https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79436
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> --- "this default behaviour produces invalid results." No. Read http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.22.6768 > The same code compiled at -O2 for x86-64 does not assert. That because the default x86_64 target does not have the fused multiple add instruction. On a newer Intel (or AMD) machine, add -march=naitve and you will see the same behavior. You are not understand what fused multiple add instruction does and how floating point works. VFMA is not just multiply and accumulate but rather fused multiply and add. The multiplication is done in infinite precision and then the add is done in that same infinite precision and then a round happens down to double. Instead of what happens with -ffp-contract=off which is not to used the fused multiple add instruction which means round between the multiply and addition.
