IPA merge part 1: localize SSA variables

Jan Hubicka jh@suse.cz
Wed Nov 15 21:14:00 GMT 2006


> Jan Hubicka wrote on 11/15/06 15:59:
> >>Jan Hubicka wrote on 11/15/06 15:49:
> >>
> >>>OK, I will use gimple_df then (or some shorter variant?).  What about
> >>>the ancestor macros/functions problem?  I
> >>>
> >>My vote is to qualify all of them with a CFUN argument.  But this will 
> >>certainly have repercussions in branches.  Let's see what other 
> >>maintainers think.
> >
> >I think we settled down that we do want to have accestors.  I meant
> >whehter we want to use something like
> >
> >#define blah(fun) (fun)->df_gimple->blah
> >or
> >#define gimple_blah(fun) (fun)->df_gimple->blah
> >
> >those might have problems if we want to embed some logic in it (such as
> >I do with in_ssa_p now testing whether fun->df_gimple is really nonzero,
> >so I can't use it as LHS anymore).  We can follow tree-flow-inline logic
> >too:
> >inline type
> >blah(fun)
> >{
> >  return (fun)->df_gimple->blah
> >}
> >inline void
> >set_blah(fun,val)
> >{
> >  fun->df_gimple->blah = val
> >}
> >alternatively with the gimple_ prefixes or some other variant.
> >
> Well, I'm biased.  I kind-of like blah/set_blah, but I think I'm in the 
> minority, so I won't push for it.

I would vote for blah/set_blah too (for consistency and better
extendability), so unless someone speaks out soon, I would go for it.
> 
> Regarding the accessor name, might as well prefix them with gimple_

This will imply more reformating for me, but it is definitly better to
do it at once if there is overall plan for the prefixing.  Shorter
prefix would be nice ;)

Honza



More information about the Gcc-patches mailing list