| |||||||||||||
Rethinking optimization for sizeRethinking optimization for sizePosted Feb 8, 2013 14:12 UTC (Fri) by jdbrandmeyer (subscriber, #85840)In reply to: Rethinking optimization for size by robert_s Parent article: Rethinking optimization for size
Other way around, actually. Acovea came well in advance of your LLVM example, and the target compiler was GCC.
(Log in to post comments)
Rethinking optimization for size Posted Feb 8, 2013 17:56 UTC (Fri) by daglwn (subscriber, #65432) [Link]
Oh, this was done long before that. See for example:
http://dl.acm.org/citation.cfm?id=1134650.1134663&col...
http://dl.acm.org/citation.cfm?id=603339.603341&coll=...
http://dl.acm.org/citation.cfm?id=314403.314414&coll=...
And that's just a small sample. In addition, there are many papers using genetic algorithms to drive the heuristic decisions within various transformation passes.
Rethinking optimization for size Posted Feb 8, 2013 22:20 UTC (Fri) by Shewmaker (subscriber, #1126) [Link] The first project that I remember seeing that seriously pursued automatically optimizing for whatever architecture you compiled it on was the ATLAS linear algebra library. It was moderately successful, but it could be beat by Goto's BLAS.I remember ACOVEA too, but it looks like it is no longer being maintained. There is a current effort, Collective Tuning that goes beyond ACOVEA's intentions. They compare it to their Continuous Collective Compilation Framework I don't know how it compares to the LLVM work.
|
|||||||||||||