From: Linus Torvalds <torvalds@linux-foundation.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: "Stanislaw Gruszka" <sgruszka@redhat.com>,
"Linux Kernel Mailing List" <linux-kernel@vger.kernel.org>,
"Ingo Molnar" <mingo@kernel.org>,
"H. Peter Anvin" <hpa@zytor.com>,
"Frédéric Weisbecker" <fweisbec@gmail.com>,
"Steven Rostedt" <rostedt@goodmis.org>,
"Andrew Morton" <akpm@linux-foundation.org>,
"Thomas Gleixner" <tglx@linutronix.de>,
linux-tip-commits@vger.kernel.org
Subject: Re: [tip:sched/core] sched: Lower chances of cputime scaling overflow
Date: Thu, 11 Apr 2013 08:38:37 -0700 [thread overview]
Message-ID: <CA+55aFway4C89fTewKvED4RvFLaNdM-BVtkJ2YChsvracLKiyA@mail.gmail.com> (raw)
In-Reply-To: <1365687946.8824.3.camel@laptop>
On Thu, Apr 11, 2013 at 6:45 AM, Peter Zijlstra <peterz@infradead.org> wrote:
> On Tue, 2013-03-26 at 15:01 +0100, Stanislaw Gruszka wrote:
>> Thoughts?
>
> Would something like the below work?
Ugh, this is hard to think about, it's also fairly inefficient.
> static cputime_t scale_stime(u64 stime, u64 rtime, u64 total)
> {
> - u64 rem, res, scaled;
> + int stime_fls = fls64(stime);
> + int total_fls = fls64(total);
> + int rtime_fls = fls64(rtime);
Doing "fls64()" unconditionally is quite expensive on some
architectures, and if I am not mistaken, the *common* case (by far) is
that all these values fit in 32 bits, no?
So let's re-think the whole thing entirely. First, let's throw away
the uncommon case, and we'll come back to it later:
if (unlikely((stime | total | rtime) >> 32)
return uncommon_case(stime, total, rtime);
and now we know we have the simple case where everything is in 32
bits, and we can just do the trivial
/* Make sure gcc understands that this is a 32x32->64 multiply,
followed by a 64/32->64 divide */
return div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total);
which is the *cheapest* possible case of scale_stime(), with the
simplified multiply and divide. Agreed? This is cheap even on 32-bit.
Well, relatively.
Do we agree that this is (a) the common case that we need to worry
about from performance and (b) simple, understandable and efficient?
Now, let's look at the uncommon case, and I lied, I'm not going to
actually do it as a "uncommon_case()" function, I'm going to do this
as a "let us simplify the uncommon case until it *looks* like the
common case". IOW, in this uncommon thing, the aim is simply to just
reduce stime/rtime/total to the point where they are 32 bits. Ok? And
let's keep it simple again.
So *now*, once we are in the uncommon case, let's start counting bits.
Like this:
/* We know one of the values has a bit set in the high 32 bits */
for (;;) {
/* Make sure "stime" is the bigger of stime/rtime */
if (rtime > stime) {
u64 tmp = stime; stime = rtime; rtime = tmp;
}
/* Do we need to balance stime/rtime bits? */
if (stime >> 32) {
if (rtime >> 31)
goto drop_precision;
/* We can grow rtime and shrink stime and try to make them
both fit */
rtime <<= 1;
stime >>= 1;
continue;
}
/* stime/rtime fits in 32 bits, how about total? */
if (!(total >> 32))
break;
drop_precision:
/* We drop from stime, it has more bits than rtime */
stime >>= 1;
total >>= 1;
}
The above is totally untested, but each step is pretty damn simple and
fairly cheap. Sure, it's a loop, but it's bounded to 32 (cheap)
iterations, and the normal case is that it's not done at all, or done
only a few times.
And the advantage is that the end result is always that simple
32x32/32 case that we started out with as the common case.
I dunno. Maybe I'm overlooking something, and the above is horrible,
but the above seems reasonably efficient if not optimal, and
*understandable*.
Linus
next prev parent reply other threads:[~2013-04-11 15:38 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <tip-d9a3c9823a2e6a543eb7807fb3d15d8233817ec5@git.kernel.org>
2013-03-26 14:01 ` [tip:sched/core] sched: Lower chances of cputime scaling overflow Stanislaw Gruszka
2013-03-26 14:19 ` Frederic Weisbecker
2013-03-26 16:54 ` Stanislaw Gruszka
2013-04-10 12:51 ` Ingo Molnar
2013-04-10 15:28 ` Frederic Weisbecker
2013-04-10 17:32 ` Ingo Molnar
2013-04-11 8:04 ` Stanislaw Gruszka
2013-04-11 13:45 ` Peter Zijlstra
2013-04-11 14:50 ` Stanislaw Gruszka
2013-04-11 17:31 ` Peter Zijlstra
2013-04-11 15:38 ` Linus Torvalds [this message]
2013-04-11 18:07 ` Peter Zijlstra
2013-04-11 18:22 ` Frederic Weisbecker
2013-04-11 18:26 ` Frederic Weisbecker
2013-04-11 18:22 ` Linus Torvalds
2013-04-12 7:55 ` Peter Zijlstra
2013-04-13 14:49 ` Stanislaw Gruszka
2013-04-13 18:44 ` Linus Torvalds
2013-04-16 10:40 ` Stanislaw Gruszka
2013-04-30 14:03 ` Stanislaw Gruszka
2013-04-13 14:55 ` Stanislaw Gruszka
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=CA+55aFway4C89fTewKvED4RvFLaNdM-BVtkJ2YChsvracLKiyA@mail.gmail.com \
--to=torvalds@linux-foundation.org \
--cc=akpm@linux-foundation.org \
--cc=fweisbec@gmail.com \
--cc=hpa@zytor.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-tip-commits@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=sgruszka@redhat.com \
--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.
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).