Project

General

Profile

Bug #394

LINCS errors, segfault on step 0 with cmake build but not autoconf

Added by Peter Kasson almost 10 years ago. Updated over 9 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Erik Lindahl
Category:
mdrun
Target version:
Affected version - extra info:
Affected version:
Difficulty:
uncategorized
Close

Description

Created an attachment (id=423)
tpr

On my OS/X 10.4 build system using gcc, fftw3, and threads but no other options (defaults to 32-bit build), I get multiple LINCS errors and a segfault on step 0 with my cmake build but not with a build using autoconf. This is the git master as of Feb 8, 2010. The system is charmm/CMAP/vsites/GBSA. Sander reported seeing a similar issue for particular systems on a windows cmake build but not linux builds (that was an explicit-solvent system using AMBER03).

tpr attached.

PS when there are constraints other than the vsites, mdrun will terminate with a fatal error. When no other constraints (different tpr), mdrun will segfault.

eq2_vsite.tpr (2.86 MB) eq2_vsite.tpr tpr Peter Kasson, 02/08/2010 08:26 PM
fail.out (61.1 KB) fail.out stdout/stderr for failure case #1 Peter Kasson, 02/08/2010 08:27 PM
fail2.out (10.5 KB) fail2.out stdout/stderr for failure case #2 (segfault; different tpr) Peter Kasson, 02/08/2010 08:28 PM

History

#1 Updated by Peter Kasson almost 10 years ago

Created an attachment (id=424)
stdout/stderr for failure case #1

#2 Updated by Peter Kasson almost 10 years ago

Created an attachment (id=425)
stdout/stderr for failure case #2 (segfault; different tpr)

#3 Updated by Erik Lindahl over 9 years ago

Hi Peter,

This system seems to be using almost every single new feature in git-head in combination, and also using an old Mac architecture we no longer build for.

Could you try to isolate which new feature is causing it by selectively disabling threads, Cmake, GBSA, vsites?

#4 Updated by Peter Kasson over 9 years ago

Update: it fails on current git master autoconf build on merlot with mpi np 8 (system default mpicc) but runs in single-processor. Will test with gbsa off and update. Is there a clean way to strip vsites from a gro/top pair? (could manually delete and rerun pdb2gmx)

#5 Updated by Erik Lindahl over 9 years ago

Test without GBSA (and then try both PD and DD). There might be something in the LINCS code in PD that is still wrong. Don't worry about vsites for now...

#6 Updated by Peter Kasson over 9 years ago

On further testing, it breaks with GB turned off (both md and sd integrators) on parallel runs with dd but runs successfully in all cases with either single-processor or pd.

#7 Updated by Berk Hess over 9 years ago

I experienced crashes as well and I have found a fixed a bug.
There could be too little memory allocated with DD with a
twin-range cut-off setup combined with constraints or vsites.
I guess this also caused you problems.
Please check if the latest git version fixed it.

Berk

#8 Updated by Peter Kasson over 9 years ago

Thanks, but unfortunately the problem persists with a git pull and rebuild. So I'm guessing it's another error somewhere.

#9 Updated by Berk Hess over 9 years ago

Indeed, this is a bug in constraints in vacuum with 2 DD cells in a dimension.
I fixed it.

Berk

#10 Updated by Peter Kasson over 9 years ago

Great--thanks. Fix verified and works.

Also available in: Atom PDF