[Bug math/14064] New: libm-test.inc ulps calculation incorrect for subnormals

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

http://sourceware.org/bugzilla/show_bug.cgi?id=14064

             Bug #: 14064
           Summary: libm-test.inc ulps calculation incorrect for
                    subnormals
           Product: glibc
           Version: 2.15
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
        AssignedTo: unassigned@sourceware.org
        ReportedBy: jsm28@gcc.gnu.org
    Classification: Unclassified


libm-test.inc:check_float_internal calculates ulps errors thus:

        case FP_NORMAL:
          ulp = diff / FUNC(ldexp) (1.0, FUNC(ilogb) (expected) - MANT_DIG);
          break;
        case FP_SUBNORMAL:
          ulp = (FUNC(ldexp) (diff, MANT_DIG)
                 / FUNC(ldexp) (1.0, FUNC(ilogb) (expected)));

(MANT_DIG is one less than the appropriate one of FLT_MANT_DIG, DBL_MANT_DIG,
LDBL_MANT_DIG.)  The calculation for subnormals has the effect of assuming they
have as many significant mantissa bits as normal values; it may have been
written on the incorrect understanding that ilogb for a subnormal returns the
smallest normal exponent rather than the mathematical exponent of the input.

-- 
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.