[Bug math/18956] New: powf inaccuracy

[Bug math/18956] New: powf inaccuracy

[jsm28 at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=18956

            Bug ID: 18956
           Summary: powf inaccuracy
           Product: glibc
           Version: 2.22
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: jsm28 at gcc dot gnu.org
  Target Milestone: ---

Even in round-to-nearest mode, powf can have errors beyond the 9ulp maximum
accepted in glibc (seen on x86_64 and mips64):

Failure: Test: pow (0xd.73035p-4, -0x1.47bb8p+8)
Result:
 is:          5.22400435890634124322e+24   0x1.148e7600000000000000p+82
 should be:   5.22399859429881820899e+24   0x1.148e6200000000000000p+82
 difference:  5.76460752303423488000e+18   0x1.40000000000000000000p+62
 ulp       :  10.0000
 max.ulp   :  3.0000
Failure: Test: pow_downward (0xd.73035p-4, -0x1.47bb8p+8)
Result:
 is:          5.22398418278001062340e+24   0x1.148e3000000000000000p+82
 should be:   5.22399801783806590556e+24   0x1.148e6000000000000000p+82
 difference:  1.38350580552821637120e+19   0x1.80000000000000000000p+63
 ulp       :  24.0000
 max.ulp   :  4.0000
Failure: Test: pow_towardzero (0xd.73035p-4, -0x1.47bb8p+8)
Result:
 is:          5.22398360631925831997e+24   0x1.148e2e00000000000000p+82
 should be:   5.22399801783806590556e+24   0x1.148e6000000000000000p+82
 difference:  1.44115188075855872000e+19   0x1.90000000000000000000p+63
 ulp       :  25.0000
 max.ulp   :  8.0000
Failure: Test: pow_upward (0xd.73035p-4, -0x1.47bb8p+8)
Result:
 is:          5.22400435890634124323e+24   0x1.148e7600000000000000p+82
 should be:   5.22399859429881820899e+24   0x1.148e6200000000000000p+82
 difference:  5.76460752303423488000e+18   0x1.40000000000000000000p+62
 ulp       :  10.0000
 max.ulp   :  8.0000

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

[Bug math/18956] powf inaccuracy

[jsm28 at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=18956

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |2.23

--- Comment #2 from Joseph Myers <jsm28 at gcc dot gnu.org> ---
Fixed for 2.23.

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

[Bug math/18956] powf inaccuracy

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=18956

--- Comment #1 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "GNU C Library master sources".

The branch, master has been updated
       via  fa752c698146ca3e9f7747d33059fbef9bb02b0e (commit)
      from  6ace39382120dd0668b2168591611f71b04ac8f3 (commit)

Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.

- Log -----------------------------------------------------------------
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=fa752c698146ca3e9f7747d33059fbef9bb02b0e

commit fa752c698146ca3e9f7747d33059fbef9bb02b0e
Author: Joseph Myers <joseph@codesourcery.com>
Date:   Sat Sep 26 00:27:06 2015 +0000

    Fix powf inaccuracy (bug 18956).

    The flt-32 version of powf can be inaccurate because of bugs in the
    extra-precision calculation of (x-1)/(x+1) or (x-1.5)/(x+1.5) as part
    of calculating log(x) with extra precision: a constant used (as part
    of adding 1 or 1.5 through integer arithmetic) is incorrect, and then
    the code fails to mask a computed high part before using it in
    arithmetic that relies on s_h*t_h being exactly representable.  This
    patch fixes these bugs.

    Tested for x86_64 and x86.  x86_64 ulps for powf removed and
    regenerated to reflect reduced ulps from the increased accuracy for
    existing tests.

        [BZ #18956]
        * sysdeps/ieee754/flt-32/e_powf.c (__ieee754_powf): Add 0x00400000
        not 0x0040000 for high bit of mantissa.  Mask with 0xfffff000 when
        extracting high part.
        * math/auto-libm-test-in: Add another test of pow.
        * math/auto-libm-test-out: Regenerated.
        * sysdeps/x86_64/fpu/libm-test-ulps: Update.

-----------------------------------------------------------------------

Summary of changes:
 ChangeLog                         |   10 ++++++++++
 NEWS                              |    4 ++--
 math/auto-libm-test-in            |    1 +
 math/auto-libm-test-out           |   25 +++++++++++++++++++++++++
 sysdeps/ieee754/flt-32/e_powf.c   |    4 +++-
 sysdeps/x86_64/fpu/libm-test-ulps |   16 ++++++++--------
 6 files changed, 49 insertions(+), 11 deletions(-)

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