> I had to drop down to O0 to see branchless be faster in any case

Did you check whether your branchy code actually still was branchy after the compiler processed it at higher optimization levels?