http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52076
--- Comment #1 from Andreas Schwab <[email protected]> 2012-02-01 11:15:57 UTC --- gcc 4.1 and before generate or.b #-128,(%a0) instead of bfins. As a consequence of this they also generate moveq.l #4,%d0 instead of move.b, but never shorten the write.
