For several important program classes (image processing, scientific computing, ...) High Level Loop Optimizations are essential to reach top performance. With Polly, we present a high-level loop optimization framework for LLVM, which provides a flexible infrastructure to develop and describe such high-level loop optimizations.
We present Polly as a classical optimizer for a C compiler, but focus on its use as an infrastructure to develop optimizations for domain specific languages and specialized hardware such as GPUs. This talks gives an overview of the overall structure of Polly itself and also insights into the components and tools essential when working with Polly (isl, ppcg, islpy, islplot).
Speakers: Tobias Grosser