LLVM and SPIRV-LLVM-Translator pulldown (WW48)

LLVM: llvm/llvm-project@225648e
SPIRV-LLVM-Translator: KhronosGroup/SPIRV-LLVM-Translator@3770469

Author: bb-sycl

.ci/monolithic-linux.sh

@@ -49,7 +49,8 @@ cmake -S ${MONOREPO_ROOT}/llvm -B ${BUILD_DIR} \
       -D LLVM_ENABLE_LLD=ON \
       -D CMAKE_CXX_FLAGS=-gmlt \
       -D BOLT_CLANG_EXE=/usr/bin/clang \
-      -D LLVM_CCACHE_BUILD=ON
+      -D LLVM_CCACHE_BUILD=ON \
+      -D MLIR_ENABLE_BINDINGS_PYTHON=ON
 
 echo "--- ninja"
 # Targets are not escaped as they are passed as separate arguments.

.ci/monolithic-windows.sh

@@ -48,7 +48,8 @@ cmake -S ${MONOREPO_ROOT}/llvm -B ${BUILD_DIR} \
       -D LLVM_LIT_ARGS="-v --xunit-xml-output ${BUILD_DIR}/test-results.xml" \
       -D COMPILER_RT_BUILD_ORC=OFF \
       -D CMAKE_C_COMPILER_LAUNCHER=sccache \
-      -D CMAKE_CXX_COMPILER_LAUNCHER=sccache
+      -D CMAKE_CXX_COMPILER_LAUNCHER=sccache \
+      -D MLIR_ENABLE_BINDINGS_PYTHON=ON
 
 echo "--- ninja"
 # Targets are not escaped as they are passed as separate arguments.

.github/workflows/docs.yml

@@ -15,6 +15,9 @@ on:
     paths:
       - 'llvm/docs/**'
       - 'clang/docs/**'
+      - 'clang/include/clang/Basic/AttrDocs.td'
+      - 'clang/include/clang/Driver/ClangOptionDocs.td'
+      - 'clang/include/clang/Basic/DiagnosticDocs.td'
       - 'clang-tools-extra/docs/**'
       - 'lldb/docs/**'
       - 'libunwind/docs/**'
@@ -29,6 +32,9 @@ on:
     paths:
       - 'llvm/docs/**'
       - 'clang/docs/**'
+      - 'clang/include/clang/Basic/AttrDocs.td'
+      - 'clang/include/clang/Driver/ClangOptionDocs.td'
+      - 'clang/include/clang/Basic/DiagnosticDocs.td'
       - 'clang-tools-extra/docs/**'
       - 'lldb/docs/**'
       - 'libunwind/docs/**'
@@ -64,6 +70,9 @@ jobs:
               - 'llvm/docs/**'
             clang:
               - 'clang/docs/**'
+              - 'clang/include/clang/Basic/AttrDocs.td'
+              - 'clang/include/clang/Driver/ClangOptionDocs.td'
+              - 'clang/include/clang/Basic/DiagnosticDocs.td'
             clang-tools-extra:
               - 'clang-tools-extra/docs/**'
             lldb:

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

@@ -22,6 +22,12 @@ on:
       - 'runtimes/**'
       - 'cmake/**'
       - '.github/workflows/libcxx-build-and-test.yaml'
+  schedule:
+    # Run nightly at 8 AM UTC (or roughly 3 AM eastern)
+    - cron: '0 3 * * *'
+
+permissions:
+  contents: read # Default everything to read-only
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number }}
@@ -157,7 +163,11 @@ jobs:
           'generic-no-unicode',
           'generic-no-wide-characters',
           'generic-static',
-          'generic-with_llvm_unwinder'
+          'generic-with_llvm_unwinder',
+          # TODO Find a better place for the benchmark and bootstrapping builds to live. They're either very expensive
+          # or don't provide much value since the benchmark run results are too noise on the bots.
+          'benchmarks',
+          'bootstrapping-build'
         ]
         machine: [ 'libcxx-runners-8' ]
         std_modules: [ 'OFF' ]

.github/workflows/pr-code-format.yml

@@ -40,9 +40,11 @@ jobs:
           path: code-format-tools
 
       - name: "Listed files"
+        env:
+          CHANGED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
         run: |
           echo "Formatting files:"
-          echo "${{ steps.changed-files.outputs.all_changed_files }}"
+          echo "$CHANGED_FILES"
 
       - name: Install clang-format
         uses: aminya/setup-cpp@v1

.github/workflows/sync-release-repo.yml

@@ -611,6 +611,9 @@ class BinaryContext {
   /// Indicates if the binary contains split functions.
   bool HasSplitFunctions{false};
 
+  /// Indicates if the function ordering of the binary is finalized.
+  bool HasFinalizedFunctionOrder{false};
+
   /// Is the binary always loaded at a fixed address. Shared objects and
   /// position-independent executables (PIEs) are examples of binaries that
   /// will have HasFixedLoadAddress set to false.
@@ -1230,6 +1233,9 @@ class BinaryContext {
   ///
   /// Return the pair where the first size is for the main part, and the second
   /// size is for the cold one.
+  /// Modify BinaryBasicBlock::OutputAddressRange for each basic block in the
+  /// function in place so that BinaryBasicBlock::getOutputSize() gives the
+  /// emitted size of the basic block.
   std::pair<size_t, size_t> calculateEmittedSize(BinaryFunction &BF,
                                                  bool FixBranches = true);
 

bolt/include/bolt/Core/BinaryContext.h

@@ -273,7 +273,8 @@ class DIEBuilder {
   void buildCompileUnits(const std::vector<DWARFUnit *> &CUs);
   /// Preventing implicit conversions.
   template <class T> void buildCompileUnits(T) = delete;
-  void buildBoth();
+  /// Builds DWO Unit. For DWARF5 this includes the type units.
+  void buildDWOUnit(DWARFUnit &U);
 
   /// Returns DWARFUnitInfo for DWARFUnit
   DWARFUnitInfo &getUnitInfoByDwarfUnit(const DWARFUnit &DwarfUnit) {

bolt/include/bolt/Core/DIEBuilder.h

@@ -23,6 +23,9 @@ 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,
@@ -40,7 +43,7 @@ class SplitStrategy {
 
   virtual ~SplitStrategy() = default;
   virtual bool canSplit(const BinaryFunction &BF) = 0;
-  virtual bool keepEmpty() = 0;
+  virtual bool compactFragments() = 0;
   virtual void fragment(const BlockIt Start, const BlockIt End) = 0;
 };
 

bolt/include/bolt/Passes/SplitFunctions.h

@@ -2331,14 +2331,36 @@ BinaryContext::calculateEmittedSize(BinaryFunction &BF, bool FixBranches) {
   MCAsmLayout Layout(Assembler);
   Assembler.layout(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 ColdSize =
-      std::accumulate(SplitLabels.begin(), SplitLabels.end(), 0ULL,
-                      [&](const uint64_t Accu, const LabelRange &Labels) {
-                        return Accu + Layout.getSymbolOffset(*Labels.second) -
-                               Layout.getSymbolOffset(*Labels.first);
-                      });
+  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);
+    FragmentSizes.push_back(Size);
+    ColdSize += Size;
+  }
+
+  // Populate new start and end offsets of each basic block.
+  uint64_t FragmentIndex = 0;
+  for (FunctionFragment &FF : BF.getLayout().fragments()) {
+    BinaryBasicBlock *PrevBB = nullptr;
+    for (BinaryBasicBlock *BB : FF) {
+      const uint64_t BBStartOffset = Layout.getSymbolOffset(*(BB->getLabel()));
+      BB->setOutputStartAddress(BBStartOffset);
+      if (PrevBB)
+        PrevBB->setOutputEndAddress(BBStartOffset);
+      PrevBB = BB;
+    }
+    if (PrevBB)
+      PrevBB->setOutputEndAddress(FragmentSizes[FragmentIndex]);
+    FragmentIndex++;
+  }
 
   // Clean-up the effect of the code emission.
   for (const MCSymbol &Symbol : Assembler.symbols()) {

bolt/lib/Core/BinaryContext.cpp

@@ -40,6 +40,7 @@
 #include "llvm/Support/Regex.h"
 #include "llvm/Support/Timer.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/xxhash.h"
 #include <functional>
 #include <limits>
 #include <numeric>
@@ -108,6 +109,13 @@ cl::opt<bool>
                             cl::desc("try to preserve basic block alignment"),
                             cl::cat(BoltOptCategory));
 
+static cl::opt<bool> PrintOutputAddressRange(
+    "print-output-address-range",
+    cl::desc(
+        "print output address range for each basic block in the function when"
+        "BinaryFunction::print is called"),
+    cl::Hidden, cl::cat(BoltOptCategory));
+
 cl::opt<bool>
 PrintDynoStats("dyno-stats",
   cl::desc("print execution info based on profile"),
@@ -510,6 +518,11 @@ void BinaryFunction::print(raw_ostream &OS, std::string Annotation) {
       OS << BB->getName() << " (" << BB->size()
          << " instructions, align : " << BB->getAlignment() << ")\n";
 
+      if (opts::PrintOutputAddressRange)
+        OS << formatv("  Output Address Range: [{0:x}, {1:x}) ({2} bytes)\n",
+                      BB->getOutputAddressRange().first,
+                      BB->getOutputAddressRange().second, BB->getOutputSize());
+
       if (isEntryPoint(*BB)) {
         if (MCSymbol *EntrySymbol = getSecondaryEntryPointSymbol(*BB))
           OS << "  Secondary Entry Point: " << EntrySymbol->getName() << '\n';
@@ -3283,7 +3296,7 @@ void BinaryFunction::fixBranches() {
       BB->eraseInstruction(BB->findInstruction(UncondBranch));
 
     // Basic block that follows the current one in the final layout.
-    const BinaryBasicBlock *NextBB =
+    const BinaryBasicBlock *const NextBB =
         Layout.getBasicBlockAfter(BB, /*IgnoreSplits=*/false);
 
     if (BB->succ_size() == 1) {
@@ -3300,39 +3313,54 @@ void BinaryFunction::fixBranches() {
       assert(CondBranch && "conditional branch expected");
       const BinaryBasicBlock *TSuccessor = BB->getConditionalSuccessor(true);
       const BinaryBasicBlock *FSuccessor = BB->getConditionalSuccessor(false);
-      // Check whether we support reversing this branch direction
-      const bool IsSupported = !MIB->isUnsupportedBranch(*CondBranch);
-      if (NextBB && NextBB == TSuccessor && IsSupported) {
+
+      // Eliminate unnecessary conditional branch.
+      if (TSuccessor == FSuccessor) {
+        BB->removeDuplicateConditionalSuccessor(CondBranch);
+        if (TSuccessor != NextBB)
+          BB->addBranchInstruction(TSuccessor);
+        continue;
+      }
+
+      // Reverse branch condition and swap successors.
+      auto swapSuccessors = [&]() {
+        if (MIB->isUnsupportedBranch(*CondBranch))
+          return false;
         std::swap(TSuccessor, FSuccessor);
-        {
-          auto L = BC.scopeLock();
-          MIB->reverseBranchCondition(*CondBranch, TSuccessor->getLabel(), Ctx);
-        }
         BB->swapConditionalSuccessors();
-      } else {
+        auto L = BC.scopeLock();
+        MIB->reverseBranchCondition(*CondBranch, TSuccessor->getLabel(), Ctx);
+        return true;
+      };
+
+      // Check whether the next block is a "taken" target and try to swap it
+      // with a "fall-through" target.
+      if (TSuccessor == NextBB && swapSuccessors())
+        continue;
+
+      // Update conditional branch destination if needed.
+      if (MIB->getTargetSymbol(*CondBranch) != TSuccessor->getLabel()) {
         auto L = BC.scopeLock();
         MIB->replaceBranchTarget(*CondBranch, TSuccessor->getLabel(), Ctx);
       }
-      if (TSuccessor == FSuccessor)
-        BB->removeDuplicateConditionalSuccessor(CondBranch);
-      if (!NextBB ||
-          ((NextBB != TSuccessor || !IsSupported) && NextBB != FSuccessor)) {
-        // If one of the branches is guaranteed to be "long" while the other
-        // could be "short", then prioritize short for "taken". This will
-        // generate a sequence 1 byte shorter on x86.
-        if (IsSupported && BC.isX86() &&
-            TSuccessor->getFragmentNum() != FSuccessor->getFragmentNum() &&
-            BB->getFragmentNum() != TSuccessor->getFragmentNum()) {
-          std::swap(TSuccessor, FSuccessor);
-          {
-            auto L = BC.scopeLock();
-            MIB->reverseBranchCondition(*CondBranch, TSuccessor->getLabel(),
-                                        Ctx);
-          }
-          BB->swapConditionalSuccessors();
-        }
-        BB->addBranchInstruction(FSuccessor);
+
+      // No need for the unconditional branch.
+      if (FSuccessor == NextBB)
+        continue;
+
+      if (BC.isX86()) {
+        // We are going to generate two branches. Check if their targets are in
+        // the same fragment as this block. If only one target is in the same
+        // fragment, make it the destination of the conditional branch. There
+        // is a chance it will be a short branch which takes 5 bytes fewer than
+        // a long conditional branch. For unconditional branch, the difference
+        // is 4 bytes.
+        if (BB->getFragmentNum() != TSuccessor->getFragmentNum() &&
+            BB->getFragmentNum() == FSuccessor->getFragmentNum())
+          swapSuccessors();
       }
+
+      BB->addBranchInstruction(FSuccessor);
     }
     // Cases where the number of successors is 0 (block ends with a
     // terminator) or more than 2 (switch table) don't require branch
@@ -3624,7 +3652,7 @@ size_t BinaryFunction::computeHash(bool UseDFS,
   for (const BinaryBasicBlock *BB : Order)
     HashString.append(hashBlock(BC, *BB, OperandHashFunc));
 
-  return Hash = std::hash<std::string>{}(HashString);
+  return Hash = llvm::xxh3_64bits(HashString);
 }
 
 void BinaryFunction::insertBasicBlocks(
@@ -4131,10 +4159,6 @@ void BinaryFunction::updateOutputValues(const BOLTLinker &Linker) {
   if (!requiresAddressMap())
     return;
 
-  // Output ranges should match the input if the body hasn't changed.
-  if (!isSimple() && !BC.HasRelocations)
-    return;
-
   // AArch64 may have functions that only contains a constant island (no code).
   if (getLayout().block_empty())
     return;
@@ -4182,6 +4206,12 @@ void BinaryFunction::updateOutputValues(const BOLTLinker &Linker) {
                                     ? FF.getAddress() + FF.getImageSize()
                                     : getOutputAddress() + getOutputSize());
   }
+
+  // Reset output addresses for deleted blocks.
+  for (BinaryBasicBlock *BB : DeletedBasicBlocks) {
+    BB->setOutputStartAddress(0);
+    BB->setOutputEndAddress(0);
+  }
 }
 
 DebugAddressRangesVector BinaryFunction::getOutputAddressRanges() const {

bolt/lib/Core/BinaryFunction.cpp

@@ -193,12 +193,6 @@ void DIEBuilder::buildTypeUnits(const bool Init) {
   if (Init)
     BuilderState.reset(new State());
 
-  unsigned int CUNum = getCUNum(DwarfContext, IsDWO);
-  getState().CloneUnitCtxMap.resize(CUNum);
-  DWARFContext::unit_iterator_range CU4TURanges =
-      IsDWO ? DwarfContext->dwo_types_section_units()
-            : DwarfContext->types_section_units();
-
   const DWARFUnitIndex &TUIndex = DwarfContext->getTUIndex();
   if (!TUIndex.getRows().empty()) {
     for (auto &Row : TUIndex.getRows()) {
@@ -208,6 +202,11 @@ void DIEBuilder::buildTypeUnits(const bool Init) {
                                        true);
     }
   }
+  const unsigned int CUNum = getCUNum(DwarfContext, IsDWO);
+  getState().CloneUnitCtxMap.resize(CUNum);
+  DWARFContext::unit_iterator_range CU4TURanges =
+      IsDWO ? DwarfContext->dwo_types_section_units()
+            : DwarfContext->types_section_units();
 
   getState().Type = ProcessingType::DWARF4TUs;
   for (std::unique_ptr<DWARFUnit> &DU : CU4TURanges)
@@ -278,11 +277,13 @@ void DIEBuilder::buildCompileUnits(const std::vector<DWARFUnit *> &CUs) {
     constructFromUnit(*DU);
 }
 
-void DIEBuilder::buildBoth() {
+void DIEBuilder::buildDWOUnit(DWARFUnit &U) {
   BuilderState.release();
   BuilderState = std::make_unique<State>();
   buildTypeUnits(false);
-  buildCompileUnits(false);
+  getState().Type = ProcessingType::CUs;
+  registerUnit(U, false);
+  constructFromUnit(U);
 }
 
 DIE *DIEBuilder::constructDIEFast(DWARFDie &DDie, DWARFUnit &U,