WIP: Fboemer/inplace ntt
- Fixes #187 (closed)
- Use in-place NTT/invNTT, yielding ~10% performance improvement in NTT
- This should speedup SwitchFormat(), and thereby most HE operations as well.
NTTTransform1024 14.3 us 14.3 us 36699
NTTTransformInPlace1024 11.2 us 11.2 us 62468
INTTTransform1024 11.8 us 11.8 us 59110
INTTTransformInPlace1024 11.1 us 11.1 us 63256
NTTTransform4096 57.2 us 57.2 us 12241
NTTTransformInPlace4096 53.0 us 53.0 us 13194
INTTTransform4096 55.8 us 55.8 us 12607
INTTTransformInPlace4096 51.6 us 51.6 us 13565
Edited by Fabian Boemer