Loop optimizer issues

[Richard Kenner]

    It is not somehow terribly difficult (a hundred or two of lines of code,
    I guess)

That's *way* large: try a dozen, at most!

    but IMO it would be nice to get rid of this feature with tree-ssa
    completely.

I just don't see why it's helpful to force that assumption on an RTL
loop optimizer: namely that any invariant LIBCALL will already have
been moved out.  Perhaps it will, but perhaps it won't.  Isn't it
simpler just to not worry about that issue and add the few extra lines?