This is probably a basic question, but is this for floating point multiplies? Isn't the part thats being multiplied less than 64 bits because you also add the exponents?
Yes, this is for floating point multiplies. But internally, the FPU uses x86 extended-precision numbers, 80-bit floats, with 64 bits of significand and 15 exponent bits. So the multiplier is dealing with 64-bit numbers, plus a few more bits to support rounding correctly.
