We have a new Semi-Lagrangian advection solver using the method of characteristics to the
src/contrib/semiLagrange, contributed by Richard Katz and Marc Spiegelman of Columbia University. Examples are included.
Added C99 complex support. [check configure changes below]
Added MatCreateScatter() and MATSCATTER to easily construct efficient “restriction operators”
Removed MatConvertRegister() and friends, added
(*convertfrom)to Mat function table to replace it
Added PCGALERKIN and PCGalerkinSetRestriction/Interpolation() to allow easy creating of Galerkin-like preconditioners.
Added support for the KSP/PC to auto-create the Mat(s) for you. (Which you usally set with KSP/PCSetOperators()). See manual page for KSP/PCGetOperators(). Note this is slightly speculative code currently.
PCSetOperators()/KSPSetOperators() now do reference counting on the operators you pass in. Current usage will work, however this means you NO LONGER need to keep a reference to the operators around until the KSP/PC object is destroyed.
Added KSPDefaultConvergedSetUIRNorm() and -ksp_default_converged_initial_residual_norm to go back to the pre PETSc 2.3.1 of determing relative tolerance for convergence.
Now c++ complex version requires the options’ ‘–with-scalar-type=complex –with-clanguage=cxx’. Specyfing only ‘–with-scalar-type=complex’ will default to ‘–with-clanguage=c’ i.e uses C99 complex support.
The convergence test functions for SNES now pass the current iteration in as the second argument. Also the convergence test routine is called after the initial function evaluation in SNES but before the first Newton step is computed.
Removed PetscRandomType from the prototype of PetscRandomCreate().
Added PetscRandomSetType(), PetscRandomGetValueReal() and PetscRandomGetValueImaginary().
Replaced RANDOM_DEFAULT, RANDOM_DEFAULT_REAL and RANDOM_DEFAULT_IMAGINARY, the type of random numbers, with PETSCRAND48, PETSCRAND etc.
AODATA object and associated routines are now deprecated and scheduled for future removal
SPRNG support added.