[llvm-branch-commits] [llvm] [BOLT] Match blocks with pseudo probes (PR #99891)

[via llvm-branch-commits]

================
@@ -223,10 +249,31 @@ class StaleMatcher {
     return Hash1.InstrHash == Hash2.InstrHash;
   }
 
+  /// Returns true if a profiled block was matched with its pseudo probe.
+  bool isPseudoProbeMatch(BlendedBlockHash YamlBBHash) {
+    return MatchedWithPseudoProbes.find(YamlBBHash.combine()) !=
+           MatchedWithPseudoProbes.end();
+  }
+
+  /// Returns the number of blocks matched with opcodes.
+  size_t getNumBlocksMatchedWithOpcodes() const { return MatchedWithOpcodes; }
+
+  /// Returns the number of blocks matched with pseudo probes.
+  size_t getNumBlocksMatchedWithPseudoProbes() const {
+    return MatchedWithPseudoProbes.size();
+  }
+
 private:
   using HashBlockPairType = std::pair<BlendedBlockHash, FlowBlock *>;
   std::unordered_map<uint16_t, std::vector<HashBlockPairType>> OpHashToBlocks;
   std::unordered_map<uint64_t, std::vector<HashBlockPairType>> CallHashToBlocks;
+  std::unordered_map<uint64_t, std::vector<const MCDecodedPseudoProbe *>>
----------------
WenleiHe wrote:

@dcci DenseMap or LLVM ADT isn't generally better. we have a ton of undered_map throughout the code base, and we have benchmarked some cases carefully between the two, and DenseMap sometimes uses a lot more memory, as a result many uses of undered_map were intentional. 

If memory is of a concern (probe decoding and holding probe forest generally isn't cheap), I'm not sure DenseMap will help us.  


https://github.com/llvm/llvm-project/pull/99891