Traditionally performance data is presented based on code regions. However, for many situations it is useful to understand performance through the data structures in a parallel program. In this talk, I will discuss our experience with developing data centric measurement techniques for the Cuda and Chapel programming models.
Slides not available for download yet
Jeff Hollingsworth is the Vice President of IT and CIO for the
university of Maryland. He is also a professor in the Computer Science
Department. His research is focused on developing tools to make it easier
for programmers to get high performance from their parallel programs.
Back to Top