[llvm] [SimplifyCFG] Fix hoisting problem in SimplifyCFG (PR #78615)

[Quentin Dian via llvm-commits]

================
@@ -1589,80 +1610,121 @@ bool SimplifyCFGOpt::hoistCommonCodeFromSuccessors(BasicBlock *BB,
   }
 
   bool Changed = false;
+  auto *SuccIterPairBegin = SuccIterPairs.begin();
+  SuccIterPairBegin++;
+  auto OtherSuccIterPairRange =
+      iterator_range(SuccIterPairBegin, SuccIterPairs.end());
+  auto OtherSuccIterRange = make_first_range(OtherSuccIterPairRange);
+  using InstrFlagPair = std::pair<Instruction *, unsigned>;
+  SmallVector<DenseMap<llvm::hash_code, InstrFlagPair>, 2> OtherSuccessorsHash;
+
+  for (auto BBItrPair : OtherSuccIterRange) {
+    // Fill the hashmap for every other successor
+    DenseMap<llvm::hash_code, InstrFlagPair> hashMap;
----------------
DianQK wrote:

I think this might improve compile times, if we ignore hash collisions.
If the hash is the same, we overwrite or ignore the previous value in a vector. Then you could just check that the length of the vector matches the number of switch cases. This may reduce a lot of hash computations.
I'm not sure whether it works, but I think it's worth a try.

https://github.com/llvm/llvm-project/pull/78615