http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35862
--- Comment #24 from Tobias Burnus <burnus at gcc dot gnu.org> --- Created attachment 30458 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30458&action=edit Proof of concept (.c file) Proof of concept test case (in C, tested on x86-64-gnu linux with libc 2.17). [Somehow printf doesn't like my long double/__float128 example and prints 0.0 (long double) or garbage numbers (__float128).] Default 0.1: 0.1000000015 / 0.100000000000000006 -0.1: -0.1000000015 / -0.100000000000000006 FE_DOWNWARD 0.1: 0.0999999940 / 0.099999999999999991 -0.1: -0.1000000015 / -0.100000000000000006 FE_TONEAREST 0.1: 0.1000000015 / 0.100000000000000006 -0.1: -0.1000000015 / -0.100000000000000006 FE_TOWARDZERO 0.1: 0.0999999940 / 0.099999999999999991 -0.1: -0.0999999940 / -0.099999999999999991 FE_UPWARD 0.1: 0.1000000015 / 0.100000000000000006 -0.1: -0.0999999940 / -0.099999999999999991