Which hashes are heading for a precompile, and is leanVM-b the path?

[aryaethn]

Following the pivot away from Poseidon for L1, I have been trying to work out from public material which byte-oriented hash leanVM is heading toward, and I could not.

What I can see from the repos:

• leanVM exposes POSEIDON, DOT_PRODUCT and MULTILINEAR_EVAL. No SHA or BLAKE instruction.
• leanVM-b is a 192-bit binary field on WHIR/Ligerito and already carries BLAKE2S, benchmarked at 434,176 compressions/s.

Two questions:

1. Which hash or hashes are expected to become instructions in leanVM?
2. Is leanVM-b the intended path, or an experiment running alongside leanVM? Byte-oriented hashes are cheap in a binary field and expensive over Koala-Bear, so from the outside these look like the same question, but I would rather ask than assume.

Happy to help with implementation or spec work on whichever hashes end up in scope.

[TomWambsgans]

Hi,
Yes, leanVM-b is the path. As soon as it's clean enough, it will be merged into the leanVM repo. Binary fields for the win. About which hash to chose, there is no consensus yet. Very likely: SHA2, SHA3, BLAKE2s or BLAKE3.

[aryaethn]

Thanks @TomWambsgans, appreciate your response.
If it would be useful while the hash choice is open, I am happy to take a run at SHA-256 as an instruction in leanVM-b, following the BLAKE2S precedent. Given the pending merge into leanVM I would rather not aim at a moving target, so say the word if and when that is worth doing.

[aryaethn]

Looks like you've already got a full SHA2 implementation on the sha2 branch in leanVM-b (primitives + flock, replacing BLAKE2s throughout) — I'll stand down on that so we don't end up with two competing implementations. Happy to help review/test that branch instead if useful, or pick up something else.