https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
Bill Schmidt changed:
What|Removed |Added
Status|NEW |RESOLVED
Resolution|---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #11 from Bill Schmidt ---
Author: wschmidt
Date: Mon Jun 26 14:19:33 2017
New Revision: 249649
URL: https://gcc.gnu.org/viewcvs?rev=249649=gcc=rev
Log:
[gcc]
2016-06-26 Bill Schmidt
PR
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #10 from Bill Schmidt ---
I re-ran benchmarks today and the results that I saw before are no longer
present. The patch is neutral with regard to SPEC cpu2006 performance on
ppc64le. So I'll plan to have this patch reviewed.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #9 from Bill Schmidt ---
I'm not comfortable with the results of the patch. Overall I see a slight
improvement for SPECint CPU2006 and a slightly larger degradation for SPECfp
CPU2006. But there are some individual slowdowns that
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #8 from Bill Schmidt ---
Created attachment 39085
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=39085=edit
Patch under test
Attaching a patch that passes regstrap. I want to do a little benchmarking
before submitting it
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #7 from Bill Schmidt ---
I have a prototype that fixes this in the obvious way and it causes both
slsr-35.c and slsr-36.c to pass again without turning off code hoisting. I'll
do a regstrap and then work on some benchmark testing.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #6 from Bill Schmidt ---
Actually, it looks like a similar problem for the unknown stride case. Again
there is logic that relies on single-reached-use for determining what
expressions go dead. We need to factor in expressions that
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #5 from Bill Schmidt ---
I'll note that in the case where the stride is known (slsr-35.c), SLSR is
making at least a somewhat rational decision based on cost not to
strength-reduce the phi candidate. In this case the stride is a
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
Bill Schmidt changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #3 from Bill Schmidt ---
OK. I'm busy wrapping up some things before a vacation, but I'll plan to look
into this when I get back.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #2 from rguenther at suse dot de ---
On Fri, 8 Jul 2016, wschmidt at gcc dot gnu.org wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
>
> --- Comment #1 from Bill Schmidt ---
> Interesting. How do we enable/disable
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71815
--- Comment #1 from Bill Schmidt ---
Interesting. How do we enable/disable code hoisting? I don't see a documented
option for this.
12 matches
Mail list logo