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

Re: [petsc-dev] MatSetDA -> MatSetDM
by Lisandro Dalcin 16 Mar '11

16 Mar '11

On 16 March 2011 22:47, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > On Mar 16, 2011, at 6:21 AM, Lisandro Dalcin wrote: > >> From petscdmda.h: >> >> extern PetscErrorCode MatSetDA(Mat,DM); >> >> Could you rename it? (sorry, I'm really busy with other stuff) > > Fixed. > Many thanks, >> >> BTW, we should have DMSetMatType() ... > > Actually my plan is to remove DMSetVecType() and change DMGetGlobalVector(), DMGGetLocalVector() and DMGetMatrix() to take an already created Vec or Mat and if the type is set inside that object then to use that type otherwise to use a default type. This will match the new style of VecLoad() and MatLoad(). > > The rational for this change is I want to remove all the MatCreate()s and VecCreates() and setting of types inside other parts of PETSc. This allows new Vec and Mat types to be introduced without also adding into DMGetMatrxi() etc new code whenever a new type is introduced. > I like it, but... Could we have DMGetComm(dm,&comm) ?... the casting required for PetscObjectGetComm((PetscObject)dm,&comm) is a bit ugly... -- 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] MatSetDA -> MatSetDM
by Barry Smith 16 Mar '11

16 Mar '11

On Mar 16, 2011, at 6:21 AM, Lisandro Dalcin wrote: > From petscdmda.h: > > extern PetscErrorCode MatSetDA(Mat,DM); > > Could you rename it? (sorry, I'm really busy with other stuff) Fixed. > > BTW, we should have DMSetMatType() ... Actually my plan is to remove DMSetVecType() and change DMGetGlobalVector(), DMGGetLocalVector() and DMGetMatrix() to take an already created Vec or Mat and if the type is set inside that object then to use that type otherwise to use a default type. This will match the new style of VecLoad() and MatLoad(). The rational for this change is I want to remove all the MatCreate()s and VecCreates() and setting of types inside other parts of PETSc. This allows new Vec and Mat types to be introduced without also adding into DMGetMatrxi() etc new code whenever a new type is introduced. 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] quad precision
by Barry Smith 16 Mar '11

16 Mar '11

On Feb 8, 2011, at 6:31 AM, Jed Brown wrote: > > GCC now ships with the LGPL-licensed libquadmath library, which provides for targets with a __float128 datatype quad-precision mathematical functions. __float128 is available for targets on 32-bit x86, x86-64 and Itanium architectures. The libquadmath library is automatically built on such targets when building the Fortran compiler. > > http://gcc.gnu.org/gcc-4.6/changes.html#libquadmath > http://gcc.gnu.org/onlinedocs/libquadmath/ > > Looks like this is something that actually works. GCC-4.6 is probably about two months away from a release, but this should be functional now. Perhaps it should be supported by petsc-3.2? Done (use ./configure --with-precision=__float128 --download-qblaslapack barry-smiths-macbook-pro:tutorials barrysmith$ petscmpiexec -n 2 ./ex19 -snes_monitor -ksp_monitor -ksp_rtol 1.e-30 -snes_rtol 1.e-28 lid velocity = Qe, prandtl # = Qe, grashof # = Qe 0 SNES Function norm 1.461936698223e-01 0 KSP Residual norm 1.461936698223e-01 1 KSP Residual norm 1.831093524630e-02 2 KSP Residual norm 3.842156479462e-03 3 KSP Residual norm 1.288406516834e-04 4 KSP Residual norm 6.151749539864e-06 5 KSP Residual norm 5.304145544729e-07 6 KSP Residual norm 1.885113023052e-08 7 KSP Residual norm 7.029816194041e-10 8 KSP Residual norm 5.197815114974e-11 9 KSP Residual norm 1.551090319297e-12 10 KSP Residual norm 6.563825928589e-14 11 KSP Residual norm 4.237666605992e-15 12 KSP Residual norm 1.991927370291e-16 13 KSP Residual norm 2.039546156320e-17 14 KSP Residual norm 1.399229447695e-17 15 KSP Residual norm 1.137432635542e-17 16 KSP Residual norm 9.814665753619e-18 17 KSP Residual norm 8.761318174145e-18 18 KSP Residual norm 7.987316427844e-18 19 KSP Residual norm 7.387832309954e-18 20 KSP Residual norm 6.905793284234e-18 21 KSP Residual norm 6.507264709320e-18 22 KSP Residual norm 6.170620606220e-18 23 KSP Residual norm 5.881342212029e-18 24 KSP Residual norm 5.629271709982e-18 25 KSP Residual norm 5.407060816706e-18 26 KSP Residual norm 5.209244641147e-18 27 KSP Residual norm 5.031663281118e-18 28 KSP Residual norm 4.871086689542e-18 29 KSP Residual norm 4.724963445290e-18 30 KSP Residual norm 1.945800083119e-17 31 KSP Residual norm 3.878029318777e-18 32 KSP Residual norm 8.021501106718e-19 33 KSP Residual norm 3.209251006205e-20 34 KSP Residual norm 9.713771954068e-22 35 KSP Residual norm 7.628163240222e-23 36 KSP Residual norm 4.108476441305e-24 37 KSP Residual norm 1.743186899274e-25 38 KSP Residual norm 1.115126800943e-26 39 KSP Residual norm 6.198813188878e-28 40 KSP Residual norm 2.869837519115e-29 41 KSP Residual norm 1.515585472682e-30 42 KSP Residual norm 5.770691059470e-32 1 SNES Function norm 2.507018921541e-05 0 KSP Residual norm 2.507018921541e-05 1 KSP Residual norm 7.131114735363e-07 2 KSP Residual norm 6.965532093127e-08 3 KSP Residual norm 8.087355337901e-09 4 KSP Residual norm 2.749048294679e-10 5 KSP Residual norm 1.113870556134e-11 6 KSP Residual norm 4.684216963041e-13 7 KSP Residual norm 4.911624900830e-14 8 KSP Residual norm 2.475960732046e-15 9 KSP Residual norm 1.012965264508e-16 10 KSP Residual norm 5.199268899437e-18 11 KSP Residual norm 2.361382990671e-19 12 KSP Residual norm 1.624218498133e-20 13 KSP Residual norm 1.626389274652e-21 14 KSP Residual norm 1.086627523198e-21 15 KSP Residual norm 8.796270682258e-22 16 KSP Residual norm 7.562576022670e-22 17 KSP Residual norm 6.738068994025e-22 18 KSP Residual norm 6.134859714231e-22 19 KSP Residual norm 5.669225669612e-22 20 KSP Residual norm 5.295704884674e-22 21 KSP Residual norm 4.987454291392e-22 22 KSP Residual norm 4.727440148233e-22 23 KSP Residual norm 4.504266287687e-22 24 KSP Residual norm 4.309981502786e-22 25 KSP Residual norm 4.138846776251e-22 26 KSP Residual norm 3.986602059865e-22 27 KSP Residual norm 3.850009759823e-22 28 KSP Residual norm 3.726559420703e-22 29 KSP Residual norm 3.614270387400e-22 30 KSP Residual norm 1.485730798507e-21 31 KSP Residual norm 5.734533095762e-23 32 KSP Residual norm 2.917983740862e-24 33 KSP Residual norm 2.415955278478e-25 34 KSP Residual norm 2.754121553256e-26 35 KSP Residual norm 1.530546077890e-27 36 KSP Residual norm 1.236527954824e-28 37 KSP Residual norm 5.968984750158e-30 38 KSP Residual norm 2.679693674423e-31 39 KSP Residual norm 2.728339059485e-32 40 KSP Residual norm 8.956127952264e-34 41 KSP Residual norm 2.732653737468e-35 42 KSP Residual norm 1.278928596713e-36 2 SNES Function norm 5.123354965471e-13 0 KSP Residual norm 5.123354965471e-13 1 KSP Residual norm 1.922138659280e-14 2 KSP Residual norm 1.916027080380e-15 3 KSP Residual norm 1.173166951366e-16 4 KSP Residual norm 6.562671109080e-18 5 KSP Residual norm 1.982740889768e-19 6 KSP Residual norm 1.593609248118e-20 7 KSP Residual norm 5.849655936465e-22 8 KSP Residual norm 2.871644899437e-23 9 KSP Residual norm 8.762292108710e-25 10 KSP Residual norm 2.883597656362e-26 11 KSP Residual norm 1.508069830626e-27 12 KSP Residual norm 9.552654188779e-29 13 KSP Residual norm 5.964437523019e-29 14 KSP Residual norm 4.773742821457e-29 15 KSP Residual norm 4.073114192114e-29 16 KSP Residual norm 3.613955807224e-29 17 KSP Residual norm 3.281398363062e-29 18 KSP Residual norm 3.026482504508e-29 19 KSP Residual norm 2.823023108355e-29 20 KSP Residual norm 2.655755272176e-29 21 KSP Residual norm 2.515082243565e-29 22 KSP Residual norm 2.394629606506e-29 23 KSP Residual norm 2.289975422292e-29 24 KSP Residual norm 2.197943014798e-29 25 KSP Residual norm 2.116183578028e-29 26 KSP Residual norm 2.042917954989e-29 27 KSP Residual norm 1.976770491487e-29 28 KSP Residual norm 1.916658520825e-29 29 KSP Residual norm 1.861716726173e-29 30 KSP Residual norm 7.847792961992e-29 31 KSP Residual norm 3.265251638392e-30 32 KSP Residual norm 2.554035610120e-31 33 KSP Residual norm 1.708749092417e-32 34 KSP Residual norm 1.088457747082e-33 35 KSP Residual norm 4.004032418947e-35 36 KSP Residual norm 3.626864425969e-36 37 KSP Residual norm 1.998928979977e-37 38 KSP Residual norm 5.929794876014e-39 39 KSP Residual norm 3.867971851833e-40 40 KSP Residual norm 2.306219660368e-41 41 KSP Residual norm 1.109502994834e-42 42 KSP Residual norm 7.010253552905e-44 3 SNES Function norm 1.676590295787e-28

1 0

Re: [petsc-dev] How to respond to incompatible right/left preconditioned/unpreconditioned?
by Jed Brown 16 Mar '11

16 Mar '11

On Wed, Mar 16, 2011 at 22:53, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > This is an oversight, it should generate an error. > It does now. > > So what if instead we had KSP_NORM_DEFAULT and PC_SIDE_DEFAULT, then > KSPSetUp_XXX() would turn the default value into whatever actually works and > error if the user has set an incompatible combination? > > Sounds reasonable. > I haven't done this yet so right now users have to change the norm type explicitly when they switch sides.

1 0

Re: [petsc-dev] How to respond to incompatible right/left preconditioned/unpreconditioned?
by Barry Smith 16 Mar '11

16 Mar '11

On Mar 16, 2011, at 4:40 PM, Jed Brown wrote: > So currently, I can do > > $ ./ex2 -ksp_pc_side right -ksp_norm_type preconditioned > > which is nonsense (it actually uses the unpreconditioned norm). We could just add a guard in KSPSolve_GMRES to handle this case, but then This is an oversight, it should generate an error. > > ./ex2 -ksp_pc_side right > > fails. It would be a pain to have two options that always need to be changed together so maybe calling KSPSetPCSide() should automatically choose which side to use. But that isn't consistent across methods and it's really bad to ignore explicit user input. Also, what if you just want to think about what norm the residuals are calculated in instead of thinking of which "side" the preconditioner is working with. > > So what if instead we had KSP_NORM_DEFAULT and PC_SIDE_DEFAULT, then KSPSetUp_XXX() would turn the default value into whatever actually works and error if the user has set an incompatible combination? Sounds reasonable. Barry

1 0

How to respond to incompatible right/left preconditioned/unpreconditioned?
by Jed Brown 16 Mar '11

16 Mar '11

So currently, I can do $ ./ex2 -ksp_pc_side right -ksp_norm_type preconditioned which is nonsense (it actually uses the unpreconditioned norm). We could just add a guard in KSPSolve_GMRES to handle this case, but then ./ex2 -ksp_pc_side right fails. It would be a pain to have two options that always need to be changed together so maybe calling KSPSetPCSide() should automatically choose which side to use. But that isn't consistent across methods and it's really bad to ignore explicit user input. Also, what if you just want to think about what norm the residuals are calculated in instead of thinking of which "side" the preconditioner is working with. So what if instead we had KSP_NORM_DEFAULT and PC_SIDE_DEFAULT, then KSPSetUp_XXX() would turn the default value into whatever actually works and error if the user has set an incompatible combination?

1 0

Re: [petsc-dev] Preconditioner side
by Jed Brown 16 Mar '11

16 Mar '11

On Wed, Mar 16, 2011 at 22:13, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > I etag hunted down one -ksp_preconditioner_side that I have fixed. Well, it was kind of the most important one, being the only non-documentation occurrence and all. src/ksp/ksp/interface/itcl.c: ierr = PetscOptionsEList("-ksp_preconditioner_side","KSP preconditioner side","KSPSetPCSide",PCSides,3,PCSides[ksp->pc_side],&indx,&flg);CHKERRQ(ierr);

1 0

Re: [petsc-dev] Preconditioner side
by Barry Smith 16 Mar '11

16 Mar '11

On Mar 16, 2011, at 4:08 PM, Jed Brown wrote: > On Fri, Feb 25, 2011 at 16:14, Hong Zhang <hzhang(a)mcs.anl.gov> wrote: > -ksp_preconditioner_side <LEFT> (choose one of) LEFT RIGHT SYMMETRIC > (KSPSetPCSide) > > Which is it going to be? > > -ksp_pc_side (KSPSetPCSide) It should always be this. > > or > > -ksp_preconditioner_side (KSPSetPreconditionerSide) > > There is still a mix and the documentation is inconsistent. I etag hunted down one -ksp_preconditioner_side that I have fixed. Etags did not find any other in correct ones. Barry

1 0

Re: [petsc-dev] Preconditioner side
by Jed Brown 16 Mar '11

16 Mar '11

On Fri, Feb 25, 2011 at 16:14, Hong Zhang <hzhang(a)mcs.anl.gov> wrote: > -ksp_preconditioner_side <LEFT> (choose one of) LEFT RIGHT SYMMETRIC > (KSPSetPCSide) > Which is it going to be? -ksp_pc_side (KSPSetPCSide) or -ksp_preconditioner_side (KSPSetPreconditionerSide) There is still a mix and the documentation is inconsistent.

1 0

Re: [petsc-dev] Should pARMS converge in parallel?
by Jed Brown 16 Mar '11

16 Mar '11

On Wed, Mar 16, 2011 at 21:49, Jose E. Roman <jroman(a)dsic.upv.es> wrote: > Yes, I am aware of this. I added a note to the PCPARMS manpage stating that > this preconditioner should be used with flexible methods (fgmres, gcr). Aha, I should have read it. Still seems like a common gotcha so defaulting to linear would be nice.

1 0