FS$$DIV and FPU32 on F28335

You have to enable generation of FP code in project options.

Hello Bartosz,

what do you intend by "enable generation of FP code " ?

I have already activated the --float_support=fpu32 option. Is it what you are talking to ?

Yes. Hmm, do you use rts2800_fpu32.lib?

No, and that 's a part of my question : The DSP doesn't know how to make directly a division and we are forced to use a library (or define a .asm function to do it).

Like many float processors, the C28x+FPU has an instruction that estimates 1/x to 16-bits.  This estimate can then be used in a Newton-Raphson Algorithm to improve precision and create a division routine.  There is an example of this in the C28x FPU FastRTS Library http://focus.ti.com/docs/toolsw/folders/print/sprc664.html

Regards

Lori

Hi Mathieu,

In general, across all processors both fixed-point and floating point, and all manufacturers (not just TI), division is an expensive and slow instruction. You should avoid dividing where ever possible if you have a performance sensitive application.

For instance, instead of dividing by a constant, multiply by it's inverse. If the divide is a power of 2, use right shifts.

Regards,

Pete