How do I get CORDIC to run faster with Fixed-Point data.

I have created two scripts (available). Each runs a program to take real and imaginary data, cordiccart2pol to find the magnitude and phase; and then take the phase only, with magnitude = 1, into cordicpol2cart to form a Constant-Modulus Limiter; with real and imaginary outputs. To do the same thing, the one with floating point data takes 6.063 seconds. The one using fixed-point data takes 129.114 seconds; over 20 times longer. Profile results show the difference is in the time the cordics take. What can I do to get the fixed-point version to run faster?