Issue with sha256 circuit benchmark

[vivijj]

Hello zkSync Team,

I am currently conducting research using the era-boojum proof system and have encountered an issue during the execution of the SHA-256 benchmark, the program appears to stall at the CSReferenceAssembly::wait_for_witness() function, both run_sha256_prover_non_recursive and run_sha256_prover_recursive_mode_poseidon2.

[arose0]

Hey @vivijj - can you share what steps you took here so that we can investigate?

[vivijj]

sure, I ran the following script on a Mac Pro M1 with 16GB of RAM:
RUSTFLAGS='-C target-cpu=native' cargo +nightly test --release -- --nocapture run_sha256_prover_non_recursive --ignored
By default, the test preimage size is set to 8KB, which causes the proving process to get stuck. I have tried changing the preimage size to 4KB and 2KB, and only when it is set to 2KB does the process not get stuck.

[canaancc]

I have the same problem

[shamatar]

There is/was a bug in the new witness resolver, and it's fixed in the internal repo, I'll ask to merge it here too. Depending on your machine configuration 1-2-4 kB tests just work (I have 4kB working on M1). It's actually funny because it looks to happen in this test only and doesn't happen in the Era circuits

[canaancc]

Hi，when this issue can be solved, when the internal repo can be published?

[shamatar]

I believe it was updated

[vivijj]

Seems alright. By the way, when do you plan to open source your GPU prover for Boojum? @shamatar

[arose0]

Closing this issue as resolved.

@vivijj we will be opening this repo very soon - I don't think we have an exact date but it is on the list of priorities.