Replace IngredientCache lock with atomic primitive

[Veykril]

No description provided.

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	6181ddf
🔍 Latest deploy log	https://app.netlify.com/sites/salsa-rs/deploys/67b81ca0b712aa0008cb436f

[Veykril]

I believe both of these checks (here and for MemoIngredientIndex) were accidentally wrong

[codspeed-hq]

CodSpeed Performance Report

Merging #687 will not alter performance

_{Comparing Veykril:veykril/push-pwqsvzurlkun (6181ddf) with master (198db98)}

Summary

✅ 9 untouched benchmarks

[ibraheemdev]

Release is sufficient here, there's no need for Acquire if the current thread wins the race (though of course both are correct).

[davidbarsky]

Suggested change

	// This is a packed representation of `Option<(Nonce<StorageNonce>, IngredientIndex)>`
	// allowing us to avoid a lock in favor of an atomic load. This works thanks to `Nonce`
	// having a niche and so the entire type can fit into a u64.
	// A packed representation of `Option<(Nonce<StorageNonce>, IngredientIndex)>`.
	//
	// This allows us to replace a lock in favor of an atomic load. This works thanks to `Nonce`
	// having a niche, which means so the entire type can fit into an `AtomicU64`.

[davidbarsky]

i think this looks good to me? the only note that i'd have is "it'd be nice to have a Loom/shuttle test for IngredientCache, but I think I committed to that...

[Veykril]

Ye I'd push the test onto the rest of the Loom work. I don't think without Loom writing a test is too useful here right now Though we can also delay this until we have loom integrated, no preference from me.

[davidbarsky]

I think we can/should merge this, but codespeed is reporting 1–2% regressions. do you happen to why they're showing up?

[Veykril]

I don't know why, I'd expect this to not regress it. Maybe its just within noise

[MichaReiser]

1-2% is something I consider noise.