opt: change polynomials to point value form #64
Conversation
|
On my machine, in a quick test,
|
|
Just added manual tables for MSM since I don't know any MSM libraries or more optimized versions off the top of my head. Current numbers are like this on my machine: (again, average of 100 iterations):
|
|
Added another optimization to MSM, now it's like this:
|
|
Last commit also should add another ~3% decrease to the time. |
Full Flow
|
|
Part of #62 |
| let mut result = vec![T::zero(); n_points]; | ||
|
|
||
| for i in 0..n_known { | ||
| let mut table = vec![]; |
There was a problem hiding this comment.
Nit: maybe worth initializing with Vec::with_capacity to avoid some reallocations?
There was a problem hiding this comment.
The main thing is that the maths is essentially correct. Next, I will refactor here to achieve greater performance.
|
I think that it is possible but would be more trouble than it's worth, since there are many special optimizations in similarly functioning parts. |
3 participants
Changes the representation and commitment of polynomials to point value form for better scaling, especially for evaluating polynomials with the domain of EC points. Assuming there are
arevealed points andbhidden points, previous algorithm used around2 * (a + b) * a * log_2(a)(log from exponentiation) EC additions, this version uses arounda * b * 256(256 from exponentiation). (Although it doesn't look like it, it might be slower with the current parameters, so further performance testing before merging would be preferable.) (Whole work is on top of #59)