[SW] Fix FFT performance

[mp-17]

HP:

• There are #FPUs in total in the architecture
• Each FPU can compute 1 DP-FMACC per cycle or 1 DP-FLOP per cycle for every other operation
• 1 DP-FMACC == 2 DP-FLOP in terms of number of operations

64-bit FFT
#samples: `NFFT`
operations: 5 * NFFT * log2(NFFT) [DP-FLOP]
performance: operations / runtime [DP-FLOP/cycle]
max_performance: 5/4 * #FPUs [DP-FLOP/cycle]
utilization: 100 * performance / max_performance [%]

Notes:
To calculate the utilization using integer numbers, avoiding precision loss, we replace the 5/4 term with 1250/1000:

1250/1000 = 1.25 = 5/4

[DiyouS]

Can we also add the reason we need to multiply with 5/4 as a comment into the utilization calculation? I feel it would be helpful for anyone who wants to understand the code.

[mp-17]

Thanks, done!

[mp-17]

I can compile it locally, so it seems okay. But we need to fix the CI