Related articles |
---|
register variables in C. eric@pencom.com (1992-04-22) |
Static vs. dynamic analysis chuck_lins@gateway.qm.apple.com (Chuck Lins) (1992-04-24) |
Re: Static vs. dynamic analysis tmb@ai.mit.edu (1992-04-26) |
Re: Static vs. dynamic analysis chambers@cs.washington.edu (1992-04-26) |
Re: Static vs. dynamic analysis pardo@cs.washington.edu (1992-04-27) |
Re: Static vs. dynamic analysis alex.zatsman@spd.analog.com (1992-05-11) |
Re: Static vs. dynamic analysis stephen@estragon.uchicago.edu (1992-05-12) |
Newsgroups: | comp.compilers |
From: | stephen@estragon.uchicago.edu (Stephen P Spackman) |
Keywords: | experiment, optimize |
Organization: | University of Chicago CILS |
References: | 92-04-108 92-05-064 |
Date: | Tue, 12 May 1992 00:26:27 GMT |
alex.zatsman@spd.analog.com (Alex Zatsman) writes:
|I've spent some time working on a prototype code generator which took into
|account profiling information (it assigned execution frequence to each arc
|of the flow graph). The impression I've got is that the code generator is
|not overly sensitive to that data, i.e. it often does not matter whether
|a particular loop is executed one million times or ten million, but it
|does matter if it is executed only once or twice. Thus even a non-typical
|run produces useful information.
If you think about the decisions that the code generator makes, you'll
realise that it could emit hypotheses that need checking, and do
subsumption analysis on them. This could radically reduce the amount
of instrumentation that is needed at runtime, and provide fascinating
reading for the programmer: "if I was sure this was executed more than
2.8 times per call, I would have done it differently".
Hm. If I had my code generator written yet, this would have been worth
a couple of papers ;-).
--
stephen p spackman Center for Information and Language Studies
stephen@estragon.uchicago.edu University of Chicago
--
Return to the
comp.compilers page.
Search the
comp.compilers archives again.