|Recognizing complicated patterns firstname.lastname@example.org (Ralph Johnson) (1990-08-01)|
|Re: Recognizing complicated patterns email@example.com (1990-08-06)|
|Re: Recognizing complicated patterns pd@complex.Eng.Sun.COM (1990-08-15)|
|From:||Ralph Johnson <firstname.lastname@example.org>|
|Date:||Wed, 01 Aug 90 13:06:05 GMT|
David Chase argued that it was already known how to find instructions that
matched complicated patterns in the original program. I am familiar with
about half of the papers that he mentioned, so perhaps I am missing
something, but it was my understanding that these techniques were not very
good for matching instructions with loops in them, like block moves.
Further, the incompleteness of arithmetic says that it may not be possible to
canonicalize all input patterns. The tree matching techniques are very good
at finding complex addressing modes, but they are not the end-all and be-all
of code generation.
I would love to be proven wrong, because we have built our own code generator
(in Smalltalk, which is why we didn't use someone else's), and are always
looking for ways to improve it.
By the way, the GNU compilers use a pattern matching code generator. Also, I
agree that those who are interested in this area should take a look at Robert
Ralph Johnson - University of Illinois at Urbana-Champaign
Return to the
Search the comp.compilers archives again.