July 2011 - petsc-dev - mailman3.cels.anl.gov

cmake problem
by Barry Smith 08 Jul '11

08 Jul '11

Reran ./configure but then cmake made me trouble. Sending the configure.log only to Jed Barry xxx=========================================================================xxx Configure stage complete. Now build PETSc libraries with: make PETSC_DIR=/Users/barrysmith/Src/petsc-dev PETSC_ARCH=arch-gnu all or (experimental with cmake): make -j4 -C /Users/barrysmith/Src/petsc-dev/arch-gnu or (experimental with python): PETSC_DIR=/Users/barrysmith/Src/petsc-dev PETSC_ARCH=arch-gnu ./config/builder.py xxx=========================================================================xxx cd /Users/barrysmith/Src/petsc-dev; make -j2 -C /Users/barrysmith/Src/petsc-dev/arch-gnu cd /Users/barrysmith/Src/petsc-dev; make -j2 -C /Users/barrysmith/Src/petsc-dev/arch-gnu Scanning dependencies of target petsc Error copying Fortran module "include/petscsys". Tried "include/PETSCSYS.mod" and "include/petscsys.mod". make[4]: *** [CMakeFiles/petsc.dir/src/sys/f90-mod/petscsysmod.F.o.provides.build] Error 1 make[3]: *** [CMakeFiles/petsc.dir/src/sys/f90-mod/petscsysmod.F.o.provides] Error 2 make[2]: *** [CMakeFiles/petsc.dir/all] Error 2 make[1]: *** [all] Error 2 make: *** [cmake] Error 2 anlextwls002-241:petsc-dev barrysmith$ echo $PETSC_DIR /Users/barrysmith/Src/petsc-dev

1 0

mechanism to prevent running of untested PETSc codes on difficult machines
by Barry Smith 08 Jul '11

08 Jul '11

I suggest we embed in the PETSc library a mechanism that does not allow an untested PETSc code to be run on machines like BlueGene and Cray (or even if possible clusters with batch systems). After a program has been successfully run a number of times on an easy to use machine (including some parallel runs) it generates a key associated with the code. Then at runtime PetscInitialize() on the BlueGene would check for the appropriate key; if it does not locate it it refuses to run and tells the user to develop on a development system. Thoughts? Barry >> >> I am running the PETSc code in Bluegene. >> >> And some errors came up like: >> "Floating point exception!" >>

1 0

TS changes
by Jed Brown 07 Jul '11

07 Jul '11

I have pushed several changes to TS. Significant order dependence has been removed in the setup phase. The user visible changes are summarized below, please report any problems. TSSetIFunction() and TSSetRHSFunction() have an additional argument for the vector to compute the function value in. If PETSC_NULL is used, then one is created by the TS. TSSetMatrices() has been removed, use TSSetIFunction() and TSSetRHSFunction(), perhaps providing Jacobian matrices and TSComputeRHSFunctionLinear() and/or TSComputeRHSJacobianConstant(). See src/ts/examples/tutorials/ex{2,4,5,6}.c for examples. Changed TSStep() interface to only perform one step. Added TSSolve() to perform multiple steps with callbacks and monitors if desired. Added TSARKIMEX: additive Runge-Kutta implicit-explicit methods for multi-rate systems. Schemes of order 2 through 5 with L-stable implicit part and SSP explicit part are available.

1 0

Re: [petsc-dev] [petsc-maint #78191] Re: [petsc-users] Reusing ML preconditioner
by John Fettig 05 Jul '11

05 Jul '11

Hi Barry, On Fri, Jul 1, 2011 at 11:48 PM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > if (i) > { > ierr = KSPSetInitialGuessNonzero(ksp,PETSC_TRUE);CHKERRQ(ierr); > PC pc; > ierr = KSPGetPC(ksp,&pc);CHKERRQ(ierr); > PetscInt nlevels; > ierr = PCMGGetLevels(pc,&nlevels);CHKERRQ(ierr); > KSP sksp; > ierr = PCMGGetSmoother(pc,nlevels-1,&sksp);CHKERRQ(ierr); > ierr = KSPSetOperators(sksp,A[i],A[i],SAME_NONZERO_PATTERN);CHKERRQ(ierr); > ierr = PCMGSetResidual(pc,nlevels-1,PCMGDefaultResidual,A[i]);CHKERRQ(ierr); > > Note that this is not technically reusing the exact old preconditioner, it is actually partially modifying the old preconditioner with new matrix information. Thanks for finding the cause, this does work for me. > Remind me why are you passing entirely new matrices to KSPSetOperators instead of the same matrix with new values in it? Our code already generated a CSR matrix so it makes sense to pass it in directly. However, in parallel we are not generating the split arrays that would bypass the copy, so instead we just re-create the matrix. I'm sure this isn't as efficient as passing in the split arrays, perhaps you have a better idea about what should be done in this case (Jed thought that the performance gain from using the "split array" matrix creation would be minimal). Thanks, John

1 0

Re: [petsc-dev] PETSc recommended visualization packages
by Matthew Knepley 05 Jul '11

05 Jul '11

On Tue, Jul 5, 2011 at 1:54 PM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > On Tue, Jul 5, 2011 at 08:46, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > >> Should we make a new viewer based on VTK binary that automatically manages >> the multiple files and headers etc? > > > Perhaps, but you still lose a lot of information with VTK binary, in the > sense that it's usually no longer useful as a checkpoint. I think we would > be better off porting/generalizing the DHM viewer I wrote for Dohp over to > PETSc and provide VisIt and ParaView plugins for viewing it. If we do that, > I would like to get away from HDF5 in the process, probably in favor of a > SQLite index with a binary object store, so it would be a rewrite as much as > a port. > > Note that with such a system, if you link to VTK, you can automatically > write in any VTK format to make a file that is viewable without a plugin > (though this isn't the lowest memory way to create such a file). > <Excerpted from discussion with Jed> I think that HDF5 is a good choice because you can write a full checkpoint, and also interface with plain vanilla visualization using the xdmf specification. This is what we do in PyLith. Matt -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener

1 0

Re: [petsc-dev] PETSc recommended visualization packages
by Jed Brown 05 Jul '11

05 Jul '11

On Tue, Jul 5, 2011 at 08:46, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > Should we make a new viewer based on VTK binary that automatically manages > the multiple files and headers etc? Perhaps, but you still lose a lot of information with VTK binary, in the sense that it's usually no longer useful as a checkpoint. I think we would be better off porting/generalizing the DHM viewer I wrote for Dohp over to PETSc and provide VisIt and ParaView plugins for viewing it. If we do that, I would like to get away from HDF5 in the process, probably in favor of a SQLite index with a binary object store, so it would be a rewrite as much as a port. Note that with such a system, if you link to VTK, you can automatically write in any VTK format to make a file that is viewable without a plugin (though this isn't the lowest memory way to create such a file).

1 0

PETSc recommended visualization packages
by Barry Smith 05 Jul '11

05 Jul '11

Should we make a new viewer based on VTK binary that automatically manages the multiple files and headers etc? Barry On Jul 4, 2011, at 10:06 PM, Jed Brown wrote: > On Mon, Jul 4, 2011 at 21:44, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > What are the recommended visualization packages for use with PETSc (for example making movies of contour plots and isosurfaces) and what are the recommended data formats to use to save Vecs for visualization? > > VisIt and ParaView are the primary choices. In my experience, VisIt is a more complete system, but ParaView is easier to install and probably easier to get started with. > > The VTK XML or binary formats are the easiest to get started with. See src/ts/examples/tutorials/ex14.c for an example of writing an XML file containing two fields on different grids (2D and 3D, structured, but deformed) in parallel. All file formats are horrible non-extensible crud, usually grown out of a particular application with structured grids of low-order/non-exotic basis functions on unstructured grids. They tend not to retain enough semantic information to reconstruct a state for further computations and also work for visualization without needing additional application-provided information.

1 0

Re: [petsc-dev] TSBEULER and TSCN
by Barry Smith 02 Jul '11

02 Jul '11

On Jul 2, 2011, at 5:19 PM, Jed Brown wrote: > On Sat, Jul 2, 2011 at 16:56, Matthew Knepley <knepley(a)gmail.com> wrote: > It will be somewhat confusing in that Backwards Euler will not appear in the -ts_view. Maybe we could get a better > name for the helper function. > > Good point. We could make it use the same data structure and implementation from TSTHETA, but have its own TSCreate and TSView (and keep its own stated type). This is somewhat weird/inconsistent with PETSc style, but it seems a shame to copy the whole damn file and delete three lines just to preserve the name TSBEULER. Sounds good. Barry

1 0

Re: [petsc-dev] TSBEULER and TSCN
by Jed Brown 02 Jul '11

02 Jul '11

On Sat, Jul 2, 2011 at 16:56, Matthew Knepley <knepley(a)gmail.com> wrote: > It will be somewhat confusing in that Backwards Euler will not appear in > the -ts_view. Maybe we could get a better > name for the helper function. > Good point. We could make it use the same data structure and implementation from TSTHETA, but have its own TSCreate and TSView (and keep its own stated type). This is somewhat weird/inconsistent with PETSc style, but it seems a shame to copy the whole damn file and delete three lines just to preserve the name TSBEULER.

1 0

Re: [petsc-dev] TSBEULER and TSCN
by Matthew Knepley 02 Jul '11

02 Jul '11

On Sat, Jul 2, 2011 at 9:39 PM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > In refactoring TS, I want these implementations to go away (they are > special cases of TSTHETA and it's easy for TSTHETA to be as efficient for > the special case). The question is whether the names should be preserved, > with implementations that look like > > TSCreate_BEuler(TS ts) { > TSSetType(ts,TSTHETA); > TSThetaSetTheta(ts,1.0); > } > > or just have users set this themselves. Thoughts? > It will be somewhat confusing in that Backwards Euler will not appear in the -ts_view. Maybe we could get a better name for the helper function. Matt -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener

1 0