September 2017 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] ParMETIS change
by Balay, Satish 13 Sep '17

13 Sep '17

What errors do you get? The primary code change was to use real_t instead of PetscReal in calls to parmetis. The change was done to make parmetis built by petsc be same as default parmetis that others/distros might build. And this enables consistent usage from quad build of petsc. I can add –download-metis-double or similar option to configure – if useful. Satish From: Eric Chamberland<mailto:[email protected]> Sent: Wednesday, September 13, 2017 8:41 AM To: petsc-dev<mailto:[email protected]> Subject: [petsc-dev] ParMETIS change Hi, having trouble to compile our code with PETSc-master new defaults options for PARMEtis. Parmetis was using double precision before... By curiosity, is there a big performance gain to use float? Thanks, Eric

1 0

Re: [petsc-dev] ParMETIS change
by Barry Smith 13 Sep '17

13 Sep '17

We made the policy change to always by default to use single precision ParMetis instead of making it depend on the precision of PETSc. The savings is about memory usage, not about speed. Plus Jed says that nothing ParMetis does is affected by using single precision. is there a reason you cannot just have your code use single precision related to ParMetis? Barry > On Sep 13, 2017, at 8:41 AM, Eric Chamberland <Eric.Chamberland(a)giref.ulaval.ca> wrote: > > Hi, > > having trouble to compile our code with PETSc-master new defaults options for PARMEtis. > > Parmetis was using double precision before... > > By curiosity, is there a big performance gain to use float? > > Thanks, > > Eric > >

1 0

ParMETIS change
by Eric Chamberland 13 Sep '17

13 Sep '17

Hi, having trouble to compile our code with PETSc-master new defaults options for PARMEtis. Parmetis was using double precision before... By curiosity, is there a big performance gain to use float? Thanks, Eric

1 0

possibility of short term subcontract work with PETSc
by Barry Smith 11 Sep '17

11 Sep '17

PETSc folks, Argonne National Laboratory has recently set up a system that may make it possible for the PETSc group at ANL to subcontract particular PETSc contribution projects to developers in most of the world. These could be from one week projects to months. If you have particular projects in mind, for example "add feature XXX to package YYY in PETSc" please contact me directly bsmith(a)mcs.anl.gov with an outline of the idea, how it could help many other PETSc users and how long you think it make take. And we can iterate to see if it is something appropriate to fund. Barry

1 0

Re: [petsc-dev] TSComputeRHSJacobian
by Barry Smith 10 Sep '17

10 Sep '17

Stefano, Thanks, but I don't really want a fix. I want a code that I can run that demonstrates a problem (then we can evaluate fixes) Barry > On Sep 8, 2017, at 11:51 AM, Stefano Zampini <stefano.zampini(a)gmail.com> wrote: > > I encountered the problem while working on the adjoint code. Here is the fix I have in my branch > > https://bitbucket.org/petsc/petsc/branch/stefano_zampini/feature-continuous… > > However, I'm not sure if this is a problem just with my code. This is why I'm asking on the list. > > > > 2017-09-08 0:02 GMT+03:00 Barry Smith <bsmith(a)mcs.anl.gov>: > > Example that demonstrates the current code is problematic in some cases? > > > On Aug 28, 2017, at 11:59 AM, Stefano Zampini <stefano.zampini(a)gmail.com> wrote: > > > > > > I was wondering if the following code in TSComputeRHSJacobian is correct when the user pass the reuse flag via TSRHSJacobianReuse > > > > if (ts->rhsjacobian.time == t && (ts->problem_type == TS_LINEAR || (ts->rhsjacobian.Xid == Uid && ts->rhsjacobian.Xstate == Ustate)) && (rhsfunction != TSComputeRHSFunctionLinear)) { > > PetscFunctionReturn(0); > > } > > > > Should the above code be replaced by the following? > > > > if (ts->rhsjacobian.time == t && (ts->problem_type == TS_LINEAR || (ts->rhsjacobian.Xid == Uid && ts->rhsjacobian.Xstate == Ustate)) && (rhsfunction != TSComputeRHSFunctionLinear)) { > > if (ts->rhsjacobian.reuse) { > > ierr = MatShift(A,-ts->rhsjacobian.shift);CHKERRQ(ierr); > > ierr = MatScale(A,1./ts->rhsjacobian.scale);CHKERRQ(ierr); > > if (B && A != B) { > > ierr = MatShift(B,-ts->rhsjacobian.shift);CHKERRQ(ierr); > > ierr = MatScale(B,1./ts->rhsjacobian.scale);CHKERRQ(ierr); > > } > > ts->rhsjacobian.shift = 0; > > ts->rhsjacobian.scale = 1.; > > } > > PetscFunctionReturn(0); > > } > > > > > > -- > > Stefano > > > > > -- > Stefano

1 0

Re: [petsc-dev] AD support in PETSc?
by Barry Smith 10 Sep '17

10 Sep '17

> On Sep 10, 2017, at 6:15 AM, Stefano Zampini <stefano.zampini(a)gmail.com> wrote: > > Should the references to ADIFOR listed below be removed? Yes > It seems PETSc does no longer provide support for automatic differentiation. I never used AD support in PETSc, but I was considering adding some reverse mode support to compute the transpose of the Jacobian matrices in adjoint computations. May ask why was it removed? It sucked. Though AD seems a good idea, the AD community, for whatever reason has never been able to produce anything usable for libraries. Barry > > > [szampini@localhost petsc]$ git grep -i adifor > bin/adiforfix.py:# $Id: adiforfix.py,v 1.3 2001/08/24 16:32:18 bsmith Exp $ > bin/adiforfix.py:# | adiforfix.py > include/petsc/private/matimpl.h: PetscErrorCode (*setvaluesadifor)(Mat,PetscInt,void*); > include/petscmat.h: MATOP_SET_VALUES_ADIFOR=74, > lib/petsc/conf/test:testexamples_ADIFOR: ${TESTEXAMPLES_ADIFOR} > lib/petsc/conf/test: -@ if test ${PRINT_PROGRESS}foo = debugtestfoo -a -n "${TESTEXAMPLES_ADIFOR}"; then printf "Running examples in `pwd`:\n ${TESTEXAMPLES_ADIFOR}\n"; fi > lib/petsc/conf/test:buildexamples_ADIFOR: > lib/petsc/conf/test: -@${OMAKE} testexamples_ADIFOR TESTEXAMPLES_ADIFOR=`echo ${TESTEXAMPLES_ADIFOR} | sed s/runex[0-9]*[a-z0-9_]*//g` > src/docs/tao_tex/manual/mathprog.bib:@TechReport{bischof.carle.ea:adifor, > src/docs/tao_tex/manual/mathprog.bib: title = "{ADIFOR} 2.0 User's Guide", > src/docs/tex/petscapp.bib: title = "Using {ADIFOR} and {ADIC} to Provide a {J}acobian for the {SNES} Component of {PETSc}", > src/mat/f90-mod/petscmat.h: PetscEnum MATOP_SET_VALUES_ADIFOR > src/mat/f90-mod/petscmat.h: parameter(MATOP_SET_VALUES_ADIFOR=74) > src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f -adifor_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type refine_always > ex5f_4.tmp 2>&1;\ > src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f -adiformf_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type refine_always > ex5f_5.tmp 2>&1;\ > src/snes/examples/tutorials/makefile:TESTEXAMPLES_ADIFOR = ex5f.PETSc runex5f_2 ex5f.rm > > > -- > Stefano

1 0

Re: [petsc-dev] AD support in PETSc?
by Matthew Knepley 10 Sep '17

10 Sep '17

In my opinion, the software engineering problem with ADIFOR(C) is that they are designed to run on whole programs. What we really want is for small pieces of code, like routines for element vectors, to be differentiated. Once you get whole programs, you need to parse system headers, etc. which is a Herculean task. It broke so often that they would maintain a machine in MCS to run on and then spit out the differentiated code. It became too much to maintain and we got rid of it. I like the FEniCS/Firedrake approach much better, in that you differentiate the intermediate representation, and then generate code. Matt On Sun, Sep 10, 2017 at 9:32 AM, Patrick Sanan <patrick.sanan(a)gmail.com> wrote: > I asked Barry about this wrt the manual a while ago, and we removed all > reference to ADIFOR there, since it was no longer supported. > Stefano Zampini <stefano.zampini(a)gmail.com> schrieb am So. 10. Sep. 2017 > um 13:15: > >> Should the references to ADIFOR listed below be removed? It seems PETSc >> does no longer provide support for automatic differentiation. I never used >> AD support in PETSc, but I was considering adding some reverse mode support >> to compute the transpose of the Jacobian matrices in adjoint computations. >> May ask why was it removed? >> >> >> [szampini@localhost petsc]$ git grep -i adifor >> bin/adiforfix.py:# $Id: adiforfix.py,v 1.3 2001/08/24 16:32:18 bsmith Exp >> $ >> bin/adiforfix.py:# | adiforfix.py >> include/petsc/private/matimpl.h: PetscErrorCode (*setvaluesadifor)(Mat, >> PetscInt,void*); >> include/petscmat.h: MATOP_SET_VALUES_ADIFOR=74, >> lib/petsc/conf/test:testexamples_ADIFOR: ${TESTEXAMPLES_ADIFOR} >> lib/petsc/conf/test: -@ if test ${PRINT_PROGRESS}foo = debugtestfoo -a >> -n "${TESTEXAMPLES_ADIFOR}"; then printf "Running examples in `pwd`:\n >> ${TESTEXAMPLES_ADIFOR}\n"; fi >> lib/petsc/conf/test:buildexamples_ADIFOR: >> lib/petsc/conf/test: -@${OMAKE} testexamples_ADIFOR >> TESTEXAMPLES_ADIFOR=`echo ${TESTEXAMPLES_ADIFOR} | sed >> s/runex[0-9]*[a-z0-9_]*//g` >> src/docs/tao_tex/manual/mathprog.bib:@TechReport{bischof.carle.ea:adifor, >> src/docs/tao_tex/manual/mathprog.bib: title = "{ADIFOR} 2.0 >> User's Guide", >> src/docs/tex/petscapp.bib: title = "Using {ADIFOR} and {ADIC} to Provide >> a {J}acobian for the {SNES} Component of {PETSc}", >> src/mat/f90-mod/petscmat.h: PetscEnum MATOP_SET_VALUES_ADIFOR >> src/mat/f90-mod/petscmat.h: parameter(MATOP_SET_VALUES_ADIFOR=74) >> src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f >> -adifor_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type >> refine_always > ex5f_4.tmp 2>&1;\ >> src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f >> -adiformf_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type >> refine_always > ex5f_5.tmp 2>&1;\ >> src/snes/examples/tutorials/makefile:TESTEXAMPLES_ADIFOR = >> ex5f.PETSc runex5f_2 ex5f.rm >> >> >> -- >> Stefano >> > -- 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 http://www.caam.rice.edu/~mk51/

1 0

Re: [petsc-dev] AD support in PETSc?
by Patrick Sanan 10 Sep '17

10 Sep '17

I asked Barry about this wrt the manual a while ago, and we removed all reference to ADIFOR there, since it was no longer supported. Stefano Zampini <stefano.zampini(a)gmail.com> schrieb am So. 10. Sep. 2017 um 13:15: > Should the references to ADIFOR listed below be removed? It seems PETSc > does no longer provide support for automatic differentiation. I never used > AD support in PETSc, but I was considering adding some reverse mode support > to compute the transpose of the Jacobian matrices in adjoint computations. > May ask why was it removed? > > > [szampini@localhost petsc]$ git grep -i adifor > bin/adiforfix.py:# $Id: adiforfix.py,v 1.3 2001/08/24 16:32:18 bsmith Exp $ > bin/adiforfix.py:# | adiforfix.py > include/petsc/private/matimpl.h: PetscErrorCode > (*setvaluesadifor)(Mat,PetscInt,void*); > include/petscmat.h: MATOP_SET_VALUES_ADIFOR=74, > lib/petsc/conf/test:testexamples_ADIFOR: ${TESTEXAMPLES_ADIFOR} > lib/petsc/conf/test: -@ if test ${PRINT_PROGRESS}foo = debugtestfoo -a > -n "${TESTEXAMPLES_ADIFOR}"; then printf "Running examples in `pwd`:\n > ${TESTEXAMPLES_ADIFOR}\n"; fi > lib/petsc/conf/test:buildexamples_ADIFOR: > lib/petsc/conf/test: -@${OMAKE} testexamples_ADIFOR > TESTEXAMPLES_ADIFOR=`echo ${TESTEXAMPLES_ADIFOR} | sed > s/runex[0-9]*[a-z0-9_]*//g` > src/docs/tao_tex/manual/mathprog.bib:@TechReport{bischof.carle.ea:adifor, > src/docs/tao_tex/manual/mathprog.bib: title = "{ADIFOR} 2.0 User's > Guide", > src/docs/tex/petscapp.bib: title = "Using {ADIFOR} and {ADIC} to Provide > a {J}acobian for the {SNES} Component of {PETSc}", > src/mat/f90-mod/petscmat.h: PetscEnum MATOP_SET_VALUES_ADIFOR > src/mat/f90-mod/petscmat.h: parameter(MATOP_SET_VALUES_ADIFOR=74) > src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f > -adifor_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type > refine_always > ex5f_4.tmp 2>&1;\ > src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f > -adiformf_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type > refine_always > ex5f_5.tmp 2>&1;\ > src/snes/examples/tutorials/makefile:TESTEXAMPLES_ADIFOR = > ex5f.PETSc runex5f_2 ex5f.rm > > > -- > Stefano >

1 0

AD support in PETSc?
by Stefano Zampini 10 Sep '17

10 Sep '17

Should the references to ADIFOR listed below be removed? It seems PETSc does no longer provide support for automatic differentiation. I never used AD support in PETSc, but I was considering adding some reverse mode support to compute the transpose of the Jacobian matrices in adjoint computations. May ask why was it removed? [szampini@localhost petsc]$ git grep -i adifor bin/adiforfix.py:# $Id: adiforfix.py,v 1.3 2001/08/24 16:32:18 bsmith Exp $ bin/adiforfix.py:# | adiforfix.py include/petsc/private/matimpl.h: PetscErrorCode (*setvaluesadifor)(Mat,PetscInt,void*); include/petscmat.h: MATOP_SET_VALUES_ADIFOR=74, lib/petsc/conf/test:testexamples_ADIFOR: ${TESTEXAMPLES_ADIFOR} lib/petsc/conf/test: -@ if test ${PRINT_PROGRESS}foo = debugtestfoo -a -n "${TESTEXAMPLES_ADIFOR}"; then printf "Running examples in `pwd`:\n ${TESTEXAMPLES_ADIFOR}\n"; fi lib/petsc/conf/test:buildexamples_ADIFOR: lib/petsc/conf/test: -@${OMAKE} testexamples_ADIFOR TESTEXAMPLES_ADIFOR=`echo ${TESTEXAMPLES_ADIFOR} | sed s/runex[0-9]*[a-z0-9_]*//g` src/docs/tao_tex/manual/mathprog.bib:@TechReport{bischof.carle.ea:adifor, src/docs/tao_tex/manual/mathprog.bib: title = "{ADIFOR} 2.0 User's Guide", src/docs/tex/petscapp.bib: title = "Using {ADIFOR} and {ADIC} to Provide a {J}acobian for the {SNES} Component of {PETSc}", src/mat/f90-mod/petscmat.h: PetscEnum MATOP_SET_VALUES_ADIFOR src/mat/f90-mod/petscmat.h: parameter(MATOP_SET_VALUES_ADIFOR=74) src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f -adifor_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type refine_always > ex5f_4.tmp 2>&1;\ src/snes/examples/tutorials/makefile: -@${MPIEXEC} -n 2 ./ex5f -adiformf_jacobian -snes_monitor_short -ksp_gmres_cgs_refinement_type refine_always > ex5f_5.tmp 2>&1;\ src/snes/examples/tutorials/makefile:TESTEXAMPLES_ADIFOR = ex5f.PETSc runex5f_2 ex5f.rm -- Stefano

1 0

Re: [petsc-dev] TSComputeRHSJacobian
by Stefano Zampini 08 Sep '17

08 Sep '17

I encountered the problem while working on the adjoint code. Here is the fix I have in my branch https://bitbucket.org/petsc/petsc/branch/stefano_zampini/feature-continuous… However, I'm not sure if this is a problem just with my code. This is why I'm asking on the list. 2017-09-08 0:02 GMT+03:00 Barry Smith <bsmith(a)mcs.anl.gov>: > > Example that demonstrates the current code is problematic in some cases? > > > On Aug 28, 2017, at 11:59 AM, Stefano Zampini <stefano.zampini(a)gmail.com> > wrote: > > > > > > I was wondering if the following code in TSComputeRHSJacobian is correct > when the user pass the reuse flag via TSRHSJacobianReuse > > > > if (ts->rhsjacobian.time == t && (ts->problem_type == TS_LINEAR || > (ts->rhsjacobian.Xid == Uid && ts->rhsjacobian.Xstate == Ustate)) && > (rhsfunction != TSComputeRHSFunctionLinear)) { > > PetscFunctionReturn(0); > > } > > > > Should the above code be replaced by the following? > > > > if (ts->rhsjacobian.time == t && (ts->problem_type == TS_LINEAR || > (ts->rhsjacobian.Xid == Uid && ts->rhsjacobian.Xstate == Ustate)) && > (rhsfunction != TSComputeRHSFunctionLinear)) { > > if (ts->rhsjacobian.reuse) { > > ierr = MatShift(A,-ts->rhsjacobian.shift);CHKERRQ(ierr); > > ierr = MatScale(A,1./ts->rhsjacobian.scale);CHKERRQ(ierr); > > if (B && A != B) { > > ierr = MatShift(B,-ts->rhsjacobian.shift);CHKERRQ(ierr); > > ierr = MatScale(B,1./ts->rhsjacobian.scale);CHKERRQ(ierr); > > } > > ts->rhsjacobian.shift = 0; > > ts->rhsjacobian.scale = 1.; > > } > > PetscFunctionReturn(0); > > } > > > > > > -- > > Stefano > > -- Stefano

1 0