How long does the 2048 point FFT computation of floating-point data in the Piccolo?

Andrey,

For 32-bit using FPU you are looking at 10 cycles per butterfly, so 20480 cycles on a Piccolo.

There is more info here comparing different strategies across our device families.  I believe the code is in library form, so the source is not shown.  Typically these are hand optimized by our SW team, so this should be the most ideal code/cycles.

http://processors.wiki.ti.com/index.php/C2000_FFT:_VCU,_FPU_or_FixedPoint