This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-profiling] fix some side corners of finalization machinery

I will apply your patch and remove my tests for the three (now two) bits, and rerun everything. If I get the orphaned node, I will send you the patch and you can investigate further. I do remember that the node had no callers and no callees.


Jan Hubicka wrote:


I think that I (actually you) found the problem. I added the code to copy my info that you suggested to the clone routine.
The problem went away. It seems that the code that I had in the remove_cgraph_node was only triggered if the master_clone had been set.

I also added code to set the master_clone to cgraph_node and to clone_cgraph_node.

Yan I will send you this patch when I am finished testing it. I also needed to add code so that I only look at nodes that have been either finalized, external or externally_visible since I hit an orphan one of the in the java libraries.

externally_visible nodes should be all finalized or external, so you only need the first two classes I would say. I also wonder how the orphan node look like? It might be fixed by the patch I sent (I also noticed little inconsistency where instead of "finalized" test I had DECL_SAVED_TREE test that is not equivalent for C++/Java)


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]