Re: 'Superoptimizers'

  bill@amber.ssd.hcsc.com (Bill Leonard) writes:
|> Trouble is, the time taken in most optimizers is not in performing the
|> optimizations, but in looking for the opportunities. ...
|>
|> If someone wants a topic for research, here it is: Come up with a fast (and
|> reliable) predictor of optimizations.


The trouble is that it is a solved problem, and is equivalent to
predicting whether a Turing machine will halt! At best, you can get
it right most of the time.


I would slightly dispute your statement that it is likely that there
are few opportunities for optimisation - it is more usual that there
are huge numbers, but insufficient information to decide which ones
will do good, and which will do harm! But I rather doubt that you
would classify such things as opportunities :-)




Nick Maclaren,
University of Cambridge Computer Laboratory,
New Museums Site, Pembroke Street, Cambridge CB2 3QG, England.
Email: nmm1@cam.ac.uk
Tel.: +44 1223 334761 Fax: +44 1223 334679
--

Return to the comp.compilers page.
Search the comp.compilers archives again.