[polly] r212080 - [FIX] Don't consider reductions which are partially outside the SCoP

[Tobias Grosser]

On 02/07/2014 20:46, Johannes Doerfert wrote:
> Just out of curiosity:
>    Is there a plan to remove independent blocks in the future?
>    I was under the impression we will hold on to that part and not migrate to
> SSA-variable dependences.

Yes, I would like to make ScopInfo produce results without any 
modifications to the IR. This is in the light of enabling Polly by 
default where we only want to modify the IR if we actually performed 
some useful changes both to not accidentally introduce performance 
regressions in the generated code and also to not invalidate any 
analysis passes. -polly-codegen-scev is a step into this direction.

The goal is not per-SSA-variable dependences, but to do the very same 
thing as today, but to only introduce the one-element arrays 
conceptually but only materialize them at AST generation time.

Cheers,
Tobias