#include <iostream>
#include <chrono>
#include <vector>
#include <intrin.h>

using namespace std;
using namespace chrono;

int main()
{ constexpr size_t
ROUNDS = 10'000,
BITS = 1'000'000;
auto tFn = []( char const *prefix, auto fn )
{
auto start = high_resolution_clock::now();
fn();
cout << prefix << duration_cast<nanoseconds>(
high_resolution_clock::now() - start ).count() / 1.0e9 << endl;
};
tFn( "BTS: ", [&]()
{
vector<char> vc( (BITS + 7) / 8, 0 );
long *data = (long *)vc.data();
for( size_t r = ROUNDS; r--; )
for( size_t i = 0; i < BITS; i += 7 )
_bittestandset( data, (long)i );
} );
tFn( "vector<bool>: ", [&]()
{
vector<bool> vb( BITS );
for( size_t r = ROUNDS; r--; )
for( size_t i = 0; i < BITS; i += 7 )
vb[i] = true;
} );

}

I'm testing a bool-vector against a vector<char> and x86's BTS
instruction through the MSVC-intinsic. Which code is faster ?