Merge branch 'main' into clang-doc-ftime-trace

Author: PeterChou1

.github/workflows/issue-write.yml

@@ -18,7 +18,11 @@ jobs:
     permissions:
       pull-requests: write
     if: >
-      github.event.workflow_run.event == 'pull_request'
+      github.event.workflow_run.event == 'pull_request' &&
+      (
+        github.event.workflow_run.conclusion == 'success' ||
+        github.event.workflow_run.conclusion == 'failure'
+      )
     steps:
       - name: 'Download artifact'
         uses: actions/download-artifact@6b208ae046db98c579e8a3aa621ab581ff575935 # v4.1.1

bolt/docs/CommandLineArgumentReference.md

@@ -688,6 +688,10 @@
 
   Use a modified clustering algorithm geared towards minimizing branches
 
+- `--name-similarity-function-matching-threshold=<uint>`
+
+  Match functions using namespace and edit distance.
+
 - `--no-inline`
 
   Disable all inlining (overrides other inlining options)

bolt/docs/OptimizingLinux.md

@@ -44,6 +44,11 @@ $ perf2bolt -p perf.data -o perf.fdata vmlinux
 
 Under a high load, `perf.data` should be several gigabytes in size and you should expect the converted `perf.fdata` not to exceed 100 MB.
 
+Profiles collected from multiple workloads could be joined into a single profile using `merge-fdata` utility:
+```bash
+$ merge-fdata perf.1.fdata perf.2.fdata ... perf.<N>.fdata > perf.merged.fdata
+```
+
 Two changes are required for the kernel build. The first one is optional but highly recommended. It introduces a BOLT-reserved space into `vmlinux` code section:
 
 

bolt/include/bolt/Core/BinaryBasicBlock.h

@@ -842,15 +842,6 @@ class BinaryBasicBlock {
   bool analyzeBranch(const MCSymbol *&TBB, const MCSymbol *&FBB,
                      MCInst *&CondBranch, MCInst *&UncondBranch);
 
-  /// Return true if iterator \p I is pointing to the first instruction in
-  /// a pair that could be macro-fused.
-  bool isMacroOpFusionPair(const_iterator I) const;
-
-  /// If the basic block has a pair of instructions suitable for macro-fusion,
-  /// return iterator to the first instruction of the pair.
-  /// Otherwise return end().
-  const_iterator getMacroOpFusionPair() const;
-
   /// Printer required for printing dominator trees.
   void printAsOperand(raw_ostream &OS, bool PrintType = true) {
     if (PrintType)

bolt/include/bolt/Core/BinaryContext.h

@@ -698,10 +698,6 @@ class BinaryContext {
 
   /// Binary-wide aggregated stats.
   struct BinaryStats {
-    /// Stats for macro-fusion.
-    uint64_t MissedMacroFusionPairs{0};
-    uint64_t MissedMacroFusionExecCount{0};
-
     /// Stats for stale profile matching:
     ///   the total number of basic blocks in the profile
     uint32_t NumStaleBlocks{0};

bolt/include/bolt/Core/BinaryFunction.h

@@ -835,10 +835,6 @@ class BinaryFunction {
   /// them.
   void calculateLoopInfo();
 
-  /// Calculate missed macro-fusion opportunities and update BinaryContext
-  /// stats.
-  void calculateMacroOpFusionStats();
-
   /// Returns if BinaryDominatorTree has been constructed for this function.
   bool hasDomTree() const { return BDT != nullptr; }
 

bolt/include/bolt/Core/DebugData.h

@@ -210,6 +210,15 @@ class DebugRangesSectionWriter {
   static bool classof(const DebugRangesSectionWriter *Writer) {
     return Writer->getKind() == RangesWriterKind::DebugRangesWriter;
   }
+  
+  /// Append a range to the main buffer.
+  void appendToRangeBuffer(const DebugBufferVector &CUBuffer);
+
+  /// Sets Unit DIE to be updated for CU.
+  void setDie(DIE *Die) { this->Die = Die; }
+
+  /// Returns Unit DIE to be updated for CU.
+  DIE *getDie() const { return Die; }
 
   /// Writes out range lists for a current CU being processed.
   void virtual finalizeSection(){};
@@ -232,6 +241,9 @@ class DebugRangesSectionWriter {
   static constexpr uint64_t EmptyRangesOffset{0};
 
 private:
+  /// Stores Unit DIE to be updated for CU.
+  DIE *Die{0};
+
   RangesWriterKind Kind;
 };
 

bolt/include/bolt/Core/MCPlusBuilder.h

@@ -930,13 +930,6 @@ class MCPlusBuilder {
   /// Return true if the instruction is encoded using EVEX (AVX-512).
   virtual bool hasEVEXEncoding(const MCInst &Inst) const { return false; }
 
-  /// Return true if a pair of instructions represented by \p Insts
-  /// could be fused into a single uop.
-  virtual bool isMacroOpFusionPair(ArrayRef<MCInst> Insts) const {
-    llvm_unreachable("not implemented");
-    return false;
-  }
-
   struct X86MemOperand {
     unsigned BaseRegNum;
     int64_t ScaleImm;

bolt/include/bolt/Profile/YAMLProfileReader.h

@@ -93,6 +93,9 @@ class YAMLProfileReader : public ProfileReaderBase {
     ProfiledFunctions.emplace(&BF);
   }
 
+  /// Matches functions with similarly named profiled functions.
+  uint64_t matchWithNameSimilarity(BinaryContext &BC);
+
   /// Check if the profile uses an event with a given \p Name.
   bool usesEvent(StringRef Name) const;
 };

bolt/include/bolt/Rewrite/DWARFRewriter.h

@@ -16,7 +16,6 @@
 #include "llvm/ADT/StringRef.h"
 #include "llvm/CodeGen/DIE.h"
 #include "llvm/DWP/DWP.h"
-#include "llvm/MC/MCAsmLayout.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/Support/ToolOutputFile.h"
 #include <cstdint>
@@ -90,6 +89,10 @@ class DWARFRewriter {
   /// Store Rangelists writer for each DWO CU.
   RangeListsDWOWriers RangeListsWritersByCU;
 
+  /// Stores ranges writer for each DWO CU.
+  std::unordered_map<uint64_t, std::unique_ptr<DebugRangesSectionWriter>>
+      LegacyRangesWritersByCU;
+
   std::mutex LocListDebugInfoPatchesMutex;
 
   /// Dwo id specific its RangesBase.
@@ -183,7 +186,7 @@ class DWARFRewriter {
   void updateDebugInfo();
 
   /// Update stmt_list for CUs based on the new .debug_line \p Layout.
-  void updateLineTableOffsets(const MCAsmLayout &Layout);
+  void updateLineTableOffsets(const MCAssembler &Asm);
 
   uint64_t getDwoRangesBase(uint64_t DWOId) { return DwoRangesBase[DWOId]; }
 

bolt/lib/Core/BinaryBasicBlock.cpp

@@ -404,45 +404,6 @@ bool BinaryBasicBlock::analyzeBranch(const MCSymbol *&TBB, const MCSymbol *&FBB,
                             CondBranch, UncondBranch);
 }
 
-bool BinaryBasicBlock::isMacroOpFusionPair(const_iterator I) const {
-  auto &MIB = Function->getBinaryContext().MIB;
-  ArrayRef<MCInst> Insts = Instructions;
-  return MIB->isMacroOpFusionPair(Insts.slice(I - begin()));
-}
-
-BinaryBasicBlock::const_iterator
-BinaryBasicBlock::getMacroOpFusionPair() const {
-  if (!Function->getBinaryContext().isX86())
-    return end();
-
-  if (getNumNonPseudos() < 2 || succ_size() != 2)
-    return end();
-
-  auto RI = getLastNonPseudo();
-  assert(RI != rend() && "cannot have an empty block with 2 successors");
-
-  BinaryContext &BC = Function->getBinaryContext();
-
-  // Skip instruction if it's an unconditional branch following
-  // a conditional one.
-  if (BC.MIB->isUnconditionalBranch(*RI))
-    ++RI;
-
-  if (!BC.MIB->isConditionalBranch(*RI))
-    return end();
-
-  // Start checking with instruction preceding the conditional branch.
-  ++RI;
-  if (RI == rend())
-    return end();
-
-  auto II = std::prev(RI.base()); // convert to a forward iterator
-  if (isMacroOpFusionPair(II))
-    return II;
-
-  return end();
-}
-
 MCInst *BinaryBasicBlock::getTerminatorBefore(MCInst *Pos) {
   BinaryContext &BC = Function->getBinaryContext();
   auto Itr = rbegin();

bolt/lib/Core/BinaryContext.cpp

@@ -20,7 +20,6 @@
 #include "llvm/DebugInfo/DWARF/DWARFCompileUnit.h"
 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
 #include "llvm/DebugInfo/DWARF/DWARFUnit.h"
-#include "llvm/MC/MCAsmLayout.h"
 #include "llvm/MC/MCAssembler.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
@@ -2404,32 +2403,23 @@ BinaryContext::calculateEmittedSize(BinaryFunction &BF, bool FixBranches) {
     Streamer->emitLabel(SplitStartLabel);
     emitFunctionBody(*Streamer, BF, FF, /*EmitCodeOnly=*/true);
     Streamer->emitLabel(SplitEndLabel);
-    // To avoid calling MCObjectStreamer::flushPendingLabels() which is
-    // private
-    Streamer->emitBytes(StringRef(""));
-    Streamer->switchSection(Section);
   }
 
-  // To avoid calling MCObjectStreamer::flushPendingLabels() which is private or
-  // MCStreamer::Finish(), which does more than we want
-  Streamer->emitBytes(StringRef(""));
-
   MCAssembler &Assembler =
       static_cast<MCObjectStreamer *>(Streamer.get())->getAssembler();
-  MCAsmLayout Layout(Assembler);
-  Assembler.layout(Layout);
+  Assembler.layout();
 
   // Obtain fragment sizes.
   std::vector<uint64_t> FragmentSizes;
   // Main fragment size.
-  const uint64_t HotSize =
-      Layout.getSymbolOffset(*EndLabel) - Layout.getSymbolOffset(*StartLabel);
+  const uint64_t HotSize = Assembler.getSymbolOffset(*EndLabel) -
+                           Assembler.getSymbolOffset(*StartLabel);
   FragmentSizes.push_back(HotSize);
   // Split fragment sizes.
   uint64_t ColdSize = 0;
   for (const auto &Labels : SplitLabels) {
-    uint64_t Size = Layout.getSymbolOffset(*Labels.second) -
-                    Layout.getSymbolOffset(*Labels.first);
+    uint64_t Size = Assembler.getSymbolOffset(*Labels.second) -
+                    Assembler.getSymbolOffset(*Labels.first);
     FragmentSizes.push_back(Size);
     ColdSize += Size;
   }
@@ -2439,7 +2429,8 @@ BinaryContext::calculateEmittedSize(BinaryFunction &BF, bool FixBranches) {
   for (FunctionFragment &FF : BF.getLayout().fragments()) {
     BinaryBasicBlock *PrevBB = nullptr;
     for (BinaryBasicBlock *BB : FF) {
-      const uint64_t BBStartOffset = Layout.getSymbolOffset(*(BB->getLabel()));
+      const uint64_t BBStartOffset =
+          Assembler.getSymbolOffset(*(BB->getLabel()));
       BB->setOutputStartAddress(BBStartOffset);
       if (PrevBB)
         PrevBB->setOutputEndAddress(BBStartOffset);

bolt/lib/Core/BinaryEmitter.cpp

@@ -38,19 +38,6 @@ extern cl::opt<bool> PreserveBlocksAlignment;
 cl::opt<bool> AlignBlocks("align-blocks", cl::desc("align basic blocks"),
                           cl::cat(BoltOptCategory));
 
-cl::opt<MacroFusionType>
-AlignMacroOpFusion("align-macro-fusion",
-  cl::desc("fix instruction alignment for macro-fusion (x86 relocation mode)"),
-  cl::init(MFT_HOT),
-  cl::values(clEnumValN(MFT_NONE, "none",
-               "do not insert alignment no-ops for macro-fusion"),
-             clEnumValN(MFT_HOT, "hot",
-               "only insert alignment no-ops on hot execution paths (default)"),
-             clEnumValN(MFT_ALL, "all",
-               "always align instructions to allow macro-fusion")),
-  cl::ZeroOrMore,
-  cl::cat(BoltRelocCategory));
-
 static cl::list<std::string>
 BreakFunctionNames("break-funcs",
   cl::CommaSeparated,
@@ -453,20 +440,7 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         Streamer.emitLabel(EntrySymbol);
     }
 
-    // Check if special alignment for macro-fusion is needed.
-    bool MayNeedMacroFusionAlignment =
-        (opts::AlignMacroOpFusion == MFT_ALL) ||
-        (opts::AlignMacroOpFusion == MFT_HOT && BB->getKnownExecutionCount());
-    BinaryBasicBlock::const_iterator MacroFusionPair;
-    if (MayNeedMacroFusionAlignment) {
-      MacroFusionPair = BB->getMacroOpFusionPair();
-      if (MacroFusionPair == BB->end())
-        MayNeedMacroFusionAlignment = false;
-    }
-
     SMLoc LastLocSeen;
-    // Remember if the last instruction emitted was a prefix.
-    bool LastIsPrefix = false;
     for (auto I = BB->begin(), E = BB->end(); I != E; ++I) {
       MCInst &Instr = *I;
 
@@ -479,16 +453,6 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         continue;
       }
 
-      // Handle macro-fusion alignment. If we emitted a prefix as
-      // the last instruction, we should've already emitted the associated
-      // alignment hint, so don't emit it twice.
-      if (MayNeedMacroFusionAlignment && !LastIsPrefix &&
-          I == MacroFusionPair) {
-        // This assumes the second instruction in the macro-op pair will get
-        // assigned to its own MCRelaxableFragment. Since all JCC instructions
-        // are relaxable, we should be safe.
-      }
-
       if (!EmitCodeOnly) {
         // A symbol to be emitted before the instruction to mark its location.
         MCSymbol *InstrLabel = BC.MIB->getInstLabel(Instr);
@@ -525,7 +489,6 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
       }
 
       Streamer.emitInstruction(Instr, *BC.STI);
-      LastIsPrefix = BC.MIB->isPrefix(Instr);
     }
   }
 

bolt/lib/Core/BinaryFunction.cpp

@@ -2279,8 +2279,6 @@ void BinaryFunction::postProcessCFG() {
     postProcessBranches();
   }
 
-  calculateMacroOpFusionStats();
-
   // The final cleanup of intermediate structures.
   clearList(IgnoredBranches);
 
@@ -2297,29 +2295,6 @@ void BinaryFunction::postProcessCFG() {
          "invalid CFG detected after post-processing");
 }
 
-void BinaryFunction::calculateMacroOpFusionStats() {
-  if (!getBinaryContext().isX86())
-    return;
-  for (const BinaryBasicBlock &BB : blocks()) {
-    auto II = BB.getMacroOpFusionPair();
-    if (II == BB.end())
-      continue;
-
-    // Check offset of the second instruction.
-    // FIXME: arch-specific.
-    const uint32_t Offset = BC.MIB->getOffsetWithDefault(*std::next(II), 0);
-    if (!Offset || (getAddress() + Offset) % 64)
-      continue;
-
-    LLVM_DEBUG(dbgs() << "\nmissed macro-op fusion at address 0x"
-                      << Twine::utohexstr(getAddress() + Offset)
-                      << " in function " << *this << "; executed "
-                      << BB.getKnownExecutionCount() << " times.\n");
-    ++BC.Stats.MissedMacroFusionPairs;
-    BC.Stats.MissedMacroFusionExecCount += BB.getKnownExecutionCount();
-  }
-}
-
 void BinaryFunction::removeTagsFromProfile() {
   for (BinaryBasicBlock *BB : BasicBlocks) {
     if (BB->ExecutionCount == BinaryBasicBlock::COUNT_NO_PROFILE)

bolt/lib/Core/DebugData.cpp

@@ -177,6 +177,12 @@ uint64_t DebugRangesSectionWriter::getSectionOffset() {
   return SectionOffset;
 }
 
+void DebugRangesSectionWriter::appendToRangeBuffer(
+    const DebugBufferVector &CUBuffer) {
+  *RangesStream << CUBuffer;
+  SectionOffset = RangesBuffer->size();
+}
+
 DebugAddrWriter *DebugRangeListsSectionWriter::AddrWriter = nullptr;
 
 uint64_t DebugRangeListsSectionWriter::addRanges(