https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78386
--- Comment #4 from joseph at codesourcery dot com <joseph at codesourcery dot com> --- On Wed, 16 Nov 2016, pinskia at gcc dot gnu.org wrote: > Most likely the use of fma. Which applies with current x86_64 just as it does with powerpc. It should be possible to reduce such an issue to a single function, with a specified argument for which that function returns different results with different optimization options. If it's an fma issue, using -ffp-contract=off should eliminate the difference between optimization levels.