Campus Event Calendar
Event Entry
What and Who
Enabling Reductions in Polyhedral Loop Nest Optimization
Johannes Dörfert
Fachrichtung Informatik - Saarbrücken
PhD Application Talk
Master studies / doctoral preparation in the
Saarbruecken Graduate School of Computer
Science
AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, MMCI
Public Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Monday, 7 October 2013
11:10
90 Minutes
E1 4
024
Saarbrücken
Abstract
If applicable, the polyhedral model provides powerful mathematical abstractions enabling effective optimization of loop nests with respect to given optimization goal, of which one of the most important ones is to enable parallelization. Nevertheless one of the most important parallelism enabling transformations, namely reduction recognition and exploitation, is out of scope of polyhedral schedulers so far.
In this paper we develop different increments of a reduction aware polyhedral scheduler. We show how to recognize opportunities, precisely model the resulting dependences and enable Polly, a state-of-the-art polyhedral optimizer based on LLVM, to take reduction possibilities into account during optimization.
All implemented approaches are evaluated on the full Polybench benchmark suite with respect to runtime, communication overhead and memory consumption.
In this paper we develop different increments of a reduction aware polyhedral scheduler. We show how to recognize opportunities, precisely model the resulting dependences and enable Polly, a state-of-the-art polyhedral optimizer based on LLVM, to take reduction possibilities into account during optimization.
All implemented approaches are evaluated on the full Polybench benchmark suite with respect to runtime, communication overhead and memory consumption.
Contact
Aaron Alsancak
068193251800
--email hidden
passcode not visible
logged in users only
Aaron Alsancak, 10/04/2013 11:45 -- Created document.