Technical Papers Archive

Interactive Debugger for Performance-Portable Python HPC Kernels


Authors: Ivan Grigorik, Gabriel Kosmacher, George Biros, and Milos Gligoric (The University of Texas at Austin)

Abstract: We propose pkdb, the first interactive debugger for performance-portable kernels written in Python. Python is widely used in HPC, with frameworks such as PyKokkos translating Python-embedded domain-specific languages to native code that runs across various CPUs and GPUs. pkdb enables standard interactive debugging like breakpoints, stepping, continuation, and variable inspection while preserving actual on-device execution without source modification. Beyond these fundamentals, pkdb introduces three advanced capabilities that exploit the dynamic nature of Python and PyKokkos: (1) live code evaluation, (2) hot-swap, and (3) concurrent kernel comparison. Our performance evaluations on Intel and AMD CPUs and NVIDIA and AMD GPUs show that pkdb introduces limited overhead and is practical for everyday use, and introduces critical debugging features to the Python HPC ecosystem.


Presentation: file

Back to Technical Papers Archive Listing