On i386 and amd64 on essentially all operating systems there is some kind of thread description structure that is accessible through one of segment registers which contains low-level thread context, some way to get to thread ID and user-accessible space for thread local variables (or pointer to such space). Getting some kind of thread ID is thus only about one "mov dest, [?S: some offset]", which is what both pthread_self() on linux and GetCurrentThreadId() on windows does (note that linux stores pointer to pthread_t in there, windows stores some opaque number that you have to convert to handle yourself if you need handle. The point is that in both cases the function does not do any kind of involved computation as would be the case if it had to traverse list of currently running threads).

This mechanism is essentially the only reason why amd64 in long mode still has limited support (complete enough to implement this, nothing much more) for segment registers.