Damian Rouson, Lawrence Berkeley National Laboratory
Partitioned Global Address Space (PGAS) programming models, languages, and libraries offer HPC software developers a set of abstractions that facilitate parallel communication and computation, including remote memory access and remote procedure calls. This talk will give a high-level overview of PGAS research and development at Berkeley Lab, covering our contributions to the PGAS software ecosystem, including our work on unit tests for the PGAS features in the LLVM Flang Fortran compiler; creating the Caffeine coarray Fortran parallel runtime library ; producing the UPC++ PGAS template library ; and developing the GASNet-EX exascale networking middleware that supports a range of PGAS languages, libraries, and frameworks . The talk will also highlight the use of the aforementioned technologies in task-scheduling frameworks, including FEATS , Legion , and DepSpawn .
Michelle Strout, The University of Arizona and HPEThe Chapel programming language provides constructs for expressing a wide range of parallelism patterns, while also remaining easy-to-use. This talk will show usage examples from machine learning, data analytics, aeronautical engineering, hydrology, and other application areas. Perspectives of how existing workflows were adjusted to leverage Chapel and the resulting performance and scaling will also be presented.
George Bosilca, University of Tennessee, KnoxvilleChallenges introduced by highly hybrid many-cores architectures have a lasting impact on the portability and performance of applications, partially due to traditional programming paradigms. These programming paradigms lack the flexibility and capabilities required to deal with large amounts of potential parallelism and a dynamic hybrid execution environment, putting the performance and scalability of applications at risk. Advances in task-based runtime have shown to provide a plausible solution to this problem, one that not only increase the domain scientists’ productivity but also deliver codes that are more efficient, more scalable, and more adaptable to various hardware architectures, and show an increased portability potential to transition from one generation of hardware to another. This talk will describe a distributed task-based runtime, PaRSEC, and highlight its data management strategies and features to allow the implementation of highly efficient and scalable algorithms at any scale.