There are single assembly language instructions to test individual bits, and you can test multiple bits at once with immediate operands, although IIRC those are slightly slower than comparing entire integers to zero (boolean). As always, time it both ways (with rdtsc, if possible) and go with the faster version. If the times are so close as to be inconclusive, go with the "easy to understand" version.

Nota bene: If you have a huge amount of items to test, using the bitfields may be faster since they won't blow out the L1 cache.

bool isBoolFasterThanBitwise()
{
    return rand() & 1;
}

(It uses both bitwise and bool!)

I wrote an eratosthenes sieve prime calculator, and I assumed that using bools (actually it was C so I used chars...) would be faster than using a bit vector. At one point I started running out of memory so I switched to using a bit vector and was astounded by how much faster it was.

But as Jonatan says, in any real program the bit vector is unlikely to be the performance bottleneck.

I would bet that boolean is faster, since at a minimum, a bit comes from a byte, and there is the extra bitwise operation to get at the bit you want plus there's the division at a higher level to get which byte you want.

I would use bitfields where you need a lot/have low memory requirements and booleans elsewhere. I doubt it'll make a big difference, except memory wise.

This is what I was saying in my post about prime calculation. If the array is big enough, then the bit vector wins on performance (as well as memory usage) because you get 8 times as much information in the cache.

Ok, you made me do it:

test1 results:

count = 52424828
time = 0.990623 seconds
count = 52432056
time = 0.990727 seconds
count = 52430310
time = 0.991179 seconds

test2 results:

count = 52423770
time = 1.33507 seconds
count = 52422236
time = 1.33386 seconds
count = 52423791
time = 1.33561 seconds

That won't even work, since it never increases byte.

count = 52426098
time = 1.5705 seconds

[edit] Offtopic but interesting just how random rand() is on my system

What about:

for (i = 0; i < NUM; i++) {
    if ((bools[i] & 1) {
        count++;
    }
}

Yes, it will only test one of the possible 8 bits, but in your test case above you're mostly measuring the speed of changing your bit variable and not what we are interested in which is the boolean check. In actual code you would also test against a fixed bit (like 1) so it's the speed we're interested in.

[edit:}

Or alternatively, have another loop where you do just your bit manipulation (but not the check) then subtract that time from the result.

Another test would be, for actual bitfields:

[edit:]

No, it makes no difference which bit you test. But you can use this instead to test all 8, but without the time comsuming code to change a variable:

You couldn't ask for less overhead!

Of course if you write suboptimal code it's going to be slower

Why don't you try this instead:

Test1: 0.34 sec
Test2: 0.145 sec

I don't have much time, maybe I've done some error with all that bitwise operations, so correct me if there's a misstake.

<edit>
Corrected small error: it was i&0x7, no i&0x8. (I want a mask of the last 3 bits).