The Fast Multipole Method (FMM) is an optimally efficient algorithm for solving N -body problems: a fundamental challenge in fields like astrophysics, plasma physics and molecular dynamics. It is particularly suited for computing$1 / \mathrm{r}$potentials present in Coulomb and gravitational particle systems. Despite the near-field phase being trivially parallelisable, the far-field phase of the$1 / \mathrm{r}$FMM currently lacks an efficient, massively parallel GPU algorithm fitting for the era of Exascale computing. Current state-of-the-art approaches either favor highly parallel but inefficient expansion shift operators or asymptotically efficient but poorly parallelisable rotation-based ones. Recently, a breakthrough was made with the re-evaluation of a rotation operator variant called fast rotation, which dramatically increases caching effectiveness and marries the advantages of both methods. Thus, this paper incorporates this approach to create fast rotation-based operators that facilitate an efficient far-field algorithm for the FMM on GPUs. Additionally, a warpcentric data access scheme is co-developed alongside a matching octree design, which yields coalesced memory access patterns for the bottleneck operators of the far-field phase. The fast rotation algorithm is enhanced with a cache-tiling mechanism, maximising GPU cache utilisation. Compared to the state-of-theart GPU FMM far-field implementation, our algorithm achieves lower running times across the board and a 2.47 x speedup for an increased precision simulation, with the performance improvement growing as precision increases, providing concrete proof of efficacy for dense particle systems.

We propose a new parallel, noncollective I/O strategy called Distant I/O that targets clustered computer systems in which disks are attached to compute nodes. Distant I/O allows one sided access to remote secondary storage without installing server processes or daemons on remote compute nodes. We implemented this model using Active Messages and demonstrated its performance advantages over the PIOFS parallel filesystem for an I/O intensive parallel application on the IBM SP.

In this paper, we describe a novel parallelization approach we developed to solve the largest multireference configuration interaction (MRCI) problem ever attempted. From the mathematical perspective, the program solves the eigenvalue problem for a very large, sparse, symmetric Hamilton matrix. Using an out-of-core approach, shared memory programming model, improved data compression algorithms, and dynamic load balancing we were able to solve a problem six times larger than previously reported. The potential curve for the chromium dimer was calculated with a Hamilton matrix of dimension 1.3 billion (1,295,937,374). This task involved moving 1.5 terabytes of data between main memory and secondary storage per MRCI iteration. Furthermore, by employing Active Messages and user-level striping to combine multiple files on local disks on the IBM SP into a single logically-shared file, the execution time of the program was reduced by a factor of three, as compared to our initial implementation on top of the IBM PIOFS parallel filesystem.