Improvements over plonk #671
Conversation
Codecov Report
@@ Coverage Diff @@
## main #671 +/- ##
=======================================
Coverage 96.77% 96.77%
=======================================
Files 118 118
Lines 28454 28461 +7
=======================================
+ Hits 27536 27543 +7
Misses 918 918
📣 Codecov offers a browser extension for seamless coverage viewing on GitHub. Try it in Chrome or Firefox today! |
| let new_factor = num / den; | ||
| let new_term = coefficients.last().unwrap() * &new_factor; | ||
| coefficients.push(new_term); | ||
| z_evaluations.push(z_evaluations.last().unwrap() * (&z_nums[i] * &z_denoms[i])); |
There was a problem hiding this comment.
I think this can be done better using a scan iterator. That avoids pushing every evaluation into the vector, which is not super efficient
| .collect(); | ||
| let powers_main_group: Vec<G1Point> = | ||
| core::iter::successors(Some(FieldElement::from(1)), |acc| Some(acc * &s)) | ||
| .take(n + 3) |
There was a problem hiding this comment.
There is no need for two maps; we can just fuse them into a single one with g1.operate_with_self(power.representative()).
Yet another version could use successors starting from the generator and having s directly as representative, so at every step we have s*previous
| let z_nums: Vec<_> = (0..&cpi.n - 1) | ||
| .map(|i| { | ||
| lp(&witness.a[i], &cpi.domain[i]) | ||
| * lp(&witness.b[i], &(&cpi.domain[i] * &cpi.k1)) |
There was a problem hiding this comment.
Maybe this could be more efficient just by just iterating over cpi.domain zipped with witness.a, witness.b, witness.c
| result | ||
| let u_powers = [&FieldElement::one(), &u, &(u * u)]; | ||
|
|
||
| (0..=2) |
There was a problem hiding this comment.
There is no need to start from zero if it is just multiplying with one
Description