Re: Any further work on superoptimizer?

In article 92-01-078 mark@hubcap.clemson.edu (Mark Smotherman) writes:
>Henry Massalin published a paper on a superoptimizer in ASPLOS II, 1987
>[1]. One important aspect of this work was the generation of *branchless*
>run-time library routines for simple functions (e.g., absolute value, max)
>that could be in-line substituted. Thus the overhead of procedure call
>and the problem of branch stalls / delay slot scheduling were
>avoided.


I don't know about the superoptimizer, but people think about these things
in their spare time, and come up with additional sequences.


Back at Olivetti people in Acorn with whom we worked would put forth the
most astounding sequences of code for things like byte-reversal.


In recent months Preston Briggs and I traded a number of increasingly
horrible tricks for doing multiplication by constants (taking advantage of
the wonders of two's complement arithmetic, and the fact that ignoring
overflow, we're really computing in a ring). We've been meaning to do
something with this (i.e., publish it), but the short answer is that you
can do better than reported in the Bernstein paper (Robert Bernstein, SP&E
16:7, 1986), but it probably doesn't matter.


David Chase
--

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