I don't mean the cmov code, I mean the normal naive testl + sete code. The CPU can do all testl's in parallel and all sete's in parallel. I think this should be

I don't mean the cmov code, I mean the normal naive testl + sete code. The CPU can do all testl's in parallel and all sete's in parallel. I think this should be faster than branches if you ever have a 0. I'm assuming your branch code looks something like: if (a == 0) { return true; } if (b == 0) { return true; } // ... return false;