Another DCE problem with array addresses

[Richard Kenner]

    Something has definitely changed upstream from the optimizers.  I
    don't get anything anywhere close to this IL.  Could you give me
    another test case?

I think RTH may have already fixed this, but I'm really confused right
now as to what is what and I've been away for a few days, which didn't
help.  Tomorrow I'll update my sources and start from the beginning to
see what works and what doesn't, though I have two patches from RTH in
my tree at the moment that were aimed at things related to this.