All of lore.kernel.org
 help / color / mirror / Atom feed
From: Oleg Nesterov <oleg@redhat.com>
To: Brian Behlendorf <behlendorf1@llnl.gov>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [PATCH] Make div64_u64() precise on 32bit platforms
Date: Thu, 14 Oct 2010 14:11:59 +0200	[thread overview]
Message-ID: <20101014121159.GA407@redhat.com> (raw)
In-Reply-To: <20101013213746.GA27248@redhat.com>

On 10/13, Oleg Nesterov wrote:
>
> On 10/12, Brian Behlendorf wrote:
> >
> >  u64 div64_u64(u64 dividend, u64 divisor)
> >  {
> > -	u32 high, d;
> > -
> > -	high = divisor >> 32;
> > -	if (high) {
> > -		unsigned int shift = fls(high);
> > +	u64 u0, quot0, quot1;
> > +	u32 rem;
> > +	int n;
> > +
> > +	if (divisor >> 32 == 0) {
> > +		if (dividend >> 32 < divisor) {
> > +			return div_u64_rem(dividend, divisor, &rem);
> > +		} else {
> > +			u0 = dividend & 0xFFFFFFFF;
> > +			quot1 = div_u64_rem(dividend >> 32, divisor, &rem);
> > +			u0 += ((u64)rem << 32);
> > +			quot0 = div_u64_rem(u0, divisor, &rem);
> > +			return (quot1 << 32) + quot0;
> > +		}
>
> Looks correct... but I can't understand these complications.
> Looks like we can just do
>
> 	if ((divisor >> 32) == 0) {
> 		div_u64(dividend, divisor);
> 	} else {
> 	...
>
> No?
>
> > +	} else {
> > +		n = __builtin_clzll(divisor);
> > +		quot1 = div_u64_rem(dividend >> 1, (divisor << n) >> 32, &rem);
> > +		quot0 = (quot1 << n) >> 31;
>
> I can't understand this "dividend >> 1". It seems to me that
>
> 		quot1 = div_u64(dividend, (divisor << n) >> 32);
> 		quot0 = (quot1 << n) >> 32;
>
> should be equally correct. Or I missed some overflow?

Thinking more about this with a fresh head, we don't event need quot1,
unless I missed something. We can do

		quot0 = div_u64((dividend << n) >> 32, (divisor << n) >> 32);

instead. Or, better,

		n = 32 - __builtin_clzll(divisor);
		quot0 = div_u64(dividend >> n, divisor >> n);

And 32 - clzll == fls.

So, I think it can be really trivial, see the test-case below,
seems to work (you need 64bit machine to test).

What do you think? I do not trust my math skills.

Oleg.

#include <stdlib.h>
#include <unistd.h>
#include <stdio.h>
#include <sys/fcntl.h>
#include <assert.h>

typedef unsigned long long u64;
typedef unsigned long u32;

static inline u64 div_u64(u64 A, u32 B)
{
	return A / B;
}

static inline unsigned long __fls(unsigned long word)
{
	asm("bsr %1,%0"
	    : "=r" (word)
	    : "rm" (word));
	return word;
}

u64 div64_u64(u64 A, u64 B)
{
	u32 high = B >> 32;
	u64 quot;

	if (high == 0) {
		quot = div_u64(A, B);
	} else {
		int n = 1 + __fls(high);
		quot = div_u64(A >> n, B >> n);

		if (quot != 0)
			quot--;
		if ((A - quot * B) >= B)
			quot++;
	}

	return quot;
}

int main(void)
{
	int fd, n;

	fd = open("/dev/urandom", O_RDONLY);
	assert(fd >= 0);

	for (n = 1;; ++n) {
		u64 xx[2], rs;

		assert(read(fd, xx, sizeof(xx)) == sizeof(xx));

		if (xx[1] == 0)
			continue;

		rs = div64_u64(xx[0], xx[1]);

		if (rs != xx[0] / xx[1]) {
			printf("ERR!! %llx / %llx = %llx : %llx\n",
				xx[0] , xx[1], xx[0] / xx[1], rs);
			return 1;
		}

		if (!(n %100000))
			printf("passed: %d\n", n);
	}
}


  reply	other threads:[~2010-10-14 12:16 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-10-12 19:26 [PATCH] Make div64_u64() precise on 32bit platforms Brian Behlendorf
2010-10-13 21:37 ` Oleg Nesterov
2010-10-14 12:11   ` Oleg Nesterov [this message]
2010-10-21 17:46     ` Brian Behlendorf
2010-10-21 18:12       ` Oleg Nesterov
2010-10-21 19:22         ` Andrew Morton
2010-10-21 19:49           ` Oleg Nesterov
  -- strict thread matches above, loose matches on Subject: below --
2010-08-02 16:09 [PATCH] trivial, document that div64_u64() is not " Oleg Nesterov
2010-08-03 22:28 ` Andrew Morton
2010-08-09 16:30   ` [PATCH] Make div64_u64() " Brian Behlendorf
2010-09-17  0:00     ` Oleg Nesterov

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=20101014121159.GA407@redhat.com \
    --to=oleg@redhat.com \
    --cc=akpm@linux-foundation.org \
    --cc=behlendorf1@llnl.gov \
    --cc=linux-kernel@vger.kernel.org \
    /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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.