Re: Compiler analysis

[dor@lanl.gov (David Rich)]

In article <8812162238.AA27888@sesame.stanford.edu>, mkatz@SESAME.STANFORD.EDU (Morris Katz) writes:
> In order to yield better performance for my parallelization system, I am trying
> to derive an algorithm for ddetermining which calls to cons, make-vector, etc.
> generate objects which are guaranteed not to be mutated anywhere in a program.
> Any ideas on this subject or references to related literature would be greatly
> appreciated...


Well, here's a start:

   %A Adrienne Bloss
   %A Paul Hudak
   %A Jonathan Young
   %T Code Optimizations For Lazy Evaluation
   %R Draft
   %I CSD, Yale UNIV
   %C New Haven, CT
   %D 1988


Also, Adrienne's PhD dissertation is a good reference (it predates this
one I believe) -- most likely available as a technical report from Yale.
She uses a technique called "Path Analysis" (a form of abstract interpretation)
to determine when destructive updates are possible.  However, if I remember
correctly, her work deals mainly with first order languages and serial evaluation.
Hope this helps!

--Dave