[PATCH] D20933: Preallocate ExplodedNode hash table

[Ben Craig via cfe-commits]

bcraig added a comment.

I'll try to get those stats.  My suspicion is that it is highly language / projects specific.  Specifically, I suspect that C (and maybe ObjC?) won't hit the analysis limit often, but that C++ will hit the limit frequently because of the large number of inline functions.

I will note that one of the reasons I like using the number of steps (or a number derived from it) is that it scales to the expected worst-case workload.  Shallow analysis only get 75000 work items per top level function, so the exploded graph reservation scales down as appropriate.

I'll make sure the next diff (and future diffs) include more context.


http://reviews.llvm.org/D20933