Parallel molecular dynamics using OPENMP on a shared memory machine

Abstract The parallelization of a sequential molecular dynamics (MD) program using OpenMP directives for a Fortran compiler is described. The strength of this approach lies in the possibility to proceed incrementally, with only a few alterations of the original, scalar code. If do-loop splitting with OpenMP is certainly less efficient than domain decomposition (DD) approaches, its fast implementation, requiring little knowledge of the source code, also involves significantly less programming effort. Efficiency of the parallel executable is probed on a distributed-shared memory Silicon Graphics Origin 2000 for MD applications of various complexities. Linear wallclock speed-up ratios using four processors, and ranging between 6.0 and 7.0 with eight, underline the cost-effectiveness of OpenMP directives for fine-grained parallel MD simulations on limited computational resources.