gomp-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gomp-discuss] Somethings to think about ....


From: Steven Bosscher
Subject: Re: [Gomp-discuss] Somethings to think about ....
Date: 11 Mar 2003 01:43:45 +0100

Op di 11-03-2003, om 01:27 schreef Pop Sébastian:
> On Mon, Mar 10, 2003 at 11:51:05PM +0100, Steven Bosscher wrote:
> > Op ma 10-03-2003, om 23:02 schreef Pop Sébastian:
> > >I really think that 
> > > for the moment what we call GENERIC is what came out from C and C++ 
> > > without 
> > > worrying much about other languages as Fortran, Java, Ada, Pascal, ...
> > 
> > Well thank you very much :#
> > 
> I'm sorry but that was true at the moment when GENERIC was designed :-(
> and I think we'll not end with a suitable generic IR by just including nodes 
> as 
> it was been done until now.  I really think we should base our work on a 
> standard 
> and avoid, if possible, to reinvent the wheel.
> 
> Sorry again Steven, I wasn't thinking bad when I wrote that statement.  I 
> just 
> meant that GENERIC was mostly the union of the C and C++ front ends and that 
> its
> conception was mostly been based on the "empiric" discovery that front-end
> representations could be factored to a generic representation.  Here it would 
> be
> a good time to formalize a little this GENERIC representation.  We'll just 
> win in 
> the adoption of this standard, since the most of the semantics documentation 
> would
> be already there: the standard itself.  

<laugh> I was just teasing.

I am all for a good documentation of the GENERIC tree codes and their
semantics.  Right now it is not even clear which tree codes will
eventually make up GENERIC.  I discussed this shortly with Diego, a few
weeks ago.  I had started documenting GENERIC, but that is on hold a bit
now because it really isn't quite clear to me (or anybody, apparently)
what trees we really need to make GENERIC generic enough.

It is an interesting idea to define GENERIC as "an IR capable of
representing CIL + GNU C/C++ extensions" and build the IR from there.  I
think it is a bad idea to re-invent the wheel.

But in the end, no IR is OK for Fortran if it cannot represent array
assignments.  I had a dream... :-)

You pick a lot of hay on that fork of yours.  You should try to get
payed for hacking on GCC ;-)
(Or is all this a voluntary effort?  Maaaan, way cooool, duuude :-D )

Greetz
Steven






reply via email to

[Prev in Thread] Current Thread [Next in Thread]