On 31-Mar-2003, Miguel de Icaza <[EMAIL PROTECTED]> wrote:
> > > The important bit about this work is the fact that the register-based
> > > RTL representation is used to map to CIL code.  The other nuisances are
> > > technical details.  
> > > 
> > > In the "solvable" universe, not in the "impossible" universe.
> > 
> > Sure, the problem is solvable.  But IMHO the approach under discussion
> > is not a useful step toward a solution -- instead, it's just exploring
> > a dead end.
> 
> So what would be your solution, do you have any alternative proposals? 
> Which part in particular you believe is the dead end?  What would be
> your proposal?

Generating code which uses a big array as a stack and has stack pointer
frame pointer registers that point into this array is definitely not
the right approach.

Going to RTL at all is probably a mistake, I think -- a GCC back-end
for CIL should probably go directly from GCC trees (e.g. the new GIMPLE
tree representation, see http://gcc.gnu.org/projects/tree-ssa/#gimple)
rather than RTL.

-- 
Fergus Henderson <[EMAIL PROTECTED]>  |  "I have always known that the pursuit
The University of Melbourne         |  of excellence is a lethal habit"
WWW: <http://www.cs.mu.oz.au/~fjh>  |     -- the last words of T. S. Garp.

--------------------------------------------------------------------
This message could not be encrypted because no digital certificate
was found for your address. However, you can encrypt your reply with
the certificate attached below. Details at www.unicrypt.com.au

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to