Attendees: Razya, Li, Ramakrishna, Sebastian, David, Christophe
* Sebastian:
- Merged the reductions patch in the graphite branch
- several testcases using reduction patterns in the autopar and autovect passes do not pass now
* Li:
- prepared a conservative alias analysis, integrated in graphite
- several ideas of how to improve the alias analysis, but none of them are elegant:
- use a graph to structure alias, not elegant as not passing the info in polyhedral form, good point is that it is not NP-complete
- use alias numbers, pass in polyhedral, costs a lot as it has to solve the clique problem that is NP-complete
- quadratic approach, pass in polyhedral, pass a graph, the base of the data references represents connections in the graph
* Ramakrishna:
- reduce from the vectorizer testsuite a testcase that shows fails due to the insertion of casts by graphite