Previous Thread
Next Thread
Print Thread
Page 1 of 2 1 2
Joined: Apr 2009
Posts: 22
Forum Member
OP Offline
Forum Member
Joined: Apr 2009
Posts: 22
Dear colleagues,

I am sorry for such a dumb and probably badly specified question about charmm compilation, but I hope that someone more experienced will have an idea of what can be the reason of my problem.

The thing is that I am trying to make c35b2 working. First in serial, then in parallel. We had an old compilation of serial and parallel c35b2, compiled as "./install.com gnu medium" for serial run and "./install.com gnu medium M mpich" for parallel. Those binaries didn't work in my case: when trying to minimize a solvent ball around the protein (22338 atoms overall) it was crashing afer printing

VEHEAP> Expanding heap size by 9879552 words.
Segmentation fault

when generating the list of nonbonded interactions. I made a look into the current forum, figured out that for c35b1 there was a bug in the install.com and asked my sys.admin. to reinstall the serial version with "x86_64 GFORTRAN" (we have intel EMT64 machines). Since I also needed SCCDFTB and CGENFF compatibility, the whole compilation command was

"./install.com gnu xlarge +CGENFF T x86_64 GFORTRAN"

The new compiled binary worked fine, without any errors.
Now I asked to compile the parallel version, and here the problems started. I must say that the old parallel version compiled with "./install.com gnu medium M mpich" was in fact working, but was giving me the same Segmentation fault as the serial one. The compilation command for the new parallel executable was

"./install.com gnu xlarge T +CGENFF M MPICH x86_64 GFORTRAN"

To my regret, when running it, I was immediately getting the error

p0_741: p4_error: interrupt SIGSEGV: 11

without charmm even printing it's version name.
I asked my sys.admin. to play around with keywords, and we got the version

"./install.com gnu xlarge T +CGENFF M MPICH AMD64 GFORTRAN"

running in parallel, but once again printing the Segmentation fault after Heap size expansion notification.

It seems like i need this "x86_64" option to create a working executable, however, it looks like there is some problem running at the same time "M MPICH" and "x86_64"...

We are using Mpich v 1.2.7p1, Gnu Fortran 4.5.0 and Open Suse 11.3.

I am once again very sorry for such a broad question, and will appreciate any hint.

Sinceely,
Iakov

Joined: Sep 2003
Posts: 8,623
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,623
Likes: 24
The MPICH library could also be an issue; was it also built with the same gcc/gfortran version?

I'm not sure CGENFF is well supported in c35b2

There can be issues with MPI when a QM program is involved

I don't think the AMD64 option is appropriate

Try a simpler parallel build, such as

./install.com gnu large x86_64 M MPICH

and add other features once that works.


Rick Venable
computational chemist

Joined: Apr 2009
Posts: 22
Forum Member
OP Offline
Forum Member
Joined: Apr 2009
Posts: 22
Dear Rick,

Thank you very much for prompt reply. I also asked my sys.admin. whether MPICH library was compiled with the same gfortran version we are using now, and he said that since both are default for Suse 11.3 and come with distribution, it SHOULD be the case smile I hope, it is the case))

I'll try what you suggest, thank you once again!

Joined: Apr 2009
Posts: 22
Forum Member
OP Offline
Forum Member
Joined: Apr 2009
Posts: 22
Unfortunately, running the binary, compiled as

./install.com gnu large x86_64 M MPICH

leads to the same error:

p0_30559: p4_error: interrupt SIGSEGV: 11

So that there indeed is some quarrel between "M MPICH" and "x86_64" frown

Joined: Apr 2009
Posts: 22
Forum Member
OP Offline
Forum Member
Joined: Apr 2009
Posts: 22
Have anyone compiled c35b2 with the same options? (Including M MPICH and x86_64) It would be great if you could share your experience...

Joined: Sep 2003
Posts: 8,623
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,623
Likes: 24
Yes, many times without problems.


Rick Venable
computational chemist

Joined: Apr 2009
Posts: 22
Forum Member
OP Offline
Forum Member
Joined: Apr 2009
Posts: 22
I see.. so there is no bug - the problem is in the way we compile it.. Then the only thing that comes to mind is having MPICH compiled with different version of GFOTRAN, right?.. Sorry for dumb question, but don't you know, how to check it?

Joined: Sep 2003
Posts: 8,623
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,623
Likes: 24
I'd probably try verifying the MPICH setup independently with the Fortran test programs provided with the MPICH distribution.

Note that MPICH-1 is considered out of date, and may not be supported in the next major release of CHARMM. Both MPICH-2 and OpenMPI work fairly well with current and developmental versions of CHARMM.


Rick Venable
computational chemist

Joined: Apr 2009
Posts: 22
Forum Member
OP Offline
Forum Member
Joined: Apr 2009
Posts: 22
So MPICH-1 is out of date already for c35b2? Or for later versions?
I succeeded in compiling c33b1 in parallel, and since I am not using currently CGENFF compatibility, I decided to freeze my efforts towards c35b2 for time being..

Joined: Sep 2003
Posts: 8,623
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,623
Likes: 24
MPICH-1 may not be supported for the beta release of c36; it should work fine for c35bN releases (current release is c35b6). MPICH-2 has been available for a number of years.


Rick Venable
computational chemist

Page 1 of 2 1 2

Moderated by  lennart, rmv 

Link Copied to Clipboard
Powered by UBB.threads™ PHP Forum Software 7.7.5
(Release build 20201027)
Responsive Width:

PHP: 7.3.31-1~deb10u1 Page Time: 0.012s Queries: 35 (0.008s) Memory: 0.7836 MB (Peak: 0.8777 MB) Data Comp: Off Server Time: 2022-12-04 20:13:35 UTC
Valid HTML 5 and Valid CSS