SIMPLE: A language-independent tree IR

[Joe Buck]

I wrote:
> Since we have the as-if rule, that's not quite the answer; we can only
> say that the visible side effects (accesses to volatile variables,
> calls to other functions that may have side effects) cannot be
> interleaved.  If f, g, and h are inlined, we could have considerable
> freedom to do reordering despite the C99 rules.

I don't bring this up just to nitpick.  If we're going to have a common
low-level tree representation, we need some way to represent which
reorderings are permitted and which aren't, in a language-independent way.