[gmx-developers] gromacs 4 optimizations / cell support

[Carsten Kutzner]

Maik VP wrote:
> Dear developers,
> 
> I found a paper about gromacs 4 which describes: 'Algorithms for Highly
> Efficient Load-Balanced, and Scalable Molecular Simulation' in the Journal
> of Chemical Theory and Computation. The paper describes that gromacs 4
> uses multithreading instead of MPI, and other interesting parallelization
That is not correct. Nothing is mentioned about multithreading in the paper.
Maybe you confused it with multiple-program multiple-data parallelism? This
you can switch on by choosing separate PME nodes with the -npme option to
mdrun.
Configuring with threads enabled will currently not work.

Carsten

> optimizations.
> 
> However, when I try to enable multithreading in the latest-greatest
> version from CVS, using ./configure --enable-threads --with-x
> --with-gnu-ld, the linker fails because gmx_thread_mutex_lock and
> gmx_thread_mutex_unlock are not defined. I have searched for these
> functions in the code but I could indeed not find their implementations.
> 
> Since multithreading was described in the paper, I am wondering where I
> can find a gromacs version in which multithreading and the other
> optimizations from the paper are implemented.
> 
> Furthermore I am interested in running Gromacs on the Cell architecture,
> since the folding at home project clearly shows that gromacs is able to
> exploit the resources in this advanced processor. However, again, the
> latest-greatest CVS version does not include Cell support, only generic
> altivec support Does anybody know where I can find the Cell
> implementation of gromacs, which uses the SPE accelators?
> 
> Maik Nijhuis