jolt
jolt copied to clipboard
Optimize Eq Poly Construction during Grand Products
During each layer of grand products we compute a progressively larger multilinear lagrange basis polynomial EqPolynomial_0(r_0), EqPolynomial_1(r_0, r_1), ... EqPolynomial_n(r_0, r_1, ... r_n)
.
The polynomial is computed fresh every time but can take as input the previous layer's EqPolynomial. EqPolynomial_n(EqPolynomial_{n-1})
to accelerate by O(n/2).
Looks like 50% of the largest layer's Eq evals is unsafe_alloc_zero_vec
so probably not worth doing. Maybe worthwhile for Spartan's sumchecks.
after discussion with Justin we decided that this is not actually possible