Floating point emulation on DSP-L without floating point arithmetic

Hi,

i am currently analyzing the FIR filter example project.

Our DSP processor supports only integer numbers. Floating point arithmetic will not be supported.

"The SoftFloat library from John Hausser is used to emulate floating point operations

on a fixed point processor (www.jhauser.us/arithmetic/SoftFloat.htm). This library is only used in combination with the CHESS front end."

As I checked the example FIR implementation the cycle add function will be realized with fmadds. The assembly command fmadds dealing with two floating point number.

 (cyclic_add function is provided by DSP-L /
 
 ST. void* cyclic_add(T* p, int offs, T* start, int len)).

The example code is dealing with floating numbers and the floating point calculation will be emulated.

Why convert the compiler the cyclic add function to a floating point assembly command?

(See attached file.)