This takes some register pressure away, and makes the FTL generate code that is performance-competitive against the DFG on crypto's am3().
Created attachment 219266 [details] the patch
Landed on the branch in http://trac.webkit.org/changeset/160610
Landed in trunk in http://trac.webkit.org/changeset/160613
Review status updated in r161938: <http://trac.webkit.org/r161938>.