What performance penalty does overflow checking have in release?

What performance penalty does overflow checking have in release?

Presumably some, since it's not enabled by default in release 😅 I didn't measure.

retrigger-benchmark

⏱️ Benchmark run finished in 1m 58s
🧠 Peak memory usage: 1611 MB

📈 Compared against baseline: 124s
🧮 Delta: -6s
🔢 Commit: 047e3a0

✅ This PR improves generation time.

You can retrigger the benchmark by commenting retrigger-benchmark.

Btw, I wouldn't necessarily trust the benchmark results - AISI, it's measuring how long the implicit cargo build for a cargo run takes?

It'd be better to cargo build --release --no-default-features first, and then benchmark a couple times with e.g. hyperfine to make things more sound statistically.

I can PR that if there's interest. See #457

retrigger-benchmark

Thanks a lot for the contribution, looks good to me! :)

What performance penalty does overflow checking have in release?

I did a few local benchmarks, doesn't seem too worse to me. But I definitely want to check that again in more detail before the next release.

Btw, I wouldn't necessarily trust the benchmark results - AISI, it's measuring how long the implicit cargo build for a cargo run takes?

It'd be better to cargo build --release --no-default-features first, and then benchmark a couple times with e.g. hyperfine to make things more sound statistically.

I can PR that if there's interest. See #457

You're definitely right with that, thanks for already opening a PR for that. Will look into that asap!