Merge branch 'main' into fixup/no-prof-table-based-cttz-basics

Author: jinhuang1102

.ci/all_requirements.txt

@@ -194,10 +194,6 @@ ml-dtypes==0.5.1 ; python_version < "3.13" \
     --hash=sha256:d13755f8e8445b3870114e5b6240facaa7cb0c3361e54beba3e07fa912a6e12b \
     --hash=sha256:fd918d4e6a4e0c110e2e05be7a7814d10dc1b95872accbf6512b80a109b71ae1
     # via -r mlir/python/requirements.txt
-nanobind==2.9.2 \
-    --hash=sha256:c37957ffd5eac7eda349cff3622ecd32e5ee1244ecc912c99b5bc8188bafd16e \
-    --hash=sha256:e7608472de99d375759814cab3e2c94aba3f9ec80e62cfef8ced495ca5c27d6e
-    # via -r mlir/python/requirements.txt
 numpy==2.0.2 \
     --hash=sha256:0123ffdaa88fa4ab64835dcbde75dcdf89c453c922f18dced6e27c90d1d0ec5a \
     --hash=sha256:11a76c372d1d37437857280aa142086476136a8c0f373b2e648ab2c8f18fb195 \
@@ -299,10 +295,6 @@ pyasn1-modules==0.4.2 \
     --hash=sha256:29253a9207ce32b64c3ac6600edc75368f98473906e8fd1043bd6b5b1de2c14a \
     --hash=sha256:677091de870a80aae844b1ca6134f54652fa2c8c5a52aa396440ac3106e941e6
     # via google-auth
-pybind11==2.13.6 \
-    --hash=sha256:237c41e29157b962835d356b370ededd57594a26d5894a795960f0047cb5caf5 \
-    --hash=sha256:ba6af10348c12b24e92fa086b39cfba0eff619b61ac77c406167d813b096d39a
-    # via -r mlir/python/requirements.txt
 pyyaml==6.0.1 \
     --hash=sha256:04ac92ad1925b2cff1db0cfebffb6ffc43457495c9b3c39d3fcae417d7125dc5 \
     --hash=sha256:062582fca9fabdd2c8b54a3ef1c978d786e0f6b3a1510e0ac93ef59e0ddae2bc \

.github/new-prs-labeler.yml

@@ -1121,3 +1121,6 @@ tablegen:
   - llvm/include/TableGen/**
   - llvm/lib/TableGen/**
   - llvm/utils/TableGen/**
+
+infrastructure:
+  - .ci/**

.github/workflows/libcxx-build-and-test.yaml

@@ -255,7 +255,6 @@ jobs:
       - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
       - name: Install dependencies
         run: |
-          choco install -y ninja
           pip install psutil
       - name: Install a current LLVM
         if: ${{ matrix.mingw != true }}
@@ -281,6 +280,10 @@ jobs:
       - name: Set up the MSVC dev environment
         if: ${{ matrix.mingw != true }}
         uses: ilammy/msvc-dev-cmd@0b201ec74fa43914dc39ae48a89fd1d8cb592756 # v1.13.0
+      - name: Add the installed Clang at the start of the path
+        if: ${{ matrix.mingw != true }}
+        run: |
+          echo "c:\Program Files\LLVM\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
       - name: Build and test
         run: |
           bash libcxx/utils/ci/run-buildbot ${{ matrix.config }}

.github/workflows/llvm-tests.yml renamed to .github/workflows/llvm-abi-tests.yml

@@ -1,4 +1,4 @@
-name: LLVM Tests
+name: LLVM ABI Tests
 
 permissions:
   contents: read

bolt/include/bolt/Core/BinaryContext.h

@@ -190,6 +190,9 @@ class BinaryContext {
   /// Unique build ID if available for the binary.
   std::optional<std::string> FileBuildID;
 
+  /// GNU property note indicating AArch64 BTI.
+  bool UsesBTI{false};
+
   /// Set of all sections.
   struct CompareSections {
     bool operator()(const BinarySection *A, const BinarySection *B) const {
@@ -384,6 +387,9 @@ class BinaryContext {
   }
   void setFileBuildID(StringRef ID) { FileBuildID = std::string(ID); }
 
+  bool usesBTI() const { return UsesBTI; }
+  void setUsesBTI(bool Value) { UsesBTI = Value; }
+
   bool hasSymbolsWithFileName() const { return HasSymbolsWithFileName; }
   void setHasSymbolsWithFileName(bool Value) { HasSymbolsWithFileName = Value; }
 

bolt/include/bolt/Passes/SplitFunctions.h

@@ -18,25 +18,6 @@
 namespace llvm {
 namespace bolt {
 
-/// Strategy used to partition blocks into fragments.
-enum SplitFunctionsStrategy : char {
-  /// Split each function into a hot and cold fragment using profiling
-  /// information.
-  Profile2 = 0,
-  /// Split each function into a hot, warm, and cold fragment using
-  /// profiling information.
-  CDSplit,
-  /// Split each function into a hot and cold fragment at a randomly chosen
-  /// split point (ignoring any available profiling information).
-  Random2,
-  /// Split each function into N fragments at a randomly chosen split points
-  /// (ignoring any available profiling information).
-  RandomN,
-  /// Split all basic blocks of each function into fragments such that each
-  /// fragment contains exactly a single basic block.
-  All
-};
-
 class SplitStrategy {
 public:
   using BlockIt = BinaryFunction::BasicBlockOrderType::iterator;

bolt/include/bolt/Rewrite/MetadataRewriters.h

@@ -27,6 +27,8 @@ std::unique_ptr<MetadataRewriter> createPseudoProbeRewriter(BinaryContext &);
 
 std::unique_ptr<MetadataRewriter> createSDTRewriter(BinaryContext &);
 
+std::unique_ptr<MetadataRewriter> createGNUPropertyRewriter(BinaryContext &);
+
 } // namespace bolt
 } // namespace llvm
 

bolt/include/bolt/Utils/CommandLineOpts.h

@@ -29,6 +29,25 @@ enum HeatmapModeKind {
   HM_Optional   // perf2bolt --heatmap
 };
 
+/// Strategy used to partition blocks into fragments.
+enum SplitFunctionsStrategy : char {
+  /// Split each function into a hot and cold fragment using profiling
+  /// information.
+  Profile2 = 0,
+  /// Split each function into a hot, warm, and cold fragment using
+  /// profiling information.
+  CDSplit,
+  /// Split each function into a hot and cold fragment at a randomly chosen
+  /// split point (ignoring any available profiling information).
+  Random2,
+  /// Split each function into N fragments at a randomly chosen split points
+  /// (ignoring any available profiling information).
+  RandomN,
+  /// Split all basic blocks of each function into fragments such that each
+  /// fragment contains exactly a single basic block.
+  All
+};
+
 using HeatmapBlockSizes = std::vector<unsigned>;
 struct HeatmapBlockSpecParser : public llvm::cl::parser<HeatmapBlockSizes> {
   explicit HeatmapBlockSpecParser(llvm::cl::Option &O)
@@ -78,6 +97,7 @@ extern llvm::cl::opt<std::string> OutputFilename;
 extern llvm::cl::opt<std::string> PerfData;
 extern llvm::cl::opt<bool> PrintCacheMetrics;
 extern llvm::cl::opt<bool> PrintSections;
+extern llvm::cl::opt<SplitFunctionsStrategy> SplitStrategy;
 
 // The format to use with -o in aggregation mode (perf2bolt)
 enum ProfileFormatKind { PF_Fdata, PF_YAML };

bolt/lib/Passes/LongJmp.cpp

@@ -895,6 +895,10 @@ void LongJmpPass::relaxLocalBranches(BinaryFunction &BF) {
 
 Error LongJmpPass::runOnFunctions(BinaryContext &BC) {
 
+  assert((opts::CompactCodeModel ||
+          opts::SplitStrategy != opts::SplitFunctionsStrategy::CDSplit) &&
+         "LongJmp cannot work with functions split in more than two fragments");
+
   if (opts::CompactCodeModel) {
     BC.outs()
         << "BOLT-INFO: relaxing branches for compact code model (<128MB)\n";

bolt/lib/Passes/SplitFunctions.cpp

@@ -86,29 +86,6 @@ static cl::opt<unsigned> SplitThreshold(
              "increase after splitting."),
     cl::init(0), cl::Hidden, cl::cat(BoltOptCategory));
 
-static cl::opt<SplitFunctionsStrategy> SplitStrategy(
-    "split-strategy", cl::init(SplitFunctionsStrategy::Profile2),
-    cl::values(clEnumValN(SplitFunctionsStrategy::Profile2, "profile2",
-                          "split each function into a hot and cold fragment "
-                          "using profiling information")),
-    cl::values(clEnumValN(SplitFunctionsStrategy::CDSplit, "cdsplit",
-                          "split each function into a hot, warm, and cold "
-                          "fragment using profiling information")),
-    cl::values(clEnumValN(
-        SplitFunctionsStrategy::Random2, "random2",
-        "split each function into a hot and cold fragment at a randomly chosen "
-        "split point (ignoring any available profiling information)")),
-    cl::values(clEnumValN(
-        SplitFunctionsStrategy::RandomN, "randomN",
-        "split each function into N fragments at a randomly chosen split "
-        "points (ignoring any available profiling information)")),
-    cl::values(clEnumValN(
-        SplitFunctionsStrategy::All, "all",
-        "split all basic blocks of each function into fragments such that each "
-        "fragment contains exactly a single basic block")),
-    cl::desc("strategy used to partition blocks into fragments"),
-    cl::cat(BoltOptCategory));
-
 static cl::opt<double> CallScale(
     "call-scale",
     cl::desc("Call score scale coefficient (when --split-strategy=cdsplit)"),
@@ -724,14 +701,14 @@ Error SplitFunctions::runOnFunctions(BinaryContext &BC) {
   // If split strategy is not CDSplit, then a second run of the pass is not
   // needed after function reordering.
   if (BC.HasFinalizedFunctionOrder &&
-      opts::SplitStrategy != SplitFunctionsStrategy::CDSplit)
+      opts::SplitStrategy != opts::SplitFunctionsStrategy::CDSplit)
     return Error::success();
 
   std::unique_ptr<SplitStrategy> Strategy;
   bool ForceSequential = false;
 
   switch (opts::SplitStrategy) {
-  case SplitFunctionsStrategy::CDSplit:
+  case opts::SplitFunctionsStrategy::CDSplit:
     // CDSplit runs two splitting passes: hot-cold splitting (SplitPrfoile2)
     // before function reordering and hot-warm-cold splitting
     // (SplitCacheDirected) after function reordering.
@@ -742,21 +719,21 @@ Error SplitFunctions::runOnFunctions(BinaryContext &BC) {
     opts::AggressiveSplitting = true;
     BC.HasWarmSection = true;
     break;
-  case SplitFunctionsStrategy::Profile2:
+  case opts::SplitFunctionsStrategy::Profile2:
     Strategy = std::make_unique<SplitProfile2>();
     break;
-  case SplitFunctionsStrategy::Random2:
+  case opts::SplitFunctionsStrategy::Random2:
     Strategy = std::make_unique<SplitRandom2>();
     // If we split functions randomly, we need to ensure that across runs with
     // the same input, we generate random numbers for each function in the same
     // order.
     ForceSequential = true;
     break;
-  case SplitFunctionsStrategy::RandomN:
+  case opts::SplitFunctionsStrategy::RandomN:
     Strategy = std::make_unique<SplitRandomN>();
     ForceSequential = true;
     break;
-  case SplitFunctionsStrategy::All:
+  case opts::SplitFunctionsStrategy::All:
     Strategy = std::make_unique<SplitAll>();
     break;
   }