Re: [AMBER] parallel installation of amber on sun

From: Atro Tossavainen <atro.tossavainen+amber.helsinki.fi>
Date: Fri, 3 Apr 2009 23:33:52 +0100

Per,

> ./configure_amber -mpi -nosanderidc sparc
>
> /lib/cpp -P -DBINTRAJ -DMPI nose_hoover.f > _nose_hoover.f
> "nose_hoover.f", line 455: Can't find include file mpif.h

I get that too. I believed the fix should have been using the mpf90
with preprocessing flags instead of /lib/cpp. In general, I get the
impression that specifying the -mpi flag should lead to the use of the
mp* compiler wrappers in order to avoid the AMBER authors having to
know where the MPI headers and libraries are supposed to reside.

man mpf95
...
DESCRIPTION
     tmf77, mpf77, tmf90, mpf90, tmf95, mpf95, tmcc, mpcc, tmCC
     and mpCC are front ends to the f77, f90, f95, cc and CC com-
     pilers. These tools simplify the creation of executables and
     object files by supplying -I, -R, -L, and -l options needed
     to use Sun HPC ClusterTools header files and libraries.

An unfortunate issue is that "mpf90 -F -fpp" writes files instead
of the standard output which is what you'd expect from fpp/cpp:

Starting installation of Amber10 (parallel) at Sat Apr 4 01:18:27 EEST 2009.
cd sander; make parallel
./makedepend > depend
./checkparconf
mpf90 -F -fpp -DBINTRAJ -DMPI nose_hoover.f > _nose_hoover.f
f90: Outfile nose_hoover.f would overwrite infile

Fixing this is too much work, so let's fall back on using cpp directly
and adding -I/opt/SUNWhpc/HPC*/include in fppflags in configure_amber
before running it. It's ugly, but it does the job.

You should also never need to include the sunperf directory with -L
explicitly, the -xlic_lib=sunperf linker parameter takes care of that.
But of course you need to have the right checks in configure_amber for
that to become possible later on, but the reason you're seeing this is
that your system has the compiler in a nonstandard place.

With these changes, I appear to be well on my way to producing working
kit on Solaris 9 with SUNWhpc 5.0. I haven't seen anything to do with
mpif-common.h yet, but I can't produce the executable:

f90 -free -o sander.MPI (tons of object files) ../netcdf/lib/libnetcdf.a -lmpi -xlic_lib=sunperf -lm
ld: fatal: library -lmpi: not found
ld: fatal: File processing errors. No output written to sander.MPI

Changing LOAD from "f90" to "mpf90" in config_amber.h fixes this
(as I suspect above). Of course it should have already been written
that way by configure_amber.

Installation of Amber10 (parallel) is complete at Sat Apr 4 01:32:38 EEST 2009.

Yay!

-- 
Atro Tossavainen (Mr.)               / The Institute of Biotechnology at
Systems Analyst, Techno-Amish &     / the University of Helsinki, Finland,
+358-9-19158939  UNIX Dinosaur     / employs me, but my opinions are my own.
< URL : http : / / www . helsinki . fi / %7E atossava / > NO FILE ATTACHMENTS
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Apr 05 2009 - 01:11:58 PDT
Custom Search