Re: CCL:CHARMm and MPICH

From: Ben Webb <ben (- at -) bellatrix.pcl.ox.ac.uk>
Subject: Re: CCL:CHARMm and MPICH
Date: Sun, 17 Dec 2000 14:15:19 +0000 (GMT)
On Sun, 17 Dec 2000, Greg Lakatos wrote:
 > Recently I've been trying to compile a parallelized CHARMm c27b2
 > executable with MPICH 1.2.0 on a two-machine RedHat 7 cluster and have
 > run into some difficulty.  In particular when compiling charmm the
 > command:
 >
 >      ./install.com gnu large mpich
 >
 > runs to completion, produces an executable, but never asks for the path
 > to the mpich libraries.
 	This is because you need to tell install.com to compile an MPI
 binary with the "M" switch. The MPICH switch simply tells install.com
 which variety of MPI you're using.
 ./install.com gnu large m mpich
 should do the trick. (BTW, if you want to use QUANTUM, you'll need c27b3,
 as there are some bugs in the c27b2 implementation on GNU platforms.)
 > Moreover when charmm is run with the command:
 >
 > charmm -p4wd . -p4pg hosts < inputfile.inp > outfile.out
 >
 > charmm issues an error over the p4wd and p4pg switches.
 	CHARMM's command line argument handling is a bit suspect, so even
 a binary linked properly to the MPICH libraries will give warnings for
 some MPICH options (see source/machdep/startup.src for details) but these
 can be safely ignored. CHARMM compiled _with_ the M switch should at least
 only treat them as warnings, not errors, though! (Oh, and another thing
 about using MPICH - I would recommend explicitly specifying the number of
 processors with -np, or, better still, running your binary via. mpirun.)
 	Ben
 --
 ben (- at -) bellatrix.pcl.ox.ac.uk           http://bellatrix.pcl.ox.ac.uk/~ben/
 "Public speaking is very easy."
 	- Vice President Dan Quayle