[LLVMdev] unwinds to in the CFG

[Duncan Sands]

> In LLVM the rule is that an instruction must be dominated by its 
> operands.

...

> We already have this issue with invoke. Consider:
>
> bb1:
>    %x = invoke i32 @f() to label %normal unwind label %unwind
> normal:
>    phi i32 [%x, %bb1]
>    ret i32 %x
> unwind:
>    phi i32 [%x, %bb1]  ; illegal
>    ret i32 %x
>
> The PHI in %unwind must mention %bb1, but may not refer to %x.

In some sense this PHI is not dominated by its operands.  There
is an edge to %unwind coming out of the invoke, but it is coming
out of the RHS, i.e. before the assignment to %x.  If I wrote
the invoke as:
  invoke i32 @f() <= this may branch to %unwind
  place the result in %x
then it is clear that the phi is dominated by the call part of
the invoke but not the result copying part.  Likewise for something
like

bb1: unwinds to %cleanup
   %x = udiv i32 @foo, @bar <= The udiv is always executed before %cleanup, but not the assignment to %x
   ret i32 %x
cleanup:
   %y = phi i32 [%x, bb1]

So maybe it is helpful to think of instructions as corresponding to
a LHS and a RHS node in the cfg, with an edge from RHS to LHS, and
with exceptional edges coming out of the RHS node.

Ciao,

Duncan.