[AMBER-Developers] solution (!?!) to the pme_setup problem?

From: case <case.biomaps.rutgers.edu>
Date: Wed, 10 Mar 2010 12:24:14 -0500

Some more recent data:

1. The symptom (for me) does not occur with ifort 11.1 (064) but does
happen with ifort 11.1 (069). Confirmation or refutation by others would
be helpful here, since when switching compilers all the time it's easy to
make some mistake.

2. The following simple patch seems effective:

casegroup% git diff pme_setup.fpp
diff --git a/src/pmemd/src/pme_setup.fpp b/src/pmemd/src/pme_setup.fpp
index 09dff4b..7859fb7 100644
--- a/src/pmemd/src/pme_setup.fpp
+++ b/src/pmemd/src/pme_setup.fpp
.@ -403,8 +403,8 @@ end subroutine get_ee_func
 subroutine vdw_correct_setup
- use pme_force_mod
- use prmtop_dat_mod
+ use pme_force_mod, only : gbl_nvdwcls
+ use prmtop_dat_mod, only : natom, atm_iac
   implicit none

Note that these modules are pretty big, and the conservative approach of only
importing what you need minimizes exposures to compiler idiosyncracies, and to
typos (where you accidentally type in a name that you don't want, but which
exists among the hundreds of variables in the modules).

In general, the use of "only" is a Good Thing (tm), somewhat like implicit
none, especially when one is only using a tiny subset of the module variables.

I am still testing the above change, but it looks pretty safe, and so will
probably add it to CVS after giving it a stress test. But others can
manually make the change and see if it works for them.


AMBER-Developers mailing list
Received on Wed Mar 10 2010 - 09:30:05 PST
Custom Search