[MPICH] Problem compiling mpich2-1.0.2p1 on FreeBSD 5.4

[Philip Sydney Lavers]

---- Original message ----
>Date: Wed, 10 Aug 2005 09:47:42 +0100
>From: Yann Golanski <yann at kierun.org>  
>Subject: [MPICH] Problem compiling mpich2-1.0.2p1 on FreeBSD
5.4  
>To: mpich-discuss at mcs.anl.gov

>Subject says it all really... the rest is just more
information.  Any

Hello Yann   and folks

I understand your problem. I have just upgraded my nodes from
FreeBSD 5.3 to 5.4 (64 bit) && MPICH2 from 2-1.0.1 to 2-1.0.2p1.

Siiiiigh - one wonders why one does these things. My perfectly
happy mpi cluster turned into a frustrating tangle.

I now have it working again, including two Linux 64 bit nodes.

The front end machine is dual opteron, the others are athlon64
, both 754 and 939 pin.

mpich2-1.0.2p1 installed straight away on the Linux machines,
but did not work until I fixed up the /etc/hosts file to make
sure that the short name (i.e. no domain qualifications) -e.g
claude6 and claude7 in my case - ONLY appeared on the line
containing their local IP adress - e.g 192.168.1.16 and
192.168.1.17 - and NOT on the localhost line where the
Mandrake installation automatically also puts it.

mpich2-1.0.2p1 would not compile on FreeBSD5.4 machines until
I used the --with-enablef90=no switch.

Here is my procedure in the bash shell, taking into account
that I do my debugging on another cluster - I am trying to get
numbercrunching speed out of this one. Replace "athlon64" with
"opteron" or whatever where appropriate:
in the downloaded unzipped and tar -xvf directory:

export CFLAGS='-O3 -march=athlon64' FFLAGS='-O3
-march=athlon64' CXXFLAGS='-O3 -march=athlon64'

./configure --prefix=/usr/local/mpich --with-fast
--with-enablef90=no

make

su

make install

This worked fine for me, and the cluster is now back to work.

The problem was that there is an inbuilt shell command fc in
the FreeBSD path that configure was tryng to use as an f90
compiler.

I also installed LAM on the Linux machines so that I can pull
some Solaris nodes into the cluster - no problems BUT
LAM is fractured for FreeBSD. It installs and boots but does
not run programmes. I even tried the oficial FreeBSD port.
Same problem - that stupid message about at least one of the
processes not using MPI_INIT (or whatever, I am not near the
machines now) and laminfo causes a core dump.

I hope this helps,

regards,

Philip Lavers