[PATCH] D112757: [llvm-reduce] optimize extractFromModule functions

[Dwight Guth via Phabricator via llvm-commits]

dwightguth added a comment.

I tried changing it to an unordered set without any of the other changes originally, but it still had the same problem of a lot of time spent inserting into the hash table. I eventually abandoned that approach because my intuition was that a hash table where the keys are addresses in memory would probably end up poorly balanced a lot of the time.

If you want I can try out the combination of pushing first into a vector and also creating a hash set from the vector, but I didn't go that route because I had already largely addressed the performance bottleneck in this function with the diff as you see it now, and it didn't seem worth the risk that it might be badly balanced in some cases when the majority of the execution time was now taking place elsewhere.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112757/new/

https://reviews.llvm.org/D112757