Re: Any further work on superoptimizer?

If memory serves me, the super-optimizer did its work by representing the
effect of each instruction in some encoded form (roughly Boolean). This
form worked well for some instructions and would "explode" for others. As
a result, the set of instructions it would use in building the
super-sequences was a subset of the machine's capabilities. I wouldn't be
surprised that someone could then code up a shorter sequence using
instructions not covered by the original optimizer. However, unless there
was a bug in Henry's implementation, I don't see how one could do better
if you kept yourself restricted to the same subset. The search was
*exhaustive*.
--
Robert M. Fuhrer, Computer Science Department, Columbia University
1117B Fairchild Building
Internet: rmf@cs.columbia.edu, UUCP: ...!rutgers!cs.columbia.edu!rmf
--

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