Thanks! 😸 I pushed significant changes - please meow if you have concerns.

nah, lgtm.
However:

I don't really like pulling the allocators out into a template parameter, as I don't think it's worth multiplying the test matrix for them (although having some benchmarks use one and some the other probably was bad).

I don't really like pulling the allocators out into a template parameter, as I don't think it's worth multiplying the test matrix for them (although having some benchmarks use one and some the other probably was bad).

If there's little difference in performance between allocators, I'd just use the worse one (not_highly_aligned_allocator), to make AVX2 or AVX-512 vectorization, if they exist here, of less advantage.

Sounds good to me since I saw no perf difference. I'll validate and push.

lgtm

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

Thanks for filling out our benchmark directory! ⏱️ 📈 😹