* commit '4db0e9dc9a86259a8795f5043b36791bb007aacb':
Fixed a problem with AUTOTEST cleantest.sh script
Cleaned up a couple of minor things
Fix a bug with 2D red/black relaxation in PFMG
fix the test errors for bigint and cmake
fix the compiling problem for example 10
fix the compiling problem for ex10 with FEI
Added a regression test for C++ compile
fix the errors when merge the master branch
fix the inconsistent function definition in hypre_blas.h
fix the compiling errors in test and the inconsistent function definition in ParaSails
make hypre compatible with c++ compiler
In the case where there are no '.err' files, this script was deleting files it
should not have. Fixed by adding a check that each .err file actually exists.
In 2D, red/black relaxation was using the Z dimension to help define "red" and
"black" points, but the Z data was not initialized so the resulting method was
not predictable and was probably not even correct on more than one process.
This differs slightly from the original pull request in that compiling with MLI
is set to on by default. This agrees with the cmake build and mirrors what is
currently done with the fei and superlu as well.
* commit '2d69b4a5184db7185b2044200ccf77b9bf34c2fd':
Changing some of the runtest saved files
Minor styling changes.
Incorrectly added this to the wrong directory
Incorrectly added this to the wrong directory
Incorrectly added this to the wrong directory
Incorrectly added this to the wrong directory
Incorrectly added this to the wrong directory
Accidentally added this to the wrong directory
Add ability to set relative tolerance
Add ability to set relative tolerance
Add ability to set relative tolerance
The HYPRE interface to LOBPCG was only able to set the absolute tolerance even though internally LOBPCG can use both relative and absolute tolerance. The new interface can set either.
HYPRE_ame was incorrectly setting both absolute and relative tolerance to the same value. This is not always acceptable and can lead to very poor convergence. The new interface sets the two separately.
The last fix caused the earlier FEMAddValues problem to reappear. Duplicate
boxes are now explicitly eliminated in the SStructGridCreateCommInfo routine.
