[tree-ssa] copy propagation and the abstraction penalty

[Joe Buck]

I wrote:
>         this->im = 0.0;
> 
> Now we know that Uva150.im is 0.0.  We do need to care about aliasing, but
> that means that we might have invalidate the value of these fields if we
> get a write through a pointer that might alias UVa150.  This program has
> no pointer writes at all, though.

This may puzzle some folks: isn't "this->im = 0" a pointer write?  It is
not, because we know the exact value of "this", so it is a write to a
known structure field.  Aliasing isn't an issue when we know exactly where
the pointer points.