2812 x 28335 performance

Silvia Ferreira said:

I'm migrating from DSP TMS320F2812 fixed point to TMS320F28335 floating point.

But, after some analisys I realize that there is no significant difference in performance.

time for a multiplication: 128ns in floating point
while the fixed point takes about 117ns.

The project is already configured to run floating point

with the appropriate libraries and run from flash memory.

Is this performance normal?

Is there a specific library that I need to use to improve the DSP performance other than:


- rts2800_fp32.lib
- rts2800_fp32_fast_supplement.lib


#include <stdlib.h>
#include <stdio.h>
#include <math.h>

#include "DSP28x_Project.h"

floating point is make your programming and debugging easier. It doens't make sense to compare the fixed point algorithm with floating point in some very straighforward algorithm ( even for FIR filtering).  you should compare the floating point IIR filter with fixed point IIR filter then u can get significant difference.

28335 give u DMA and FPU which 2812 can't provide. FPU will significantly decrease the debugging cost for complicated algorithm like high order IIR filtering. It provides bigger dyanmic range while doesn't lower the speed significantly. Fixed point debugging is always much difficult than floating point.