That is strange. The only docs I can find suggest this should fit comfortably on a GitHub-hosted runner: https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners/about-github-hosted-runners#standard-github-hosted-runners-for-public-repositories

Note that on Windows we're using a 32bit target as the host. Not sure how Windows manages the address space there, maybe it keeps half of it for the kernel so there's only 2GB total for the program?

So probably it would make sense to decrease the allocation size to 1GB if the pointer_width is 32.

Right. I was trying to avoid doing that because the effect size shrinks, but of course a little loop fixes that.

How long does the benchmark take (with the fix)?

rust-lang/rust#125633 landed, but a ./miri rustc-pull will be needed before we have it available in the repo.

How long does the benchmark take (with the fix)?

200 ms per iteration with the fix. Without the fix, it's 800 ms per iteration. So I think it would be fair to land this without the fix, because it's already one of our faster benchmarks.

Makes sense, thanks!
@bors r+

📌 Commit 88a2496 has been approved by RalfJung

It is now in the queue for this repository.

⌛ Testing commit 88a2496 with merge 952aa4d...

☀️ Test successful - checks-actions
Approved by: RalfJung
Pushing 952aa4d to master...