Use movapd instead of movsd to move between xmm registers. Avoids partial
register writes and potential stalls.

Use xorps instead of xorpd to zero a register. Same effect and latency, but
one byte smaller.

Review URL: http://codereview.chromium.org/6873006

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7631 ce2b1a6d-e550-0410-aec6-3dcde31c8c00