ccastingintegerabsolute-valuebranchless

What does (unsigned) do in this abs() function?

There are many ways to implement abs() with or without branching. I came across this one:

r = (v < 0) ? -(unsigned)v : v;

Does this use branching or is it branchless?

And why can it not just use -v, instead of -(unsigned)v? What does the (unsigned) do here? Why does it give more correct results than -v?

Solution 
  •  r=(v<0)?-(unsigned)v : v

    There are some ways to compute abs without branching but this one is not one of them as < and ?: operators are usually translated to branching instructions.

    Assuming r is of type unsigned int, the cast to unsigned is here to get a specified behavior when v value is INT_MIN. Without the cast, -v would be undefined behavior when v is INT_MIN (as -INT_MIN is undefined behavior in C).