28 July 2026 UTC

UXL Language SIG: Twenty Years of TBB

Virtual

TBB is 20! by Michael Voss (Intel)

Abstract: Threading Building Blocks (TBB) 1.0 was released in 2006, just as Intel was introducing its first mainstream multicore processor. This talk traces the evolution of TBB from its first release to today, highlighting its history, key features, application domains, and the computing trends and inflection points that shaped its role in today’s heterogeneous computing environment. The talk concludes with an overview of the library’s current focus areas and an invitation for the community to help guide its future through its open design process.
Bio: Michael Voss is a Senior Principal Engineer of Middleware Architecture at Intel and the current oneTBB architect. He has co-authored two books on TBB: Pro TBB: C++ Parallel Programming with Threading Building Blocks, released in 2019, and Today’s TBB, released in 2025. He works on parallel programming models and libraries, participates in the C++ standards committee, and frequently consults with customers on their use of parallelism and threading tools.

oneTBB Meets the Community, with two 10-minute short talks by Kyle Knoepfel (Fermilab) and Federico Ficarelli (CINECA)

Title: oneTBB use in DUNE/Phlex and in High-Energy Physics

Abstract: High-energy physics experiments rely on processing vast amounts of statistical events to obtain conclusions about fundamental physics. The events are almost always statistically independent of each other and thus well-suited for parallel processing with a library like oneTBB. We will briefly cover how oneTBB has been used by the HEP community to improve the efficiency of its workflows. We will focus particularly on a recent fruitful collaboration between the Phlex framework team and oneTBB developers
Bio: Kyle Knoepfel is a senior software developer at Fermilab, focused on developing computing frameworks to meet the data-processing needs of many of Fermilab’s experiments. He leads the development of Phlex, a framework designed to support the Fermilab DUNE experiment’s scientific workflows by leveraging data-flow-processing techniques.

Title: Scaling Drug Discovery via oneTBB: Portable Exascale Conformer Generation

Abstract: High-throughput virtual screening (HTVS) is the automated computational evaluation of massive compound to identify molecules likely to bind and inhibit a biological target. It combines fast 3D conformer generation, docking, scoring, and property prediction to triage millions to trillions of compounds. In our HTVS platform, conformer generation—enumerating a molecule’s physically feasible 3D structures—is a key bottleneck at extreme scale. To keep GPUs busy, we process work at both fine-grained and batched levels. This talk covers our production use of oneTBB flow graphs and the challenges of targeting heterogeneous systems that pair GPUs with Intel and ARM multicore CPUs.
Bio: Federico Ficarelli leads the team at CINECA Supercomputing, the Italian national HPC center, working on pharmaceutical R&D with pharma companies; their home-grown virtual screening platform is currently the world’s fastest on accelerated HPC systems. His research spans GPU-optimized molecular docking, energy-efficient heterogeneous computing, and performance portability across GPUs and x86, ARM and RISC-V CPUs