[R] installing Rmpi

[Prof Brian Ripley]

On Fri, 5 Oct 2007, Erin Hodgess wrote:

> Hi R people:
>
> I installed R-2.6.0 on a Red Hat Linux system and am now trying
> to install the Rmpi package.
>
> The installation cannot seem to find the mpi.h file, even though it exists.
>
> Is there a variable that I should be exporting into the process, please?

It's tricky, and we may need to know a bit more about your OS.  I presume 
you have the latest Rmpi 0.5-4: if not get it from the CRAN master (it 
may still be propagating).

On the RH systems I have seen the mpi.h file is in /usr/include/lam. So 
you need to ensure you have that installed, and it is in the lam-devel RPM 
(which has other dependencies, including lam and lam-libs).  If that is 
installed and this is still failing you need to look at the config.log to 
see why -- that's not R-help territory and I think you need to find some 
local expertise to help you: if that is not possible, use R-devel.

The problems I usually find are with the next step, finding the libraries. 
I've seen systems where /usr/lib64/lam was not in the ld.so cache, and 
others where you needed to link against -lutil as well as -lmpi -llam. 
Only someone with essentially the same OS on the same cputype will be able 
to help you find things like this.  (For some reaons Rmpi tests for static 
libraries on 32-bit systems and dynamic ones on 64-bit systems.)

Incidentally, the lam RPMs have things in quite different places from a 
vanilla source install of lam, so you do need to tell us exactly when your 
MPI libraries came from.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595