I've measured some cranial traits in a monkey species and now I want to test if males and females are statistically differents. To do so, the first step is to test normality, right?

So here comes the real question: Do I have to to perform two separate Shapiro-Wilk tests, one for male and another for females?

The answer seems pretty obvious, you can't unite possibly different things in the same sample. But you know, I need some certainty.