Looks like we're not linking up all of the slow cases in the baseline JIT. put_to_base is the culprit due to some weird mismatch in the switch statement logic of the normal case and the slow case.
<rdar://problem/13445011>
Created attachment 198752 [details] Patch
Comment on attachment 198752 [details] Patch Patch looks good, but it needs a regression test.
Created attachment 198776 [details] Patch
Comment on attachment 198776 [details] Patch r=me
Committed r148711: <http://trac.webkit.org/changeset/148711>