Closed LindaGuiga closed 8 months ago
Kudos, no new issues were introduced!
0 New issues
0 Security Hotspots
No data about Coverage
No data about Duplication
Update: With the new optimization PRs, there is no additional column: when there are only 1 or 2 columns for a STARK, we do not create helper columns.
Why so complicated? Why introduce a completely new helper, instead of just bundling eg every two lookups into a single partial sum?
Is opening a single z polynomial only, on the first row cheaper, than opening many of them on the first row?
This PR bundles, within a CTL, the looking tables associated to the same STARK together. This is done using helper columns.
Even though this adds helper columns when there are only 1 or 2 looking tables to bundle together, we can almost halve the number of commited columns for large groups of looking tables. For example,
MemoryStark
has 136 looking tables coming fromKeccakSponge
reads, but we only have to commit to 69 polynomials (68 helper columns and 1 Z polynomial) with the bundling.