Re: [AMBER-Developers] compilation issues with RC 3/31

From: Tyler Luchko <tluchko.rci.rutgers.edu>
Date: Thu, 1 Apr 2010 16:32:47 -0400

On 2010-04-01, at 3:55 PM, Jason Swails wrote:

> Hello,
>
> I've been having a heck of a time testing out compilation and such
> today, mostly arising due to indecision about what to do with
> rism-mpi. I have since stopped specifying -rismmpi when building in
> parallel because the tests would take about 20 to 30 minutes to
> produce a "program error".

This happens if you set -rismmpi and do not set XTRA_FLIBS. AFAIK, the name of the MPI fortran library may be different for any given MPI distribution and needs to be set by the user. The attached patch modifies configure to check if XTRA_FLIBS is set when -rismmpi is specified. If it is not set, an error message is issued that reiterates the instructions in the configure help message.

> Looking at the message from configure, I
> found that I would have to specify another extra-library environment
> variable, and -rismmpi was put under a "infrequently tested" header,
> so I began to leave that out.
>
> However, I had to remove librism.a from the lib folder after I decided
> not to build rismmpi since that library housed routines requiring MPI
> calls. Since I believe nab will wind up depending on this, would it
> be wise to try and separate librism.MPI.a and librism.a (or something
> to that effect?) In any case, getting rid of that and recompiling
> fixed those errors. However, when I then went to build amber in
> parallel, it died trying to build sander.RISM.MPI since the line
>
> SANDER_RISM_MPI=sander.RISM.MPI$(SFX)
>
> was still present.

There seems to be an error in configure that was leaving out the FFTW libraries when -rismmpi was missing. (I'm sure this was working correctly before.) This is also corrected in the patch.

> This is put in whenever "-mpi" is given to
> configure (even if -rismmpi is not specified). Removing this
> definition allowed compilation to finish. All a bit long-winded, but
> my end question is: is this all reasonable? Should we not build
> sander.RISM.MPI unless -rismmpi is specified? In which case, I've
> attached a patch to configure that will fix this issue.

Only MPI NAB requires user intervention for RISM support. sander.RISM.MPI should always be built.

Tyler



_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers


Received on Thu Apr 01 2010 - 14:00:02 PDT
Custom Search