Optimize transform of vector<bool> for more predicates

[AlexGuteniev]

Towards #625, specifically #625 (comment) item 3

Follow up to #5769

➡️ Optimization

@rodiers asked for set subtraction.

3. A substract like method
   /// Subtract the sets of true flags in both CFlags.
   /// S1 = {i where *this[i] == true}, S2 = {i where ac_Other[i] == true}, Result[i] == true <=> i element of S with S=S1\S2

Out of the standard predicates std::greater does the trick: it is only true if the element is in first set and not in second set. Or std::less if converse. For completeness, let's add all four remaining comparisons.

❓ What it even is

Apparently, there's no universally accepted name for the corresponding integer operation, unlike xnor for equal_to.

In the benchmark, we can keep comparison names, like less_equal. But we also need to name functors. Here are some ideas:

• In Wikipedia, these operations defined as implication and nonimplication, and the gates are imply and nimply. The other two, when the other input is negated, are converse implication and converse nonimplication,
• In our favorites ISA there are instructions to do one of these operation: andn (scalar, BMI1), pandn (MMX, SSE2), and more flavors of pandn for bigger vectors and element masks, all accessible as intrinsics, and generated automatically by the compiler. We can call these ops andn and orn therefore. With slight confusion with nand / nor that we don't use anyway, and with no obvious way how to mark one vs the other inputs as negated.
• A historian in a Stack Overflow comment mentions such names as "selective clear" and "selective set". Still without obvious way to define which arg is inverted.

In the PR I went with the Wikipedia naming, but I'm open to any other option

⏱️ Benchmark results

Benchmark	Before	After	Speedup
transform_two_inputs_aligned<less<>>/64	130 ns	3.03 ns	42.9
transform_two_inputs_aligned<less<>>/4096	11462 ns	10.8 ns	1060
transform_two_inputs_aligned<less<>>/65536	264332 ns	141 ns	1870

🚗 Drive-by

There's a concise way to test for void, done that.

[StephanTLavavej]

The 1870x speedup exceeds all but the literally exponential one 😹 😻

Thanks for the thorough test coverage and I agree with the names!

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo. Please notify me if any further changes are pushed, otherwise no action is required.

[StephanTLavavej]

👻 🧛 🦇