[PATCH] D113688: [CSSPGO] Fix a hash code truncating issue in ContextTrieNode.

[Wenlei He via Phabricator via llvm-commits]

wenlei added inline comments.


================
Comment at: llvm/lib/Transforms/IPO/SampleContextTracker.cpp:183-185
+  uint64_t NameHash = std::hash<std::string>{}(ChildName.str());
+  uint64_t LocId = (Callsite.LineOffset << 16) | Callsite.Discriminator;
   return NameHash + (LocId << 5) + LocId;
----------------
This `nodeHash` is the blend function similar to DJB and it's assuming 32-bit length, see the shift 16. If we are fully switching to 64 bit hash, we should change the blend function too.

However, does the collision come from the blend function (`nodeHash`), or lower part of `std::hash`? I guess it's the later, then I'm curious what strings actually lead to collision on the lower 32 bit? 




Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113688/new/

https://reviews.llvm.org/D113688