VMD-L Mailing List
From: John Stone (johns_at_ks.uiuc.edu)
Date: Thu Mar 29 2012 - 09:58:15 CDT
- Next message: Phelan Jr, Frederick R. Dr.: "Is there any way to stop the display of bonds for bonded atoms on opposite sides of a periodic boundary without unwrapping?"
- Previous message: John Stone: "Re: autopsf: can't read "logfilename""
- In reply to: Christoph Willing: "Possible MPI code problem (and fix)"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Christoph,
This update (and several others) for the MPI code is already
in the VMD 1.9.2 test versions. If you want to run VMD with MPI, you
will want to use the latest source code rather than the earlier VMD 1.9.1
version, particularly if you're using MPI 2.x, and MPICH in particular.
I made several changes to the MPI code in VMD to make it get along with
MPI implementations that now require MPI_IN_PLACE, and to prevent a conflict
with Tcl and MPICH that could cause a deadlock during MPI_Finalize().
You can get the latest version of the code via CVS here:
http://www.ks.uiuc.edu/Research/vmd/doxygen/cvsget.html
Let me know if you need any assistance.
Cheers,
John Stone
vmd_at_ks.uiuc.edu
On Thu, Mar 29, 2012 at 10:30:12PM +1000, Christoph Willing wrote:
> I just rebuilt vmd 1.9.1 with MPI support (mpich2-1.2.1-p1). However
> at startup, running a single instance of this build aborts, reporting
> the error:
>
> Info) Initializing parallel VMD instances via MPI...
> Assertion failed in file helper_fns.c at line 337: 0
> memcpy argument memory ranges overlap, dst_=0x9a4b80 src_=0-x9a4b80
> len_=544
> Internal ABORT - process 0
>
> Googling for this sort of error turned up:
> http://trac.mcs.anl.gov/projects/mpich2/ticket/1006
> where a fix is suggested.
>
> I applied a similar fix (patch below) and the resulting build now runs
> as expected i.e. without aborting at startup.
>
>
> chris
>
>
> --- src/VMDMPI.C.orig 2012-02-02 17:54:56.000000000 +1000
> +++ src/VMDMPI.C 2012-03-29 21:56:58.756600825 +1000
> @@ -95,7 +95,7 @@
> strncpy(nodename, namebuf,
> (((namelen + 1) < (maxnodenamelen-1)) ? (namelen+1) :
> (maxnodenamelen-1)));
>
> - MPI_Allgather(&nodes[mynode], sizeof(nodeinfo), MPI_BYTE,
> + MPI_Allgather(mynode ? &nodes[mynode] : MPI_IN_PLACE,
> sizeof(nodeinfo), MPI_BYTE,
> &nodes[ 0], sizeof(nodeinfo), MPI_BYTE,
> MPI_COMM_WORLD);
>
>
> Christoph Willing +61 7 3365 8316
> Research Computing Centre
> University of Queensland
>
>
-- NIH Resource for Macromolecular Modeling and Bioinformatics Beckman Institute for Advanced Science and Technology University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801 http://www.ks.uiuc.edu/~johns/ Phone: 217-244-3349 http://www.ks.uiuc.edu/Research/vmd/ Fax: 217-244-6078
- Next message: Phelan Jr, Frederick R. Dr.: "Is there any way to stop the display of bonds for bonded atoms on opposite sides of a periodic boundary without unwrapping?"
- Previous message: John Stone: "Re: autopsf: can't read "logfilename""
- In reply to: Christoph Willing: "Possible MPI code problem (and fix)"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]