[Bug math/14793] New: Further spurious underflows from fma when a * b is small

[Bug math/14793] New: Further spurious underflows from fma when a * b is small

[jsm28 at gcc dot gnu.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=14793

             Bug #: 14793
           Summary: Further spurious underflows from fma when a * b is
                    small
           Product: glibc
           Version: 2.16
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
        AssignedTo: unassigned@sourceware.org
        ReportedBy: jsm28@gcc.gnu.org
    Classification: Unclassified


Given my patch http://sourceware.org/ml/libc-alpha/2012-11/msg00001.html for
bug 14784 there are still some underflows from fma (a, b, c) when a * b is
small, because the logic has a gap of two exponents between the cases when a *
b is so small that its exact value doesn't matter, and the cases where the
scaling-up for small a * b is enough to avoid underflow when computing the
Dekker multiplication.  Example, on x86 or x86_64: fmal
(0x1.0000000000000002p-16382L, 0x1.0000000000000002p-66L, 0x1p16383L) raises a
spurious underflow exception.

-- 
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/14793] Further spurious underflows from fma when a * b is small

[jsm28 at gcc dot gnu.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=14793

--- Comment #1 from Joseph Myers <jsm28 at gcc dot gnu.org> 2012-11-05 22:04:38 UTC ---
Actually, there are two cases causing problems here: cases such as the one
given, where scaling down happens rather than scaling up, and cases where e.g.
0x1p16319L is used instead of 0x1p16383L, where scaling up happens but not by
quite enough.

-- 
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/14793] Further spurious underflows from fma when a * b is small

[jsm28 at gcc dot gnu.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=14793

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> 2012-11-06 14:13:53 UTC ---
Fixed for 2.17 by:

commit 82477c28f46c579a149a8333c07233e9f4e43408
Author: Joseph Myers <joseph@codesourcery.com>
Date:   Tue Nov 6 14:12:54 2012 +0000

    Fix fma underflows with small x * y (bug 14793).

-- 
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/14793] Further spurious underflows from fma when a * b is small

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=14793

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.