University of Houston

Fortran remains a very widely used programming language for technical computing; for instance, the Department of Energy reported that a majority of new code written in their labs in 2013 was created in Fortran. Fortran coarrays are new elements in the Fortran standard that aim to enable the use of parallel Fortran programs without the need to introduce external constructs such as libraries (e.g. MPI) or directives (such as OpenMP). The new features provide a Partitioned Global Address Space (PGAS) approach to parallel programming in which the programmer explicitly creates arrays that are partitioned across the different “images” that will collaboratively perform the computations: they provide an abstraction that simplifies the task of parallel program creation, yet also give critical information to the compiler to enable an efficient implementation.

Yet to date, coarrays are not used in commercial code. The primary reason for this is that only one robust implementation has been produced (by Cray Research). Thus there is very little experience with coarrays or reporting of performance data, and a clear lack of portability. The work in this thesis aims to make a fundamental contribution to the state of the art of parallel programming by providing a robust, open source implementation of the coarray features in the Fortran standard. It describes an efficient implementation technology for coarrays that was developed, focusing on key performance for coarray programs at scale, evaluates the performance of a near-commercial strength implementation on developed benchmarks and codes from real-world applications, and describes several potential enhancements to the coarray features in the Fortran standard that were identified during the course of this work.