This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] (Partial) fix for PR middle-end/13392
- From: Jakub Jelinek <jakub at redhat dot com>
- To: Roger Sayle <roger at eyesopen dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Thu, 8 Jan 2004 06:19:57 -0500
- Subject: Re: [PATCH] (Partial) fix for PR middle-end/13392
- References: <20031231153849.GC2020@sunsite.ms.mff.cuni.cz> <Pine.LNX.4.44.0401070658350.22521-100000@www.eyesopen.com>
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
On Wed, Jan 07, 2004 at 07:12:45AM -0700, Roger Sayle wrote:
> My completely uninformed opinion, is to avoid falling back to SCC
> entirely, then you'll no longer have to worry about the NOTES and
But then we risk that expand_builtin_expect_jump will not be able
to add branch probability note to the jump when it for some reason
does not understand what do_jump created.
This can happen e.g. for if (__builtin_expect (1, 1)),
if (__builtin_expect ((foo (), 0), 0)) (where it probably doesn't
matter if falling back or not setting probability note - the jump
is unconditional anyway), but perhaps could happen on some architectures
even on code where it does matter.
The question is what is worse, if possibly pessimizing the code
with SCC but ensuring the probability note is there (then e.g.
block reordering could make advantage of it) or emitting best
jump code but not emitting the note.
Jakub