When expanding fabs inline, GCC attempts to use a sequence without jumps. If
that sequence fails, GCC falls back to the equivalent of
if (x >= 0.0)
return x;
return -x;
This sequence is incorrect for the IEEE floating point value -0.0.
This problem normally does not occur because most architectures have an
instruction to compute fabs or one can perform bit-twiddling, especially for
software floating point emulation. However, for the cases such as software
floating point emulation of long double, the jump sequence is produced, leading
to wrong results for the corner case.
--
Summary: expand_abs wrong default code for floating point
Product: gcc
Version: 4.2.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: middle-end
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: dje at gcc dot gnu dot org
GCC build triplet: *-*-*
GCC host triplet: *-*-*
GCC target triplet: *-*-*
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29253
