code review 5467060: math: fix special cases in Nextafter

math: fix special cases in Nextafter

Nextafter(0, -1) != -0.

[Charlie Dorian, 2011-12-09 22:12:55]

Hello rsc@golang.org, golang-dev@googlegroups.com (cc:
golang-dev@googlegroups.com),

I'd like you to review this change to
https://go.googlecode.com/hg/

[rsc, 2011-12-09 22:42:51]

I'm sorry, but I take back what I said.
I tested the C library on my machine,
and it has nextafter(0, -1) == -0.
We should probably match that.

It may not be worth it, but if you take the
floating point bits as an int64, and then do:
   if negative { xor with 1<<63 - 1 }
then I think that maps all the non-NaN
floating-point values to int64 in a way
that preserves the ordering (the old float64
order is now the order of the int64).
If that's right, then the function can
simplify to

if x != x || y != y {
    return NaN()
}

ix := int64(Float64bits(x))
ix ^= int64(uint64(ix>>63)>>1)

iy := int64(Float64bits(y))
iy ^= int64(uint64(iy>>63)>>1)

if ix < iy {
    ix++
} else if ix > iy {
   ix--
}

ix ^= int64(uint64(ix>>63)>>1)
return Float64frombits(uint64(ix))

Or it might not.
Russ

[rsc, 2011-12-09 22:45:18]

On Fri, Dec 9, 2011 at 17:42, Russ Cox <rsc@golang.org> wrote:
> I'm sorry, but I take back what I said.

Never mind: the current code is fine.
My C program was using %f and showing -0.00000
but that's not really -0.  Changing it to say %.17g
shows the number we expected.

$ cat x.c
#include <stdio.h>
#include <math.h>

int
main(void)
{
	printf("%.17g\n", nextafter(0, -1));
}
$ a.out
-4.9406564584124654e-324
$

This CL is fine as is, except please add this as a test case.

Russ

[Charlie Dorian, 2011-12-10 00:38:24]

Hello rsc@golang.org, golang-dev@googlegroups.com (cc:
golang-dev@googlegroups.com),

Please take another look.

[Charlie Dorian, 2011-12-10 00:40:19]

Oops, ignore the 7:39 PM CL. I'll change it back.

On Fri, Dec 9, 2011 at 5:45 PM, Russ Cox <rsc@golang.org> wrote:
> On Fri, Dec 9, 2011 at 17:42, Russ Cox <rsc@golang.org> wrote:
>> I'm sorry, but I take back what I said.
>
> Never mind: the current code is fine.
> My C program was using %f and showing -0.00000
> but that's not really -0.  Changing it to say %.17g
> shows the number we expected.
>
> $ cat x.c
> #include <stdio.h>
> #include <math.h>
>
> int
> main(void)
> {
>        printf("%.17g\n", nextafter(0, -1));
> }
> $ a.out
> -4.9406564584124654e-324
> $
>
> This CL is fine as is, except please add this as a test case.
>
> Russ

[Charlie Dorian, 2011-12-10 01:00:59]

Hello rsc@golang.org, golang-dev@googlegroups.com (cc:
golang-dev@googlegroups.com),

Please take another look.

[Charlie Dorian, 2011-12-10 01:01:49]

Hello rsc@golang.org, golang-dev@googlegroups.com (cc:
golang-dev@googlegroups.com),

Please take another look.

[rsc, 2011-12-12 20:51:13]

*** Submitted as http://code.google.com/p/go/source/detail?r=f57006515d93 ***

math: fix special cases in Nextafter

Nextafter(0, -1) != -0.

R=rsc, golang-dev
CC=golang-dev
http://codereview.appspot.com/5467060

Committer: Russ Cox <rsc@golang.org>