Here's an example of using the wrong byte:

movb r9, @>8C02 -> should be copying low byte
mov r9, r2
ori r2, >4000
movb r2, @>8C02

I'm not proud of the fix I made, but here it is:

There is a function named "reload_inner_reg_of_subreg" which allowed a subreg expression which assumes the byte is stored in the low byte of a register. I made a change to that function to force a reload in that case. So now we see code like this:

mov r9, r3
mov r3, r2 <-- Unnecessary MOV
swpb r2
movb r2, @>8C02
mov r9, r2
ori r2, >4000
movb r2, @>8C02

This will work, but there is an unnecessary "mov r3, r2" instruction. What's going on here is that we are reallocating the subreg subject. (In this example, allocate R3 instead of using R9). During the register reallocaction process, we find that we need to do extra work to get the byte value, which causes the "swpb" instruction to be emitted.

I think I can find a better way to fix this.