I think you are jumping to conclusions. The C6678 device has floating point instructions. So while a change to fixed point operations is likely to increase performance, it is unlikely to be a big difference. A rough guess is things might get 2X faster, not 20X.
So I recommend you take more time to work out where you are losing cycles and why. Please see if this wiki article on profiling is helpful.
Here is one wild guess to consider. You might have a few loops that are very slow, and applying the techniques from this wiki article could make a big difference.
Thanks and regards,
-George