The biannual SIAM Conference on Computational Science and Engineering (CSE) was conducted virtually between March 1 to 5, 2021. Theo Mary and I organised a two-part minisymposium on recent algorithmic and software advances of mixed precision methods in scientific computing. Below are the links to the slides of the talk.
Minisymposium description: The increasing support of lower precision arithmetics in hardware, such as fp16 and bfloat16, provides new opportunities for high performance scientific computing. However, even though low precision arithmetics can provide significant speed, communication, and energy benefits, their use in scientific computing poses the challenge of preserving the accuracy and stability of the computation. To address this issue, a variety of mixed precision algorithms that combine low and high precisions have emerged.
- Part 1
- Mixed Precision Low Rank Compression and its Application to BLR Matrix Factorization. Patrick R. Amestoy, Mumps Technologies, France; Olivier Boiteau, EDF, France; Alfredo Buttari, CNRS-IRIT, France; Matthieu Gerest, EDF R&D, France; Fabienne Jezequel, Sorbonne Universités, France; Jean-Yves L’Excellent, Mumps Technologies, France; Theo Mary, Sorbonne Universités and CNRS, France.
- Tile-Centric Mixed Precision Computations for HPC Applications. Sameh Abdulah, Kadir Akbudak, and Rabab Alomairy, King Abdullah University of Science & Technology (KAUST), Saudi Arabia; George Bosilca and Qinglei Cao, University of Tennessee, Knoxville, U.S.; Jack J. Dongarra, University of Tennessee and Oak Ridge National Laboratory, U.S.; David E. Keyes and Hatem Ltaief, King Abdullah University of Science & Technology (KAUST), Saudi Arabia; Yu Pei, University of Tennessee, U.S.
- Mixed Precision LU Factorization using GPU Tensor Cores: Improving Performance, Accuracy, and Memory Footprint. Florent Lopez, University of Tennessee, Knoxville, U.S.; Theo Mary, Sorbonne Universités and CNRS, France.
- Mixed Precision Cholesky-QR Algorithm with Applications. Francoise Tisseur and Srikara Pranesh, University of Manchester, United Kingdom
- TSQR on Tensor Cores with Error Correction. Hiroyuki Ootomo and Rio Yokota, Tokyo Institute of Technology, Japan.
- Part 2
- Three-Precision GMRES-Based Iterative Refinement for Least Squares Problems. Srikara Pranesh, University of Manchester, United Kingdom; Erin C. Carson, Charles University, Czech Republic; Nicholas J. Higham, University of Manchester, United Kingdom.
- How NVIDIA Tensor Cores can Help HPC Scientific Application Unleash the Power of GPUs using Mixed Precision Solvers. Azzam Haidar and Harun Bayraktar, NVIDIA, U.S.; Abeynaya Gnanasekaran, Stanford University, U.S.
- Iterative Refinement in up to Five Precisions for the Solution of Large Sparse Linear Systems. Patrick R. Amestoy, Mumps Technologies, France; Alfredo Buttari, CNRS-IRIT, France; Nicholas J. Higham, University of Manchester, United Kingdom; Jean-Yves L’Excellent, Mumps Technologies, France; Theo Mary, Sorbonne Universités and CNRS, France; Bastien Vieublé, Universite de Toulouse, France.
- Compressed Basis GMRES on High Performance GPUs. José I. Aliaga, Universitat Jaume I, Spain; Hartwig Anzt, University of Tennessee, U.S.; Thomas Gruetzmacher, Karlsruhe Institute of Technology, Germany; Enrique S. Quintana-Orti, Universidad Jaume I, Spain; Andres Tomas, Universidad Politecnica de Valencia, Spain.
- DGEMM using Tensor Cores. Daichi Mukunoki, RIKEN, Japan; Katsuhisa Ozaki, Shibaura Institute of Technology, Japan; Takeshi Ogita, Tokyo Woman’s Christian University, Japan; Toshiyuki Imamura, RIKEN, Japan.