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

International audience—We describe how 2-level memory hierarchies can be exploited to optimize the implementation of teams in the parallel facet of the upcoming Fortran 2015 standard. We focus on reducing the cost associated with moving data within a computing node and between nodes, finding that this distinction is of key importance when looking at performance issues. We introduce a new hardware-aware 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 into the CAF OpenUH compiler, this methodology to three important collective operations, namely barrier, all-to-all reduction and one-to-all broadcast; this is the first Fortran compiler that both provides teams and handles such a memory hierarchy methodology within teams I. INTRODUCTION The emergence of many-core computing nodes in large-scale distributed systems requires programming model im-plementers to consider more carefully memory hierarchy when looking at performance issues. Most parallel applications are programmed using the Message Passing Interface (MPI) [1], where multiple processes execute in a coordinated manner, communicating by performing send and receive operations. More recently, several languages and libraries have added support for explicit or implicit remote memory access (RMA) using so-called " one-sided communication " , including languages following the Partitioned Global Address Space (PGAS) paradigm as well as MPI (MPI-2 added RMA to the interface and MPI-3 made significant refinements to better support it). Of special note is the Fortran 2008 addition for supporting coarrays, a language mechanism that enables RMA as a natural extension to Fortran's array syntax, informally named CAF 1. In this paradigm, an image is an executing process in a Single Program Multiple Data (SPMD) program with its own cop