dorsal team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Aug 19, 2010 at 04:39:11PM -0400, Paul Barbone wrote:
> Paul Barbone wrote:
> >Anders Logg wrote:
> >>On Thu, Aug 19, 2010 at 03:48:12PM -0400, Paul Barbone wrote:
> >>>Anders Logg wrote:
> >>>>On Thu, Aug 19, 2010 at 02:46:47PM -0400, Paul Barbone wrote:
> >>>>>Anders Logg wrote:
> >>>>>>On Thu, Aug 19, 2010 at 07:03:51PM +0100, Garth N. Wells wrote:
> >>>>>>>On Thu, 2010-08-19 at 17:58 +0000, noreply@xxxxxxxxxxxxx wrote:
> >>>>>>>>------------------------------------------------------------
> >>>>>>>>revno: 390
> >>>>>>>>committer: Anders Logg <logg@xxxxxxxxx>
> >>>>>>>>branch nick: dorsal
> >>>>>>>>timestamp: Thu 2010-08-19 19:55:20 +0200
> >>>>>>>>message:
> >>>>>>>>Remove hypre, scalapack and blacs from petsc.package for now,
> >>>>>>>>not working on this shitty OS becuase of Fortran issues.
> >>>>>>>They're nice to have - can you just install a Fortran compiler?
> >>>>>>>
> >>>>>>>Garth
> >>>>>>I tried installing g95 from macports but it just hung. But even if I
> >>>>>>do that, I probably need to build MPI with Fortran support which
> >>>>>>probably will lead to new problems.
> >>>>>>
> >>>>>>Has anyone tried getting PETSc with Fortfran (for Hypre)
> >>>>>>to work on OS
> >>>>>>X?
> >>>>>>
> >>>>>>If we can't get it to work, we'll configure it so that a OS X gets a
> >>>>>>stripped down version and the rest of us can enjoy a full-featured
> >>>>>>PETSc.
> >>>>>>
> >>>>>>
> >>>>>Maybe....  With Garth's recent changes, dolfin just compiled on my
> >>>>>leopard system.  (Thanks Garth!)  Still working through the rest of
> >>>>>the packages...
> >>>>>
> >>>>>About petsc:  I'm not sure what you mean by "work", but if "work"
> >>>>>means "compile", then yes.  With...
> >>>>>
> >>>>>export CC=gcc-mp-4.4
> >>>>>export CXX=g++-mp-4.4
> >>>>>
> >>>>>...defined in the *.platform file.  As I'm not yet in a position to
> >>>>>test, I can't tell if it "works".
> >>>>>
> >>>>>-Paul
> >>>>Don't you need MPI compilers?
> >>>>
> >>>>Are you using -download-hypre=1?
> >>>>
> >>>>That fails for me since Hypre requires that MPI was compiled with
> >>>>Fortran support. Did you compile MPI yourself?
> >>>>
> >>>So, take whatever I tell you with a pinch of salt, because I don't
> >>>quite know what I'm doing.  Having said that, the following
> >>>*.platform file using stable_build=false "installed" and eventually
> >>>gave the message:
> >>>
> >>>---------------------------------------------------------
> >>>DOLFIN successfully compiled and installed in
> >>>
> >>> /Users/barbone/Research/Computation/FEniCS/
> >>>
> >>>---------------------------------------------------------
> >>>
> >>>I have yet to test it...
> >>>
> >>>Here's the *.platform file that I used on my leopard system:  (Note
> >>>the ugly hardwired ARMADILLO_DIR setting.)
> >>>
> >>>---------------------------------------------------------
> >>>
> >>># Mac OS X 10.5 with MacPorts
> >>>
> >>># This build script assumes that you have installed Xcode.app and
> >>># X11.app from Mac OS's original install disc. This gives you access
> >>># to a recent gcc, among other things. In order to have easy access to
> >>># a lot of useful Free Software, this script also assumes you have
> >>># installed MacPorts (from http://www.macports.org/).
> >>>#
> >>># Via Macports, you need to install the following packages:
> >>>#
> >>># sudo port install wget autogen readline pkgconfig libxml2 libtool \
> >>># cgal gmp zlib python26 swig-python py26-numpy bzr bzrtools git-core \
> >>># scons ginac boost +python26 vtk-devel +python26 suitesparse \
> >>># py26-scientific
> >>>#
> >>># Before proceeding, you need to make sure that Python 2.6 installed
> >>># by MacPorts is the default one on your machine:
> >>>#
> >>># sudo port install python_select # sudo python_select python26
> >>>#
> >>>##
> >>>
> >>>
> >>># Platform specific variables
> >>>default PETSC_ARCH=darwin9.2.0-cxx-opt
> >>>default GMP_ARCH=none-apple-darwin
> >>># default SCOTCH_ARCH=i686_mac_darwin8
> >>># default UMFPACK_DIR=/sw
> >>># default BOOST_DIR=/sw
> >>>
> >>># default PETSC_ARCH=darwin10.0.0-cxx-opt
> >>>default SCOTCH_ARCH=i686_mac_darwin8
> >>>default BOOST_DIR=/opt/local
> >>>default UMFPACK_DIR=/opt/local
> >>>default CGAL_DIR=/opt/local
> >>>default GMP_DIR=/opt/local
> >>># PEB ADDED - SYSTEM SPECIFIC:
> >>># default ARMADILLO_DIR=/opt/local
> >>>default ARMADILLO_DIR=/Users/barbone/Research/Computation/FEniCS
> >>># END PEB ADDED-----------------------------------------
> >>>
> >>>
> >>># export CC=gcc-4
> >>># export CXX=g++4
> >>>export CC=gcc-mp-4.4
> >>>export CXX=g++-mp-4.4
> >>>
> >>>
> >>>export PKG_CONFIG_PATH=/opt/local/lib/pkgconfig:${PKG_CONFIG_PATH}
> >>># export PKG_CONFIG_PATH=/sw/lib/pkgconfig
> >>>
> >>>
> >>># Define the packages this platform needs (for 10.5 ?)
> >>>PACKAGES=(
> >>>petsc
> >>>slepc
> >>>armadillo
> >>>mtl
> >>>fiat
> >>>ferari
> >>>ufc
> >>>ufl
> >>>ffc
> >>>viper
> >>>instant
> >>>dolfin
> >>>)
> >>>
> >>>---------------------------------------------------------
> >>>
> >>>
> >>>
> >>
> >>What's in your petsc.package? Which options does it use for PETSc?
> >>
> >>
> >
> >No hypre, apparently.  Here it is:
> >
> >
> >NAME=petsc-3.0.0-p12
> >SOURCE=http://ftp.mcs.anl.gov/pub/petsc/release-snapshots/
> >PACKING=.tar.gz
> >BUILDCHAIN=autotools
> >
> >package_specific_setup () {
> >   export PETSC_DIR=${PWD}
> >   ./config/configure.py COPTFLAGS=-O3 --with-debugging=0
> >--with-shared=1 --with-clanguage
> >=cxx --download-umfpack=1 --prefix=${INSTALL_PATH}
> >}
> >
> >package_specific_install () {
> >   if [ -e ${INSTALL_PATH}/lib/${PETSC_ARCH} ]
> >   then
> >       ln -sf ${INSTALL_PATH}/lib/${PETSC_ARCH}/* ${INSTALL_PATH}/lib/
> >   fi
> >}
> >
> >package_specific_register () {
> >   export PETSC_DIR=${INSTALL_PATH}
> >   export PETSC_ARCH_TEMP=${PETSC_ARCH}
> >   unset PETSC_ARCH
> >}
> >
> >
>
> Seems to be working :-)...  Again, this is on leopard (10.5), not
> snowleopard (10.6).   I had to add to dolfin.conf the line:
>
> export BOOST_DIR=/opt/local
>
> Afterward, it seemed to run ok.  I've tried only Poisson and
> nonlinear poisson so far, and only through python.   The 'q' didn't
> work on the interactive graphics, but I can live with that... :-)
>
> Thanks Garth and Harish for endless patience.   Anders - let me know
> if I can provide more info...
>
> Note that my installation doesn't use scotch, swiginac, trilinos, or
> syfi.

Good to hear!

If you want, you could experiment with adding --download-hypre=1 in
the petsc.package file. I assume you will run into the same problem
that I couldn't solve (needing to get an MPI build with Fortran). Let
me know if/how it works.

--
Anders

Attachment: signature.asc
Description: Digital signature