Thanks for the link, @hobovsky!

I'll give it a read :)

Except this is UB: https://stackoverflow.com/a/28783339/1756361

This is the approach I took too, except I moved mine into a macro to keep it a little more readable and re-usable:

#define SWAP(x, y) (x ^= y ^= x ^= y)

While I agree with the idea behind moving the strlen into a variable to avoid it being repeatedly called, I don't agree with having that live in the loop. Having multiple variable declarations in a loop makes the loop look more complicated than it is, without any significant pay-off at compilation.

It reads much better to have the line moved out of the loop entirely, and also given a more descriptive name such as string_length.

Division can be an incredibly expensive operation, so you'll want to avoid doing it as much as possble. To help with this, you can move the result of (int)strlen(str)/2 into a variable and re-use that variable, otherwise the division operation will be performed every iteration.

Also, while we're on the subject of "division by 2," it's a lot more efficient to use bit-shifting whenever you're dividing or multiplying by multiples of two. Division could take up to 30-60 CPU cycles to perform, compared to bit-shifting which only takes a single cycle.

n >> 1 // Divides by 2
n << 1 // Multiplies by 2

This works because you're shifting the bits over into the next adjacent digit place, and since each digit place in binary acts as a multiple of two, it can divide or multiply by two:

1 << 0 // Results in 1 (0001). You'd never really do this, it's just to show that shifting by 0 doesn't perform a shift.
1 << 1 // Results in 2 (0010)
1 << 2 // Results in 4 (0100)
1 << 3 // Results in 8 (1000)

Gaming the system!