You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While benchmarking @AlexanderLemmens's modular reduction procedure I observed isolated cases where performance improved by switching from Montgomery representation and multiplication to canonical representation and Alexander reduction. Specifically, for large enough $N$, NTT-based interpolation seems to be faster. The following benchmark snippets are obtained after switching back from Alexander's method to Montgomery.
While benchmarking @AlexanderLemmens's modular reduction procedure I observed isolated cases where performance improved by switching from Montgomery representation and multiplication to canonical representation and Alexander reduction. Specifically, for large enough$N$ , NTT-based interpolation seems to be faster. The following benchmark snippets are obtained after switching back from Alexander's method to Montgomery.
All other benchmarks are green, implying that Montgomery is faster.
Focusing on these red ones, I would say it indicates some suboptimal processing. I estimate that there is some performance gain to be had here.
Note that this task has very low priority since NTT-based interpolation is not used anywhere. (We use INTT in Triton.)
The text was updated successfully, but these errors were encountered: