Topic: Performance, Portability, and Productivity

A Livermore-developed programming approach helps software to run on different platforms without major disruption to the source code.

Supported by the Advanced Simulation and Computing program, the open-source Axom project focuses on developing software infrastructure components that can be shared by HPC applications running on diverse computing platforms.

LLNL participates in the digital ISC High Performance Conference (ISC21) on June 24 through July 2.

Computing relies on engineers like Stephanie Brink to keep the legacy codes running smoothly. “You’re only as fast as your slowest processor or your slowest function,” says Brink, who works in CASC. By analyzing a legacy code’s performance, Brink and her team can reduce the amount of time it takes to run and allow for more critical science to be accomplished.

The latest issue of LLNL's Science & Technology Review magazine showcases Computing in the cover story alongside a commentary by Bruce Hendrickson.

Highlights include scalable deep learning, high-order finite elements, data race detection, and reduced order models.

Our researchers will be well represented at the virtual SIAM Conference on Computational Science and Engineering (CSE21) on March 1–5. SIAM is the Society for Industrial and Applied Mathematics with an international community of more than 14,500 individual members.

Proxy apps serve as specific targets for testing and simulation without the time, effort, and expertise that porting or changing most production codes would require.

Highlights include response to the COVID-19 pandemic, high-order matrix-free algorithms, and managing memory spaces.

After years of preparation, LLNL’s upgraded Ares code runs a 98-billion-element simulation on 16,384 GPUs on the Sierra supercomputer.

Computational Scientist Ramesh Pankajakshan came to LLNL in 2016 directly from the University of Tennessee at Chattanooga. But unlike most recent hires from universities, he switched from research professor to professional researcher.

FGPU provides code examples that port FORTRAN codes to run on IBM OpenPOWER platforms like LLNL's Sierra supercomputer.

The 2019 Department of Energy (DOE) Performance, Portability and Productivity meeting is slated for April 2–4, 2019, in Denver, CO, where attendees will have the opportunity to share ideas and updates on performance portability.

Computer scientist Greg Becker contributes to HPC research and development projects for LLNL’s Livermore Computing division.

LLNL's Advanced Simulation Computing program formed the Advanced Architecture and Portability Specialists team to help LLNL code teams identify and implement optimal porting strategies.

Highlights include the latest work with RAJA, the Exascale Computing Project, algebraic multigrid preconditioners, and OpenMP.

A new software model helps move million-line codes to various hardware architectures by automating data movement in unique ways.

Apollo, an auto-tuning extension of RAJA, improves performance portability in adaptive mesh refinement, multi-physics, and hydrodynamics codes via machine learning classifiers.

LLNL computer scientists use machine learning to model and characterize the performance and ultimately accelerate the development of adaptive applications.

LLNL researchers are finding some factors are more important in determining HPC application performance than traditionally thought.

Performance analysis of parallel scientific codes is becoming increasingly difficult, and existing tools fall short in revealing the root causes of performance problems. We have developed the HAC model, which allows us to directly compare the data across domains and use data visualization and analysis tools available in other domains.

AutomaDeD is a tool that automatically diagnoses performance and correctness faults in MPI applications. It has two major functionalities: identifying abnormal MPI tasks and code regions and finding the least-progressed task. The tool produces a ranking of MPI processes by their abnormality degree and specifies the regions of code where faults are first manifested.

To overcome the shortcomings of the analytical and architectural approaches to performance modeling and evaluation, we are developing techniques that emulate the behavior of anticipated future architectures on current machines.

Olga Pearce studies how to detect and correct load imbalance in high performance computing applications.

Todd Gamblin leads the PAVE project, which develops performance data visualization techniques that are more intuitive for application scientists.