Optimize all_of / any_of / none_of for vector<bool> with some predicates

[AlexGuteniev]

Towards #625, specifically #625 (comment) items 5, 6, 7

Follow up to #5769

⚙️ Optimization

Similarly to transform, we can map bool functor to underlying integer vector functor. Although we have to do the operation on integer of bits after this.

identity looking like ideal functor to use, as bools are already boolean, and we just need to apply the algorithm. We can also support logical_not, which we have already mapped, although it is obscure to use.

Sure identity is C++20, but that's fine to optimize only new standard modes.

As the algorithm uses one range, we can handle any offset without introducing too much complexity. Some complexity is still introduced by this though.

As the resulting flow is not very simple, but is the same for all three, the traits approach from the vector algorithms seems suitable.

Vectorization

Auto vectorization does not work here. @Alcaro suggested that from abstract machine view point the memory load would be invented if they happen past the early return point.

We can vectorize manually. It will be inefficient if early return happens really early, but may be a good idea overall. Not this time though.

⏱️ Benchmark results

Benchmark	Before	After	Speedup
meow_of<alg::any_, content::zeros_then_ones>/64	26.8 ns	1.42 ns	18.9
meow_of<alg::any_, content::zeros_then_ones>/4096	1767 ns	26.3 ns	67.2
meow_of<alg::any_, content::zeros_then_ones>/65536	25751 ns	253 ns	102
meow_of<alg::any_, content::ones_then_zeros, not_>/64	27.0 ns	1.40 ns	19.3
meow_of<alg::any_, content::ones_then_zeros, not_>/4096	1859 ns	32.7 ns	56.9
meow_of<alg::any_, content::ones_then_zeros, not_>/65536	27850 ns	257 ns	108
meow_of<alg::all_, content::ones_then_zeros>/64	33.4 ns	1.39 ns	24
meow_of<alg::all_, content::ones_then_zeros>/4096	1919 ns	27.0 ns	71
meow_of<alg::all_, content::ones_then_zeros>/65536	30449 ns	251 ns	121
meow_of<alg::none_, content::zeros_then_ones>/64	28.7 ns	1.45 ns	19.8
meow_of<alg::none_, content::zeros_then_ones>/4096	1916 ns	25.9 ns	74
meow_of<alg::none_, content::zeros_then_ones>/65536	29903 ns	248 ns	121

[AlexGuteniev]

This is not currently vectorized, so not_highly_aligned_allocator currently serves no purpose, as not-overaligning only matters for vectorization, but:

• It can be vectorized manually in the future
• gcc with -O3 would vectorize this, so the target compilers someday also might