fix: record actual bench phase durations instead of TimerInfo start instants

[0oyun]

Problem

BaseTester::bench_builder (in halo2-base/src/utils/testing.rs) stores
ark_std TimerInfo values in BenchStats:

pub struct BenchStats {
    pub vk_time: TimerInfo,
    pub pk_time: TimerInfo,
    pub proof_time: TimerInfo,
    pub verify_time: TimerInfo,
    ...
}

A TimerInfo only holds the Instant at which the timer started. The
downstream benches read the timing later like this:

stats.proof_time.time.elapsed(),
stats.verify_time.time.elapsed(),

Because .elapsed() is called on the start instant at report time, each
value measures "phase start → report time", not the duration of the phase
itself. Concretely:

• proof_time.time.elapsed() includes the whole verification step (and any
  work between proving and the CSV write).
• Earlier phases are inflated by everything that ran after them.

So the timings written to the *.csv bench outputs are wrong (overlapping and
inflated).

Affected bench writers: bn254/tests/{msm, fixed_base_msm, ec_add, pairing}.rs
and secp256k1/tests/ecdsa.rs.

Fix

Measure each phase with an explicit Instant and store std::time::Duration
in BenchStats:

let vk_timer = start_timer!(|| "Generating vkey");
let vk_start = Instant::now();
let vk = keygen_vk(&params, &builder).unwrap();
let vk_time = vk_start.elapsed();
end_timer!(vk_timer);

start_timer!/end_timer! are kept purely for their log output; the recorded
value is now the actual operation time. Downstream writers use the fields
directly (stats.proof_time / stats.verify_time), which are still
Duration, so the {:?} formatting is unchanged.

Testing

cargo check -p halo2-base and cargo check -p halo2-ecc --tests both pass.

[Copilot]

Pull request overview

This PR fixes benchmark timing collection in BaseTester::bench_builder by recording per-phase durations instead of storing ark_std::TimerInfo start instants (which caused inflated/overlapping timings when .elapsed() was evaluated later).

Changes:

• Switch BenchStats timing fields from TimerInfo to std::time::Duration, measured via Instant::now() + .elapsed() around each phase.
• Keep start_timer!/end_timer! for log output, but decouple recorded stats from TimerInfo.
• Update downstream bench CSV writers to use stats.proof_time / stats.verify_time directly.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
halo2-base/src/utils/testing.rs	Record true phase durations (vk/pk/prove/verify) into BenchStats as Duration values.
halo2-ecc/src/secp256k1/tests/ecdsa.rs	Write Duration fields directly to CSV instead of calling .time.elapsed() on stored start instants.
halo2-ecc/src/bn254/tests/pairing.rs	Same CSV writer update for proof/verify timing fields.
halo2-ecc/src/bn254/tests/msm.rs	Same CSV writer update for proof/verify timing fields.
halo2-ecc/src/bn254/tests/fixed_base_msm.rs	Same CSV writer update for proof/verify timing fields.
halo2-ecc/src/bn254/tests/ec_add.rs	Same CSV writer update for proof/verify timing fields.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.