Re: BUG in libm's powf

From: Mark Murray <markm_at_FreeBSD.org>
Date: Mon, 6 Sep 2021 18:55:07 +0100
Thanks!

And it's committed!

M

> On 6 Sep 2021, at 18:53, Steve Kargl <sgk_at_troutmask.apl.washington.edu> wrote:
> 
> Fine with me.  I don't have a phabricator account and
> bugzilla reports seems to get lost in the ether.
> 
> --
> steve
> 
> On Mon, Sep 06, 2021 at 06:45:11PM +0100, Mark Murray wrote:
>> Hi
>> 
>> I've opened a Phab ticket for this. I hope that's OK?
>> 
>> https://reviews.freebsd.org/D31865
>> 
>> M
>> 
>>> On 6 Sep 2021, at 16:28, Steve Kargl <sgk_at_troutmask.apl.washington.edu> wrote:
>>> 
>>> Paul Zimmermann has identified a bug in Openlibm's powf(),
>>> which is identical to FreeBSD's libm.  Both derived from
>>> fdlibm. https://github.com/JuliaMath/openlibm/issues/212.
>>> 
>>> Consider
>>> 
>>> % cat h.c
>>> #include <math.h>
>>> #include <stdio.h>
>>> int
>>> main(void)
>>> {
>>>  float x, y, z;
>>>  x =  0x1.ffffecp-1F;
>>>  y = -0x1.000002p+27F;
>>>  z =  0x1.557a86p115F;
>>>  printf("%e %e %e <-- should be %e\n", x, y, powf(x,y), z);
>>>  return 0;
>>> }
>>> 
>>> % cc -o h -fno-builtin h.c -lm && ./h
>>> 9.999994e-01 -1.342177e+08 inf <-- should be 5.540807e+34
>>> 
>>> Note, clang seems to have a builtin for powf(), but one cannot
>>> count of clang being the only consumer of libm.  With the patch
>>> at the end of this email, I get
>>> 
>>> % cc -o h -fno-builtin h.c -L/home/kargl/trunk/math/libm/msun -lmath && ./h
>>> 9.999994e-01 -1.342177e+08 5.540807e+34 <-- should be 5.540807e+34
>>> 
>>> Watch for copy and paste whitespace corruption.
>>> 
>>> --- /usr/src/lib/msun/src/e_powf.c	2021-02-21 03:29:00.956878000 -0800
>>> +++ src/e_powf.c	2021-09-06 08:17:09.800008000 -0700
>>> _at__at_ -136,7 +136,7 _at__at_
>>>    /* |y| is huge */
>>> 	if(iy>0x4d000000) { /* if |y| > 2**27 */
>>> 	/* over/underflow if x is not close to one */
>>> -	    if(ix<0x3f7ffff7) return (hy<0)? sn*huge*huge:sn*tiny*tiny;
>>> +	    if(ix<0x3f7ffff6) return (hy<0)? sn*huge*huge:sn*tiny*tiny;
>>> 	    if(ix>0x3f800007) return (hy>0)? sn*huge*huge:sn*tiny*tiny;
>>> 	/* now |1-x| is tiny <= 2**-20, suffice to compute
>>> 	   log(x) by x-x^2/2+x^3/3-x^4/4 */
>>> 
>>> 
>>> --
>>> Steve
>>> 
>> 
>> --
>> Mark R V Murray
>> 
> 
> 
> 
> --
> Steve
> 

--
Mark R V Murray


Received on Mon Sep 06 2021 - 17:55:07 UTC

Original text of this message