This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Rewrite of pure/const pass and nothrow stuff
- From: Rafael Espindola <espindola at google dot com>
- To: Jan Hubicka <jh at suse dot cz>
- Cc: gcc-patches at gcc dot gnu dot org, dnovillo at google dot com, zadeck at naturalbridge dot com
- Date: Fri, 14 Nov 2008 12:42:42 +0000
- Subject: Re: Rewrite of pure/const pass and nothrow stuff
- References: <20081113190834.GA11360@kam.mff.cuni.cz>
Sorry for taking so long to reply. Just a small comment.
> 2) There is local variant of the pass. It can be done during early optimization
> and at the place original RTL pass executed. Especially in early
> optimization we are effective. Main motivation for doing so early is to strip down EH
> and dead calls to get early inliner more informed; main motivation for late pass is to avoid
> dwarf2out from unnecesarily outputting unwind info and being stupid.
> IPA pass is the only able to handle non-trivial recursion cycles.
The current pass has a hidden ordering problem. If we have
void f(void) {g()};
the pass will mark g as nothrow and that will invalidate f's eh
regions since the call will be in a eh region but will not throw. An
IPA pass could update f to fix it. With your patch, don't we still
have the same problem in the early and RTL passes?
Thanks!
--
Rafael Avila de Espindola
Google | Gordon House | Barrow Street | Dublin 4 | Ireland
Registered in Dublin, Ireland | Registration Number: 368047