Lucas Xia

Results 41 issues of Lucas Xia

I've measured through the double_verify_proof ACIR bb.js test that Honk uses right about 4GiB of memory, while Plonk uses 3.6GiB of memory, potentially less. These numbers could be inaccurate, but...

We use size_t in way too many places. We should probably just use uint32_t or uint64_t by default and cast to size_t when needed. This leads to Mac build serialization...

Currently, write_vk, prove, and verify flows are implemented, but not the write_pk flow since it wasn't necessary for the prove-then-verify flow.

I added a VerificationKey constructor purely for MsgPack deserialization as part of the write_vk/verify flow in the bb binary, but it leads to a lot of boilerplate code.

Brought up in a discussion with Zac over potential slowdowns in Sumcheck. Currently we iterate over each polynomial and for each one, we call `add_scaled`, which loops over the same...

optimization

Circuit generation is not parallelized in protogalaxy tests or benchmarks. They're not necessary for the actual benchmarks, but would be nice to have for faster iteration.

This script replaces the gen_inner_proof_inputs_ultra_honk.sh with regenerate_verify_honk_proof_inputs.sh, which does the resetting of test_programs and input copying automatically so that people don't have to know intricate details on what to do...

This would involve modifying the verify_proof interfaces to take in an aggregation object and aggregating the pairing points from KZG. The tube circuit would then have proper aggregation. (but that's...

This will break many tests likely. The fix is adding a `builder.add_recursive_proof(default_agg_obj)` to many of the circuits. One solution is to refactor such that a call to finalize adds the...

We currently just add some (correct amount of) public inputs as new variables to the builder when constructing the Tube circuit. However, we need to set witnesses (from the folding...