[polly] r214200 - IslAst: Enhance parallelism detection test

[Johannes Doerfert]

>> If you believe there are reasons why we would like to change this, feel
free to open a discussion.
Didn't I do that ;)? 

For me printing the Ast and parallelizing are two completely different
things. Here a few points I have in mind when thinking about those two
things:

Parallel codegen: 
     - Different parallel backends could:
         - Not handle certain kinds of loops, thus annotations of more
innermost loops would be helpful (this actually becomes an issue with
privatization).
         - Be able to parallelize multiple loops (OpenMP basically is). 
     - Even at the moment there is the implicit rule that #simd will win
over #omp if both are present, thus #omp doesn't always mean OpenMP
parallelized code.
     - An annotation backend (we actually have one) could annotate memory
accesses parallel in different loops (regarding also the first point here)
The printed Ast:
     - Is feedback for programmers and developers.
     - Annotating all loops would sometimes be helpful to understand e.g.,
scheduling decisions


--

Johannes Doerfert
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by
The Linux Foundation