Re: [9fans] Rant (was Re: Plan9 and Ada95?)

[forsyth@vitanuova.com]

[-- Attachment #1: Type: text/plain, Size: 914 bytes --]

sorry, i wasn't clear.  i meant that you'd need to change the gnat code generator
to generate a different representation, not that you'd need another
code generator working from the RTL.  as you say, gcc's RTL
is machine-specific.

by `if gnat is structured so as to allow another ... in principle',
i meant, it has some internal representation of the Ada program that
is at or near a useful level for code generation, before it produces RTL.
if it is producing RTL directly from a basic parse tree,
you'll have your work cut out.

i don't remember dewar's description of gnat's structure,
except that it converts its internal representation of Ada
to RTL at some point, so i don't know the answer
and i can't be much more specific without looking
at Gnat's code (which i've also seen but forgotten as regards this point).
it was a good five years ago and i was curious
rather than interested.


[-- Attachment #2: Type: message/rfc822, Size: 2368 bytes --]

To: 9fans@cse.psu.edu
Subject: Re: [9fans] Rant (was Re: Plan9 and Ada95?)
Date: Fri, 9 Nov 2001 10:17:07 GMT
Message-ID: <874ro5vz8b.fsf@becket.becket.net>

forsyth@vitanuova.com writes:

> the gnat code generator doesn't generate code
> directly (or it didn't), it generates the gcc internal representation,
> so it's a non-trivial modification, even if gnat is structured
> so as to allow another code generator in principle.

The "gcc internal representation" (RTL) is not something a different
code generator could work with.  The definition of "valid RTL" is
entirely machine-specific and so forth.  When a GCC front-end begins
writing RTL, it is tightly coupled with the back-end.  (Indeed, it
generates RTL mostly by asking the back-end to spit out RTL for
various operations.)

The value of RTL in GCC is not to firmly decouple the front end from
the back end, but rather to decouple optimizations from the back end.

Thomas