http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52480
--- Comment #2 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-08 18:00:25 UTC --- Created attachment 28391 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28391 Proposed patch Hm, the line && INTVAL (operands[3]) == -24 * (BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN) in the "extv" and "extzv" expanders looks wrong. operands[3] is always zero, regardless of the endianess, for the cases in gcc/testsuite/gcc.target/sh/sh4a-bitmovua.c. Changing the check to == 0 seems to be a simple fix for this. Although I didn't dig any further. It might be that there are some more missed unaligned-load opportunities. I've queued up the attached patch for full testing.