[gmx-users] problem with porting excutables to another site (IBM AIX)

[Erik Lindahl]

>
Hi,

> I am having trouble when trying to port the compiled gromacs  
> excutables to another site (a remote site). I changed PATHs in
> GMXRC and GMARC.bash etc to proper new localtions, and the program  
> pdb2gmx found the right location for top lib files, and it ran.
> However, it core dumps at the end:
>
Actually, in most cases you shouldn't even have to do that; the  
binaries look in places like ../../share/top and can almost always find  
the topology stuff automatically if you move the whole tree. The GMXRC  
files are mainly intended to provide option and filename completion.

> Back Off! I just backed up 1a45.top to ./#1a45.top.3#
> Processing chain 1 (1481 atoms, 173 residues)
> Opening library file  
> /dfs/protein/programs/gromacs/share/top/specbond.dat
> 5 out of 5 lines of specbond.dat converted succesfully
> Illegal instruction (core dumped)
> /dfs/protein/ruhongz/gromacs>
>
>
Whoa, that looks strange - "Illegal instruction". That means the CPU  
cannot execute the binary, which I cannot really explain if it was  
compiled on identical nodes. In general it should be trivial to move  
the executables. And grompp worked perfectly on the other system?

> Am I missing something here? Do I have to do something else, or do I  
> have to recompile/install in the new enviornment?The OS is the same,  
> both are IBM AIX 4.3.3.0 (IBM SP Power3 nodes)

Are the nodes identical, or? I know there are slightly different  
generations of power3 (nighthawk, winterhawk, etc), but as far as I  
know the instructions set is the same... FFTW has been having some  
problems with not recognizing power3 nodes (and compiling for pwr2  
caused crashes) , but that should be fixed in the latest version of  
their sources (2.1.5). However, the FFTs are only called if you do PME,  
and definitely not by grompp.

Are you running in parallel? I do remember that the IBM SP switch  
libraries can be a little picky, and there might be different  
communication hardware on the two systems. Still, I don't _think_ that  
should cause problems until you call MPI.


If you want to, you can try the 3.1.5_pre1 version that's available on  
the ftp site (don't remember if it's in the tmp or beta directory). In  
that version I've added a --all-static flag that will link all binaries  
completely statically. (We also just decided that we should get around  
and release a new stable version (3.1.5) so people don't have to wait  
for 4.0 for things like faster altivec support, minor OPLS fixes, and  
the static linking.)


Cheers,

Erik

------------------------------------------------------------------------ 
-----
Erik Lindahl, MSc, PhD     <lindahl at stanford.edu>
D109, Fairchild Building
Dept. Structural Biology, Stanford University School of Medicine
Tel. 650-7250754    Fax. 650-7238464

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/enriched
Size: 2906 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20030420/cabd1c31/attachment.bin>