Re: [tree-ssa vs lno] who is right?

[Andrew MacLeod <amacleod at redhat dot com>]

On Fri, 2004-03-26 at 10:30, law@redhat.com wrote:
> In message <1080308082.12528.27.camel@p4>, Andrew MacLeod writes:
>  >I have some minor changes to copyrename in the queue, but I doubt it
>  >will affect this. Since there is a constant in the arguments of the PHI,
>  >it must have been renamed due to a copy between a temp and maxmin_Result
>  >elsewhere in the program. The hope was that we'd be able to get rid of
>  >the copy.  Looks like not, unless one or more of the PHIs can be
>  >optimized away. :-)
> I'd expect DOM to propagate the constants so that there were no uses
> of maxmin_Result_140, maxmin_Result_142, then I'd expect DCE to zap the
> useless PHIs.
> 
> Alternatively Zdenek's block trivial PHI removal code may zap them.
> 
> But again, I think step #1 is to figure out how we got them in the first
> place.
> jeff
> 

It would be pretty easy, something like:

maxmin_Result_140 = PHI <1(10)>;
T.55_142 = PHI <2(10)>;
<...>
maxmin_Result_133 = T.55_142

could result in the 2 PHI nodes since we'd rename T.55_142 to
maxmin_Result_133 in hopes of getting rid of the copy.

It could also happen if T.55_142 is later used as a PHI argument to a
maxmin_Result PHI result.

Andrew