2024 R&D 100 Award for IRIS-SDK: Intelligent Runtime System for Extremely Heterogeneous Computer Architectures

IRIS is a unique runtime system designed to work with various types of computer hardware, boosting performance, portability and productivity. It’s the world’s only framework capable of concurrently executing application tasks on a diverse array of devices simultaneously, including CPUs, NVIDIA GPUs, AMD GPUs, field-programmable gate arrays, or FPGAs, and Hexagon digital signal processors, or DSPs, without any extra work from developers. IRIS is available open source in github. It comes with a software development kit, or SDK, composed of comprehensive suite of tools: Hunter – task scheduling policies hub, Dagger – robust tool validating IRIS runtime and task scheduling policies, and MatRIS – a genuinely heterogeneous and portable linear algebra library, which delivers BLAS, LAPACK and math operations.

IRIS represents a pioneering leap in intelligent runtime systems tailored for extremely heterogeneous computer architectures. It makes programming for diverse computer setups much easier by handling tasks such as heterogeneous kernel execution, memory management, data transfers and scheduling. All this is made automatically and transparently to the users, elevating and taking the programming productivity to a different level. IRIS-SDK enables the application code to write once and deploy everywhere in the heterogeneous computing environment. It improves the performance of application with heterogeneous execution and makes it easy for them to run on both high-performance and embedded systems without needing changes to the code.

IRIS research was funded by the DOE Office of Science.

The development team, all from ORNL, includes Jeffrey S. Vetter, Beau Johnston, Jungwon Kim, Seyong Lee, Narasinga Rao Miniskar, Mohammad Alaul Haque Monil, Pedro Valero-Lara and Aaron Young.