This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: loop.c fix
- To: Joern Rennecke <amylaar at cygnus dot co dot uk>
- Subject: Re: loop.c fix
- From: Jeffrey A Law <law at hurl dot cygnus dot com>
- Date: Fri, 05 Feb 1999 23:31:13 -0700
- cc: egcs-patches at cygnus dot com, gcc-local at cygnus dot com
- Reply-To: law at cygnus dot com
In message <199902060610.GAA21300@phal.cygnus.co.uk>you write:
> For loops like where the current ppc failure is, we got a
> NOTE_INSN_LOOP_CONT that is quite a few instructions - and givs - after
> the last label.
Yes. Similarly for the bug which got me looking at this to start with. As
far as I can tell this is a relatively common case
> So if we also had a label after the NOTE_INSN_LOOP_CONT, we'd miss the
> always_executed for a number of givs.
And more importantly bivs -- which can make it impossible to determine the
number of iterations for the loop. Which is precisely the problem I fixed.
> Or maybe in the test if we pass the last label in the loop, use the
> NOTE_INSN_LOOP_CONT instead of the NOTE_INSN_LOOP_END to delimit the insns
> where we search for another label.
Clearing not_every_iteration at the last label before the LOOP_CONT note
or at the LOOP_CONT note itself has the same net effect as we'd get if we
moved the notes.
jeff