Re: RTL-based compilers

I write to clarify a recent posting in which someone said:


The RTL intermediate language work was originally done by Davidson &
Fraser, then at University of New Mexico, I think. Anyway, Davidson is
now at the University of Virginia, and is continuing to work on that base,
called VPCC (Very Portable C Compiler). I think Fraser long-since
abandoned this line of research. Davidson's students continue to publish
papers on various minor aspects of this technology.


The work was originally done at the University of Arizona, not the
University of New Mexico.


The phrase "long-since abandoned" might be misconstrued. I have indeed
been working with other intermediate representations for the last three
years, but that's because the priorities for my current compiler -- "lcc",
written with Dave Hanson at Princeton -- emphasized simplicity and
compilation speed over hairy optimization, and I feared that an
intermediate step like RTL would cost too much. If, however, I were
working on a more ambitious optimizer, I would probably be using register
transfers. I think that they're the representation of choice when a
machine-independent optimizer needs to represent the effect of a
machine-specific instruction.


Chris Fraser
AT&T Bell Laboratories
cwf@research.att.com
--

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