remove PCFactorSetShiftNonzero(), PCFactorSetShiftPd() and PCFactorSetShiftInBlocks()

Hg-commit: 7003e666813cb29616df228e25ece92955c33033

petscblaslapack.h now included by petscsys.h
some fixes to builder.py

Hg-commit: 56bbc3637d96c6c99b427f2d14f38bcf58262255

continue replacing PCFactorSetShiftxxx() with PCFactorSetShiftType() and PCFactorSetShiftAmount()

Hg-commit: b64c17c9fbc009eefb8072446fc4ef9e80146566

replace shiftpd, shiftnz, shiftinblocs with shifttype and shiftamount
only partially done for seqaij format.

Hg-commit: e8ba899db4d012e7c2e1e0b2afe23bf80c595415

Add PCFactorSetShiftType() and PCFactorSetShiftAmount()

Hg-commit: 11b2d9da5dd711b89de2445adb08f29eaa980990

Bugfix for shift schemes used in numerical factorization.
Turn-off default shiftnz/shiftpd in icc and ilu until a better design is determined
Only one of shiftnz, shiftpd and shiftinblocks is allowed

Bugfix for shift schemes used in numerical factorization.
Turn-off default shiftnz/shiftpd in icc and ilu until a better design is determined
Only one of shiftnz, shiftpd and shiftinblocks is allowed to be turned on.

Hg-commit: 9fab294fdd73e23ef93a1c65bfd18772aaab2b91

show more ...

petsc.h --> petscsys.h (fortran equivalents also)
new petsc.h that includes all PETSc includes is added

Hg-commit: a82a9bcb80ed39c69991db706672838a5b792d44

moved petscsys.h stuff into petsc.h files (Fortran and C) removed petscsys.h files fortran and C
eliminate use of petscsys.h (Fortran and C)
eliminated redundant use of #include "petscfix" all over t

moved petscsys.h stuff into petsc.h files (Fortran and C) removed petscsys.h files fortran and C
eliminate use of petscsys.h (Fortran and C)
eliminated redundant use of #include "petscfix" all over the place because included by petsc.h

Hg-commit: dea4bc8a57aa0490c3a40fdee08c4b41e42e03ce

show more ...

dropped the use from PCFactorSetUseDropTolerance() and -pc_factor_use_drop_tolerance

Hg-commit: 38537a65d99317cb3764af5eac895154a7e2f605

fix extern "C" error icc.c

Hg-commit: 36824b7f3a6877fb76d2b75cd2cf21d6835aa7a4

Fixed missing extern C

Hg-commit: b10df184561da314aea2bb7b4d4703a833d1316d

added support for drop tolerance to PCICC
first cut of Bas ICC, not yet completely working.

Hg-commit: c81908c7b1f7211c4c768df4016c15cc490189b8

removed public interface for MatILUDTFactor*, this will now be handled by regular MatILU/ICCFactor* interface

Hg-commit: b130c3ab430985359fffc88a9c93110f07b98d13

basic running of bas factorization
typo in hdf5 manaul page
credit Dave May for the GCR

Hg-commit: a91a506f6e78f1175807298f0b0acb3a753ac329

Added in hypre sstruct and syspfmg linkage

Hg-commit: 63e941bd7b8b26d9b300cd92d4c3bf42bc09b5cb

improvements to PetscOpenMPMerge() to allow it to be run inside the program and an example of its use
updates to lsqr.c with preconditioning to more stable form

Hg-commit: 2d51898f92e225dc9c39f6963f

improvements to PetscOpenMPMerge() to allow it to be run inside the program and an example of its use
updates to lsqr.c with preconditioning to more stable form

Hg-commit: 2d51898f92e225dc9c39f6963fd52521b394fa6f

show more ...

minor clean up of ML interface code, making private functions static, added __FUNC__ name

Hg-commit: d447c91ad9627c8804abd2623256eaf296f373ef

removed totally out of date feti code from pc/impls/is/feti
it still used SLES :-(

Hg-commit: b2165070c15dbadc9a645a2222700f48551747e3

bug fix for ML preconditioner, since ML could change vector sizes need to totally destroy PC_MG object and rebuild it for each new ML build.
Re: [petsc-maint #40882] Re: ML changing the size of the c

bug fix for ML preconditioner, since ML could change vector sizes need to totally destroy PC_MG object and rebuild it for each new ML build.
Re: [petsc-maint #40882] Re: ML changing the size of the coarse-level problem

Hg-commit: 37aef68e14f14c948b13164c3e47037069bee603

show more ...

removed some dead unreachable code
removed meshpflotran.c since not used

Hg-commit: 62a79e979f9ab663f4833a5b897c938e2fcd870a

Set operators on mg_levels_0_ksp for single-level PCMG

This patch should fix the following problem:

$ ./ex19 -lidvelocity 100 -dmmg_grid_sequence -mg_levels_0_ksp_monitor -mg_levels_0_ksp_converged

Set operators on mg_levels_0_ksp for single-level PCMG

This patch should fix the following problem:

$ ./ex19 -lidvelocity 100 -dmmg_grid_sequence -mg_levels_0_ksp_monitor -mg_levels_0_ksp_converged_reason -snes_converged_reason -snes_monitor
lid velocity = 100, prandtl # = 1, grashof # = 1
0 SNES Function norm 1.414215308315e+02
Residual norms for mg_levels_0_ solve.
0 KSP Residual norm 3.774528917219e+02
1 KSP Residual norm 6.346139612437e-14
Linear solve converged due to CONVERGED_RTOL iterations 1
1 SNES Function norm 1.275449881702e+02
Residual norms for mg_levels_0_ solve.
0 KSP Residual norm 3.391497818757e+02
1 KSP Residual norm 3.987425425283e+01
Linear solve did not converge due to DIVERGED_ITS iterations 1
2 SNES Function norm 1.182495633342e+02
Residual norms for mg_levels_0_ solve.
0 KSP Residual norm 3.053413184804e+02
1 KSP Residual norm 6.330531581879e+01
Linear solve did not converge due to DIVERGED_ITS iterations 1
3 SNES Function norm 1.158742303075e+02
Residual norms for mg_levels_0_ solve.
0 KSP Residual norm 2.759245022712e+02
1 KSP Residual norm 7.545828124034e+01
Linear solve did not converge due to DIVERGED_ITS iterations 1
Nonlinear solve did not converge due to DIVERGED_LS_FAILURE
0 SNES Function norm 2.019656949307e+02
1 SNES Function norm 1.714196530675e+02
2 SNES Function norm 1.193750806523e+02
3 SNES Function norm 6.464354333802e+01
4 SNES Function norm 3.034725334143e-01
5 SNES Function norm 1.014781939751e-05
6 SNES Function norm 2.242251270896e-11
Nonlinear solve converged due to CONVERGED_FNORM_RELATIVE
Number of Newton iterations = 6


Note that SNES does not converge on level 0, which is because the
operator never gets updated (one stale LU isn't good enough for the KSP
to converge). That is, KSPSetOperators for mg_levels_0_ksp is only
called on the first Newton iteration.


This patch just adds a special case for when the coarse level is equal
to the fine level.

Hg-commit: 9c24c36f89bf0df309dc885082720d7fc7a15097

show more ...

PCML no longer tries to use data structures for SAME_NONZERO_PATTERN because ML may construct different size and nonzero structures for coarser levels even with the same nonzero pattern on finest. Bu

PCML no longer tries to use data structures for SAME_NONZERO_PATTERN because ML may construct different size and nonzero structures for coarser levels even with the same nonzero pattern on finest. Bug report Jed Brown ./ex48 -M 6 -P 4 -thi_mat_type aij -pc_type ml -pc_ml_maxnlevels 3

Hg-commit: fdad64742f328b16a4c4868cd879727a07ae5b7e

show more ...

general cleanup of ml.c source code

Hg-commit: ff75809292e9a437d2b0aab78697cd2bf674b0d3

updated PCML to use innerctx of PC_MG instead of PetscContainer

Hg-commit: 510c3c0a85a55c107431fbc0fde1b79e6ac1a748

error in PCREDISTRIBUTE when preconditioner is reused with same nonzero pattern, tmat was mistakenly destroyed

Hg-commit: 7062b3cc2ae5af39cbecc8e94d10d46c856c2397