Skip to content

perf: Compute $g_j(1)$ for Zero-Check using wire assignment #1722

@Tabaie

Description

@Tabaie

There is no need to evaluate the gate again for these values. It can instead be computed as a dot product of the second half of the assignment vector $w(1, -)$ and the second half of the evaluation point eq vector.
If we clone the assignments or decide to use it destructively we can fold it as we do the eq vector and continue computing $g_j(1)$ this way for $j > 0$ as well.

It may be that for low degree gates the extra work of folding the assignment (causing a total of 2 multiplication passes to compute $g_j(1)$ may not be worthwhile for linear or quadratic gates) It would probably be best to count the number of multiplication operations in the gate to decide this.

Metadata

Metadata

Assignees

No one assigned

    Labels

    feat: gkrPRs related to GKR

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions