Skip to content

bench: add vanilla revm/op-revm baselines to bench suite#301

Open
RealiCZ wants to merge 3 commits into
mainfrom
cz/perf/revm-bench-baselines
Open

bench: add vanilla revm/op-revm baselines to bench suite#301
RealiCZ wants to merge 3 commits into
mainfrom
cz/perf/revm-bench-baselines

Conversation

@RealiCZ
Copy link
Copy Markdown
Collaborator

@RealiCZ RealiCZ commented May 28, 2026

Summary

Add vanilla revm and op-revm baseline rows to the criterion bench suite at both the currently-pinned versions and the latest crates.io releases, so each group shows how far the mega specs sit from upstream — at versions we depend on today and at versions we'll likely move to next.

What's new

  • Four baseline rows per group: revm_pinned, revm_latest, op_revm_pinned, op_revm_latest. Joined by the three mega rows (equivalence, mini_rex, rex4) so each criterion group emits seven comparable rows.
  • Shared adapters in benches/common/baseline_adapters.rs, pulled into mega_bench, revm_bench, and transact via #[path = ...]. Also exposes SPEC_IDS, make_mega_evm, and build_mega_tx so the operator-fee zero-out and the MegaTransaction envelope-wrap exist in one place instead of seven.
  • revm 38 + op-revm 20 added as renamed dev-dependencies (revm_latest / op_revm_latest). op-revm 20 transitively pulls revm 38, so both _latest baselines share one revm tree.
  • Criterion HTML reports opted in (html_reports, plotters, cargo_bench_support) for local violin plots under target/criterion/report/.

Notes

  • transact_call_revm_latest pins Cancun to dodge EIP-7825's tx_gas_limit_cap against the multi-G gas_limit workloads. op_revm_latest doesn't need this — its default Isthmus spec lacks the cap; the comment notes what to do if a future op-revm release inherits it.
  • bench_transfer_multi reuses one EVM across 1000 txns, so it inlines the baseline setup instead of using the single-shot transact_call_* adapters.
  • No production code touched — dev-dependencies and benches only.

Test plan

  • cargo check -p mega-evm --benches
  • cargo fmt --all --check
  • cargo clippy -p mega-evm --benches --all-features --locked
  • cargo bench -p mega-evm --bench transact -- --quick — 7/7 rows pass per
    group
  • cargo bench -p mega-evm --bench mega_bench -- --quick
  • cargo bench -p mega-evm --bench revm_bench -- --quick

RealiCZ added 3 commits May 26, 2026 15:14
@RealiCZ
bench: add vanilla revm/op-revm baselines to bench suite
0a20a75 
Add four baseline rows -- revm_pinned, revm_latest, op_revm_pinned,
op_revm_latest -- to the criterion workloads so each group shows how
far the mega specs sit from vanilla revm/op-revm, at both the
currently-depended versions and the latest crates.io releases
(revm 38 / op-revm 20).

The shared adapters and CacheDB builder live in
benches/common/baseline_adapters.rs and are reused by revm_bench,
transact, and mega_bench. revm_latest / op_revm_latest are added as
renamed dev-dependencies (op-revm 20 transitively pulls revm 38, so
both _latest baselines share one revm tree). criterion gains
html_reports for local violin plots.
@RealiCZ
bench: dedupe mega-evm setup and polish baseline adapters
6424530 
Extract `SPEC_IDS`, `make_mega_evm`, and `build_mega_tx` into
`benches/common/baseline_adapters.rs` so the `MegaContext` +
`modify_chain` + `MegaEvm::new` dance and the `MegaTransaction`
envelope-wrap live in one place instead of seven across the three
bench files (net -35 lines).

Also drop now-stale version numbers from adapter docstrings,
document why `transact_call_op_revm_latest` does not pin a spec,
and merge the split import block in `revm_bench.rs`.
@RealiCZ
bench: polish comments across bench files
4f9b459 
Drop the verbose Cargo.toml criterion-features comment, strip rotty
version numbers and implementation counts from baseline-adapter
docstrings, remove redundant WHAT comments around CREATE/CREATE2
helpers, and fix a stale group name in the mega_bench module doc.
@RealiCZ RealiCZ added spec:unchanged No change to any `mega-evm`'s behavior api:unchanged No change to the public interface or API dependencies Pull requests that update a dependency file rust Pull requests that update rust code comp:core Changes to the `mega-evm` core crate labels May 28, 2026
@claude
Copy link
Copy Markdown
Contributor

claude Bot commented May 28, 2026

LGTM. The abstraction in baseline_adapters.rs is clean, the #[path] sharing pattern is the right approach for criterion binaries, and the spec-pinning rationale is well documented.

One minor note: bench_transfer_multi inlines the four baseline setups rather than using the shared adapters (justified, since it holds one EVM across 1000 transactions). The CANCUN pin on the revm_latest block there is missing the inline rationale that transact_call_revm_latest carries — easy for a future reader to remove without knowing why it was added (see inline).

claude[bot]
claude Bot reviewed May 28, 2026
View reviewed changes
Comment thread crates/mega-evm/benches/revm_bench.rs
group.bench_function("revm_latest", |b| {
b.iter(|| {
let mut evm = ContextLatest::mainnet()
.modify_cfg_chained(|cfg| cfg.set_spec_and_mainnet_gas_params(SpecIdLatest::CANCUN))
Copy link
Copy Markdown
Contributor

@claude claude Bot May 28, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The CANCUN pin here silently carries the EIP-7825 tx_gas_limit_cap avoidance rationale from transact_call_revm_latest, but this workload uses gas_limit(100_000) so the cap (2^24 ≈ 16 M) is never reached anyway. A brief inline comment (or just dropping the pin) would make future maintenance easier:

Suggested change
.modify_cfg_chained(|cfg| cfg.set_spec_and_mainnet_gas_params(SpecIdLatest::CANCUN))
.modify_cfg_chained(|cfg| cfg.set_spec_and_mainnet_gas_params(SpecIdLatest::CANCUN)) // avoid EIP-7825 tx_gas_limit_cap on default Osaka spec

@RealiCZ
Copy link
Copy Markdown
Collaborator Author

RealiCZ commented May 28, 2026

/benchmark

Troublor
Troublor requested changes May 28, 2026
View reviewed changes
Comment thread crates/mega-evm/benches/transact.rs
Comment on lines +18 to +19
#[path = "common/baseline_adapters.rs"]
mod common;
Copy link
Copy Markdown
Collaborator

@Troublor Troublor May 28, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: since common/baseline_adapters.rs is the only file under common/, renaming it to common/mod.rs would let all three bench files drop the #[path] attribute — mod common; would resolve via the standard sibling-folder lookup. Same change in mega_bench.rs:34-35 and revm_bench.rs:59-60.

Trade-off: requires renaming the file, so probably only worth doing if you're already touching this area. If a second shared file ever joins, you'd switch to a mod.rs that re-exports submodules anyway.

Suggested change
#[path = "common/baseline_adapters.rs"]
mod common;
mod common;

Comment thread crates/mega-evm/benches/common/baseline_adapters.rs
Comment on lines +247 to +276
pub fn add_baseline_rows_suffixed<FP, FL>(
group: &mut criterion::BenchmarkGroup<'_, criterion::measurement::WallTime>,
variant: &str,
params: &CallParams,
make_pinned_db: &FP,
make_latest_db: &FL,
) where
FP: Fn() -> MemoryDatabase,
FL: Fn() -> CacheDBLatest<EmptyDBLatest>,
{
let with = |base: &str| -> String {
if variant.is_empty() {
base.to_string()
} else {
format!("{base}/{variant}")
}
};
group.bench_function(with("revm_pinned"), |b| {
b.iter(|| transact_call_revm_pinned(make_pinned_db(), params))
});
group.bench_function(with("revm_latest"), |b| {
b.iter(|| transact_call_revm_latest(make_latest_db(), params))
});
group.bench_function(with("op_revm_pinned"), |b| {
b.iter(|| transact_call_op_revm_pinned(make_pinned_db(), params))
});
group.bench_function(with("op_revm_latest"), |b| {
b.iter(|| transact_call_op_revm_latest(make_latest_db(), params))
});
}
Copy link
Copy Markdown
Collaborator

@Troublor Troublor May 28, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All three bench files repeat the same pipeline per subject — build the DB, build a tx, transact, assert — written seven times per group because each subject needs its own DB type, EVM constructor, and tx envelope. The current add_baseline_rows_* helper plus the parallel mega-row loop in each caller is the closest we can get to deduping under this shape.

Consider a follow-up refactor that hoists this into a portable Workload + Subject trait:

pub struct Workload {
    pub accounts: Vec<Account>,   // address → { balance, code, storage }
    pub txs:      Vec<TxSpec>,    // single-tx benches pass vec![tx]; transfer_multi
                                  // passes 1000 and a Subject loops internally
}

pub trait Subject {
    fn name(&self) -> &str;
    fn run(&self, workload: &Workload);  // owns DB build, EVM build, tx loop, assert
}

The bench file shrinks to declaring a Workload and calling one of register_all / register_baselines / register_mega. Each Subject impl absorbs its own quirks — Cancun pin, operator-fee zero-out, MegaTransaction envelope wrap, is_success() assert.

Two design points worth flagging:

  • Keep TxSpec minimal (today's CallParams, ~5 fields). Reusing revm::TxEnv directly would tie the abstraction to one revm version — both _latest subjects would have to translate it into their own TxEnv anyway, since revm 27 and revm 38 are separate crates.
  • Vec<TxSpec> absorbs bench_transfer_multi without a parallel "sequence workload" type — the subject decides build-once-loop.

Not for this PR — #301 is a clean introduction of the baselines and is better merged as-is, with the refactor landing on top.

Comment thread crates/mega-evm/benches/revm_bench.rs
Comment on lines +539 to 540
let r = evm.transact(build_mega_tx(tx)).expect("transfer should succeed");
black_box(&r);
Copy link
Copy Markdown
Collaborator

@Troublor Troublor May 28, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Missing the is_success() assertion that all four baseline blocks above have (lines 437-442, 461-467, 488-494, 516-522). expect() on transact() only catches internal errors — a transaction that halts (out-of-gas, a per-tx resource limit being exceeded under one of the mega specs, etc.) returns Ok(ResultAndState) with result.is_success() == false, and the bench keeps spinning. The mega rows would then report numbers from "1000 transfers where some fraction actually halted," which is silently incomparable to the baseline rows that always assert success — defeating the cross-row comparison this bench exists to enable.

Suggested change
let r = evm.transact(build_mega_tx(tx)).expect("transfer should succeed");
black_box(&r);
let r = evm.transact(build_mega_tx(tx)).expect("transfer should succeed");
assert!(
r.result.is_success(),
"mega transfer should succeed: {:?}",
r.result
);
black_box(&r);

Comment thread crates/mega-evm/Cargo.toml
criterion = { workspace = true, features = ["html_reports", "plotters", "cargo_bench_support"] }
hex.workspace = true
mega-evm = { path = ".", features = ["test-utils"] }
op_revm_latest = { package = "op-revm", version = "20.0.0", default-features = false, features = ["dev", "serde", "std"] }
Copy link
Copy Markdown
Collaborator

@Troublor Troublor May 28, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The two renamed packages use snake_case (op_revm_latest, revm_latest) while the rest of [dev-dependencies] follows the crate's upstream name — kebab-case for mega-evm and revm-inspectors, snake_case only for serde_json because that's literally the crate's name on crates.io. Since these are renamed packages, the local key is whatever we choose — kebab-case matches house style.

Cargo auto-converts dashes to underscores in Rust identifiers, so use op_revm_latest::... and use revm_latest::... keep working unchanged across all three bench files. Only Cargo.toml changes.

Suggested change
op_revm_latest = { package = "op-revm", version = "20.0.0", default-features = false, features = ["dev", "serde", "std"] }
op-revm-latest = { package = "op-revm", version = "20.0.0", default-features = false, features = ["dev", "serde", "std"] }

Same change on line 52:

revm-latest = { package = "revm", version = "38.0.0", default-features = false, features = ["dev", "serde", "std"] }

@github-actions
Copy link
Copy Markdown

github-actions Bot commented May 28, 2026

Criterion Benchmark Comparison

Comparing baseline → feature

Baseline gap (HEAD only — how far is each EVM layer from revm_pinned)

rex4 is the current production spec — that's the row to read for "user-visible mega gap". The other rows are diagnostic: they show which layer adds cost.

analysis

spec time × vs revm_pinned
revm_pinned 20.46 µs 1.00× (baseline)
revm_latest 22.63 µs 1.11×
op_revm_pinned 21.24 µs 1.04×
op_revm_latest 23.80 µs 1.16×
equivalence 22.00 µs 1.08×
mini_rex 27.33 µs 1.34×
rex4 27.79 µs 1.36×

empty_transaction

spec time × vs revm_pinned
revm_pinned 2.11 µs 1.00× (baseline)
revm_latest 2.60 µs 1.23×
op_revm_pinned 3.23 µs 1.53×
op_revm_latest 3.71 µs 1.76×
equivalence 3.66 µs 1.73×
mini_rex 4.01 µs 1.90×
rex4 4.03 µs 1.91×

mixed_workload

spec time × vs revm_pinned
revm_pinned 15.89 µs 1.00× (baseline)
revm_latest 14.78 µs 0.93×
op_revm_pinned 16.90 µs 1.06×
op_revm_latest 15.71 µs 0.99×
equivalence 17.45 µs 1.10×
mini_rex 26.51 µs 1.67×
rex4 27.25 µs 1.72×

simple_ether_transfer

spec time × vs revm_pinned
revm_pinned 2.36 µs 1.00× (baseline)
revm_latest 2.92 µs 1.24×
op_revm_pinned 3.47 µs 1.47×
op_revm_latest 3.89 µs 1.65×
equivalence 3.92 µs 1.66×
mini_rex 4.26 µs 1.81×
rex4 4.28 µs 1.82×

snailtracer

spec time × vs revm_pinned
revm_pinned 52.39 ms 1.00× (baseline)
revm_latest 50.83 ms 0.97×
op_revm_pinned 53.23 ms 1.02×
op_revm_latest 50.55 ms 0.96×
equivalence 52.89 ms 1.01×
mini_rex 124.13 ms 2.37×
rex4 167.99 ms 3.21×

subcall_1000_nested

spec time × vs revm_pinned
revm_pinned 947.51 µs 1.00× (baseline)
revm_latest 671.97 µs 0.71×
op_revm_pinned 1.00 ms 1.06×
op_revm_latest 680.73 µs 0.72×
equivalence 1.27 ms 1.34×
mini_rex 1.83 ms 1.93×
rex4 2.01 ms 2.12×

subcall_1000_no_value

spec time × vs revm_pinned
revm_pinned 488.14 µs 1.00× (baseline)
revm_latest 352.50 µs 0.72×
op_revm_pinned 511.73 µs 1.05×
op_revm_latest 359.50 µs 0.74×
equivalence 654.76 µs 1.34×
mini_rex 981.33 µs 2.01×
rex4 1.05 ms 2.15×

subcall_1000_transfer_1wei

spec time × vs revm_pinned
revm_pinned 342.63 µs 1.00× (baseline)
revm_latest 273.92 µs 0.80×
op_revm_pinned 347.36 µs 1.01×
op_revm_latest 279.66 µs 0.82×
equivalence 447.57 µs 1.31×
mini_rex 755.41 µs 2.20×
rex4 851.88 µs 2.49×

transfer_multi_1000

spec time × vs revm_pinned
revm_pinned 1.13 ms 1.00× (baseline)
revm_latest 763.93 µs 0.67×
op_revm_pinned 1.74 ms 1.54×
op_revm_latest 1.25 ms 1.10×
equivalence 1.85 ms 1.63×
mini_rex 2.08 ms 1.83×
rex4 2.07 ms 1.83×

weth9_transfer

spec time × vs revm_pinned
revm_pinned 17.62 µs 1.00× (baseline)
revm_latest 19.86 µs 1.13×
op_revm_pinned 18.66 µs 1.06×
op_revm_latest 20.60 µs 1.17×
equivalence 19.35 µs 1.10×
mini_rex 23.14 µs 1.31×
rex4 23.77 µs 1.35×

PR-base → PR-head regression check

147 benchmarks total, 18 with >5% change
Benchmark Baseline Feature Delta
add/mini_rex 16.0 µs 15.9 µs -0.6% ✅
address/mini_rex 7.8 µs 7.8 µs +0.9% ✅
analysis/equivalence 22.2 µs 22.0 µs -1.1% ✅
analysis/mini_rex 27.9 µs 27.3 µs -2.1% ✅
analysis/rex4 28.7 µs 27.8 µs -3.0% ✅
batch_transfer_high_complexity_1x1/equivalence 260.8 µs 264.9 µs +1.5% ✅
batch_transfer_high_complexity_1x1/mini_rex 437.8 µs 441.0 µs +0.7% ✅
batch_transfer_high_contention_10x10/equivalence 458.4 µs 456.9 µs -0.3% ✅
batch_transfer_high_contention_10x10/mini_rex 921.7 µs 883.4 µs -4.2% ✅
batch_transfer_little_complexity_1x1/equivalence 29.5 µs 29.2 µs -0.9% ✅
batch_transfer_little_complexity_1x1/mini_rex 52.8 µs 51.6 µs -2.2% ✅
batch_transfer_low_contention_1x1/equivalence 17.1 µs 16.8 µs -1.7% ✅
batch_transfer_low_contention_1x1/mini_rex 31.9 µs 30.7 µs -3.7% ✅
batch_transfer_medium_complexity_1x1/equivalence 138.7 µs 139.2 µs +0.4% ✅
batch_transfer_medium_complexity_1x1/mini_rex 238.3 µs 231.2 µs -3.0% ✅
batch_transfer_medium_contention_5x5/equivalence 132.3 µs 132.2 µs -0.1% ✅
batch_transfer_medium_contention_5x5/mini_rex 260.6 µs 251.1 µs -3.6% ✅
batch_transfer_some_complexity_5x5/equivalence 838.9 µs 844.9 µs +0.7% ✅
batch_transfer_some_complexity_5x5/mini_rex 1462.3 µs 1425.7 µs -2.5% ✅
block_executor_deploy/rex4/deploy_1 27.5 µs 27.2 µs -1.1% ✅
block_executor_empty_txs/rex4/10_txs 70.5 µs 70.5 µs +0.0% ✅
block_executor_empty_txs/rex4/1_txs 26.9 µs 26.5 µs -1.4% ✅
block_executor_empty_txs/rex4/50_txs 258.9 µs 266.5 µs +2.9% ✅
block_executor_mixed_txs/rex4/10_txs 220.0 µs 222.7 µs +1.2% ✅
block_executor_mixed_txs/rex4/1_txs 49.0 µs 48.9 µs -0.3% ✅
block_executor_spec_comparison/equivalence/5_mixed_txs 93.3 µs 93.6 µs +0.4% ✅
block_executor_spec_comparison/mini_rex/5_mixed_txs 68.0 µs 67.9 µs -0.1% ✅
block_executor_spec_comparison/rex4/5_mixed_txs 125.3 µs 126.0 µs +0.6% ✅
block_executor_spec_comparison/rex5/5_mixed_txs 125.6 µs 126.5 µs +0.8% ✅
call_value_empty_account/equivalence/empty_account_50 37.7 µs 34.6 µs -8.2% ✅
call_value_empty_account/equivalence/existing_account_50 38.1 µs 34.6 µs -9.0% ✅
call_value_empty_account/mini_rex/empty_account_50 52.5 µs 47.9 µs -8.8% ✅
call_value_empty_account/mini_rex/existing_account_50 51.8 µs 47.4 µs -8.5% ✅
call_value_empty_account/rex4/empty_account_50 55.1 µs 50.3 µs -8.9% ✅
call_value_empty_account/rex4/existing_account_50 54.3 µs 50.1 µs -7.6% ✅
create_deploy/equivalence/create2_10 26.4 µs 25.8 µs -2.0% ✅
create_deploy/equivalence/create_10 20.8 µs 20.6 µs -1.1% ✅
create_deploy/mini_rex/create2_10 39.2 µs 39.6 µs +0.9% ✅
create_deploy/mini_rex/create_10 28.5 µs 28.9 µs +1.3% ✅
create_deploy/rex4/create2_10 39.9 µs 39.8 µs -0.3% ✅
create_deploy/rex4/create_10 29.2 µs 29.2 µs +0.2% ✅
delegatecall_system_contract/rex4/delegatecall_not_intercepted 6.0 µs 6.2 µs +2.4% ✅
delegatecall_system_contract/rex4/staticcall_intercepted 6.3 µs 6.4 µs +1.0% ✅
empty_transaction/equivalence 3.5 µs 3.7 µs +4.6% ✅
empty_transaction/mini_rex 3.9 µs 4.0 µs +4.1% ✅
exp_256bit/mini_rex 304.2 µs 303.2 µs -0.3% ✅
gas_detention_computation/equivalence/compute_only_500 45.1 µs 43.9 µs -2.7% ✅
gas_detention_computation/equivalence/volatile_then_compute_500 45.6 µs 44.3 µs -2.9% ✅
gas_detention_computation/mini_rex/compute_only_500 61.1 µs 59.3 µs -2.9% ✅
gas_detention_computation/mini_rex/volatile_then_compute_500 61.2 µs 59.6 µs -2.5% ✅
gas_detention_computation/rex4/compute_only_500 68.1 µs 66.7 µs -2.1% ✅
gas_detention_computation/rex4/volatile_then_compute_500 68.4 µs 66.9 µs -2.2% ✅
keccak256_1KB/mini_rex 372.0 µs 360.9 µs -3.0% ✅
keccak256_32B/mini_rex 62.3 µs 63.8 µs +2.5% ✅
log_opcodes/equivalence/log0_256b 10.8 µs 11.1 µs +3.4% ✅
log_opcodes/equivalence/log0_32b 11.0 µs 10.6 µs -3.8% ✅
log_opcodes/equivalence/log2_32b 16.5 µs 16.1 µs -2.3% ✅
log_opcodes/equivalence/log4_256b 20.5 µs 20.8 µs +1.4% ✅
log_opcodes/equivalence/log4_32b 20.5 µs 20.1 µs -1.9% ✅
log_opcodes/mini_rex/log0_256b 12.8 µs 12.6 µs -1.7% ✅
log_opcodes/mini_rex/log0_32b 12.9 µs 12.7 µs -1.5% ✅
log_opcodes/mini_rex/log2_32b 19.1 µs 18.7 µs -2.0% ✅
log_opcodes/mini_rex/log4_256b 25.2 µs 25.0 µs -0.9% ✅
log_opcodes/mini_rex/log4_32b 25.4 µs 24.9 µs -1.9% ✅
log_opcodes/rex4/log0_256b 13.5 µs 13.4 µs -1.1% ✅
log_opcodes/rex4/log0_32b 13.5 µs 13.3 µs -1.1% ✅
log_opcodes/rex4/log2_32b 20.2 µs 19.9 µs -1.4% ✅
log_opcodes/rex4/log4_256b 26.6 µs 26.7 µs +0.4% ✅
log_opcodes/rex4/log4_32b 26.6 µs 26.4 µs -0.8% ✅
mixed_workload/equivalence 18.0 µs 17.5 µs -2.9% ✅
mixed_workload/mini_rex 27.8 µs 26.5 µs -4.6% ✅
mixed_workload/rex4 28.0 µs 27.3 µs -2.6% ✅
oracle_sload/equivalence/oracle_sload_50 16.3 µs 16.4 µs +0.5% ✅
oracle_sload/equivalence/regular_sload_50 16.2 µs 15.8 µs -2.8% ✅
oracle_sload/mini_rex/oracle_sload_50 17.8 µs 17.9 µs +0.8% ✅
oracle_sload/mini_rex/regular_sload_50 17.6 µs 17.4 µs -1.2% ✅
oracle_sload/rex4/oracle_sload_50 19.1 µs 18.0 µs -5.8% ✅
oracle_sload/rex4/regular_sload_50 18.6 µs 18.6 µs +0.1% ✅
precompile_blake2f_1round/mini_rex 114.1 µs 113.9 µs -0.2% ✅
precompile_bls12_381_g1add/mini_rex 608.6 µs 610.6 µs +0.3% ✅
precompile_bls12_381_g1msm/mini_rex 22468.9 µs 22815.5 µs +1.5% ✅
precompile_bls12_381_g2add/mini_rex 855.9 µs 855.9 µs -0.0% ✅
precompile_bls12_381_g2msm/mini_rex 43171.0 µs 44276.8 µs +2.6% ✅
precompile_bls12_381_map_fp2_to_g2/mini_rex 19769.6 µs 19790.8 µs +0.1% ✅
precompile_bls12_381_map_fp_to_g1/mini_rex 6169.2 µs 6170.9 µs +0.0% ✅
precompile_bls12_381_pairing/mini_rex 137216.9 µs 136893.3 µs -0.2% ✅
precompile_ecadd/mini_rex 530.0 µs 532.2 µs +0.4% ✅
precompile_ecmul/mini_rex 540.6 µs 540.8 µs +0.0% ✅
precompile_ecpairing/mini_rex 133136.0 µs 130689.9 µs -1.8% ✅
precompile_ecrecover/mini_rex 5200.5 µs 5190.7 µs -0.2% ✅
precompile_kzg_point_evaluation/mini_rex 160645.1 µs 160769.2 µs +0.1% ✅
precompile_modexp_32B/mini_rex 3751.4 µs 3718.7 µs -0.9% ✅
precompile_ripemd160_1KB/mini_rex 546.8 µs 548.9 µs +0.4% ✅
precompile_ripemd160_32B/mini_rex 139.5 µs 141.9 µs +1.7% ✅
precompile_sha256_1KB/mini_rex 194.0 µs 195.8 µs +0.9% ✅
precompile_sha256_32B/mini_rex 116.9 µs 117.9 µs +0.8% ✅
rex5_pre_block/bootstrap 19.3 µs 19.7 µs +2.0% ✅
rex5_pre_block/no_change 12.1 µs 12.7 µs +4.6% ✅
selfdestruct/equivalence 4.9 µs 4.9 µs +0.5% ✅
selfdestruct/rex2 5.2 µs 5.2 µs -0.2% ✅
selfdestruct/rex4 5.3 µs 5.3 µs +0.4% ✅
simple_ether_transfer/equivalence 3.7 µs 3.9 µs +4.6% ✅
simple_ether_transfer/mini_rex 4.0 µs 4.3 µs +5.2% ✅
snailtracer/equivalence 53882.4 µs 52893.5 µs -1.8% ✅
snailtracer/mini_rex 130587.6 µs 124130.6 µs -4.9% ✅
snailtracer/rex4 159593.9 µs 167987.6 µs +5.3% ✅
sstore_heavy/equivalence/sload_100 27.3 µs 25.8 µs -5.7% ✅
sstore_heavy/equivalence/sstore_100 38.3 µs 34.8 µs -9.0% ✅
sstore_heavy/equivalence/sstore_sload_100 49.3 µs 46.2 µs -6.4% ✅
sstore_heavy/mini_rex/sload_100 29.5 µs 28.4 µs -4.1% ✅
sstore_heavy/mini_rex/sstore_100 74.4 µs 70.0 µs -5.8% ✅
sstore_heavy/mini_rex/sstore_sload_100 89.9 µs 83.1 µs -7.5% ✅
sstore_heavy/rex4/sload_100 31.1 µs 29.5 µs -5.1% ✅
sstore_heavy/rex4/sstore_100 70.5 µs 68.8 µs -2.4% ✅
sstore_heavy/rex4/sstore_sload_100 84.5 µs 83.1 µs -1.6% ✅
subcall_1000_nested/equivalence 1222.1 µs 1265.2 µs +3.5% ✅
subcall_1000_nested/mini_rex 1753.4 µs 1831.0 µs +4.4% ✅
subcall_1000_nested/rex4 1926.6 µs 2005.4 µs +4.1% ✅
subcall_1000_no_value/equivalence 611.5 µs 654.8 µs +7.1% ✅
subcall_1000_no_value/mini_rex 900.1 µs 981.3 µs +9.0% ✅
subcall_1000_no_value/rex4 987.8 µs 1048.3 µs +6.1% ✅
subcall_1000_transfer_1wei/equivalence 436.3 µs 447.6 µs +2.6% ✅
subcall_1000_transfer_1wei/mini_rex 776.5 µs 755.4 µs -2.7% ✅
subcall_1000_transfer_1wei/rex4 840.5 µs 851.9 µs +1.4% ✅
system_contract_100x/rex4/access_control 109.4 µs 111.0 µs +1.4% ✅
system_contract_100x/rex4/limit_control 103.7 µs 102.9 µs -0.8% ✅
system_contract_100x/rex4/regular_contract 119.6 µs 115.6 µs -3.4% ✅
system_contract_single/rex4/access_control 6.4 µs 6.3 µs -0.8% ✅
system_contract_single/rex4/limit_control 6.3 µs 6.3 µs -0.3% ✅
system_contract_single/rex4/regular_contract 7.1 µs 7.1 µs +0.7% ✅
transfer_multi_1000/equivalence 1852.8 µs 1845.1 µs -0.4% ✅
transfer_multi_1000/mini_rex 2100.4 µs 2075.2 µs -1.2% ✅
transfer_multi_1000/rex4 2119.2 µs 2070.1 µs -2.3% ✅
volatile_data/equivalence/baseline_add 12.6 µs 12.2 µs -3.0% ✅
volatile_data/equivalence/coinbase 6.0 µs 6.0 µs +0.2% ✅
volatile_data/equivalence/number 6.0 µs 6.0 µs -0.5% ✅
volatile_data/equivalence/timestamp 6.0 µs 6.0 µs -0.4% ✅
volatile_data/mini_rex/baseline_add 16.1 µs 15.8 µs -2.3% ✅
volatile_data/mini_rex/coinbase 8.2 µs 8.3 µs +1.2% ✅
volatile_data/mini_rex/number 8.0 µs 8.2 µs +2.0% ✅
volatile_data/mini_rex/timestamp 8.0 µs 8.1 µs +1.4% ✅
volatile_data/rex4/baseline_add 17.5 µs 17.2 µs -2.1% ✅
volatile_data/rex4/coinbase 8.9 µs 9.1 µs +1.5% ✅
volatile_data/rex4/number 8.9 µs 8.9 µs -0.0% ✅
volatile_data/rex4/timestamp 9.3 µs 8.9 µs -3.7% ✅
weth9_transfer/equivalence 20.1 µs 19.4 µs -3.5% ✅
weth9_transfer/mini_rex 23.5 µs 23.1 µs -1.5% ✅

Significant changes (>5%):

Benchmark Baseline Feature Delta
call_value_empty_account/equivalence/empty_account_50 37.7 µs 34.6 µs -8.2% ✅
call_value_empty_account/equivalence/existing_account_50 38.1 µs 34.6 µs -9.0% ✅
call_value_empty_account/mini_rex/empty_account_50 52.5 µs 47.9 µs -8.8% ✅
call_value_empty_account/mini_rex/existing_account_50 51.8 µs 47.4 µs -8.5% ✅
call_value_empty_account/rex4/empty_account_50 55.1 µs 50.3 µs -8.9% ✅
call_value_empty_account/rex4/existing_account_50 54.3 µs 50.1 µs -7.6% ✅
oracle_sload/rex4/oracle_sload_50 19.1 µs 18.0 µs -5.8% ✅
simple_ether_transfer/mini_rex 4.0 µs 4.3 µs +5.2% ✅
snailtracer/rex4 159593.9 µs 167987.6 µs +5.3% ✅
sstore_heavy/equivalence/sload_100 27.3 µs 25.8 µs -5.7% ✅
sstore_heavy/equivalence/sstore_100 38.3 µs 34.8 µs -9.0% ✅
sstore_heavy/equivalence/sstore_sload_100 49.3 µs 46.2 µs -6.4% ✅
sstore_heavy/mini_rex/sstore_100 74.4 µs 70.0 µs -5.8% ✅
sstore_heavy/mini_rex/sstore_sload_100 89.9 µs 83.1 µs -7.5% ✅
sstore_heavy/rex4/sload_100 31.1 µs 29.5 µs -5.1% ✅
subcall_1000_no_value/equivalence 611.5 µs 654.8 µs +7.1% ✅
subcall_1000_no_value/mini_rex 900.1 µs 981.3 µs +9.0% ✅
subcall_1000_no_value/rex4 987.8 µs 1048.3 µs +6.1% ✅

147 benchmarks: 0 regressions, 0 warnings, 0 improvements

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@claude claude[bot] claude[bot] left review comments

@Troublor Troublor Troublor requested changes

@flyq flyq Awaiting requested review from flyq flyq is a code owner

@vincent-k2026 vincent-k2026 Awaiting requested review from vincent-k2026 vincent-k2026 is a code owner

Requested changes must be addressed to merge this pull request.

Assignees

No one assigned

Labels

api:unchanged No change to the public interface or API comp:core Changes to the `mega-evm` core crate dependencies Pull requests that update a dependency file rust Pull requests that update rust code spec:unchanged No change to any `mega-evm`'s behavior

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@RealiCZ @Troublor