HIPERFITTalk by Amos Robinson on Fusion
Fusing Filters with Integer Linear Programming
Amos Robinson, University of New South Wales, Sydney, Australia.
Time and location: August 26, 12:15-13:00. The APL meeting room, DIKU, Universitetsparken 5, Building B, 2100 Copenhagen Ø.
To attend the talk, please signup by email to Martin Elsman (mael at di.ku.dk).
Abstract:
The key to compiling functional, collection oriented array programs into efficient code is to minimise memory traffic. Greedily fusing array operations together can reduce some memory traffic, but finding the fusion clustering that produces the least amount of memory traffic turns out to be NP-hard. Integer linear programming (ILP) can be used to solve NP-hard problems, by finding the minimal solution for a given cost metric. Previous work has used ILP to find fusion clusterings for certain graphs, however these approaches cannot handle size-changing operations such as filter. In this talk, I will show how size-changing operations differ, and how we have extended previous work to support such operations.
Bio:
Amos Robinson is a PhD student at University of New South Wales, Sydney, Australia. His focus is on optimisation and fusion in purely functional programming languages such as Haskell. He was also the winner of the inaugural Sydney Coq Fight, a theorem proving tournament.