We summarize the latest news and updates from the S4PST community from this past summer, May-August 2024. The work reported is partially funded by S4PST in collaboration with other stakeholders and showcases the impact of the activities around the software portfolio. Summer highlights include new publications, major software improvements, events related to software, and awards.

Publications

  1. Christian Munley, Aaron Jarmusch, and Sunita Chandrasekaran. “LLM4VV: Developing LLM-driven testsuite for compiler validation.” Future Generation Computer Systems (2024) doi:10.1016/j.future.2024.05.034.

  2. William Godoy, Pedro Valero-Lara, Keita Teranishi, Prasana Balaprakash, Jeffrey S Vetter. “Large Language Model evaluation for High-Performance Computing Software Development” Concurrency and Computations: Practice and Experience (2024) doi:10.1002/cpe.8269.

Major Software Improvements

  1. Topology-aware MPI Shared Memory Communication for Aurora: The MPICH team has developed a new topology-aware MPI shared memory communication for Aurora, which is a key component for the upcoming exascale system.

  2. Improvements to LLVM offloading: the LLVM team has made significant improvements to the offloading capabilities of the C++ and Fortran (Flang) compilers.

  3. Improvements to Flang testing: the Fortran team is investigating whether agile techniques, including test-driven development (TDD), accelerate Flang’s progress toward compliance with the Fortran 2023 standard. Tests have identified bugs that led to 41 patches to the LLVM Project main branch.

  4. Improvements in GASNet-EX HPC communication. The development team has made significant improvements to the GASNet-EX library.

Events

  1. Workshop The “Second Julia for ORNL Science” Workshop at Oak Ridge National Laboratory (ORNL) attracted 120 registrations and 16 invited talks from the community interested in the language for science.

  2. Tutorial Julia for HPC and Intro to Julia was offered in collaboration with facilities: OLCF, SNS, and NERSC. Each tutorial had more than 100 registrations and showcased Julia HPC code on Odo (Frontier test bed - AMD GPUs) and NERSC (NVIDIA GPUs). This is part of NERSC’s performance portability training series.

Awards

  1. Yuguo Kelly Tang, UC Berkeley student, received the best poster award (2 out of 29) for “Comparing JACC in modern CPU/GPU for Performance Portability in Julia” at ORNL’s summer intern poster session.