The following gives me a nasty bug. The l.sw instruction is not
assembled correctly - the immediate value is not split up into 5 bits
and 11 bits as the or1k spec says, but is treated as every other
instruction with a 16-bit immediate value. Can someone verify this bug
with their tools?

1) as -o bad.o bad.s
2) objdump --source bad.o
3) compare the output of objdump to the original source / compare the
instruction coding from as to the expected encoding

------------------------------ file bad.s
---------------------------------------
.text
.org 0x88880
_test:
l.movhi r3,hi(_x)
l.sw lo(_x)(r3),r0
_x:
.long 0
------------------------------ file bad.s
---------------------------------------

Best regards,
Jared Momose