This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Make simple loop peeling to happen at gimple level
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: Steven Bosscher <stevenb dot gcc at gmail dot com>
- Cc: Jan Hubicka <hubicka at ucw dot cz>, gcc-patches at gcc dot gnu dot org, rguenther at suse dot de
- Date: Tue, 19 Nov 2013 18:14:26 +0100
- Subject: Re: Make simple loop peeling to happen at gimple level
- Authentication-results: sourceware.org; auth=none
- References: <20131119124206 dot GA10503 at kam dot mff dot cuni dot cz> <CABu31nPnH_ZnifOJBdJzZRPAF1iE67yZJrmdHnKSWgmKK-JKhA at mail dot gmail dot com>
> On Tue, Nov 19, 2013 at 1:42 PM, Jan Hubicka wrote:
> > The patch also removes simple peeling at RTL level that is mostly obsoleted by
> > gimple level (in few cases in our testsuite the RTL code is able to determine
> > loop bounds better than gimple code, but these seems to be just weird cases that
> > should be handled independently at gimple level)
>
> No targets that emit small-count loops for block moves?
Well, we have move_by_pieces and set_by_pieces for that reason. simple peeling
works only for loop with profile, so those expand-generated loop won't count.
I can imagine someone would emit loop with known count small enough so complete
peeling may match. I would however declare this to be expansion work to choose
proper loop/non-loop consruct in these cases rather than relying on rtl
peeling. It is what happens for i386.
Honza
>
> Ciao!
> Steven