GCC in-line assembly and the removal of -mcx16

[Toebs Douglass]

Hej, all.

I have been compiling on aarch64.

Here I also use double-word CAS, via __atomic.

As far as I know, all aarch64 systems offer double word CAS (i.e.
128-bit CAS).

However, it seems GCC also here as on x64_86 now requires linking to
libatomic.

I thought this was x86_64 only, and existed there because of the initial
Opertons lacking cmpxchg16b.