[polly] r264118 - [ScopInfo] Fix domains after loops.

[Tobias Grosser via llvm-commits]

On 04/04/2016 04:09 PM, Johannes Doerfert wrote:
> If this listing of events is "accurate" I am not convinced that the problem was
> actually the domain. Instead I think it might have been in the code generation
> that was triggered before but not after your patch.  While code generation
> should not change the semantics as the RTCs should be false anyway, it is
> possible that this happend. A bug in the RTCs or any other code generation
> part could cause problems even if the domains are fine and it would be hidden
> if we can now statically drop the SCoPs.

Right.

In fact, it might make sense to not bail out early in case 
-process-unprofitable is set. This will give us more test coverage for 
the code generator and will also expose the bug Michael has seen 
independently of which domain we generate. I had some patches that made 
the code generator robust against such issues. These will not "improve" 
the modeling, but will prevent the crash.

Best,
Tobias