Newlib's memcpy() (MIPS-optimized) uses this for unaligned copy. Consectutive load instructions without nops using same target register (which is legal) like lw $1, 0(mem) lwl $1, 1(mem) or lwl $1, 0(mem) lwr $1, 1(mem) or similar, lead to incorrect result. Now it's fixed. git-svn-id: http://moon:8086/svn/vhdl/trunk@519 cc03376c-175c-47c8-b038-4cd826a8556b