Axel,
I find that structure generation becomes very slow, (exponentially slow), as the number of atoms increases, as expected. For example, setting values of the option ‘-g=’ to anything more than 25 becomes impossible to deal with in my cluster, takes days to enumerate the structures, and no way to restart maps if it gets booted from the queue.
Is there a parallel (MPI) implementation of this part of the code or at least one that uses multithreading to make it somewhat faster ?
Thanks,
Ganesh
There is an mpi version of mmaps called mpimmaps
(to create it:
uncomment CXX=mpiCC -DATAT_MPI in makefile
then type
make mpi
)
BUT I recommend using the serial version with a new, faster enumeration algorithm:
remove the -DSLOWENUMALGO switch from the CXXFLAGS=… line that is not commented out in the make file. Then make clean ; make
This faster algorithm, does not wait enumerate all structures of certain size before generating output, which makes the code more responsive.
This faster algo will be the default in future versions.
Hi,
Why can’t I find the "-DSLOWENUMALGO" and the "CXXFLAGS" in the makefile. On my machine, The vesion of the ATAT is the latest one which I could download form the homepage .
Use the link "Beta version of the latest update".
Sorry for not making this clear!