[R] Rmpi building
sdavis2 at mail.nih.gov
Mon Oct 15 21:35:14 CEST 2007
Dirk Eddelbuettel wrote:
> On 15 October 2007 at 12:03, Sean Davis wrote:
> | I am trying to build Rmpi on Suse 10.2 linux. While I would like to use
> | the RPM version of liblam-7.1.2, I have not been able to do so. It
> | seems that Rmpi makes some pretty strong assumptions when trying to
> | guess the MPI version. Has anyone successfully built Rmpi against
> | lam-7.1.2 recently using R-2.6.0? If so, can you give some pointers
> | about where the libraries and header files need to be to have Rmpi "do
> | the right thing"? I have perused the configure file in the Rmpi
> | distribution without much success.
> On Debian/Ubuntu, I had to 'freeze' LAM at 7.1.1. It failed with the newer
> ones, incl 7.1.4. I never really looked too hard why ... as the plan was to
> migrate to Open MPI as soon as feasible. And I am happy to report that the
> newest Rmpi, with one small additional change that should appear on CRAN
> soon, builds fine against Open MPI. If you can, consider migrating from LAM
> to Open MPI.
It does. I would like to move over to OpenMPI but have been a bit
hesitant because of the support from most third-party softwares with LAM
and because of lack of documentation for OpenMPI. However, I have to
admit that I haven't looked hard at this option yet.
After some playing around, I got lam-7.1.3 compiled from source into
default locations (after assuring that -fPIC was used) to work with
Rmpi_0.5-4 by doing:
CC=mpicc R CMD INSTALL --clean Rmpi_0.5-4.tar.gz
Thanks for Martin Morgan for a reply off-list to use mpicc, which
removed the headache of trying to line up the directory locations correctly.
More information about the R-help