Commits

Show all
Author Commit Message Labels Comments Date
Stefano Zampini
MatMUMPS: forgot to stash the leading dimension of the previous rhs (if any) and set a new one
Stefano Zampini
PCHYPRE: in case of zero conductivities in the whole domain, we need to pass null to HYPRE_AMSSetBetaPoisson
Karl Rupp
MatrixMarket: Fixed output for complex-valued matrices. Output was not conforming to standard (spurious comma). Reported-by: Torquil Sørensen <torquil@gmail.com> via petsc-maint
Stefano Zampini
MatIS: move PetscLayoutSetUp out of MatCreate_IS the code for MatIS was broken using the sequence MatCreate(comm,&A); MatSetType(A,MATIS);
BarryFSmith
If number of local rows is zero then short-circuite the calculations for brows and bcols since otherwise leads to divide by zero Reported-by: Marcus Day <msday@lbl.gov>
BarryFSmith
Merge branch 'maint' of bitbucket.org:petsc/petsc into maint
BarryFSmith
--download-cmake did not work correctly with --prefix Reported-by: Roberto Torres <roberto.torres@uam.es>
Satish Balay
Merge branch 'balay/fix-sowing-valgrind' into maint
sarich
Merge branch 'sarich/fix-tao-armijo-ls-reason' into maint
Satish Balay
petsc4py: update to snapshot with cygwin build fixes
BarryFSmith
documentation for PetscViewerASCIIOpen() and MatView() and VecView() was WRONG for use with a viewer that comes from PETSC_COMM_SELF. The documentation mistakenly said that the output was independent for each process when in fact the communicator of the viewer must match that of the object. Reported-by: Longyin Cui <cuilongyin@gmail.com>
BarryFSmith
change PetscLogStageGetId() to return -1 if no stage exists
Satish Balay
sowing: fix valgrind warnings
Emil Constantinescu
added one more change to 3.6
BarryFSmith
added PCCompositeGetNumberPC() to provide number of PC in a composite PC Provided-by: Andrew Cramer <a.cramer@uq.edu.au>
BarryFSmith
mistakenly passed a complex number into a printf()
BarryFSmith
-eigest_ksp_type (and other -eigest_options) did not work if the user did not specifically set Chebyshev to use an eigenestimator since the setting of Chebyshev to use the eigest happened in KSPSetUp_Chebyshev(). Reported-by: Garth N. Wells <gnw20@cam.ac.uk>
BarryFSmith
Fix for DMLocalToLocalBegin/End() introduced by commit ef605946e53f07e3ad39ab377ff50dceade0a2df the code mistakenly used to->local.n_nonmatching even if it had not been initialed Reported in issue #100 Reported-by: Constantine Khroulev <ckhroulev@alaska.edu>
BarryFSmith
DMDACnvertToCell spelled incorrectly in function macro definition Reported-by: Mark L. Stowell <stowell1@llnl.gov>
BarryFSmith
Merged in qulogic/petsc/pcshell-fortran-indices (pull request #333) Fix incorrect indices in PCShell Fortran wrappers.
Elliott Sales de Andrade
Fix incorrect indices in PCShell Fortran wrappers. Some indices overlapped, and in some cases, the size of the allocated function point array was too small.
BarryFSmith
The web address given in the LU zero pivot error message has the wrong case, and doesn't lead me directly to the correct FAQ item. To fix it, change from "ZeroPivot" to "zeropivot" at the ends of the lines 234 and 244 in src/sys/error/err.c Reported-by: Torquil Macdonald Sørensen <torquil@gmail.com>
BarryFSmith
Comment documentation in example was missing term in Jacobian Reported-by: Fahad Hasan fahadhasanchem@gmail.com
BarryFSmith
fixed bound used to decide when iterative refinement is needed for classical Gram-Shmidt algorithm: Line 93 of ./src/ksp/ksp/impls/gmres/borthog2.c reads: if (wnrm < 1.0286 * hnrm) { I suggest to change it to: if (wnrm < hnrm) { The constant 1.0286 is kind of weird in this context. I think I can trace the 1.0286 in PETSc ( instead of 1 ) back to ARPACK. ARPACK intended to use SQRT( 2 ) which is a standard value, that lots of numerical codes tend to agree upon. Someone in the ARPACK team har…
BarryFSmith
removed discussion of threads + mpi from website home page and features/threads.html since that functionality has been removed from PETSc
BarryFSmith
In the users guide, section 11.4 mentions Matlab classes, which have been removed. The changes file does not mention that user makefiles have to be updated since the include path has changed. Related to this, in chapter 16 of the Users manual some paths have been updated but not others. Reported-by: Jose E. Roman <jroman@dsic.upv.es>
sarich
return successful ls codes
Satish Balay
openmpi: fix version check. do not define HAVE_OMPI_MAJOR_VERSION if cpp output is not parsable Reported-by: Florian Lindner <mailinglists@xgm.de> Tested-by: Florian Lindner <mailinglists@xgm.de>
BarryFSmith
remove nonexistent -pc_mg_monitor from documentation
Stefano Zampini
PCBDDC: fixes when reusing the preconditioner with adaptive selection
  1. Prev
  2. Next