🚀 go-pugleaf

In article <4061@hubcap.UUCP> mcvax!etive.ed.ac.uk!gvw@uunet.UU.NET (MOBY) writes:
>I'm looking for "clever" programming tricks.

A year or so ago someone showed me a paper in which the author described
a semi-automated hack generator (my phrasing, not the author's).
Essentially, the generator would be given a description of a common
function (say, min() or max() or some such), and would then attempt to
generate the shortest-length sequence of assembly language instructions
(or perhaps, shortest sequence with fewest branch instructions) that
would compute the function.  Yes, I know this is uncomputable in
general.  The point was that 1) the functions given were all both simple
and reasonably common, thus being perhaps worth a certain amount of
effort (from the point of view of writers of code generators), and 2)
the program didn't actually try to prove its answers.  It would test its
generated functions for equality with the supplied function at (say) 100
scattered points.  If the hack worked on those points it would be
returned to the user, who could then attempt to prove it correct.

Some of the code generated by this thing was extremely non-intuitive,
even when it was correct, making use of obscure flag settings and such,
and certainly qualified for the "hack" title.

Does any one recognize this work?  I'll try to dig up the reference.



============================================================================
David Wald                                              wald-david@yale.UUCP
waldave@yalevm.bitnet                                 wald-david@cs.yale.edu
"A monk, a clone and a ferengi decide to go bowling together..."
============================================================================