[Bug math/18210] New: clog10 inaccuracy

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

https://sourceware.org/bugzilla/show_bug.cgi?id=18210

            Bug ID: 18210
           Summary: clog10 inaccuracy
           Product: glibc
           Version: 2.21
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: jsm28 at gcc dot gnu.org

clog10, using the dbl-64 version of atan2 in non-default rounding modes, can be
inaccurate beyond the 9ulp maximum considered acceptable by libm-test.inc.  For
example, on x86_64:

Failure: Test: Imaginary part of: clog10_upward (0x1.f4755cp+0 - 0x4.29411p-4
i)
Result:
 is:         -5.74390044257056600618e-02  -0x1.d68a52b0be1890000000p-5
 should be:  -5.74390044257055767951e-02  -0x1.d68a52b0be17d0000000p-5
 difference:  8.32667268468867405318e-17   0x1.80000000000000000000p-54
 ulp       :  12.0000
 max.ulp   :  5.0000

(I suspect there may in fact be cases where atan2 by itself is that inaccurate,
like atan in bug 18197, though I haven't yet found them in my random testing.)

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