[tree-ssa] Overlapping live range patch.

[Andrew MacLeod]

On Wed, 2003-06-04 at 13:56, Andrew Pinski wrote:
> 
> On Wednesday, Jun 4, 2003, at 13:44 US/Eastern, Andrew MacLeod wrote:
> 
> > RSN for sure. no, now. just about. now. no, not quite... Thats been my
> > last week or so :-)
> >
> > So, It now bootstraps and appears to do everything fine, with one minor
> > exception. The stage 1 compiler miscompiles the fortran file 
> > f/intrin.c,
> > and this results in a bootstrapped fortran compiler which fails a lot 
> > of
> > the fortran tests. If you dont bootstrap, it works great :-|
> >
> > The problem is that the into SSA pass is generating a structure copy
> > which has a VDEF as a real use. This is a no-no. The out of SSA pass
> > creates a new temporary which is undefined on entry since there is no
> > defintion present. This of course causes problem :-)
> >
> > Diego is looking at is today.
> 
> 
> Sounds like the problem I having while bootstaping on 
> powerpc-apple-darwin6.6 where it miscompiles something (on the first 
> stage, I think rs6000.c) then it miscompiles cppexp.c and then it 
> causes not being able to compile libgcc and fails to bootstrap see my 
> pervious emails to see how ccpexp.c is miscompiled.
> 

We had half a dozen minor things which all showed the same symptoms.
 This patch fixes all of them except the one Diego is looking at,
because it is an *into* ssa issue. The others were mostly related to
unexercised paths in the insert on edge routines which split basic
blocks, or PHI node processing.

I'd give it another try with this.

Andrew