A Team-Based Methodology of Memory Hierarchy-Aware Runtime Support in Coarray Fortran

Abstract

Abstract-In this paper, we describe how a 2-level memory hierarchy can be exploited to optimize the implementation of teams in the parallel facet of the upcoming Fortran 2015 standard. Teams have been suggested to leverage the hierarchical parallelism present in many applications. They are supposed to be used as an extension to Coarray Fortran (CAF), the explicitly-parallel part of the Fortran 2008 standard that adheres to the Partitioned Global Address Space (PGAS) programming model. Specifically, we focus on reducing the cost associated with moving data within a node and between nodes, finding that this distinction is of key importance when looking at performance issues. We introduce a new hardwareaware approach for PGAS, to be used within a runtime system, to optimize the communications in the virtual topologies and clusters that are binding different teams together. We have applied and implemented this methodology to three important collective operations, namely barrier, all-to-all reduction and one-to-all broadcast.To validate our approach, we implemented full support for teams into the CAF OpenUH compiler. To the best of our knowledge, the resulting platform is the first Fortran compiler that both provides teams and handles such a memory hierarchy methodology within teams. We illustrate the benefits of our methodology on a new Team Microbenchmark suite we specifically developed for this research and also on High Performance Linpack (HPL). Our memory hierarchy-awareness approach for barrier, reduction and broadcast operations shows up to 26-, 74-and 3-fold performance improvements over the default approach, respectively.

A Framework for Unit Testing with Coarray Fortran

“…Unfortunately, at the time of this writing, the only compiler supporting teams is the OpenUH one (Khaldi, Eachempati, Ge, Jouvelot, and Chapman 2015). However, this compiler does not support other standard Fortran features; most importantly, it does not support the OO programming features, and therefore cannot be used with pFUnit.…”