http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #22 from Teresa Johnson tejohnson at google dot com 2013-01-11
18:18:48 UTC ---
Hi Honza,
I ran a number of experiments at different thresholds, and found that
performance starts dropping pretty quickly as the working set
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #21 from Jan Hubicka hubicka at ucw dot cz 2012-12-22 23:20:37
UTC ---
I'll give this patch a try and let you know how it affects the
performance I see. But unrolling shouldn't affect inlining, since all
unrolling is after
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #19 from Jan Hubicka hubicka at ucw dot cz 2012-12-21 16:15:34
UTC ---
As another data point, in our internal benchmarks I had tried a few
values and 99.9% gave the best performance. Just going down to 99.0%
reduced the
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #20 from Teresa Johnson tejohnson at google dot com 2012-12-21
16:26:17 UTC ---
On Fri, Dec 21, 2012 at 8:15 AM, hubicka at ucw dot cz
gcc-bugzi...@gcc.gnu.org wrote:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
---
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #18 from Teresa Johnson tejohnson at google dot com 2012-12-19
16:44:21 UTC ---
On Tue, Dec 18, 2012 at 9:25 AM, hubicka at ucw dot cz
gcc-bugzi...@gcc.gnu.org wrote:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
---
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #15 from Jan Hubicka hubicka at ucw dot cz 2012-12-18 15:40:34
UTC ---
It's hard to say in case of Firefox, because the only thing
that one can reliably measure is the JavaScript performance.
And this varies only very
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #16 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-18 17:03:52 UTC ---
I did some measurements with tramp3d and in this case
the default (999) gives the best performance:
par. sizetime
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #17 from Jan Hubicka hubicka at ucw dot cz 2012-12-18 17:25:37
UTC ---
I did some measurements with tramp3d and in this case
the default (999) gives the best performance:
par. sizetime
999
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #13 from tejohnson at gcc dot gnu.org 2012-12-14 15:11:00 UTC ---
Author: tejohnson
Date: Fri Dec 14 15:10:45 2012
New Revision: 194502
URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=194502
Log:
2012-12-14 Teresa
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
Markus Trippelsdorf markus at trippelsdorf dot de changed:
What|Removed |Added
Status|UNCONFIRMED
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
Richard Biener rguenth at gcc dot gnu.org changed:
What|Removed |Added
CC|
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #1 from Teresa Johnson tejohnson at google dot com 2012-12-13
14:45:01 UTC ---
I'm really surprised that using --param hot-bb-count-ws-permille=950 didn't
help, since even fewer things should look hot enough to inline than
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #2 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-13 14:46:31 UTC ---
In the tramp3d-v4 case, when I run with this simple debug patch:
diff --git a/gcc/predict.c b/gcc/predict.c
index 5d3de29..bf3a259 100644
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #3 from Teresa Johnson tejohnson at google dot com 2012-12-13
14:49:19 UTC ---
Hi Markus,
Are you sure you have my subsequent fixes patched in, to make sure the
histogram is getting streamed through the LTO files? This was
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #4 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-13 14:52:37 UTC ---
(In reply to comment #3)
Hi Markus,
Are you sure you have my subsequent fixes patched in, to make sure the
histogram is getting
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #5 from Teresa Johnson tejohnson at google dot com 2012-12-13
15:02:55 UTC ---
Ok, I will download tramp3d-v4 right now and see what is going on. Can
you send me the full set of options you are using to compile it?
Teresa
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #6 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-13 15:06:12 UTC ---
(In reply to comment #5)
Ok, I will download tramp3d-v4 right now and see what is going on. Can
you send me the full set of options you are
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #7 from Teresa Johnson tejohnson at google dot com 2012-12-13
15:50:05 UTC ---
Reproduced. Looks like somehow my fix to stream this through LTO is
not working properly. I see that the min count is valid when
generating the .o
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #8 from Teresa Johnson tejohnson at google dot com 2012-12-13
18:23:08 UTC ---
Dumb mistake in my previous fix to the lto support. Here is the patch
that fixes it, I will submit for review after regression testing
completes:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #9 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-13 19:10:40 UTC ---
*** Bug 55669 has been marked as a duplicate of this bug. ***
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #10 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-13 22:14:01 UTC ---
(In reply to comment #8)
Please let me know how this affects the mozilla size.
Looks much better now:
39748288
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #11 from Teresa Johnson tejohnson at google dot com 2012-12-13
22:16:19 UTC ---
Do you happen to know what it was with lto/pgo before the change? Should be
roughly equivalent to hot-bb-count-ws-permille=970 from what I saw in
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55674
--- Comment #12 from Markus Trippelsdorf markus at trippelsdorf dot de
2012-12-13 22:35:33 UTC ---
(In reply to comment #11)
Do you happen to know what it was with lto/pgo before the change? Should be
roughly equivalent to
23 matches
Mail list logo
