5 0.5 > > < ( 6) dim 2 offset 8 1 0.5 > > < ( 7) dim 2 offset 10 0 1 > > < ( 8) dim 2 offset 12 0.5 1 > > < ( 9) dim 2 offset 14 0.5 0.25 > > < ( 10) dim 2 offset 16 0.25 0.5 > > --- > >> ( 2) dim 0 offset 2 0.5 0 > >> ( 3) dim 0 offset 2 1 0 > >> ( 4) dim 0 offset 2 0 0.5 > >> ( 5) dim 0 offset 2 0.5 0.5 > >> ( 6) dim 0 offset 2 1 0.5 > >> ( 7) dim 0 offset 2 0 1 > >> ( 8) dim 0 offset 2 0.5 1 > >> ( 9) dim 0 offset 2 0.5 0.25 > >> ( 10) dim 0 offset 2 0.25 0.5 > > > > The numbers that printf is making 0 (or 2) clearly aren't, because > > they are used to iterate over a section and display its contents. But > > other PetscInt's are printed fine elsewhere. Weird. > > > > Toby > > > >

1 0

Re: [petsc-dev] Open solaris 64bit printf bug?
by Barry Smith 03 Sep '15

03 Sep '15

And just to annoy me I cannot get -start_in_debugger to work on that machine. Valgrind has no complaints on a Mac with the examples. I've found some miss-use of PetscViewerASCIIPushSynchronized() and %d in parts of plex, I don't see how they could cause this problem but I am fixing them and will push that. Maybe we'll get lucky. Barry > On Sep 3, 2015, at 10:06 AM, Tobin Isaac <tisaac(a)ices.utexas.edu> wrote: > > > Here's a partial diff from http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2015/09/03/examples_ne…: > > 232c232 > < field 0 with 2 components > --- >> field 0 with 0 components > 234,242c234,242 > < ( 2) dim 2 offset 0 0.5 0 > < ( 3) dim 2 offset 2 1 0 > < ( 4) dim 2 offset 4 0 0.5 > < ( 5) dim 2 offset 6 0.5 0.5 > < ( 6) dim 2 offset 8 1 0.5 > < ( 7) dim 2 offset 10 0 1 > < ( 8) dim 2 offset 12 0.5 1 > < ( 9) dim 2 offset 14 0.5 0.25 > < ( 10) dim 2 offset 16 0.25 0.5 > --- >> ( 2) dim 0 offset 2 0.5 0 >> ( 3) dim 0 offset 2 1 0 >> ( 4) dim 0 offset 2 0 0.5 >> ( 5) dim 0 offset 2 0.5 0.5 >> ( 6) dim 0 offset 2 1 0.5 >> ( 7) dim 0 offset 2 0 1 >> ( 8) dim 0 offset 2 0.5 1 >> ( 9) dim 0 offset 2 0.5 0.25 >> ( 10) dim 0 offset 2 0.25 0.5 > > The numbers that printf is making 0 (or 2) clearly aren't, because > they are used to iterate over a section and display its contents. But > other PetscInt's are printed fine elsewhere. Weird. > > Toby >

1 0

Re: [petsc-dev] problem with LSQR and CGNE
by Barry Smith 03 Sep '15

03 Sep '15

Well it is still trying to do a block Jacobi preconditioner on a nonsquare matrix so you are not setting the PC type to none on the appropriate PC. Are you sure you recompiled correctly, are you sure that this is the PC that should be set to PCNONE? Barry > On Sep 3, 2015, at 1:53 PM, Raphaël Couturier <raphael.couturier(a)univ-fcomte.fr> wrote: > > Oups, you are right, thank you > But the problem remains... > > [0]PETSC ERROR: --------------------- Error Message -------------------------------------------------------------- > [0]PETSC ERROR: No support for this operation for this object type > [0]PETSC ERROR: Only square matrices supported. > [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. > [0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 > [0]PETSC ERROR: ./ex15 on a arch-linux2-c-debug named extinction by couturie Thu Sep 3 20:50:49 2015 > [0]PETSC ERROR: Configure options --download-openmpi --download-hypre --download-f2cblaslapack --with-fc=0 > [0]PETSC ERROR: #2 MatGetDiagonalBlock_MPIDense() line 78 in /home/couturie/work/petsc-3.6.1/src/mat/impls/dense/mpi/mpidense.c > [0]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c > [0]PETSC ERROR: #4 PCSetUp_BJacobi() line 126 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/impls/bjacobi/bjacobi.c > [0]PETSC ERROR: #5 PCSetUp() line 982 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/interface/precon.c > [0]PETSC ERROR: #6 KSPSetUp() line 332 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #7 KSPSolve() line 546 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: [1]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c > > > Raphaël >> I think you mean to have >> >> ierr = PCSetType(ksp3,PCNONE); >> >> not >> >> ierr = PCSetType(ksp,PCNONE); >> >> in the code fragment below. Block diagonal (which is the default) doesn't make sense for rectangular matrices. >> >> >> Yes, PETSc is weak on solvers for nonsquare matrices so users would benefit from more choices. See https://bitbucket.org/petsc/petsc/wiki/pull-request-instructions-git and http://www.mcs.anl.gov/petsc/developers/developers.pdf for information on contributing code. >> >> Barry >> >> >>> On Sep 3, 2015, at 1:21 PM, Raphaël Couturier <raphael.couturier(a)univ-fcomte.fr> wrote: >>> >>> Hello all, >>> >>> First of all, thank you Matthew and Barry for your answers in July. >>> Your solution using PCKSP is fine. >>> Now it works perfectly with the examples I have tried for ksp and snes. I made many tests with many processors. >>> >>> I have implemented my own CGLS or LSQR method. In Petsc, I have seen that CGNE and LSQR are already implemented. However, I cannot use them because I have a problem of non square matrix. >>> In previous posts, I saw that the problem can come from the preconditioner. >>> In the attached code, I have my new solver (without preconditioner if I made no mistake) which is based on >>> ierr = PCKSPGetKSP(pc,&sub_ksp);CHKERRQ(ierr); >>> >>> and I need to use another ksp to minimize the residual (instead of my own CGLS). The matrix AS is of size (n,10) with n very large. >>> I have tried with that (line 262 in the code) >>> { >>> KSP ksp3; >>> ierr = KSPCreate(PETSC_COMM_WORLD, &ksp3); CHKERRQ(ierr); >>> ierr = KSPSetType(ksp3, KSPLSQR); CHKERRQ(ierr); >>> ierr = KSPSetOperators(ksp3, AS, AS); CHKERRQ(ierr); >>> ierr = KSPSetTolerances(ksp3, PETSC_DEFAULT, PETSC_DEFAULT, PETSC_DEFAULT, iter_max_minimization); >>> ierr = PCSetType(ksp,PCNONE); >>> ierr = KSPSolve(ksp3, b, Alpha); CHKERRQ(ierr); >>> } >>> >>> The problem on all the processors is: >>> [0]PETSC ERROR: No support for this operation for this object type >>> [0]PETSC ERROR: Only square matrices supported. >>> [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. >>> [0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 >>> [0]PETSC ERROR: ./ex15 on a arch-linux2-c-debug named extinction by couturie Thu Sep 3 13:22:03 2015 >>> [0]PETSC ERROR: Configure options --download-openmpi --download-hypre --download-f2cblaslapack --with-fc=0 >>> [0]PETSC ERROR: #2 MatGetDiagonalBlock_MPIDense() line 78 in /home/couturie/work/petsc-3.6.1/src/mat/impls/dense/mpi/mpidense.c >>> [0]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c >>> [0]PETSC ERROR: #4 PCSetUp_BJacobi() line 126 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/impls/bjacobi/bjacobi.c >>> [0]PETSC ERROR: #5 PCSetUp() line 982 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/interface/precon.c >>> [0]PETSC ERROR: #6 KSPSetUp() line 332 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c >>> [0]PETSC ERROR: #7 KSPSolve() line 546 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c >>> [0]PETSC ERROR: #8 KSPSolve_TSIRM() line 269 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/impls/gmres/tsirm.c >>> [0]PETSC ERROR: #9 KSPSolve() line 604 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c >>> [0]PETSC ERROR: #10 main() line 200 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/examples/tutorials/ex15.c >>> >>> Do you have an idea of the problem? >>> Is it a good idea to use the LSQR or CGLS method of Petsc? >>> >>> >>> Moreover I have another question. Do you think that the Petsc community would be interested to use this solver. If yes, I would be glad to improve it in order to integrate it in the source. >>> >>> Thanks >>> >>> Raphaël >>> <tsirm.c> >

1 0

Re: [petsc-dev] problem with LSQR and CGNE
by Raphaël Couturier 03 Sep '15

03 Sep '15

Oups, you are right, thank you But the problem remains... [0]PETSC ERROR: --------------------- Error Message -------------------------------------------------------------- [0]PETSC ERROR: No support for this operation for this object type [0]PETSC ERROR: Only square matrices supported. [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. [0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 [0]PETSC ERROR: ./ex15 on a arch-linux2-c-debug named extinction by couturie Thu Sep 3 20:50:49 2015 [0]PETSC ERROR: Configure options --download-openmpi --download-hypre --download-f2cblaslapack --with-fc=0 [0]PETSC ERROR: #2 MatGetDiagonalBlock_MPIDense() line 78 in /home/couturie/work/petsc-3.6.1/src/mat/impls/dense/mpi/mpidense.c [0]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c [0]PETSC ERROR: #4 PCSetUp_BJacobi() line 126 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/impls/bjacobi/bjacobi.c [0]PETSC ERROR: #5 PCSetUp() line 982 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/interface/precon.c [0]PETSC ERROR: #6 KSPSetUp() line 332 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c [0]PETSC ERROR: #7 KSPSolve() line 546 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c [0]PETSC ERROR: [1]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c Raphaël > I think you mean to have > > ierr = PCSetType(ksp3,PCNONE); > > not > > ierr = PCSetType(ksp,PCNONE); > > in the code fragment below. Block diagonal (which is the default) doesn't make sense for rectangular matrices. > > > Yes, PETSc is weak on solvers for nonsquare matrices so users would benefit from more choices. See https://bitbucket.org/petsc/petsc/wiki/pull-request-instructions-git and http://www.mcs.anl.gov/petsc/developers/developers.pdf for information on contributing code. > > Barry > > >> On Sep 3, 2015, at 1:21 PM, Raphaël Couturier <raphael.couturier(a)univ-fcomte.fr> wrote: >> >> Hello all, >> >> First of all, thank you Matthew and Barry for your answers in July. >> Your solution using PCKSP is fine. >> Now it works perfectly with the examples I have tried for ksp and snes. I made many tests with many processors. >> >> I have implemented my own CGLS or LSQR method. In Petsc, I have seen that CGNE and LSQR are already implemented. However, I cannot use them because I have a problem of non square matrix. >> In previous posts, I saw that the problem can come from the preconditioner. >> In the attached code, I have my new solver (without preconditioner if I made no mistake) which is based on >> ierr = PCKSPGetKSP(pc,&sub_ksp);CHKERRQ(ierr); >> >> and I need to use another ksp to minimize the residual (instead of my own CGLS). The matrix AS is of size (n,10) with n very large. >> I have tried with that (line 262 in the code) >> { >> KSP ksp3; >> ierr = KSPCreate(PETSC_COMM_WORLD, &ksp3); CHKERRQ(ierr); >> ierr = KSPSetType(ksp3, KSPLSQR); CHKERRQ(ierr); >> ierr = KSPSetOperators(ksp3, AS, AS); CHKERRQ(ierr); >> ierr = KSPSetTolerances(ksp3, PETSC_DEFAULT, PETSC_DEFAULT, PETSC_DEFAULT, iter_max_minimization); >> ierr = PCSetType(ksp,PCNONE); >> ierr = KSPSolve(ksp3, b, Alpha); CHKERRQ(ierr); >> } >> >> The problem on all the processors is: >> [0]PETSC ERROR: No support for this operation for this object type >> [0]PETSC ERROR: Only square matrices supported. >> [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. >> [0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 >> [0]PETSC ERROR: ./ex15 on a arch-linux2-c-debug named extinction by couturie Thu Sep 3 13:22:03 2015 >> [0]PETSC ERROR: Configure options --download-openmpi --download-hypre --download-f2cblaslapack --with-fc=0 >> [0]PETSC ERROR: #2 MatGetDiagonalBlock_MPIDense() line 78 in /home/couturie/work/petsc-3.6.1/src/mat/impls/dense/mpi/mpidense.c >> [0]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c >> [0]PETSC ERROR: #4 PCSetUp_BJacobi() line 126 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/impls/bjacobi/bjacobi.c >> [0]PETSC ERROR: #5 PCSetUp() line 982 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/interface/precon.c >> [0]PETSC ERROR: #6 KSPSetUp() line 332 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c >> [0]PETSC ERROR: #7 KSPSolve() line 546 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c >> [0]PETSC ERROR: #8 KSPSolve_TSIRM() line 269 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/impls/gmres/tsirm.c >> [0]PETSC ERROR: #9 KSPSolve() line 604 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c >> [0]PETSC ERROR: #10 main() line 200 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/examples/tutorials/ex15.c >> >> Do you have an idea of the problem? >> Is it a good idea to use the LSQR or CGLS method of Petsc? >> >> >> Moreover I have another question. Do you think that the Petsc community would be interested to use this solver. If yes, I would be glad to improve it in order to integrate it in the source. >> >> Thanks >> >> Raphaël >> <tsirm.c>

1 0

Re: [petsc-dev] problem with LSQR and CGNE
by Barry Smith 03 Sep '15

03 Sep '15

I think you mean to have ierr = PCSetType(ksp3,PCNONE); not ierr = PCSetType(ksp,PCNONE); in the code fragment below. Block diagonal (which is the default) doesn't make sense for rectangular matrices. Yes, PETSc is weak on solvers for nonsquare matrices so users would benefit from more choices. See https://bitbucket.org/petsc/petsc/wiki/pull-request-instructions-git and http://www.mcs.anl.gov/petsc/developers/developers.pdf for information on contributing code. Barry > On Sep 3, 2015, at 1:21 PM, Raphaël Couturier <raphael.couturier(a)univ-fcomte.fr> wrote: > > Hello all, > > First of all, thank you Matthew and Barry for your answers in July. > Your solution using PCKSP is fine. > Now it works perfectly with the examples I have tried for ksp and snes. I made many tests with many processors. > > I have implemented my own CGLS or LSQR method. In Petsc, I have seen that CGNE and LSQR are already implemented. However, I cannot use them because I have a problem of non square matrix. > In previous posts, I saw that the problem can come from the preconditioner. > In the attached code, I have my new solver (without preconditioner if I made no mistake) which is based on > ierr = PCKSPGetKSP(pc,&sub_ksp);CHKERRQ(ierr); > > and I need to use another ksp to minimize the residual (instead of my own CGLS). The matrix AS is of size (n,10) with n very large. > I have tried with that (line 262 in the code) > { > KSP ksp3; > ierr = KSPCreate(PETSC_COMM_WORLD, &ksp3); CHKERRQ(ierr); > ierr = KSPSetType(ksp3, KSPLSQR); CHKERRQ(ierr); > ierr = KSPSetOperators(ksp3, AS, AS); CHKERRQ(ierr); > ierr = KSPSetTolerances(ksp3, PETSC_DEFAULT, PETSC_DEFAULT, PETSC_DEFAULT, iter_max_minimization); > ierr = PCSetType(ksp,PCNONE); > ierr = KSPSolve(ksp3, b, Alpha); CHKERRQ(ierr); > } > > The problem on all the processors is: > [0]PETSC ERROR: No support for this operation for this object type > [0]PETSC ERROR: Only square matrices supported. > [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. > [0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 > [0]PETSC ERROR: ./ex15 on a arch-linux2-c-debug named extinction by couturie Thu Sep 3 13:22:03 2015 > [0]PETSC ERROR: Configure options --download-openmpi --download-hypre --download-f2cblaslapack --with-fc=0 > [0]PETSC ERROR: #2 MatGetDiagonalBlock_MPIDense() line 78 in /home/couturie/work/petsc-3.6.1/src/mat/impls/dense/mpi/mpidense.c > [0]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c > [0]PETSC ERROR: #4 PCSetUp_BJacobi() line 126 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/impls/bjacobi/bjacobi.c > [0]PETSC ERROR: #5 PCSetUp() line 982 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/interface/precon.c > [0]PETSC ERROR: #6 KSPSetUp() line 332 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #7 KSPSolve() line 546 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #8 KSPSolve_TSIRM() line 269 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/impls/gmres/tsirm.c > [0]PETSC ERROR: #9 KSPSolve() line 604 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c > [0]PETSC ERROR: #10 main() line 200 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/examples/tutorials/ex15.c > > Do you have an idea of the problem? > Is it a good idea to use the LSQR or CGLS method of Petsc? > > > Moreover I have another question. Do you think that the Petsc community would be interested to use this solver. If yes, I would be glad to improve it in order to integrate it in the source. > > Thanks > > Raphaël > <tsirm.c>

1 0

problem with LSQR and CGNE
by Raphaël Couturier 03 Sep '15

03 Sep '15

Hello all, First of all, thank you Matthew and Barry for your answers in July. Your solution using PCKSP is fine. Now it works perfectly with the examples I have tried for ksp and snes. I made many tests with many processors. I have implemented my own CGLS or LSQR method. In Petsc, I have seen that CGNE and LSQR are already implemented. However, I cannot use them because I have a problem of non square matrix. In previous posts, I saw that the problem can come from the preconditioner. In the attached code, I have my new solver (without preconditioner if I made no mistake) which is based on ierr = PCKSPGetKSP(pc,&sub_ksp);CHKERRQ(ierr); and I need to use another ksp to minimize the residual (instead of my own CGLS). The matrix AS is of size (n,10) with n very large. I have tried with that (line 262 in the code) { KSP ksp3; ierr = KSPCreate(PETSC_COMM_WORLD, &ksp3); CHKERRQ(ierr); ierr = KSPSetType(ksp3, KSPLSQR); CHKERRQ(ierr); ierr = KSPSetOperators(ksp3, AS, AS); CHKERRQ(ierr); ierr = KSPSetTolerances(ksp3, PETSC_DEFAULT, PETSC_DEFAULT, PETSC_DEFAULT, iter_max_minimization); ierr = PCSetType(ksp,PCNONE); ierr = KSPSolve(ksp3, b, Alpha); CHKERRQ(ierr); } The problem on all the processors is: [0]PETSC ERROR: No support for this operation for this object type [0]PETSC ERROR: Only square matrices supported. [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for trouble shooting. [0]PETSC ERROR: Petsc Release Version 3.6.1, Jul, 22, 2015 [0]PETSC ERROR: ./ex15 on a arch-linux2-c-debug named extinction by couturie Thu Sep 3 13:22:03 2015 [0]PETSC ERROR: Configure options --download-openmpi --download-hypre --download-f2cblaslapack --with-fc=0 [0]PETSC ERROR: #2 MatGetDiagonalBlock_MPIDense() line 78 in /home/couturie/work/petsc-3.6.1/src/mat/impls/dense/mpi/mpidense.c [0]PETSC ERROR: #3 MatGetDiagonalBlock() line 159 in /home/couturie/work/petsc-3.6.1/src/mat/interface/matrix.c [0]PETSC ERROR: #4 PCSetUp_BJacobi() line 126 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/impls/bjacobi/bjacobi.c [0]PETSC ERROR: #5 PCSetUp() line 982 in /home/couturie/work/petsc-3.6.1/src/ksp/pc/interface/precon.c [0]PETSC ERROR: #6 KSPSetUp() line 332 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c [0]PETSC ERROR: #7 KSPSolve() line 546 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c [0]PETSC ERROR: #8 KSPSolve_TSIRM() line 269 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/impls/gmres/tsirm.c [0]PETSC ERROR: #9 KSPSolve() line 604 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/interface/itfunc.c [0]PETSC ERROR: #10 main() line 200 in /home/couturie/work/petsc-3.6.1/src/ksp/ksp/examples/tutorials/ex15.c Do you have an idea of the problem? Is it a good idea to use the LSQR or CGLS method of Petsc? Moreover I have another question. Do you think that the Petsc community would be interested to use this solver. If yes, I would be glad to improve it in order to integrate it in the source. Thanks Raphaël

1 0

Open solaris 64bit printf bug?
by Tobin Isaac 03 Sep '15

03 Sep '15

Here's a partial diff from http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2015/09/03/examples_ne…: 232c232 < field 0 with 2 components --- > field 0 with 0 components 234,242c234,242 < ( 2) dim 2 offset 0 0.5 0 < ( 3) dim 2 offset 2 1 0 < ( 4) dim 2 offset 4 0 0.5 < ( 5) dim 2 offset 6 0.5 0.5 < ( 6) dim 2 offset 8 1 0.5 < ( 7) dim 2 offset 10 0 1 < ( 8) dim 2 offset 12 0.5 1 < ( 9) dim 2 offset 14 0.5 0.25 < ( 10) dim 2 offset 16 0.25 0.5 --- > ( 2) dim 0 offset 2 0.5 0 > ( 3) dim 0 offset 2 1 0 > ( 4) dim 0 offset 2 0 0.5 > ( 5) dim 0 offset 2 0.5 0.5 > ( 6) dim 0 offset 2 1 0.5 > ( 7) dim 0 offset 2 0 1 > ( 8) dim 0 offset 2 0.5 1 > ( 9) dim 0 offset 2 0.5 0.25 > ( 10) dim 0 offset 2 0.25 0.5 The numbers that printf is making 0 (or 2) clearly aren't, because they are used to iterate over a section and display its contents. But other PetscInt's are printed fine elsewhere. Weird. Toby

1 0

Re: [petsc-dev] [Bitbucket] Pull request #363: cleaned up cheby eig estimates and gamg rand (petsc/petsc)
by Barry Smith 03 Sep '15

03 Sep '15

I have cleaned up this branch a bit and merged in Matt's portable random number generator. It needs some work (updating output files etc) before being merged into next for general testing. > On Sep 2, 2015, at 6:26 PM, Mark Adams <pullrequests-reply(a)bitbucket.org> wrote: > > > Mark Adams updated pull request #363: > cleaned up cheby eig estimates and gamg rand > Author Commit Message Date > Mark Adams 361911d resolved conflicts > 2 Sep 2015 > Mark Adams 936b047 small fix > 2 Sep 2015 > View this pull request or add a comment by replying to this email. > Unsubscribe from pull request emails for this repository.

1 0