Batch multi poly multi point #1
Conversation
src/ipa_pc/mod.rs
Outdated
| for ( | ||
| (_point_label, (point, _labels)), | ||
| labeled_polynomial | ||
| ) in query_to_labels_map.into_iter().zip(polys_iter) { |
There was a problem hiding this comment.
Plus, can we parallelize this ?
There was a problem hiding this comment.
I'll unresolve this just to keep track of it (maybe we can open a separate issue). The prover is quite slow, and it's highly affected also by this portion of computation.
We need to parallelize the underlying polynomial arithmetic ( HorizenOfficial/ginger-lib#87 ), use the Horner scheme to evaluate the polynomial (arkworks has a function horner_evaluate) and other optimizations to the underlying arithmetic, see the performances and compare them by doing everything in serial and parallelize the for loop instead.
This is not urgent for now but it will be in the future.
There was a problem hiding this comment.
We need to decide whether we keep just with the Fiat-Shamir transform split between outside and inside the batch_open_individual_opening_challenges() (with a "quick and dirty" modification of what to hash when producing the new query point), or directly move to a clean separation. For detailed explanation, see the inline comment for batch_open_individual_opening_challenges().
No description provided.