[Bug 298260] [libm] fmaf() mishandled some subnormal input

From: <bugzilla-noreply_at_freebsd.org>
Date: Thu, 10 Sep 2026 19:56:55 UTC 
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=298260

Steve Kargl <kargl@FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #274517|0                           |1
        is obsolete|                            |

--- Comment #7 from Steve Kargl <kargl@FreeBSD.org> ---
Created attachment 274611
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=274611&action=edit
New diff with corrected magic numbers

The attached diff corrects a few magic numbers in my original
patch.  Compiling test5.c yields

% ./test5
fmaf: 0x1.0001p-133   expecting 0x1.0001p-133
fmaf: 0x1.d1179ep-29, expecting 0x1.d1179ep-29
fmaf: 0x1p-149,       expecting 0x1p-149

Note, this patch is against msun/src/s_fmaf.c at git revision
0dd5a5603e7a3, so it does not consider kib's recent commit to
use a hardware fma instruction of amd64 if detected.

-- 
You are receiving this mail because:
You are the assignee for the bug.