All integral arguments are passed as eight-byte values.
Floating-point arguments are passed in floating-point registers when possible.
