Re: [AMBER-Developers] Current amber CVS test fails

From: <>
Date: Wed, 10 Mar 2010 12:50:15 -0500 (EST)

Hi everyone,

>> and with "-np 4", the following *extra* are real fails:
>> test/softcore/ (pretty much all the restarts)
> Thomas should look at these; it's not just roundoff error, but it's weird
> that the output files match, and not the restarts.

interestingly, this occurs at step 20 exactly and involves some problem
with the velocities on the non-master nodes. step 19 is fine with 2 or 4
cpu, step 20 has half the velocities wrong and from step 21 on, x and v
differ in the rst.

My first guess is that it was caused by Rev. 10.35 of runmd.f,
specifically the last three lines introduced there. Commenting those out
seems to fix the problem. That was actually an experimental fix to some
other problem and apparently, while I ran the tests in parallel, I didnt
do so on more than 2 procs. My mistake, the offending code should not have
been commited at all. I do some more tests tomorrow and check in a fix.



Dr. Thomas Steinbrecher
BioMaps Institute
Rutgers University
610 Taylor Rd.
Piscataway, NJ 08854

AMBER-Developers mailing list
Received on Wed Mar 10 2010 - 10:00:06 PST
Custom Search