Sam Ragsdale
Sam Ragsdale
It's quite possible this is a bug in the custom Jolt MSMs. We use the same function in prove and verify so I don't think it would cause a verification...
For reference we had some conversion logic between `ark_ff::PrimeField` and `ff::PrimeField` back when we were using Spartan2 externally: #188
Current perf on an H100 accelerates sha2-chain 40% E2E for {Zeromorph, HyperKZG}. Currently slows down hyrax due to the tiny MSMs and call overhead. Remaining waste: - `max_num_bits` check at...
Looks like the `else` branch of the loop can be removed as well: https://github.com/a16z/jolt/blob/main/jolt-core/src/jolt/vm/mod.rs#L533
Please! Feel free to ask questions here.
242 cycle count is too small to get an idea of relative asymptotic performance. Can you try a bigger example – maybe in the 128k-512k range? Also thought sha2-chain was...
That would be great @MatteoMer! Good point @Maddiaa0 – new step 1 would be to determine if [RustCrypto/sponges](https://github.com/RustCrypto/sponges/blob/master/keccak/src/lib.rs) `keccak::f1600(...)` (used by Merlin) is the same as the EVM's Keccak256 and...
Current thinking is that we can use some preprocessing on the uniform A, B, C matrices to determine what they'll end up reading and writing. A given sparse entry's *column*...
Fixed in #344
Unit test parity between `R1CSInputs::clone_to_trace_len_chunks` and `R1CSProof::format_commitments`
I believe we'd want to match this up with the hard coded constants in `r1cs.rs` as well.