Merge branch 'main' into amd-trunk-dev

Author: skatrak

bolt/include/bolt/Profile/BoltAddressTranslation.h

@@ -141,15 +141,13 @@ class BoltAddressTranslation {
                          uint64_t FuncOutputAddress) const;
 
   /// Write the serialized address translation table for a function.
-  template <bool Cold>
-  void writeMaps(std::map<uint64_t, MapTy> &Maps, uint64_t &PrevAddress,
-                 raw_ostream &OS);
+  template <bool Cold> void writeMaps(uint64_t &PrevAddress, raw_ostream &OS);
 
   /// Read the serialized address translation table for a function.
   /// Return a parse error if failed.
   template <bool Cold>
-  void parseMaps(std::vector<uint64_t> &HotFuncs, uint64_t &PrevAddress,
-                 DataExtractor &DE, uint64_t &Offset, Error &Err);
+  void parseMaps(uint64_t &PrevAddress, DataExtractor &DE, uint64_t &Offset,
+                 Error &Err);
 
   /// Returns the bitmask with set bits corresponding to indices of BRANCHENTRY
   /// entries in function address translation map.
@@ -161,6 +159,9 @@ class BoltAddressTranslation {
 
   std::map<uint64_t, MapTy> Maps;
 
+  /// Ordered vector with addresses of hot functions.
+  std::vector<uint64_t> HotFuncs;
+
   /// Map a function to its basic blocks count
   std::unordered_map<uint64_t, size_t> NumBasicBlocksMap;
 

bolt/lib/Core/BinaryFunction.cpp

@@ -3684,9 +3684,8 @@ BinaryFunction::BasicBlockListType BinaryFunction::dfs() const {
     BinaryBasicBlock *BB = Stack.top();
     Stack.pop();
 
-    if (Visited.find(BB) != Visited.end())
+    if (!Visited.insert(BB).second)
       continue;
-    Visited.insert(BB);
     DFS.push_back(BB);
 
     for (BinaryBasicBlock *SuccBB : BB->landing_pads()) {
@@ -3879,11 +3878,8 @@ void BinaryFunction::disambiguateJumpTables(
       JumpTable *JT = getJumpTable(Inst);
       if (!JT)
         continue;
-      auto Iter = JumpTables.find(JT);
-      if (Iter == JumpTables.end()) {
-        JumpTables.insert(JT);
+      if (JumpTables.insert(JT).second)
         continue;
-      }
       // This instruction is an indirect jump using a jump table, but it is
       // using the same jump table of another jump. Try all our tricks to
       // extract the jump table symbol and make it point to a new, duplicated JT

bolt/lib/Core/DIEBuilder.cpp

@@ -57,11 +57,9 @@ getDWOName(llvm::DWARFUnit &CU,
          "DW_AT_dwo_name/DW_AT_GNU_dwo_name does not exist.");
   if (DwarfOutputPath) {
     DWOName = std::string(sys::path::filename(DWOName));
-    auto Iter = NameToIndexMap.find(DWOName);
-    if (Iter == NameToIndexMap.end())
-      Iter = NameToIndexMap.insert({DWOName, 0}).first;
-    DWOName.append(std::to_string(Iter->second));
-    ++Iter->second;
+    uint32_t &Index = NameToIndexMap[DWOName];
+    DWOName.append(std::to_string(Index));
+    ++Index;
   }
   DWOName.append(".dwo");
   return DWOName;

bolt/lib/Passes/Instrumentation.cpp

@@ -109,9 +109,8 @@ static bool hasAArch64ExclusiveMemop(
     BinaryBasicBlock *BB = BBQueue.front().first;
     bool IsLoad = BBQueue.front().second;
     BBQueue.pop();
-    if (Visited.find(BB) != Visited.end())
+    if (!Visited.insert(BB).second)
       continue;
-    Visited.insert(BB);
 
     for (const MCInst &Inst : *BB) {
       // Two loads one after another - skip whole function
@@ -126,8 +125,7 @@ static bool hasAArch64ExclusiveMemop(
       if (BC.MIB->isAArch64ExclusiveLoad(Inst))
         IsLoad = true;
 
-      if (IsLoad && BBToSkip.find(BB) == BBToSkip.end()) {
-        BBToSkip.insert(BB);
+      if (IsLoad && BBToSkip.insert(BB).second) {
         if (opts::Verbosity >= 2) {
           outs() << "BOLT-INSTRUMENTER: skip BB " << BB->getName()
                  << " due to exclusive instruction in function "

bolt/lib/Passes/LongJmp.cpp

@@ -324,9 +324,8 @@ uint64_t LongJmpPass::tentativeLayoutRelocColdPart(
 uint64_t LongJmpPass::tentativeLayoutRelocMode(
     const BinaryContext &BC, std::vector<BinaryFunction *> &SortedFunctions,
     uint64_t DotAddress) {
-
   // Compute hot cold frontier
-  uint32_t LastHotIndex = -1u;
+  int64_t LastHotIndex = -1u;
   uint32_t CurrentIndex = 0;
   if (opts::HotFunctionsAtEnd) {
     for (BinaryFunction *BF : SortedFunctions) {
@@ -351,19 +350,20 @@ uint64_t LongJmpPass::tentativeLayoutRelocMode(
   // Hot
   CurrentIndex = 0;
   bool ColdLayoutDone = false;
+  auto runColdLayout = [&]() {
+    DotAddress = tentativeLayoutRelocColdPart(BC, SortedFunctions, DotAddress);
+    ColdLayoutDone = true;
+    if (opts::HotFunctionsAtEnd)
+      DotAddress = alignTo(DotAddress, opts::AlignText);
+  };
   for (BinaryFunction *Func : SortedFunctions) {
     if (!BC.shouldEmit(*Func)) {
       HotAddresses[Func] = Func->getAddress();
       continue;
     }
 
-    if (!ColdLayoutDone && CurrentIndex >= LastHotIndex) {
-      DotAddress =
-          tentativeLayoutRelocColdPart(BC, SortedFunctions, DotAddress);
-      ColdLayoutDone = true;
-      if (opts::HotFunctionsAtEnd)
-        DotAddress = alignTo(DotAddress, opts::AlignText);
-    }
+    if (!ColdLayoutDone && CurrentIndex >= LastHotIndex)
+      runColdLayout();
 
     DotAddress = alignTo(DotAddress, Func->getMinAlignment());
     uint64_t Pad =
@@ -382,6 +382,11 @@ uint64_t LongJmpPass::tentativeLayoutRelocMode(
     DotAddress += Func->estimateConstantIslandSize();
     ++CurrentIndex;
   }
+
+  // Ensure that tentative code layout always runs for cold blocks.
+  if (!ColdLayoutDone)
+    runColdLayout();
+
   // BBs
   for (BinaryFunction *Func : SortedFunctions)
     tentativeBBLayout(*Func);

bolt/lib/Profile/BoltAddressTranslation.cpp

@@ -143,8 +143,8 @@ void BoltAddressTranslation::write(const BinaryContext &BC, raw_ostream &OS) {
 
   // Output addresses are delta-encoded
   uint64_t PrevAddress = 0;
-  writeMaps</*Cold=*/false>(Maps, PrevAddress, OS);
-  writeMaps</*Cold=*/true>(Maps, PrevAddress, OS);
+  writeMaps</*Cold=*/false>(PrevAddress, OS);
+  writeMaps</*Cold=*/true>(PrevAddress, OS);
 
   BC.outs() << "BOLT-INFO: Wrote " << Maps.size() << " BAT maps\n";
   BC.outs() << "BOLT-INFO: Wrote " << FuncHashes.getNumFunctions()
@@ -182,8 +182,7 @@ size_t BoltAddressTranslation::getNumEqualOffsets(const MapTy &Map,
 }
 
 template <bool Cold>
-void BoltAddressTranslation::writeMaps(std::map<uint64_t, MapTy> &Maps,
-                                       uint64_t &PrevAddress, raw_ostream &OS) {
+void BoltAddressTranslation::writeMaps(uint64_t &PrevAddress, raw_ostream &OS) {
   const uint32_t NumFuncs =
       llvm::count_if(llvm::make_first_range(Maps), [&](const uint64_t Address) {
         return Cold == ColdPartSource.count(Address);
@@ -213,16 +212,17 @@ void BoltAddressTranslation::writeMaps(std::map<uint64_t, MapTy> &Maps,
             : 0;
     uint32_t Skew = 0;
     if (Cold) {
-      auto HotEntryIt = Maps.find(ColdPartSource[Address]);
-      assert(HotEntryIt != Maps.end());
-      size_t HotIndex = std::distance(Maps.begin(), HotEntryIt);
+      auto HotEntryIt = llvm::lower_bound(HotFuncs, ColdPartSource[Address]);
+      assert(HotEntryIt != HotFuncs.end());
+      size_t HotIndex = std::distance(HotFuncs.begin(), HotEntryIt);
       encodeULEB128(HotIndex - PrevIndex, OS);
       PrevIndex = HotIndex;
       // Skew of all input offsets for cold fragments is simply the first input
       // offset.
       Skew = Map.begin()->second >> 1;
       encodeULEB128(Skew, OS);
     } else {
+      HotFuncs.push_back(Address);
       // Function hash
       size_t BFHash = getBFHash(HotInputAddress);
       LLVM_DEBUG(dbgs() << "Hash: " << formatv("{0:x}\n", BFHash));
@@ -311,17 +311,15 @@ std::error_code BoltAddressTranslation::parse(raw_ostream &OS, StringRef Buf) {
     return make_error_code(llvm::errc::io_error);
 
   Error Err(Error::success());
-  std::vector<uint64_t> HotFuncs;
   uint64_t PrevAddress = 0;
-  parseMaps</*Cold=*/false>(HotFuncs, PrevAddress, DE, Offset, Err);
-  parseMaps</*Cold=*/true>(HotFuncs, PrevAddress, DE, Offset, Err);
+  parseMaps</*Cold=*/false>(PrevAddress, DE, Offset, Err);
+  parseMaps</*Cold=*/true>(PrevAddress, DE, Offset, Err);
   OS << "BOLT-INFO: Parsed " << Maps.size() << " BAT entries\n";
   return errorToErrorCode(std::move(Err));
 }
 
 template <bool Cold>
-void BoltAddressTranslation::parseMaps(std::vector<uint64_t> &HotFuncs,
-                                       uint64_t &PrevAddress, DataExtractor &DE,
+void BoltAddressTranslation::parseMaps(uint64_t &PrevAddress, DataExtractor &DE,
                                        uint64_t &Offset, Error &Err) {
   const uint32_t NumFunctions = DE.getULEB128(&Offset, &Err);
   LLVM_DEBUG(dbgs() << "Parsing " << NumFunctions << (Cold ? " cold" : "")

bolt/test/AArch64/split-funcs-lite.s

@@ -0,0 +1,27 @@
+# This test checks that tentative code layout for cold blocks always runs.
+# It commonly happens when using lite mode with split functions.
+
+# REQUIRES: system-linux, asserts
+
+# RUN: %clang %cflags -o %t %s
+# RUN: %clang %s %cflags -Wl,-q -o %t
+# RUN: link_fdata --no-lbr %s %t %t.fdata
+# RUN: llvm-bolt %t -o %t.bolt --data %t.fdata -split-functions \
+# RUN:   -debug 2>&1 | FileCheck %s
+
+  .text
+  .globl  foo
+  .type foo, %function
+foo:
+.entry_bb:
+# FDATA: 1 foo #.entry_bb# 10
+    cmp x0, #0
+    b.eq .Lcold_bb1
+    ret
+.Lcold_bb1:
+    ret
+
+## Force relocation mode.
+.reloc 0, R_AARCH64_NONE
+
+# CHECK: foo{{.*}} cold tentative: {{.*}}

bolt/tools/driver/llvm-bolt.cpp

@@ -202,9 +202,9 @@ int main(int argc, char **argv) {
 
   ToolName = argv[0];
 
-  if (llvm::sys::path::filename(ToolName) == "perf2bolt")
+  if (llvm::sys::path::filename(ToolName).starts_with("perf2bolt"))
     perf2boltMode(argc, argv);
-  else if (llvm::sys::path::filename(ToolName) == "llvm-boltdiff")
+  else if (llvm::sys::path::filename(ToolName).starts_with("llvm-boltdiff"))
     boltDiffMode(argc, argv);
   else
     boltMode(argc, argv);

clang-tools-extra/clang-doc/tool/ClangDocMain.cpp

@@ -300,8 +300,7 @@ Example usage for a project using a compile commands database:
   llvm::StringMap<std::vector<StringRef>> USRToBitcode;
   Executor->get()->getToolResults()->forEachResult(
       [&](StringRef Key, StringRef Value) {
-        auto R = USRToBitcode.try_emplace(Key, std::vector<StringRef>());
-        R.first->second.emplace_back(Value);
+        USRToBitcode[Key].emplace_back(Value);
       });
 
   // Collects all Infos according to their unique USR value. This map is added

clang-tools-extra/clang-tidy/ClangTidy.cpp

@@ -458,7 +458,6 @@ ClangTidyASTConsumerFactory::createASTConsumer(
   if (!AnalyzerOptions.CheckersAndPackages.empty()) {
     setStaticAnalyzerCheckerOpts(Context.getOptions(), AnalyzerOptions);
     AnalyzerOptions.AnalysisDiagOpt = PD_NONE;
-    AnalyzerOptions.eagerlyAssumeBinOpBifurcation = true;
     std::unique_ptr<ento::AnalysisASTConsumer> AnalysisConsumer =
         ento::CreateAnalysisConsumer(Compiler);
     AnalysisConsumer->AddDiagnosticConsumer(