From: Andreas Ericsson <ae@op5.se>
To: "Bernhard R. Link" <brl+ccmadness@pcpool00.mathematik.uni-freiburg.de>
Cc: "Ralf Baechle" <ralf@linux-mips.org>,
"Junio C Hamano" <gitster@pobox.com>,
"Ingo Molnar" <mingo@elte.hu>,
git@vger.kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
"Thomas Gleixner" <tglx@linutronix.de>,
"Peter Zijlstra" <a.p.zijlstra@chello.nl>,
"Arnaldo Carvalho de Melo" <acme@redhat.com>,
"Frédéric Weisbecker" <fweisbec@gmail.com>,
"Pekka Enberg" <penberg@cs.helsinki.fi>
Subject: Re: [PATCH] git gc: Speed it up by 18% via faster hash comparisons
Date: Thu, 28 Apr 2011 11:42:19 +0200 [thread overview]
Message-ID: <4DB9367B.2050607@op5.se> (raw)
In-Reply-To: <20110428081817.GA29344@pcpool00.mathematik.uni-freiburg.de>
On 04/28/2011 10:18 AM, Bernhard R. Link wrote:
> * Ralf Baechle<ralf@linux-mips.org> [110428 02:35]:
>> On Wed, Apr 27, 2011 at 04:32:12PM -0700, Junio C Hamano wrote:
>>
>>>> +static inline int is_null_sha1(const unsigned char *sha1)
>>>> {
>>>> - return memcmp(sha1, sha2, 20);
>>>> + const unsigned long long *sha1_64 = (void *)sha1;
>>>> + const unsigned int *sha1_32 = (void *)sha1;
>>>
>>> Can everybody do unaligned accesses just fine?
>>
>> Misaligned accesses cause exceptions on some architectures which then
>> are fixed up in software making these accesses _very_ slow. You can
>> use __attribute__((packed)) to work around that but that will on the
>> affected architectures make gcc generate code pessimistically that is
>> slower than not using __attribute__((packed)) in case of proper
>> alignment. And __attribute__((packed)) only works with GCC.
>
> Even __attribute__((packed)) usually does not allow arbitrary aligned
> data, but can intruct the code to generate code to access code
> misaligned in a special way. (I have already seen code where thus
> accessing a properly aligned long caused a SIGBUS, because it was
> aligned because being in a misaligned packed struct).
>
> In short: misaligning stuff works on x86, everywhere else it is disaster
> waiting to happen. (And people claiming compiler bugs or broken
> architectures, just because they do not know the basic rules of C).
>
Given that the vast majority of user systems are x86 style ones, it's
probably worth using this patch on such systems and stick to a
partially unrolled byte-by-byte comparison that finishes early on
the rest of them. Properly pipelined, it will just mean that the early
return undoes the fetch steps for the 3-4 unrolled bytes that it
computes in advance, so if the diff comes in the first 10-12 bytes,
it will still be a win.
For bonus points, check if both bytestrings are equally (un)aligned
first and, if they are, half-Duff it out with a fallthrough switch
statement (without the while() loop) to compare byte-by-byte first
and then word-for-word on the rest of it. The setup and complexity
is probably not worth it for our meager 20-byte strings though.
--
Andreas Ericsson andreas.ericsson@op5.se
OP5 AB www.op5.se
Tel: +46 8-230225 Fax: +46 8-230231
Considering the successes of the wars on alcohol, poverty, drugs and
terror, I think we should give some serious thought to declaring war
on peace.
next prev parent reply other threads:[~2011-04-28 9:42 UTC|newest]
Thread overview: 45+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-04-27 22:51 [PATCH] git gc: Speed it up by 18% via faster hash comparisons Ingo Molnar
2011-04-27 23:10 ` Ingo Molnar
2011-04-27 23:18 ` Jonathan Nieder
2011-04-28 6:36 ` Ingo Molnar
2011-04-28 9:31 ` Jonathan Nieder
2011-04-28 10:36 ` Ingo Molnar
2011-04-28 9:32 ` Dmitry Potapov
2011-04-27 23:32 ` Junio C Hamano
2011-04-28 0:35 ` Ralf Baechle
2011-04-28 8:18 ` Bernhard R. Link
2011-04-28 9:42 ` Andreas Ericsson [this message]
2011-04-28 9:55 ` Erik Faye-Lund
2011-04-28 20:19 ` H. Peter Anvin
2011-04-28 6:27 ` Ingo Molnar
2011-04-28 9:17 ` Erik Faye-Lund
2011-04-28 9:33 ` Ingo Molnar
2011-04-28 9:37 ` Ingo Molnar
2011-04-28 9:50 ` Erik Faye-Lund
2011-04-28 10:10 ` Pekka Enberg
2011-04-28 10:19 ` Erik Faye-Lund
2011-04-28 10:30 ` Pekka Enberg
2011-04-28 11:59 ` Erik Faye-Lund
2011-04-28 12:12 ` Pekka Enberg
2011-04-28 12:36 ` Jonathan Nieder
2011-04-28 12:40 ` Erik Faye-Lund
2011-04-28 13:37 ` Ingo Molnar
2011-04-28 15:14 ` Ingo Molnar
2011-04-28 16:00 ` Erik Faye-Lund
2011-04-28 20:32 ` Ingo Molnar
2011-04-29 7:05 ` Alex Riesen
2011-04-29 16:24 ` H. Peter Anvin
2011-04-28 12:16 ` Tor Arntsen
2011-04-28 20:23 ` H. Peter Anvin
2011-04-28 12:17 ` Andreas Ericsson
2011-04-28 12:28 ` Erik Faye-Lund
2011-04-28 10:19 ` Ingo Molnar
2011-04-28 12:02 ` Nguyen Thai Ngoc Duy
2011-04-28 12:18 ` Erik Faye-Lund
2011-04-28 20:20 ` Junio C Hamano
2011-04-28 16:36 ` Dmitry Potapov
2011-04-28 8:52 ` Dmitry Potapov
2011-04-28 9:11 ` Ingo Molnar
2011-04-28 9:31 ` Dmitry Potapov
2011-04-28 9:44 ` Ingo Molnar
2011-04-28 9:38 ` Ingo Molnar
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
List information: http://vger.kernel.org/majordomo-info.html
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4DB9367B.2050607@op5.se \
--to=ae@op5.se \
--cc=a.p.zijlstra@chello.nl \
--cc=acme@redhat.com \
--cc=brl+ccmadness@pcpool00.mathematik.uni-freiburg.de \
--cc=fweisbec@gmail.com \
--cc=git@vger.kernel.org \
--cc=gitster@pobox.com \
--cc=hpa@zytor.com \
--cc=mingo@elte.hu \
--cc=penberg@cs.helsinki.fi \
--cc=ralf@linux-mips.org \
--cc=tglx@linutronix.de \
/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.
Code repositories for project(s) associated with this public inbox
https://80x24.org/mirrors/git.git
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).