[llvm] Adding Matching and Inference Functionality to Propeller-PR3: Read basic block hashes from propeller profile. (PR #164223)

[via llvm-commits]

https://github.com/wdx727 created https://github.com/llvm/llvm-project/pull/164223

Adding Matching and Inference Functionality to Propeller. For detailed information, please refer to the following RFC: https://discourse.llvm.org/t/rfc-adding-matching-and-inference-functionality-to-propeller/86238.
This is the third PR, which is used to read basic block hashes from the propeller profile. The associated PRs are:
PR1: https://github.com/llvm/llvm-project/pull/160706
PR2: https://github.com/llvm/llvm-project/pull/162963

co-authors: lifengxiang1025 [lifengxiang at kuaishou.com](mailto:lifengxiang at kuaishou.com); zcfh [wuminghui03 at kuaishou.com](mailto:wuminghui03 at kuaishou.com)

>From 8e8b119802a2339b1a175a38d76e85d04ff37a3a Mon Sep 17 00:00:00 2001
From: wudexin <wudexin at kuaishou.com>
Date: Mon, 20 Oct 2025 16:53:46 +0800
Subject: [PATCH] Adding Matching and Inference Functionality to Propeller-PR3:
 Read basic block hashes from propeller profile.

Co-authored-by: lifengxiang1025 <lifengxiang at kuaishou.com>
Co-authored-by: zcfh <wuminghui03 at kuaishou.com>
---
 .../CodeGen/BasicBlockSectionsProfileReader.h |  9 +++++
 .../BasicBlockSectionsProfileReader.cpp       | 34 +++++++++++++++++++
 2 files changed, 43 insertions(+)

diff --git a/llvm/include/llvm/CodeGen/BasicBlockSectionsProfileReader.h b/llvm/include/llvm/CodeGen/BasicBlockSectionsProfileReader.h
index 82dd5feb31dba..497d751b3b26e 100644
--- a/llvm/include/llvm/CodeGen/BasicBlockSectionsProfileReader.h
+++ b/llvm/include/llvm/CodeGen/BasicBlockSectionsProfileReader.h
@@ -54,6 +54,8 @@ struct FunctionPathAndClusterInfo {
   DenseMap<UniqueBBID, uint64_t> NodeCounts;
   // Edge counts for each edge, stored as a nested map.
   DenseMap<UniqueBBID, DenseMap<UniqueBBID, uint64_t>> EdgeCounts;
+  // Hash for each basic block.
+  DenseMap<unsigned, uint64_t> BBHashes;
 };
 
 class BasicBlockSectionsProfileReader {
@@ -86,6 +88,10 @@ class BasicBlockSectionsProfileReader {
   uint64_t getEdgeCount(StringRef FuncName, const UniqueBBID &SrcBBID,
                         const UniqueBBID &SinkBBID) const;
 
+  // Return the complete function path and cluster info for the given function.
+  std::pair<bool, FunctionPathAndClusterInfo>
+  getFunctionPathAndClusterInfo(StringRef FuncName) const;
+
 private:
   StringRef getAliasName(StringRef FuncName) const {
     auto R = FuncAliasMap.find(FuncName);
@@ -195,6 +201,9 @@ class BasicBlockSectionsProfileReaderWrapperPass : public ImmutablePass {
   uint64_t getEdgeCount(StringRef FuncName, const UniqueBBID &SrcBBID,
                         const UniqueBBID &DestBBID) const;
 
+  std::pair<bool, FunctionPathAndClusterInfo>
+  getFunctionPathAndClusterInfo(StringRef FuncName) const;
+
   // Initializes the FunctionNameToDIFilename map for the current module and
   // then reads the profile for the matching functions.
   bool doInitialization(Module &M) override;
diff --git a/llvm/lib/CodeGen/BasicBlockSectionsProfileReader.cpp b/llvm/lib/CodeGen/BasicBlockSectionsProfileReader.cpp
index fbcd614b85d18..696da00457002 100644
--- a/llvm/lib/CodeGen/BasicBlockSectionsProfileReader.cpp
+++ b/llvm/lib/CodeGen/BasicBlockSectionsProfileReader.cpp
@@ -91,6 +91,15 @@ uint64_t BasicBlockSectionsProfileReader::getEdgeCount(
   return EdgeIt->second;
 }
 
+std::pair<bool, FunctionPathAndClusterInfo>
+BasicBlockSectionsProfileReader::getFunctionPathAndClusterInfo(
+    StringRef FuncName) const {
+  auto R = ProgramPathAndClusterInfo.find(getAliasName(FuncName));
+  return R != ProgramPathAndClusterInfo.end()
+             ? std::pair(true, R->second)
+             : std::pair(false, FunctionPathAndClusterInfo());
+}
+
 // Reads the version 1 basic block sections profile. Profile for each function
 // is encoded as follows:
 //   m <module_name>
@@ -287,6 +296,25 @@ Error BasicBlockSectionsProfileReader::ReadV1Profile() {
       }
       continue;
     }
+    case 'h': { // Basic block hash secifier.
+      // Skip the profile when we the profile iterator (FI) refers to the
+      // past-the-end element.
+      if (FI == ProgramPathAndClusterInfo.end())
+        continue;
+      for (auto BBIDHashStr : Values) {
+        auto [BBIDStr, HashStr] = BBIDHashStr.split(':');
+        unsigned long long BBID = 0, Hash = 0;
+        if (getAsUnsignedInteger(BBIDStr, 10, BBID))
+          return createProfileParseError(Twine("unsigned integer expected: '") +
+                                         BBIDStr + "'");
+        HashStr.consume_front("0x");
+        if (getAsUnsignedInteger(HashStr, 16, Hash))
+          return createProfileParseError(Twine("unsigned integer expected: '") +
+                                         HashStr + "'");
+        FI->second.BBHashes[BBID] = Hash;
+      }
+      continue;
+    }
     default:
       return createProfileParseError(Twine("invalid specifier: '") +
                                      Twine(Specifier) + "'");
@@ -493,6 +521,12 @@ uint64_t BasicBlockSectionsProfileReaderWrapperPass::getEdgeCount(
   return BBSPR.getEdgeCount(FuncName, SrcBBID, SinkBBID);
 }
 
+std::pair<bool, FunctionPathAndClusterInfo>
+BasicBlockSectionsProfileReaderWrapperPass::getFunctionPathAndClusterInfo(
+    StringRef FuncName) const {
+  return BBSPR.getFunctionPathAndClusterInfo(FuncName);
+}
+
 BasicBlockSectionsProfileReader &
 BasicBlockSectionsProfileReaderWrapperPass::getBBSPR() {
   return BBSPR;