[WIP] Use pocketfft for Nesterov solver

[mkurc-ant]

This PR is intended to replace the FFT library used in ePlace Nesterov solver with pocketfft. Preliminary tests have shown that the library is faster.

Runtime measurements of the 3_3_place_gp stage with stock FFT and pocketff on an Intel(R) i7-8700 @ 3.20GHz CPU:

stock FFT lib stage runtime [s]	pocketfft stage runtime [s]
554	550
553	545
554	540
546	539
544	542
average
550.2	543.2

The improvement is expected to increase as soon as dsct and dcst functions are replaced which is a work in progress.

[maliberty]

It is helpful to state on what design you made these measurements.

[mkurc-ant]

Sorry, I forgot to mention. I've been testing on BlackParrot from OpenROAD-flow-scripts

[mkurc-ant]

Hmm. I can see regression tests failing. I need to revisit my changes.

[maliberty]

Are you still working on this or should it be closed?

[mkurc-ant]

@maliberty Hi. I think we can close it for now. I encountered issues with data layout used in pocketfft which required some conversion that ate up all performance gain.