[patch] for PR 18040

[Richard Kenner]

    But this problem goes away if you start manipulating all the
    aggregates by reference.  

See my last mesasge on that topic.

    Sure, but this is decidedly _not_ the way the tree optimizers want to
    think about it.  Where RTL requires something to be done a certain
    way, we should not make that into a design constraint on trees; at
    most, there might be an impedance mismatch to be resolved in the
    expander.

Sure, but the point isn't RTL-specific.  It relates to code
generation.  When generating code, you only have to do *one* operation
no matter how many dereferences there are: all the extra defererences
do is change the parameters (offsets in this case) of that operation.
That's why I like to view this nest as conceptually one operation.

    Hmm.  That might turn out to be equivalent to my handwavey suggestion
    of inventing some way to make unaddressable things addressable for
    tree purposes.  Do you have a pointer to that proposal?

I don't think so, but the proposal is at
	http://gcc.gnu.org/ml/gcc/2004-06/msg01376.html