[BoppreH]

You can also use math to avoid most of the jumps:

    int run_switches(char *input) {
      int res = 0;
      while (true) {
        char c = *input++;
        if (c == '\0') return res;
        // Here's the trick:
        res += (c == 's') - (c == 'p');
      }
    }

This gives a 3.7x speed compared to loop-1.c. The lower line count is also nice.

[svachalek]

Nice. The way I read the cmove version, it's more or less this except the trick line goes

    res += (c == 's') ? 1 : (c == 'p') ? -1 : 0

I haven't done C in decades so I don't trust myself to performance test this but I'm curious how it compares. Pretty disappointed that TFA didn't go back and try that in C.

[414owen]

So I actually did try that, but and IIRC it didn't produce a CMOV with either gcc or clang. I didn't put it in the repo because it wasn't an improvement (on my machine) and I decided not to write about it.

Maybe you get different results though?