ssa restore

[Stephen Weeks]

> How "broken" an ssa function should the ssa restore pass handle?
... 
> The knownCase pass will probably introduce multiple let and arg bindings
> of a variable, every path through the CFG to a use of a variable will pass
> through a binding (but, no single binding dominates all uses and other
> bindings).  Here's the transformation I have in mind:
...
> There are now three bindings of x, two let bindings and one arg binding.

Makes sense.

> What other kinds of optimizations do we think will be sent at the ssa
> restore?  And how will they break the ssa invariants?

One other optimization that breaks SSA is loop reordering (to move the
test to the end of a loop so that the loop only has one jump).  I
can't think of any others offhand.

How about this as a required property of the input to SSA restore?

  Every path from the root to a use of a variable (excluding globals)
  passes through a def of that variable.