[PATCH] [Polly][Refactor][NFC] Allow to nicely bail from ScopInfo.

[Johannes Doerfert]

On 11/01, Tobias Grosser wrote:
> On 01.11.2014 00:32, Johannes Doerfert wrote:
> >Hi grosser, sebpop, dpeixott, simbuerg, zinob,
> >
> >   Each SCoP statement has a flag to indicate if the construction was
> >   successful and if one statement couldn't be constructed the SCoP
> >   is invalidated and removed.
> 
> Hi Johannes,
> 
> I very much like your approach of submitting individual self contained
> patches. In this case, it seems the feature is fully unused.
Part of it is unused, that is true.

> I have a couple
> of comments on bailing out in the ScopStmt, but would prefer to discuss this
> when we have an actual use case where this could
> be needed. Would it be OK to revisit this change if/when we have a case
> where we might want to bail out?
I'd appreciate it if you could at least explain what the problem is
instead of only stating one. I don't want to use my spare time to write
patches that are not committed in the end, therefore I will just give you
one use case right now:

  In the future we will not rely on ScalarEvlolution to compute the loop
  bounds but do it ourselves using the affine conditions in the SCoP.
  However, as we can then allow more regions to be SCoPs (the ones for
  which ScalarEvlolution cannot figure out a trip count) we run into
  the problem that the loop trip count might be infinite (endless loop).
  For those the scheduler will complain and there is no real benefit in
  optimizing them anyway. Thus, we have to bail __after__ we created the
  statement domains using the affine conditions in a SCoP.

> Cheers,
> Tobias

-- 

Johannes Doerfert
Researcher / PhD Student

Compiler Design Lab (Prof. Hack)
Saarland University, Computer Science
Building E1.3, Room 4.26

Tel. +49 (0)681 302-57521 : doerfert at cs.uni-saarland.de
Fax. +49 (0)681 302-3065  : http://www.cdl.uni-saarland.de/people/doerfert
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 213 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141101/5b1f2cf5/attachment.sig>