X86_64: Replace x86_64 xchg instruction use

Replacing 'xchg' to exchange two registers with a three instruction move
sequence using the 'TMP' register r10 seems to be a big win.  This is
because xchg is a serializing instruction, even when used on registers.

Change-Id: I1c0f7687630936e7f3d2efc4b30ad11233bd484c
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>