Will tree-ssa be GCC 3.5?

[Richard Kenner]

    It seems to me that you still want CSE to handle little stuff like
    parts of addresses on machines which need to build up addresses by
    parts (e.g., MIPS), or for that matter parts of constants on machines
    which need to build constants by parts (e.g., ARM).

Actually, in most cases combine does more of that than CSE.  The only place
in CSE that does that is find_best_addr.
    
    Conversely, I would think that tree-ssa might be able to get rid of
    the RTL loop optimization pass.  On the other hand, then how do you
    handle prefetch generation?

It's not just prefetch.  Unless you can be sure that the lowering process
won't generate any invariant, you have to do an RTL-level invariant movement
pass.  Also, giv handling is best done at the RTL level so that it can take
the various possible addressing modes into account (yes, we don't do that as
well as we can: I once went a long way into a major rework of loop.c to do that
better, but it never got completely).