[gmx-developers] Assembler loops on Itanium2 Montecito.

[Erik Lindahl]

Hi Rafael,

First - which Gromacs version are you running?

We experienced something similar a while ago, but I think that was a  
compiler bug I never resolved.

For reference, ia64 isn't particularly fast compared to x86 since we  
have assembly loops for both :-)

Cheers,

Erik

On Feb 21, 2008, at 6:13 PM, Rafael R. Pappalardo wrote:

> Dear all,
> I am trying to install gromacs_cvs in a system with Itanium2  
> (Montecito) CPUs.
> Configuring without assembler loops seems to work but when I compile  
> with
> assembler loops mdrun complaints about:
>
> Configuring nonbonded kernels...
> Testing ia64 CPU family...Unknown Itanium
> You might want to test the speed without assembly kernels
> by setting the NOASSEMBLYLOOPS environment variable.
> Using single precision ia64 assembly kernels.
>
> Looking at src/gmxlib/nonbonded/nb_kernel_ia64_single/ 
> nb_kernel_ia64_single.c
> I find that the test to recognize an Itanium2 assumes that the  
> cpu_family is:
> 0x1f (31). This value is correct for Madison Itanium2 but the  
> Montecito's are
> 0x20 (32).
>
> If I change the test and include the 0x20 family as Itanium2 the  
> program seems
> to enter an infinite loop. The last lines in the md.log file is:
>
> Configuring nonbonded kernels...
> Testing ia64 CPU family...Itanium 2
> Using single precision ia64 assembly kernels.
>
> Removing pbc first time
> Done rmpbc
> There are 3000 flexible constraints
>
> ++++ PLEASE READ AND CITE THE FOLLOWING REFERENCE ++++
> B. Hess, H. Saint-Martin and H.J.C. Berendsen
> Flexible constraints: an adiabatic treatment of quantum degrees of  
> freedom,
> with application to the flexible and polarizable MCDHO model for water
> J. Chem. Phys. 116 (2002) pp. 9602-9610
> -------- -------- --- Thank You --- -------- --------
>
>
> Initializing LINear Constraint Solver
>
> ++++ PLEASE READ AND CITE THE FOLLOWING REFERENCE ++++
> B. Hess
> P-LINCS: A Parallel Linear Constraint Solver for molecular simulation
> J. Chem. Theory Comput. 4 (2008) pp. 116-122
> -------- -------- --- Thank You --- -------- --------
>
> After that the cpu time increases and the program never finish. The  
> same test
> with the NOASSEMBLYLOOPS variable set in the environment end in less  
> that 2
> minutes.
>
> I have tried with gcc 4.1, gcc 3.4 and ifort 10.1 with the same  
> results. I
> have access to two different machines one is running SLES 10 and the  
> other
> some Redhat variation.
>
> Any suggestion about how to discover what's going on?
>
> Thanks a lot for the attention.
>
> Rafael R. Pappalardo
>
>
>
> -- 
> Dr. Rafael R. Pappalardo
> Dept. Physical Chemistry, Univ. de Sevilla (Spain)
> e-mail: rafapa at us.es
> _______________________________________________
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.

------------
Erik Lindahl   <lindahl at cbr.su.se>  Backup: <erik.lindahl at gmail.com>
Assistant Professor, Computational Structural Biology
Center for Biomembrane Research, Dept. Biochemistry & Biophysics
Stockholm University, SE-106 91 Stockholm, Sweden
Tel: +46(0)8164675  Mobile: +46(0)704218767  Fax: mail a PDF instead