A patch for a bad combine bug

[Richard Kenner]

    > Didn't my May 13 change fix this?

    No, it didn't.

I'm surprised you didn't mention it then.

    Have you tried the testcase on x86 yourself? If SUBREG is paradoxical,
    I don't see how you can commute AND with SUBREG since the rest of
    SUBREG may have some garbage, which is shown in the testase.

I think the case prompting the May 13 fix was on 68k.  I didn't try the
case you just sent me and now can't find it anymore.  Can you send it
and your proposed patch again?   I had convinced myself that this can be
done in some paradoxical cases, but I can easily be shown wrong.  Sorry
for losing the case.