Question: Do we have any information on extended-precision math for any fixed-point platform?

Answer:
To improve the cycle time for performing extended
        precision math when you have a processor that
        doesn't have signed * unsigned multiply operations
        (such as on the C2X, C5X devices).

        The theory behind it goes roughly like this:

        Lets start with a number that requires 32-bit resolution:

                X = 0.34578934567

        Scale the number into an integer and fractional part:

                X = X * 2pwr15
                    ----------
                      2pwr15

                X = 11330.82528
                    -----------
                       2pwr15

        Split the number into an integer and fractional component:

                X"   = 11330.82528 = Xi + Xf
                Xi   = 11330
                Xf   = 0.82528

        Now perform extended precision math using signed operations
        only:

                Yi:Yf
                Xi:Xf
                -----
                Xf*Yf
               Xf*Yi
               Xi*Yf
               Xi*Yi
           -----------
          W3:W2:W1:W0
          -----------

        The final result (W) is actually in Q format.