Builtin popcount, clz, ctz

[konsumlamm]

I'm interested in adding builtins for popcount, count leading zeros (clz) and count trailing zeros (ctz), since these have specialized CPU instructions on many platforms. They would be used in the Bits/FiniteBits instances for Int/Word etc. Looking at the runtime, it seems builtins are either implemented as primops or as FFI functions. Which would be preferable here?

[augustss]

I think I'd do it as primops.

There will have to be C versions of these functions for the platforms that don't have instructions.
You can see how FFS is done.