[PATCH] D32538: [Polly][NewPM] Port ScopInfo to the new PassManager

[Tobias Grosser via Phabricator via llvm-commits]

grosser added a comment.

Hi Philip,

this looks technically correct, assuming we want to go this route. As noted earlier, there is a risk that we may invalidate some of the analysis results when code generating, hence I have some concerns here. However, as already pointed out in earlier discussions on this topic, as this code path is independent of all our existing functionality this might indeed be a great opportunity to explore a different approach of scheduling passes without risking to regress existing functionality. Assuming we can add a test case for this, I would be glad to see this patch upstream!


Repository:
  rL LLVM

https://reviews.llvm.org/D32538