I wouldn't be surprised if the difference in rounding is an innate property of how the math behind an FMA instruction works and that guaranteeing the same round

I wouldn't be surprised if the difference in rounding is an innate property of how the math behind an FMA instruction works and that guaranteeing the same rounding (which is essentially rounding twice, once after mul and once after add) heavily impacts the performance benefits. Thus implementing FMA with an inconsistent rounding to separate mul and add instruction is in the interest of performance. I'm just guessing here tho.