ZIH-Colloquium: "Mitigating Numerical Inconsistencies and Exceptions in Heterogeneous HPC Systems"

Scientific computing applications running on heterogeneous systems may suffer from numerical reproducibility and correctness issues, which can take a significant amount of time to detect and fix. This talk will present several tools and techniques to isolate numerical errors that arise when scientific software is ported to GPUs. The approaches include isolating compiler-induced numerical inconsistencies, detecting numerical exceptions, and finding inputs that trigger exceptions in GPU software.

Dr. Ignacio Laguna is a Computer Scientists and a Group Leader at the Center for Applied Scientific Computing at the Lawrence Livermore National Laboratory (LLNL), California. His research is focused on programing models for HPC, software correctness, debugging tools, and fault tolerance. He received his PhD in Computer Engineering from Purdue University in 2012 and has been at LLNL for more than ten years.

The colloquium is free of charge. Language: German/English
The event will be offered in HYBRID. You can join the ZIH-Colloquium as following:

• LIVE: Andreas-Pfitzmann-Bau, Room 1004
• ONLINE: Link ZIH-Colloquia