Re: [lno] [PATCH] Loop Versioning

[Devang Patel <dpatel at apple dot com>]

On Jan 29, 2004, at 3:28 PM, Zdenek Dvorak wrote:

> Hello,
>
> > > this patch should fix this.
> >
> > It causes abort() at
> >
> > find_induction_variable() --> find_givs() -->
> > get_loop_body_in_dom_order()
> >
> > > But first, could you please check with what
> > > statement the preheader of the loop in this case ends?
> >
> > conditional statement
> > if (cond) goto L1; else goto L2;
> >
> > > I assumed that
> > > since I force the preheader to have just a single successor,
> >
> > hmm... in loop versioning I create a pre-header with two
> > successors. One is original loop and second is new version of
> > original loop.
>
> could you please change this (call loop_split_edge_with (e, NULL)
> on both created edges), so that this property is fulfilled?

OK. I did not know about this restriction. Now I am able to make 
progress.
But it seems find_optimal_iv_set_1() goes in infinite loop while 
compiling
rtl.c on powerpc-darwin, with or without loop versioning. (without your
patch to handle loop preheader with two successors)

> It makes
> some loop manipulation functions simpler, and some of them thus rely
> on the fact that it is so.
>
> Zdenek
--
Devang