[libcxx-commits] [libcxx] [libc++] Drop the unrepresentative search_n benchmark (PR #184783)

[Nikolas Klauser via libcxx-commits]

philnik777 wrote:

> Wouldn't it make more sense to instead make it so that the compiler can't constant-fold the algorithm?
> 
> Or do you mean that the benchmark is always pointless (for libc++) because we jump forward by the needle size and so it's very fast?

This is indeed what I mean.

> If that's what you mean, I still think this benchmark makes sense: it's not because our implementation can do that efficiently right now that we should get rid of the coverage for it. I kinda view these benchmarks as regression tests for our performance.

I don't want to get rid of it because we do it efficiently now, but because the other benchmark would still catch it if we regressed the skipping again while being a bit more representative.

https://github.com/llvm/llvm-project/pull/184783