add providing -DMPI_INCLUDE_PATH for parmetis cmake if the MPI compilers are not being used

Merge branch 'maint'

Conflicts:
config/BuildSystem/config/packages/Chaco.py

ctetgen: fix build to work with both single-library=0/1
Also fix to build library in ctetgen/ not LIBDIR of PETSc.
And use 'make clean install' targets from ctetgen makefile

Use nvcc found by PETSc to generate guess for cuda dir.

This way we avoid inconsistencies between the cuda found by
generateCUDACompilerGuesses and cuda.py.

Increase minimum cusp version to 0.4

This makes the checks for smoothed aggregation obsolete so I'm removing
those.

Increase minimum cuda version to 4.2

Try to find nvcc based on --with-cuda-dir.

With this code we try to find nvcc in the location specified with
--with-cuda-dir before trying the system nvcc in $PATH. This is to
make sure that nvcc i

Try to find nvcc based on --with-cuda-dir.

With this code we try to find nvcc in the location specified with
--with-cuda-dir before trying the system nvcc in $PATH. This is to
make sure that nvcc is consistent with the version of the cuda toolkit
we're configuring with. nvcc's for different versions of the cuda
toolkit are incompatible with one another.

show more ...

Guess cuda directory based on system nvcc.

Check whether thrust/version.h is available in cuda.py

Remove thrust and cusp traces from cuda.py

Remove thrust.py

thrust is part of the cuda toolkit.

Merge branch 'barry/hwloc'

when configure fails on hwloc just keep on going anyways

Merge branch 'balay/update-hdf5'

download hwloc by default if not a batch system
try to use hwloc to print information on the machine when make streams is run

added run of hwloc with make streams when possible

Merge branch 'barry/fix-prefix-installs'

Conflicts:
config/BuildSystem/config/packages/__init__.py
config/BuildSystem/config/packages/hdf5.py
config/BuildSystem/config/packages/make.py

Thrust.py in correctly always required cuda even if thrust was not enabled

Merge branch 'fix-thrust-configuration' into barry/fix-prefix-installs

Use cuda device architecture sm_20 by default.

This is needed because nvcc in cuda 6.5 generates a deprecation warning
with older compute architectures which makes the nvcc tests fail.

Fix thrust configuration

Thrust is part of the cuda distribution. Thus the thrust location can
be found based on the cuda path. In addition, this commit fixes the
dependencies between cuda, thrust

Fix thrust configuration

Thrust is part of the cuda distribution. Thus the thrust location can
be found based on the cuda path. In addition, this commit fixes the
dependencies between cuda, thrust, and cusp.

- cuda.py:
. Fix instructions for getting cusp
. Delete instructions for getting thrust since it's shipped with cuda
. Remove obsolete -with-cuda-only argument (thrust always comes along
for the ride)
. Remove functions for checking for cusp and thrust (Having these in
cuda inverts the dependencies between the packages)
- thrust.py:
. Add cuda as a dependency
. Use cuda location as first guess for finding thrust
. Automatically set -with-thrust if cuda has been found.

show more ...

define PETSC_DIR when making ctetgen.

ctetgen now uses ${PETSC_DIR}/conf/rules, etc.

fixed passing of Fortran compiler name to CMAKE
and CXX and Fortran MPI compilers to CMAKE for elemental

update Zoltan.py to build for 64 bit indices.
Since no one was nice enough to include any tests for Zoltan I do not know if it in fact works for 64 bit indices or not :-(

Update to patched version of Sundials that always uses ${MAKE}
Do not use parallel make for sundials since it uses recursive make