Re: Teaching compilers backwards?

Rome Samo <romesamo@earthlink.net> wrote:
>Sure would be good if we could generate some kind of pcode, then have
>all the guru's optimize it, and produce assembly for various processors.
> I have looked at a lot of pcode, and, frankly am not too excited about
>what's out there. Rather than pcode generation, if we put our heads
>together, and create a plain English verion of this, it sure would go a
>long way. I guess what this means is to create a new pcode language
>that is readable by human beings (text), and does not assume processor
>architecture. The language should assume that there are registers,
>stack, memory, and should be easy to look at.


OK, I can't resist putting in a plug for C--, the compiler target
language designed primarily by Simon Peyton Jones and me. C-- is
somewhat readable, supports multiple hardware platforms, and has
direct support for registers and memory. (Memory may be in
initialized data, on the stack, or in a dynamically allocated heap.)
C-- is designed to support a constellation of low-level optimizations,
but the only one we've actually implemented is a modern,
graph-coloring register allocator (thanks Joao Dias).


We currently generate bad code for x86-linux and ppc-osx.
More targets and more optimizations will be coming this summer, but if
you want to check us out, go to www.cminusminus.org.


One more thing: from an academic, intellectual point of view, the real
novelty behind C-- is a *run-time* interface that is intended to
support garbage collectors, exception dispatch, user-level threads,
and other run-time services. We are very interested in working with
people who wish to compile languages that require such services.
--
Norman Ramsey
http://www.eecs.harvard.edu/~nr

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