August 2012 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] [petsc-maint #128159] MatGetArray
by Jed Brown 13 Aug '12

13 Aug '12

On Mon, Aug 13, 2012 at 10:45 AM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > I propose to remove MatGetArray() from PETSc and insert > MatSeqDenseGetArray() for all the uses with in that case (where it > appears often and is useful). > > What say you? > Yes, remove MatGetArray and add MatSeqDenseGetArray(), MatSeqAIJGetArray(), and possibly MatSeqBAIJGetArray(). I don't want these to have the same name because they return very different things. I don't want any MatMPI*GetArray() routines.

1 0

Re: [petsc-dev] [petsc-maint #128159] MatGetArray
by Lisandro Dalcin 13 Aug '12

13 Aug '12

On 13 August 2012 13:45, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > Jed, > > > I propose to remove MatGetArray() from PETSc and insert MatSeqDenseGetArray() for all the uses with in that case (where it appears often and is useful). > > What say you? > Big +1. What about "a" values from SeqAIJ? (ignore my question if PETSc already has something better to get the a,i,j data) -- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169

1 0

Re: [petsc-dev] why does MatCreate_MPIDense still exist?
by Jed Brown 13 Aug '12

13 Aug '12

On Mon, Aug 13, 2012 at 10:40 AM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > Should I remove all the MPIDense stuff from PETSc now? It is uses > PLAPACK which is buggy and unsupported. That is still a good distribution for MxN matrices with M>>N. We can do them with Elemental, but that would use a different distribution so it will be more complicated to interact with. (The current Elemental interface uses a squarish distribution, but we can tell Elemental to use the [VC,*] distribution (for which fewer operations are supported). The main thing I care about for that distribution is QR. The best format is dense and row-aligned. I don't care whether it uses MPIDense or a new "multi-vector" thing, but that concept should be somewhere.

1 0

Re: [petsc-dev] why does MatCreate_MPIDense still exist?
by Lisandro Dalcin 13 Aug '12

13 Aug '12

On 13 August 2012 13:47, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > On Aug 13, 2012, at 11:44 AM, Lisandro Dalcin <dalcinl(a)gmail.com> wrote: > >> On 13 August 2012 13:40, Barry Smith <bsmith(a)mcs.anl.gov> wrote: >>> >>> Should I remove all the MPIDense stuff from PETSc now? It is uses PLAPACK which is buggy and unsupported. >>> >> >> Do you mean to completely kill the subtype implementation? > > Yes, people will use the new elemental interface for dense matrices. Will that not work for your needs? Could you live with a MPIDense that did not use PLAPACK? So had very limited functionality? > As long as we have some kind of dense implementation (mostly for debugging purposes), I'm fine. -- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169

1 0

Re: [petsc-dev] why does MatCreate_MPIDense still exist?
by Barry Smith 13 Aug '12

13 Aug '12

On Aug 13, 2012, at 11:44 AM, Lisandro Dalcin <dalcinl(a)gmail.com> wrote: > On 13 August 2012 13:40, Barry Smith <bsmith(a)mcs.anl.gov> wrote: >> >> Should I remove all the MPIDense stuff from PETSc now? It is uses PLAPACK which is buggy and unsupported. >> > > Do you mean to completely kill the subtype implementation? Yes, people will use the new elemental interface for dense matrices. Will that not work for your needs? Could you live with a MPIDense that did not use PLAPACK? So had very limited functionality? Barry > > -- > Lisandro Dalcin > --------------- > CIMEC (INTEC/CONICET-UNL) > Predio CONICET-Santa Fe > Colectora RN 168 Km 472, Paraje El Pozo > 3000 Santa Fe, Argentina > Tel: +54-342-4511594 (ext 1011) > Tel/Fax: +54-342-4511169

1 0

Re: [petsc-dev] [petsc-maint #128159] MatGetArray
by Barry Smith 13 Aug '12

13 Aug '12

Jed, I propose to remove MatGetArray() from PETSc and insert MatSeqDenseGetArray() for all the uses with in that case (where it appears often and is useful). What say you? Barry On Aug 13, 2012, at 10:48 AM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > On Mon, Aug 13, 2012 at 9:06 AM, Phani Motamarri <phanim(a)umich.edu> wrote: > >> Just like VecGetArray and MatGetArray gives local entries of a vector and >> matrix on a given processor, >> > > In general, this is not true of matrices. The MPI*AIJ formats return the > "diagonal" part, but it's just a side-effect of the storage format. > > >> is there any function that gives me the local >> entries of a matrix which is of type "MATELEMENTAL". Unfortunately >> MatGetArray does not work in this case. >> > > Every time this comes up, I say that PETSc should remove the MatGetArray() > call. If you want direct access to entries, include the private header and > remember that what you are not using a public API. If you want structured > access to entries, use MatGetSubMatrices()/MatGetSubMatrix() and use a > format-specific accessor. In general, it is valuable to stop thinking about > matrices as arrays of numbers and start thinking about them as linear > operators. >

1 0

Re: [petsc-dev] why does MatCreate_MPIDense still exist?
by Lisandro Dalcin 13 Aug '12

13 Aug '12

On 13 August 2012 13:40, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > Should I remove all the MPIDense stuff from PETSc now? It is uses PLAPACK which is buggy and unsupported. > Do you mean to completely kill the subtype implementation? -- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169

1 0

why does MatCreate_MPIDense still exist?
by Barry Smith 13 Aug '12

13 Aug '12

Should I remove all the MPIDense stuff from PETSc now? It is uses PLAPACK which is buggy and unsupported. Barry

1 0

Re: [petsc-dev] calling petscinitialize from fortran code
by Satish Balay 11 Aug '12

11 Aug '12

I don't see this. Perhaps you can run in a debugger to see what is hapenning on your end? Satish ---- asterix:/home/balay/spetsc/src/ksp/ksp/examples/tutorials>./ex2f Norm of error 0.1192E-05 iterations 4 asterix:/home/balay/spetsc/src/ksp/ksp/examples/tutorials>./ex2f -options_table Norm of error 0.1192E-05 iterations 4 #PETSc Option Table entries: -display :0.0 -fp_trap -malloc_dump -options_table #End of PETSc Option Table entries asterix:/home/balay/spetsc/src/ksp/ksp/examples/tutorials> On Sat, 11 Aug 2012, Stefano Zampini wrote: > Hi, > > I'm experiencing some boring behaviour with my working copy of petsc-dev. > During the call to PetscInitialize(PETSC_NULL_CHARACTER,ierr), it seems it > calls PetscOptionsView. > This does not happen with C code, and did not happen some time ago. So, I > think it's related with some recent Fortran bindings. > > I tried a pull right now, cleaning libraries and recompiling (recreating > the fortran stubs) but the problem remains. > > Can you fix it? > > Thanks, >

1 0

calling petscinitialize from fortran code
by Stefano Zampini 11 Aug '12

11 Aug '12

Hi, I'm experiencing some boring behaviour with my working copy of petsc-dev. During the call to PetscInitialize(PETSC_NULL_CHARACTER,ierr), it seems it calls PetscOptionsView. This does not happen with C code, and did not happen some time ago. So, I think it's related with some recent Fortran bindings. I tried a pull right now, cleaning libraries and recompiling (recreating the fortran stubs) but the problem remains. Can you fix it? Thanks, -- Stefano

1 0