From: Enzo Matsumiya <ematsumiya@suse.de>
To: "Aurélien Aptel" <aurelien.aptel@gmail.com>
Cc: linux-cifs@vger.kernel.org, smfrench@gmail.com, pc@cjr.nz,
ronniesahlberg@gmail.com, nspmangalore@gmail.com
Subject: Re: [PATCH] cifs: perf improvement - use faster macros ALIGN() and round_up()
Date: Wed, 7 Sep 2022 15:15:04 -0300 [thread overview]
Message-ID: <20220907181504.slrof5u42vijakje@cyberdelia> (raw)
In-Reply-To: <CA+5B0FNBJVFC6-SfVodctu0DkyyZ9DzJM8OJsDBbVb453Mvfsw@mail.gmail.com>
Hi Aurelien,
On 09/07, Aurélien Aptel wrote:
>Hi,
>
>Changes are good from a readability stand point but like the others
>I'm very skeptical about the perf improvement claims.
I also am/was. That's why I ran 5 times each test after sending the
patch. For each run, I made sure to have the mount point clean, both
server and client freshly booted, and aside from the patch, the build
env/compile options were exactly the same.
The server had no parallel workload at all as it's only a test server.
>Many of these compile to the same division-less instructions
>especially if any of the values are known at compile time.
<snip>
>> @@ -2826,9 +2819,7 @@ SMB2_open_init(struct cifs_tcon *tcon, struct TCP_Server_Info *server,
>> - copy_size = uni_path_len;
>> - if (copy_size % 8 != 0)
>> - copy_size = roundup(copy_size, 8);
>> + copy_size = round_up(uni_path_len, 8);
>> @@ -4090,7 +4081,7 @@ smb2_new_read_req(void **buf, unsigned int *total_len,
>> - *total_len = DIV_ROUND_UP(*total_len, 8) * 8;
>> + *total_len = ALIGN(*total_len, 8);
>
>These 2 are also hot paths, but skeptical about the optimizations.
As you point out, SMB2_open_init() was indeed the function with greater
improvement, and as per my measurements, the one that actually impacted
the general latency.
>I've looked at those macros in Compiler Explorer and sure enough they
>compile to the same thing on x86_64.
>Even worse, the optimized versions compile with extra instructions for some:
>
>https://godbolt.org/z/z1xhhW9sj
I did the same comparison on a userspace program and got similar
results, but I didn't bother to check the kernel objects as testing it
was quicker to me. But I'll do it today.
>I suspect the improvements are more likely to be related to caches,
>system load, server load, ...
See above.
>You can try to use perf to make a flamegraph and compare.
>
>Cheers,
Not really used to perf, but will check it out, thanks!
Cheers,
Enzo
next prev parent reply other threads:[~2022-09-07 18:15 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-09-06 1:30 [PATCH] cifs: perf improvement - use faster macros ALIGN() and round_up() Enzo Matsumiya
2022-09-06 2:36 ` ronnie sahlberg
2022-09-06 14:41 ` Enzo Matsumiya
2022-09-06 15:47 ` Tom Talpey
2022-09-06 16:05 ` Enzo Matsumiya
2022-09-07 8:37 ` Aurélien Aptel
2022-09-07 18:15 ` Enzo Matsumiya [this message]
2022-09-07 20:41 ` Enzo Matsumiya
2022-09-14 2:50 ` Steve French
2022-09-14 14:38 ` Enzo Matsumiya
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=20220907181504.slrof5u42vijakje@cyberdelia \
--to=ematsumiya@suse.de \
--cc=aurelien.aptel@gmail.com \
--cc=linux-cifs@vger.kernel.org \
--cc=nspmangalore@gmail.com \
--cc=pc@cjr.nz \
--cc=ronniesahlberg@gmail.com \
--cc=smfrench@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 NNTP newsgroup(s).