This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: constructing webs from ssa.
- From: Sanjiv Kumar Gupta <sanjivkumargupta at yahoo dot com>
- To: law at redhat dot com
- Cc: gcc at gcc dot gnu dot org
- Date: Sun, 24 Oct 2004 23:18:53 -0700 (PDT)
- Subject: Re: constructing webs from ssa.
One more question Jeff.
If all operands (src and dest) of a phinode are
essentially in a single
web, then they all are assigned same hard register.
In that case, I think we don't need to insert copy
insns at the end of predecessor blocks while
restoring back from SSA from.
Is that correct?
Let me know if my question in not clear enough.
Thanks
--Sanjiv
--- Jeffrey A Law <law@redhat.com> wrote:
> On Tue, 2004-10-12 at 04:06, Sanjiv Kumar Gupta
> wrote:
> > 1. make a web for each du-chain.
> > 2. combine those webs that have a 'common use',
> and
> > this
> > 'common use' can only be a phi-function.
> >
> > that's all we need to do for computing webs from
> ssa.
> Right. However, it's often beneficial to be able to
> construct
> webs in just a region of the function. ie,
> construct the webs
> for the innermost loops, use the webs for some kind
> of optimization,
> then construct webs for the parent loop nest and so
> on.
>
> So, if you're pondering writing some code to create
> webs, you might
> keep keep that in mind.
> jeff
>
>
>
_______________________________
Do you Yahoo!?
Declare Yourself - Register online to vote today!
http://vote.yahoo.com