[Bug math/14034] New: acos (-1.0) gives -pi in round-downwards mode on x86

[Bug math/14034] New: acos (-1.0) gives -pi in round-downwards mode on x86

[jsm28 at gcc dot gnu.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=14034

             Bug #: 14034
           Summary: acos (-1.0) gives -pi in round-downwards mode on x86
           Product: glibc
           Version: 2.15
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
        AssignedTo: unassigned@sourceware.org
        ReportedBy: jsm28@gcc.gnu.org
    Classification: Unclassified


This was discussed in bug 13942, but is really a separate issue:

On x86, acosf (-1.0f), acos (-1.0) and acosl (-1.0L) all give -pi instead of pi
in round-downwards mode, although the result is specified by ISO C to be in the
range [0, pi].  The same applies to acosl (-1.0L) on x86_64.  I'm testing a
patch for this.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14034] acos (-1.0) gives -pi in round-downwards mode on x86

[bugdal at aerifal dot cx]

http://sourceware.org/bugzilla/show_bug.cgi?id=14034

Rich Felker <bugdal at aerifal dot cx> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bugdal at aerifal dot cx

--- Comment #1 from Rich Felker <bugdal at aerifal dot cx> 2012-04-29 17:32:06 UTC ---
One quick (but probably wrong; see below) solution I can think of is adding a
single fabs instruction just before return.

With that said, the value of -π returned by glibc's acos(-1) in
downward-rounding mode also differs from the value of -π obtained by -acos(1),
with a difference of 1ulp. What's significant (and probably bad) about this is
that the value is actually less than -π (the actual value), and thus if you
just take the absolute value, you get a result that's greater than π, which
would be rather unexpected in downward-rounding mode. It's within 1ulp so it
may technically be conformant, but it seems very bad.

On the other hand, inserting the fabs instruction just after the fsqrt
instruction (where the only thing it can do is convert -0 to +0) seems to work,
and the value I get trying it is the correctly-rounded representation of π.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14034] acos (-1.0) gives -pi in round-downwards mode on x86

[jsm28 at gcc dot gnu.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=14034

Joseph Myers <jsm28 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #2 from Joseph Myers <jsm28 at gcc dot gnu.org> 2012-04-30 09:42:01 UTC ---
Fixed by:

commit 5ba3cc691c856e5c67a7d4cd4713f20a79f7ba81
Author: Joseph Myers <joseph@codesourcery.com>
Date:   Mon Apr 30 09:38:06 2012 +0000

    Fix acos (-1) in round-downwards mode on x86 (bug 14034).

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14034] acos (-1.0) gives -pi in round-downwards mode on x86

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=14034

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|                            |security-

-- 
You are receiving this mail because:
You are on the CC list for the bug.