ICL 2023/24 Report
The most recent edition of the ICL Report is now available to download in PDF format. This year’s issue featured a retrospective of ICL’s involvement in the Exascale Computing Project, as well as highlights from 2023, project summaries of ICL’s research portfolio, and highlights of various research activities from 2023. Thanks to all the contributors!
2024 ICL Winter Reception
The 2024 ICL Winter Reception was held at Zoo Knoxville’s Amphibian and Reptile Conservation Campus on the evening of February 23, 2024. Attendees enjoyed dining amongst the animal exhibits, which included pythons, crocodiles, dart frogs, and Zoo Knoxville’s sloth. As part of the evening, Stan Tomov and Asim Yarkhan were recognized for their years of service to ICL and presented with gifts.
SLATE Highlighted in OLCF’s Exascale’s New Frontier Series
Oak Ridge Leadership Facility featured a profile of SLATE on the Exascale’s New Frontier blog series, which “explores the applications and software technology for driving scientific discoveries in the exascale era.”
“We wrote SLATE from scratch so that we could have a nicer, more modern interface that’s easier for applications to use and that targets GPUs from the ground up. We also built this to be portable, so it works on all the different DOE machines, like Frontier and El Capitan [Lawrence Livermore National Laboratory] with the AMD GPUs and Aurora [Argonne National Laboratory] with its Intel GPUs,” said Mark Gates, principal investigator for the SLATE project and a research assistant professor at the Innovative Computing Laboratory at the University of Tennessee, Knoxville.
Conference Reports
ACS Spring 2024 Panel Features ICL Associate Director Deborah Penchoff
ICL’s Deborah Penchoff was one of the featured panelists at the Spring 2024 meeting of the American Chemical Society. The panel, titled Artificial Intelligence and its Applications in Nuclear and Radiochemistry, featured discussion of a wide range of topics encompassing the broad spectrum of AI such as machine learning, neural networks, algorithm design, and data science and analytics, as well as how breakthroughs in high performance computing are enabling AI to provide unprecedented insights in the field of chemistry.
ICL Hosts 3 Workshops to start 2024
2nd Sparse BLAS Workshop
Working towards a common goal: ICL hosted the 2nd Sparse BLAS Workshop March 13-15th. Bringing together representatives from industry, academia, and government institutions, the workshop’s focus is on the design of an API for sparse linear algebra functionality.
WAMTA 2024
ICL hosted the 2024 Workshop on Asynchronous Many-Task Systems and Applications February 14th – 16th at the University of Tennessee Student Union. WAMTA 2024 brought together experts in asynchronous many-task frameworks, developers of science codes, performance experts, and hardware vendors to discuss the state-of-the-art techniques needed to to achieve maximum performance possible from modern machines. The lead organizers for this workshop included ICL’s George Bosilca, Thomas Herault, and Joseph Schuchart.
A group of WAMTA attendees took part in an outing to Fort Dickerson Quarry.
OpenMP Meeting
Between Jan 29 and Feb 2, ICL hosted the face-to-face meeting by the OpenMP ARB (Architecture Review Members). The ARB members came from academia, industry, and government to make progress towards OpenMP 6.0 that has an early release date in late 2024 and is currently previewed in the TR12 document available on the OpenMP’s web site. Among the new features considered for the release are haul tasks (sometimes described as weak dependencies between taskas), free-agent threads, and “taskgraph” directive related to similar functionality in CUDA or Kokkos. The follow up meetings will take place throughout the year across the globe to finalize 6.0 version of the standard document. The implementations of the new features are also ongoing to be available across multiple compiler toolchains.
Recent Releases
MAGMA 2.8
MAGMA 2.8 was released March 28, 2024 and is available at the MAGMA website. From the release notes:
- New functionality: band LU factorization and solve
- magma_{s,d,c,z}gbtrf_native computes the LU factorization of a band matrix using partial pivoting with row interchanges. This is equivalent to the LAPACK GBTRF routine.
- magma_{s,d,c,z}gbsv_native computes the solution to a system of linear equations A * X = B, where A is a band matrix and X and B are general dense matrices. This is equivalent to the LAPACK GBSV routine.
- magma_{s,d,c,z}gbtrfbatched and magma{s,d,c,z}gbtrf_batched_strided are the batched and the stride-batched versions of GBTRF, respectively.
- magma_{s,d,c,z}gbsvbatched and magma{s,d,c,z}gbsv_batched_strided are the batched and the stride-batched versions of GBSV, respectively.
- Native Cholesky factorization, magma_{s,d,c,z}potrf_native, now supports uplo = MagmaUpper
- Bug fixes:
- Batch QR factorization: fix numerical behavior for some corner cases
- Variable-size batch GEMM: fix numerical behavior when k = 0 and beta != 1
- GESV: fix failures for very large matrices (beyond 46k)
- Batch GESV: fix failure when the number of right hand sides is larger than 1024
- Fix compilation for rocm-6
- Multi-GPU syevd: fix failures on very large matrices
- Multi-GPU potrf: fix failures on 4 or more GPUs
Interview
Where are you from originally?
I was born and raised in Cairo, the capital of Egypt
Can you summarize your educational and professional background?
I studied computer engineering at Ain Shams University, Egypt, and completed my masters degree at the same institution in 2010. I then moved to Saudi Arabia to pursue a PhD degree in computer science at King Abdullah University of Science and Technology (KAUST). After graduating, I joined ICL as a postdoc in April 2015, and have enjoyed my time here since then.
How did you first hear about ICL, and what made you want to work here?
I have to give credit to David Keyes, a good friend of ICL, and Hatem Ltaief, a former ICLer well-known to many, for influencing and shaping my career. In particular, Hatem is the one who approached me and convinced me to join David’s group and pursue a degree in HPC. He introduced me to the work of ICL, and encouraged me to follow a similar path at KAUST, by designing optimized numerical software for GPUs. I was given the opportunity to visit ICL as an intern in Summer 2012, during which I got to know more about the great work and the high standards ICL strives for. I continued to maintain my connections with ICL, and fortunately received an offer to join the group as a postdoc. Thanks to a strong commitment and support from Jack and the entire LA group, my post-PhD career has been entirely at ICL. I love being here.
What are your main research interests and what do you work on at ICL?
My interests include parallel computing, numerical linear algebra, and general purpose GPU computing. I love developing optimized GPU software for the standard BLAS and LAPACK algorithms. Most of my work at ICL has been in the MAGMA project, especially its batch linear algebra component. I have also contributed to SLATE and indirectly to xSDK, and recently started to work on the heFFTe project.
What are your interests/hobbies outside of work?
Spending as much time as possible with my family is my first priority outside of work. I used to spend a lot of time reading different genres of novels in my native language (Arabic), but that has changed since I became a father.
Tell us something about yourself that might surprise people.
Sometimes I get obsessed with playing video games from the 90’s that I used to play as a kid. I have no interest in trying new games.
If you weren’t working at ICL, where do you think you would like to be working and why?
I have always considered myself an academic, so I will have probably pursued an academic career to eventually become a professor and direct my own group. Joining the industry is another likely possibility, since ICL Alumni are highly regarded by many HPC-oriented companies.