[Bug math/16620] New: [ldbl-128ibm] exp10l spurious overflows / bad directed rounding results

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

https://sourceware.org/bugzilla/show_bug.cgi?id=16620

            Bug ID: 16620
           Summary: [ldbl-128ibm] exp10l spurious overflows / bad directed
                    rounding results
           Product: glibc
           Version: 2.19
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: jsm28 at gcc dot gnu.org
              Host: powerpc*-*-linux*

The ldbl-128ibm implementation of exp10l uses a version of log(10) split into
high and low parts - but the low part is negative, so probably causing spurious
overflows from __ieee754_expl (exp_high) in cases close to the overflow
threshold (and definitely, given libgcc fixed for directed rounding modes,
results below LDBL_MAX for overflow when rounding downward / to zero just above
that threshold, which is the case where I found this problem).  The split
log(10) needs to be such that both parts are positive to avoid this issue.

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