From: "H.J. Lu" <hjl.tools@gmail.com>
To: Bernd Schmidt <bernds@codesourcery.com>
Cc: Eric Botcazou <ebotcazou@adacore.com>,
gcc-patches@gcc.gnu.org, Mark Mitchell <mark@codesourcery.com>,
Richard Guenther <richard.guenther@gmail.com>,
David Daney <ddaney@caviumnetworks.com>,
Andi Kleen <andi@firstfloor.org>,
Steven Bosscher <stevenb.gcc@gmail.com>
Subject: Re: Combine four insns
Date: Fri, 20 Aug 2010 00:21:00 -0000 [thread overview]
Message-ID: <AANLkTimZN58V24J+hX5JADLNgv=VWp+QTd0thgwooKhv@mail.gmail.com> (raw)
In-Reply-To: <4C6D6BEC.1020700@codesourcery.com>
On Thu, Aug 19, 2010 at 10:37 AM, Bernd Schmidt <bernds@codesourcery.com> wrote:
> On 08/19/2010 09:38 AM, Eric Botcazou wrote:
>>> Mark said the plan was sensible, so I think there is no tie.
>>
>> Sorry, this is such a bad decision in my opinion, as it will set a precedent
>> for one-percent-slowdown-for-very-little-benefit patches, that I think an
>> explicit OK is in order.
>
> We're no longer discussing the 1% slower patch. I'll even agree that
> that's a bit excessive, and the approval for it surprised me, but it
> served to get a discussion going. Several people provided datapoints
> indicating that time spent in the optimizers at -O2 or higher is
> something that just isn't on the radar as a valid concern based both on
> usage patterns and profiling results which show that most time is spent
> elsewhere.
>
> As for the patch itself, Michael Matz provided constructive feedback
> which led to a heuristic that eliminated a large number of combine-4
> attempts. I conclude that either you didn't read the thread before
> attempting once again to block one of my patches, or the above is more
> than a little disingenuous.
>
> The following is a slightly updated variant of the previous patch I
> posted. I fixed a bug and slightly cleaned up the in_feeds_im logic
> (insn_a_feeds_b isn't valid for insns that aren't consecutive in the
> combination), and I found a way to slightly relax the heuristic in order
> to use Michael's suggestion of allowing combinations if there are two or
> more binary operations with constant operand.
>
> On i686, the heuristic reduces the combine-4 attempts to slightly over a
> third of the ones tried by the first patch (and presumably disallows
> them in most cases where we'd generate overly large RTL). Hence, I
> would have expected this patch to cause slowdowns in the 0.4% range, but
> when I ran a few bootstraps today, I had a few results near 99m5s user
> time with the patch, and the best run without the patch came in at
> 99m10s. I don't have enough data to be sure, but some test runs gave me
> the impression that there is one change, using insn_a_feeds_b instead of
> reg_overlap_mentioned_p, which provided some speedup, and may explain
> this result. It still seems a little odd.
>
> Allowing unary ops in the heuristic as well made hardly a difference in
> output, and appeared to cost around 0.3%, so I left it out.
>
> Bootstrapped and regression tested on i686-linux. Committed.
>
This may have caused:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45350
--
H.J.
next prev parent reply other threads:[~2010-08-19 22:53 UTC|newest]
Thread overview: 129+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-08-06 14:49 Bernd Schmidt
2010-08-06 15:04 ` Richard Guenther
2010-08-06 20:08 ` Bernd Schmidt
2010-08-06 20:37 ` Richard Guenther
2010-08-06 21:53 ` Jeff Law
2010-08-06 22:41 ` Bernd Schmidt
2010-08-06 23:47 ` Richard Guenther
2010-08-07 8:11 ` Eric Botcazou
2010-08-09 12:29 ` Bernd Schmidt
2010-08-09 12:39 ` Steven Bosscher
2010-08-09 13:48 ` Bernd Schmidt
2010-08-10 2:51 ` Laurynas Biveinis
2010-08-09 12:41 ` Michael Matz
2010-08-09 14:34 ` Bernd Schmidt
2010-08-09 14:39 ` Toon Moene
2010-08-09 14:50 ` Steven Bosscher
2010-08-09 14:58 ` The speed of the compiler, was: " Toon Moene
2010-08-09 15:00 ` Paul Koning
2010-08-09 15:33 ` Diego Novillo
2010-08-09 15:53 ` Mark Mitchell
2010-08-09 17:15 ` Toon Moene
2010-08-09 17:19 ` Diego Novillo
2010-08-09 17:29 ` Toon Moene
2010-08-09 23:24 ` Chris Lattner
2010-08-10 13:02 ` Toon Moene
2010-08-10 15:36 ` Chris Lattner
2010-08-10 14:58 ` Andi Kleen
2010-08-10 15:03 ` Richard Guenther
2010-08-10 15:32 ` Andi Kleen
2010-08-10 20:09 ` Tom Tromey
2010-08-10 20:23 ` Andi Kleen
2010-08-10 22:40 ` Mike Stump
2010-08-10 23:16 ` Tom Tromey
2010-08-12 21:09 ` Nathan Froyd
2010-08-17 15:14 ` Mark Mitchell
2010-08-10 20:15 ` H.J. Lu
2010-08-12 21:38 ` Vectorized _cpp_clean_line Richard Henderson
2010-08-12 22:18 ` Andi Kleen
2010-08-12 22:32 ` Richard Henderson
2010-08-12 23:10 ` Richard Henderson
2010-08-12 23:13 ` Richard Henderson
2010-08-13 8:33 ` Andi Kleen
2010-08-13 7:26 ` Andi Kleen
2010-08-14 17:14 ` [CFT, v4] " Richard Henderson
2010-08-17 16:59 ` Steve Ellcey
2010-08-17 17:21 ` Richard Henderson
2010-08-17 20:32 ` Steve Ellcey
2010-08-18 17:14 ` Steve Ellcey
2010-08-17 17:32 ` Jakub Jelinek
2010-08-18 3:23 ` Tom Tromey
[not found] ` <1281998097.3725.3.camel@gargoyle>
[not found] ` <4C69C317.2080207@redhat.com>
[not found] ` <1282142212.3725.6.camel@gargoyle>
[not found] ` <4C6BF5F7.7040100@redhat.com>
[not found] ` <1282149264.3725.15.camel@gargoyle>
[not found] ` <4C6C0D92.7080100@redhat.com>
[not found] ` <1282151361.3725.19.camel@gargoyle>
[not found] ` <4C6C166A.90306@redhat.com>
[not found] ` <1282152938.3725.27.camel@gargoyle>
[not found] ` <4C6C39DB.8070409@redhat.com>
2010-08-18 21:50 ` Richard Henderson
2010-08-19 14:12 ` Luis Machado
2010-08-09 17:27 ` The speed of the compiler, was: Re: Combine four insns Joseph S. Myers
2010-08-09 18:23 ` Diego Novillo
2010-08-10 6:20 ` Chiheng Xu
2010-08-10 7:22 ` Chiheng Xu
2010-08-09 17:34 ` Steven Bosscher
2010-08-09 17:36 ` Diego Novillo
2010-08-09 23:13 ` Cary Coutant
2010-08-09 18:59 ` The speed of the compiler Ralf Wildenhues
2010-08-09 19:04 ` Diego Novillo
2010-08-09 21:12 ` The speed of the compiler, was: Re: Combine four insns Mike Stump
2010-08-09 23:48 ` Cary Coutant
2010-08-10 14:37 ` Bernd Schmidt
2010-08-10 14:40 ` Richard Guenther
2010-08-10 14:49 ` Bernd Schmidt
2010-08-10 15:06 ` Steven Bosscher
2010-08-10 15:06 ` Steven Bosscher
2010-08-10 16:27 ` Andi Kleen
2010-08-10 16:47 ` Steven Bosscher
2010-08-10 16:55 ` Andi Kleen
2010-08-10 17:03 ` David Daney
2010-08-11 8:53 ` Richard Guenther
2010-08-16 20:42 ` Mark Mitchell
2010-08-16 20:45 ` Bernd Schmidt
2010-08-16 21:03 ` Mark Mitchell
2010-08-18 20:50 ` Eric Botcazou
2010-08-18 22:03 ` Bernd Schmidt
2010-08-19 8:04 ` Eric Botcazou
2010-08-19 15:44 ` Mark Mitchell
2010-08-19 18:13 ` Bernd Schmidt
2010-08-19 18:25 ` Mark Mitchell
2010-08-19 18:42 ` Richard Henderson
2010-08-19 22:14 ` Eric Botcazou
2010-08-19 22:19 ` Bernd Schmidt
2010-08-19 22:21 ` Eric Botcazou
2010-08-19 22:37 ` Bernd Schmidt
2010-08-19 22:53 ` Steven Bosscher
2010-08-19 23:34 ` Andrew Pinski
2010-08-19 23:40 ` Bernd Schmidt
2010-08-20 1:59 ` Mark Mitchell
2010-08-20 3:56 ` Diego Novillo
2010-08-20 4:13 ` Mark Mitchell
2010-08-20 14:14 ` Eric Botcazou
2010-08-20 14:36 ` Paolo Bonzini
2010-08-21 13:46 ` Eric Botcazou
2010-08-23 17:03 ` Richard Guenther
2010-08-23 17:08 ` Mark Mitchell
2010-08-20 14:44 ` Mark Mitchell
2010-08-20 10:21 ` Paolo Bonzini
2010-08-20 0:21 ` H.J. Lu [this message]
2010-08-20 0:36 ` Bernd Schmidt
2010-08-20 14:07 ` H.J. Lu
2010-08-11 12:32 ` Michael Matz
2010-08-06 15:08 ` Steven Bosscher
2010-08-06 16:45 ` Paolo Bonzini
2010-08-06 17:22 ` Steven Bosscher
2010-08-06 18:02 ` Jeff Law
2010-08-06 20:44 ` Steven Bosscher
2010-08-06 20:48 ` Richard Guenther
2010-08-06 21:49 ` Jeff Law
2010-08-06 18:56 ` Vladimir N. Makarov
2010-08-06 19:02 ` Steven Bosscher
2010-08-06 21:11 ` Chris Lattner
2010-08-08 11:40 ` Paolo Bonzini
2010-08-12 5:53 ` Ian Lance Taylor
2010-08-06 19:20 ` Bernd Schmidt
2010-08-06 19:37 ` Jeff Law
2010-08-06 19:43 ` Bernd Schmidt
2010-08-06 21:46 ` Jeff Law
2010-08-09 14:54 ` Mark Mitchell
2010-08-09 15:04 ` Bernd Schmidt
2010-08-09 16:02 ` Chris Lattner
2010-08-09 16:07 ` Richard Guenther
2010-08-09 17:28 ` Joseph S. Myers
2010-08-09 16:19 ` Mark Mitchell
2010-08-09 17:02 ` Chris Lattner
2010-08-10 2:50 ` Mark Mitchell
2010-08-10 15:35 ` Chris Lattner
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='AANLkTimZN58V24J+hX5JADLNgv=VWp+QTd0thgwooKhv@mail.gmail.com' \
--to=hjl.tools@gmail.com \
--cc=andi@firstfloor.org \
--cc=bernds@codesourcery.com \
--cc=ddaney@caviumnetworks.com \
--cc=ebotcazou@adacore.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=mark@codesourcery.com \
--cc=richard.guenther@gmail.com \
--cc=stevenb.gcc@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).