POP partner profile: The HPCAS group of INESC-ID R&D Institute

Friday, February 16, 2024

INESC-ID is a Lisbon-based R&D+I (Research, Development and Innovation) center, mainly owned by Instituto Superior Técnico (IST), the school of engineering of University of Lisbon. Being one of the largest and highest valued in Portugal, it is organized into 4 thematic lines and 11 areas covering various topics in Computer Science and Engineering and in Electrical and Computer. INESC-ID is committed to make contributions to address current and future challenges in societal digital transformation, life and health technology, energy transition, and security and privacy.

In particular, the High-Performance Computing Architectures and Systems (HPCAS) research area, which is that of the INESC-ID members in POP3, is specialized on several topics related to high-performance and energy-efficient computing. HPCAS addresses computational challenges at the software level — development of parallel algorithms, scheduling and load balancing methods, modeling of hardware / software to support codesign of heterogeneous architectures — and at the hardware level — design of accelerators and application-specific processors and systems. The HPCAS group tackles different types of applications with potential for high societal impact.

The INESC-ID HPCAS core team working in POP3 has expertise resulting from years of collaborative research in different research topics. It is composed of two senior researchers and faculty members,Leonel Sousa and Aleksandar Ilic, and a full time researcher,Ricardo Nobre. The topics of expertise include performance modeling and application optimization, energy-efficient and heterogeneous computing, acceleration with GPU and tensor core units, software/hardware security, bioinformatics, diverse toolchains and domain-specific languages. The team's high degree of multidisciplinarity is one of its main strengths, as it enables to holistically tackle a wide range of computational challenges.

In particular, the work that is being performed at INESC-ID on performance and energy efficiency modeling for multicore CPU and GPU architectures has evidenced impact in the scientific community and industry. We have relied on the Cache-Aware Roofline Model (CARM) — one of the contributions of members of the HPCAS INESC-ID group — for a wide range of objectives and use cases (different applications, metrics, microarchitectures), often in collaboration with key industry players. For example, CARM has been part of the Intel Advisor design and analysis tool since 2017. CARM comes prepackaged with the Intel openAPI Base Toolkit, which includes tools and optimized libraries for application development based on open-standards (SYCL / C++) with support for diverse Intel / AMD / NVIDIA architectures and types of devices (CPUs, GPUs, FPGAs).

Combining the expertise on performance / energy-efficiency with that on many other high-performance computing related topics enables providing precise in-sights on code efficiency, scalability and refactoring in interactions with users, which can materialize in suggesting / demonstrating code improvements in an intuitive manner.