From my research you can always fit it in a 32x32 multiply, you only need the extra bit at compile time. The extra bit tells you how to adjust the result in the end, but the adjustment is also a constant.