FFT Multiplication (HP-48G/GX/G+)

[Gerson W. Barbosa]

(02-14-2014 12:34 AM)Raymond Del Tondo Wrote:  The most time-consuming parts are now the two FFT and the IFFT calls.
The final loop, which was the slowest part in the original program, now needs about 3 seconds only;-)

I think this has been fixed in my lastest version above, but optimization is still possible in the final loop, I admit. It appears the use of the lists, especially in the final loop, was causing the lack of linearity in the size vs. running times table we would expect from FFT multiplication. The previous hp 50g program would not complete a 128-element size multiplication, at least not under 30 minutes, when I quit waiting. Now it takes only 65 s. Here is a more complete table:

Code:


# elements     HP-71B     HP-48GX     hp 50g    
     16         24.11      10.72        5.35
     32         53.41      23.97       12.45      
     64        116.74      52.11       26.47
    128        269.89     119.82 (*)   65.35       
    256                                148.84

(*) estimated

Perhaps 5-digit groups are possible up to that size, considering IFFT is performed with full-accuracy. If not in UserRPL at least in your SysRPL version. This is worth checking.

Cheers,

Gerson.