Ulrich Drepper — Ulrich worked on free software for more than 20 years and contributed to many different projects, mostly in the system infrastructure area. He currently works at Goldman Sachs in the area of data science.
Vectorization, as opposed to parallelization, is less utilized as a means of exploiting the full capabilities of a processor. This is a problem since even today this means only ¼ to ½ of the performance of the CPU is used. This is only getting worse in future, especially as accelerators are becoming more prevalent.
After an introduction to the basics and history of vectorization the talk will introduce various techniques available for vectorization of compiled code. This talk will focus on gcc and for some details on Linux but the knowledge should be transferable if the features are fully implemented elsewhere.