Principles of runtime support for parallel processors

TLDR

Scientific problems exhibit substantial data‑level parallelism, yet this can preclude compiler‑based detection of certain types of parallelism. The PARTY runtime system aims to produce efficient parallel implementations for scientific computations whose data dependencies manifest only at runtime. The system selects an appropriate granularity, builds a directed‑acyclic‑graph representation of the program, applies aggregation techniques to generate efficient schedules, and maps those schedules onto the target machine. Initial experiments on the Intel Hypercube and Encore Multimax demonstrate the usefulness of the approach.

Abstract

There exists substantial data level parallelism in scientific problems. The PARTY runtime system is an attempt to obtain efficient parallel implementations for scientific computations, particularly those where the data dependencies are manifest only at runtime. This can preclude compiler based detection of certain types of parallelism. The automated system is structured as follows: An appropriate level of granularity is first selected for the computations. A directed acyclic graph representation of the program is generated on which various aggregation techniques may be employed in order to generate efficient schedules. These schedules are then mapped onto the target machine. We describe some initial results from experiments conducted on the Intel Hypercube and the Encore Multimax that indicate the usefulness of our approach.