[Bug math/14783] New: Incorrect subnormal fma results in non-default rounding modes

["jsm28 at gcc dot gnu.org" <sourceware-bugzilla@sourceware.org>]

http://sourceware.org/bugzilla/show_bug.cgi?id=14783

             Bug #: 14783
           Summary: Incorrect subnormal fma results in non-default
                    rounding modes
           Product: glibc
           Version: 2.16
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
        AssignedTo: unassigned@sourceware.org
        ReportedBy: jsm28@gcc.gnu.org
    Classification: Unclassified


The various fma implementations have logic for scaling down results, in the
case of subnormal results that are at least half the smallest positive normal
value in magnitude, that assumes round-to-nearest and so is incorrect in other
rounding modes.  (The code at fault is the same as in bug 14152, but from the
user perspective they are separate bugs, missing exceptions (any rounding mode)
versus incorrect results (non-default rounding modes).)

For example, on x86, fmal (0x1.4p-16382L, 0x1.0000000000000004p-1L,
0x1p-16384L) returns 0x7.000000000000001p-16385L independent of rounding mode,
but this should be 0x7.000000000000002p-16385L in FE_UPWARD mode.

Working on a patch.

-- 
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.