Compare packed single-precision (32-bit) floating-point elements in a and b for not-greater-than.