Performance of Adaptive Fast Multipole Methods In Three Dimensions For Time-Dependent Problem

In the last two decades, physical constraints in chip design have spawned a paradigm shift in computer architecture. The availability of multi-core threaded systems has increased the parallel execution which has provided an opportunity for analysing computationallydemanding problems. The N-Body problem is a fundamental problem in the field of computational science and in this project, we havereviewed one of the N-body algorithms named the Fast Multipole Methods(FMM). The implementation of FMM is pretty complex and is dependent on multiple subroutines and in each subroutine, communication, and computation patterns are different. In the previous implementation, the FMM algorithm has been tested with multiple MPI process and highlighted that P2P (point to point) subroutine lacks in execution time. The P2P is the last subroutine in FMM algorithm and its a core function in FMM algorithm that deals with point to point potentials. This project analysed the working of the FMM implementation where computational time is lacking andbenchmark using OpenMP threads on P2P subroutine. The P2P subroutine is implemented with two level OpenMP approach inside MPI process.First, the behaviour of P2P subroutine by tuning parameter settings of the FMM algorithm is analysed. Finally, in the end the speedups of P2P subroutine having multiple threads with different number of MPI processes are analyzed. It achieved a reasonable absolute efficiency by increasing the number of threads with MPI processes.Tryckt av