revm v1.9 bytecode hash

[rakita]

Updating revm to v1.9

more on the change here: bluealloy/revm#165

It removes repetitive hashing and speeds up the fuzzing tests in the forge.

Additionally I moved the flag for limit_contract_code_size in CREATE/CREATE2 from Inspector override_spec to cfg and I set it to max value in foundry. Please check if this is okay or maybe in some places it should be consensus value 0x6000?

[mattsse]

ty!

[mattsse]

broke a test that compared used json object of cfg

updated

[gakonst]

@onbjerg time to run new bench on your box..

[onbjerg]

Nightly:

solmate on  main is 📦 v6.5.0 via  v18.7.0
❯ hyperfine 'forge test --no-match-test "testSafeTransferFromToEOA"'
Benchmark 1: forge test --no-match-test "testSafeTransferFromToEOA"
  Time (mean ± σ):      2.989 s ±  0.194 s    [User: 34.802 s, System: 0.425 s]
  Range (min … max):    2.740 s …  3.283 s    10 runs

This PR:

solmate on  main is 📦 v6.5.0 via  v18.7.0 took 2s
❯ hyperfine 'forge test --no-match-test "testSafeTransferFromToEOA"'
Benchmark 1: forge test --no-match-test "testSafeTransferFromToEOA"
  Time (mean ± σ):      2.504 s ±  0.218 s    [User: 24.660 s, System: 0.429 s]
  Range (min … max):    2.189 s …  2.839 s    10 runs

🔥

[gakonst]

Amazing. Was this also using RUSTFLAGS="-C target-cpu=native" for both?

[onbjerg]

None of them used that, can try with those flags

[gakonst]

Yep, try RUSTFLAGS="-C target-cpu=native" cargo b --release

[onbjerg]

@gakonst Results were about the same for me, very minor/insignificant change

[rakita]

15-20% sounds right.

I don't see any major low hanging fruits that would give as more performance boosts.
calldataload is probably the biggest one not optimized, but that is just a few percentages of execution.
It is maybe worth looking at run_fuzz_test, it uses HashMap::insert a lot.

[mattsse]

yeh definitely (invariant)fuzzing,
there's also a lot of Abi cloning involved that could be optimized