question about the loop carried dependences routine in machine pipeliner

Hi there!


This is regarding the addLoopCarriedDependences function in the MachinePipeliner:
SwingSchedulerDAG::addLoopCarriedDependences(AliasAnalysis *AA)

What are the thoughts about attempts to refine this step (I am thinking about skipping LCDs when possible, for instance if parallel_accesses metadata is available)?

Please let me know if there are such attempts being worked out currently.