Optimize std::count even more for vector<👻>

[Alex Guteniev (AlexGuteniev)]

Continuation of #2201

Benchmark (compare run with and without changes):

// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

#include <chrono>
#include <limits>
#include <iostream>
#include <vector>

volatile int result;

int main(int argc, char** argv)
{
    const int N = 20000;
    std::vector<bool> boo(1000000);

    std::chrono::steady_clock::duration d{};
    std::chrono::steady_clock::time_point t;

    t = std::chrono::steady_clock::now();
    for (int i = 0; i < N; ++i) {
        result = std::count(boo.begin(), boo.end(), true);
    }
    d = std::chrono::steady_clock::now() - t;

    std::cout << "t\t" << std::chrono::duration_cast<std::chrono::duration<double>>(d) << "\n";
}

[Stephan T. Lavavej (StephanTLavavej)]

Moving back to WIP until #2201 is merged, to avoid any confusion for maintainers. (This is just so we'll be able to review a simpler diff; no concerns with the actual changes at this time.)

[Casey Carter (CaseyCarter)]

Also from our weekly meeting: vector<👻> is hilarious and we're keeping the nickname.

[Charlie Barto (barcharcraz)]

These look good, still a bit frustrated the optimizer can't figure this out.

[Alex Guteniev (AlexGuteniev)]

still a bit frustrated the optimizer can't figure this out

I was able to convey how to reproduce the optimizer bug, from DevCom-1527995 :

We can reproduce your issue. We’ve filed a bug for this issue on the C++ team here.

It is in Under investigation status now. So maybe it will be fixed, and we may avoid some of the tricks.

[Charlie Barto (barcharcraz)]

yeah, I wish the compiler was better at hoisting the checks out of the loops by itself, oh well.

[Stephan T. Lavavej (StephanTLavavej)]

Looks good, I'll push a small comment change!

[Stephan T. Lavavej (StephanTLavavej)]

I'm mirroring this to an MSVC-internal PR; please notify me if any further changes are pushed.

[Stephan T. Lavavej (StephanTLavavej)]

Thanks for this haunted PR! 👻 🦇 😸