Hybrid MPI/OpenMP parallelization of the effective fragment potential method in the libefp software library.

A new hybrid MPI/OpenMP parallelization scheme is introduced for the Effective Fragment Potential (EFP) method implemented in the libefp software library. The new implementation employs dynamic load balancing algorithm that uses a master/slave model. The software shows excellent parallel scaling up to several hundreds of CPU-cores across multiple nodes. The code uses functions only from the well-established MPI-1 standard that simplifies portability of the library. This new parallel EFP implementation greatly expands the applicability of the EFP and QM/EFP methods by extending attainable time- and length-scales.