AMDGPU stepthomas atomic csub no rtn forms ver2

.github/CODEOWNERS

@@ -1,4 +1,34 @@
+# This file lists reviewers that are auto-assigned when a pull request modifies
+# certain files or directories. If you add yourself to this file, you commit to
+# reviewing a large fraction of pull requests in the relevant area.
+#
+# The GitHub "code owners" mechanism is used exclusively to auto-assign
+# reviewers and does not carry significance beyond that. It is not necessary
+# to receive an approval from a "code owner" in particular -- any LLVM project
+# member can approve pull requests.
+#
+# Note that GitHub's concept of "code owner" is independent from LLVM's own
+# "code owner" concept, they merely happen to share terminology. See
+# https://llvm.org/docs/DeveloperPolicy.html#code-owners, as well as the
+# CODE_OWNERS.txt files in the respective subproject directories.
+
 /libcxx/ @llvm/reviewers-libcxx
 /libcxxabi/ @llvm/reviewers-libcxxabi
 /libunwind/ @llvm/reviewers-libunwind
 /runtimes/ @llvm/reviewers-libcxx
+
+/llvm/lib/Analysis/BasicAliasAnalysis.cpp @nikic
+/llvm/lib/Analysis/InstructionSimplify.cpp @nikic
+/llvm/lib/Analysis/LazyValueInfo.cpp @nikic
+/llvm/lib/Analysis/ScalarEvolution.cpp @nikic
+/llvm/lib/Analysis/ValueTracking.cpp @nikic
+/llvm/lib/IR/ConstantRange.cpp @nikic
+/llvm/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp @nikic
+/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp @nikic
+/llvm/lib/Transforms/InstCombine/ @nikic
+
+/clang/test/CXX/drs/ @Endilll
+/clang/www/cxx_dr_status.html @Endilll
+/clang/www/make_cxx_dr_status @Endilll
+
+/lldb/ @JDevlieghere

bolt/include/bolt/Core/MCPlus.h

@@ -66,6 +66,7 @@ class MCAnnotation {
     kTailCall,            /// Tail call.
     kConditionalTailCall, /// CTC.
     kOffset,              /// Offset in the function.
+    kLabel,               /// MCSymbol pointing to this instruction.
     kGeneric              /// First generic annotation.
   };
 

bolt/include/bolt/Core/MCPlusBuilder.h

@@ -160,6 +160,7 @@ class MCPlusBuilder {
   const MCInstrAnalysis *Analysis;
   const MCInstrInfo *Info;
   const MCRegisterInfo *RegInfo;
+  const MCSubtargetInfo *STI;
 
   /// Map annotation name into an annotation index.
   StringMap<uint64_t> AnnotationNameIndexMap;
@@ -331,8 +332,8 @@ class MCPlusBuilder {
 
 public:
   MCPlusBuilder(const MCInstrAnalysis *Analysis, const MCInstrInfo *Info,
-                const MCRegisterInfo *RegInfo)
-      : Analysis(Analysis), Info(Info), RegInfo(RegInfo) {
+                const MCRegisterInfo *RegInfo, const MCSubtargetInfo *STI)
+      : Analysis(Analysis), Info(Info), RegInfo(RegInfo), STI(STI) {
     // Initialize the default annotation allocator with id 0
     AnnotationAllocators.emplace(0, AnnotationAllocator());
     MaxAllocatorId++;
@@ -1179,6 +1180,13 @@ class MCPlusBuilder {
   /// Remove offset annotation.
   bool clearOffset(MCInst &Inst);
 
+  /// Return the label of \p Inst, if available.
+  std::optional<MCSymbol *> getLabel(const MCInst &Inst) const;
+
+  /// Set the label of \p Inst. This label will be emitted right before \p Inst
+  /// is emitted to MCStreamer.
+  bool setLabel(MCInst &Inst, MCSymbol *Label);
+
   /// Return MCSymbol that represents a target of this instruction at a given
   /// operand number \p OpNum. If there's no symbol associated with
   /// the operand - return nullptr.
@@ -2079,15 +2087,18 @@ class MCPlusBuilder {
 
 MCPlusBuilder *createX86MCPlusBuilder(const MCInstrAnalysis *,
                                       const MCInstrInfo *,
-                                      const MCRegisterInfo *);
+                                      const MCRegisterInfo *,
+                                      const MCSubtargetInfo *);
 
 MCPlusBuilder *createAArch64MCPlusBuilder(const MCInstrAnalysis *,
                                           const MCInstrInfo *,
-                                          const MCRegisterInfo *);
+                                          const MCRegisterInfo *,
+                                          const MCSubtargetInfo *);
 
 MCPlusBuilder *createRISCVMCPlusBuilder(const MCInstrAnalysis *,
                                         const MCInstrInfo *,
-                                        const MCRegisterInfo *);
+                                        const MCRegisterInfo *,
+                                        const MCSubtargetInfo *);
 
 } // namespace bolt
 } // namespace llvm

bolt/include/bolt/Core/Relocation.h

@@ -98,6 +98,10 @@ struct Relocation {
   /// Return true if relocation type is for thread local storage.
   static bool isTLS(uint64_t Type);
 
+  /// Return true of relocation type is for referencing a specific instruction
+  /// (as opposed to a function, basic block, etc).
+  static bool isInstructionReference(uint64_t Type);
+
   /// Return code for a NONE relocation
   static uint64_t getNone();
 

bolt/include/bolt/Rewrite/RewriteInstance.h

@@ -584,7 +584,8 @@ class RewriteInstance {
 MCPlusBuilder *createMCPlusBuilder(const Triple::ArchType Arch,
                                    const MCInstrAnalysis *Analysis,
                                    const MCInstrInfo *Info,
-                                   const MCRegisterInfo *RegInfo);
+                                   const MCRegisterInfo *RegInfo,
+                                   const MCSubtargetInfo *STI);
 
 } // namespace bolt
 } // namespace llvm

bolt/lib/Core/BinaryContext.cpp

@@ -1888,6 +1888,8 @@ void BinaryContext::printInstruction(raw_ostream &OS, const MCInst &Instruction,
   }
   if (std::optional<uint32_t> Offset = MIB->getOffset(Instruction))
     OS << " # Offset: " << *Offset;
+  if (auto Label = MIB->getLabel(Instruction))
+    OS << " # Label: " << **Label;
 
   MIB->printAnnotations(Instruction, OS);
 

bolt/lib/Core/BinaryEmitter.cpp

@@ -498,6 +498,9 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         BB->getLocSyms().emplace_back(Offset, LocSym);
       }
 
+      if (auto Label = BC.MIB->getLabel(Instr))
+        Streamer.emitLabel(*Label);
+
       Streamer.emitInstruction(Instr, *BC.STI);
       LastIsPrefix = BC.MIB->isPrefix(Instr);
     }

bolt/lib/Core/BinaryFunction.cpp

@@ -1173,6 +1173,13 @@ bool BinaryFunction::disassemble() {
   // basic block.
   Labels[0] = Ctx->createNamedTempSymbol("BB0");
 
+  // Map offsets in the function to a label that should always point to the
+  // corresponding instruction. This is used for labels that shouldn't point to
+  // the start of a basic block but always to a specific instruction. This is
+  // used, for example, on RISC-V where %pcrel_lo relocations point to the
+  // corresponding %pcrel_hi.
+  LabelsMapType InstructionLabels;
+
   uint64_t Size = 0; // instruction size
   for (uint64_t Offset = 0; Offset < getSize(); Offset += Size) {
     MCInst Instruction;
@@ -1329,9 +1336,23 @@ bool BinaryFunction::disassemble() {
                 ItrE = Relocations.lower_bound(Offset + Size);
            Itr != ItrE; ++Itr) {
         const Relocation &Relocation = Itr->second;
+        MCSymbol *Symbol = Relocation.Symbol;
+
+        if (Relocation::isInstructionReference(Relocation.Type)) {
+          uint64_t RefOffset = Relocation.Value - getAddress();
+          LabelsMapType::iterator LI = InstructionLabels.find(RefOffset);
+
+          if (LI == InstructionLabels.end()) {
+            Symbol = BC.Ctx->createNamedTempSymbol();
+            InstructionLabels.emplace(RefOffset, Symbol);
+          } else {
+            Symbol = LI->second;
+          }
+        }
+
         int64_t Value = Relocation.Value;
         const bool Result = BC.MIB->replaceImmWithSymbolRef(
-            Instruction, Relocation.Symbol, Relocation.Addend, Ctx.get(), Value,
+            Instruction, Symbol, Relocation.Addend, Ctx.get(), Value,
             Relocation.Type);
         (void)Result;
         assert(Result && "cannot replace immediate with relocation");
@@ -1366,6 +1387,13 @@ bool BinaryFunction::disassemble() {
     addInstruction(Offset, std::move(Instruction));
   }
 
+  for (auto [Offset, Label] : InstructionLabels) {
+    InstrMapType::iterator II = Instructions.find(Offset);
+    assert(II != Instructions.end() && "reference to non-existing instruction");
+
+    BC.MIB->setLabel(II->second, Label);
+  }
+
   // Reset symbolizer for the disassembler.
   BC.SymbolicDisAsm->setSymbolizer(nullptr);
 
@@ -1761,7 +1789,8 @@ bool BinaryFunction::postProcessIndirectBranches(
   uint64_t LastJT = 0;
   uint16_t LastJTIndexReg = BC.MIB->getNoRegister();
   for (BinaryBasicBlock &BB : blocks()) {
-    for (MCInst &Instr : BB) {
+    for (BinaryBasicBlock::iterator II = BB.begin(); II != BB.end(); ++II) {
+      MCInst &Instr = *II;
       if (!BC.MIB->isIndirectBranch(Instr))
         continue;
 
@@ -1789,7 +1818,7 @@ bool BinaryFunction::postProcessIndirectBranches(
         const MCExpr *DispExpr;
         MCInst *PCRelBaseInstr;
         IndirectBranchType Type = BC.MIB->analyzeIndirectBranch(
-            Instr, BB.begin(), BB.end(), PtrSize, MemLocInstr, BaseRegNum,
+            Instr, BB.begin(), II, PtrSize, MemLocInstr, BaseRegNum,
             IndexRegNum, DispValue, DispExpr, PCRelBaseInstr);
         if (Type != IndirectBranchType::UNKNOWN || MemLocInstr != nullptr)
           continue;
@@ -4488,7 +4517,7 @@ void BinaryFunction::addRelocation(uint64_t Address, MCSymbol *Symbol,
   uint64_t Offset = Address - getAddress();
   LLVM_DEBUG(dbgs() << "BOLT-DEBUG: addRelocation in "
                     << formatv("{0}@{1:x} against {2}\n", *this, Offset,
-                               Symbol->getName()));
+                               (Symbol ? Symbol->getName() : "<undef>")));
   bool IsCI = BC.isAArch64() && isInConstantIsland(Address);
   std::map<uint64_t, Relocation> &Rels =
       IsCI ? Islands->Relocations : Relocations;

bolt/lib/Core/MCPlusBuilder.cpp

@@ -268,6 +268,17 @@ bool MCPlusBuilder::clearOffset(MCInst &Inst) {
   return true;
 }
 
+std::optional<MCSymbol *> MCPlusBuilder::getLabel(const MCInst &Inst) const {
+  if (auto Label = tryGetAnnotationAs<MCSymbol *>(Inst, MCAnnotation::kLabel))
+    return *Label;
+  return std::nullopt;
+}
+
+bool MCPlusBuilder::setLabel(MCInst &Inst, MCSymbol *Label) {
+  getOrCreateAnnotationAs<MCSymbol *>(Inst, MCAnnotation::kLabel) = Label;
+  return true;
+}
+
 bool MCPlusBuilder::hasAnnotation(const MCInst &Inst, unsigned Index) const {
   const MCInst *AnnotationInst = getAnnotationInst(Inst);
   if (!AnnotationInst)

bolt/lib/Core/Relocation.cpp

@@ -20,6 +20,13 @@
 using namespace llvm;
 using namespace bolt;
 
+namespace ELFReserved {
+enum {
+  R_RISCV_TPREL_I = 49,
+  R_RISCV_TPREL_S = 50,
+};
+} // namespace ELFReserved
+
 Triple::ArchType Relocation::Arch;
 
 static bool isSupportedX86(uint64_t Type) {
@@ -111,6 +118,13 @@ static bool isSupportedRISCV(uint64_t Type) {
   case ELF::R_RISCV_LO12_I:
   case ELF::R_RISCV_LO12_S:
   case ELF::R_RISCV_64:
+  case ELF::R_RISCV_TLS_GOT_HI20:
+  case ELF::R_RISCV_TPREL_HI20:
+  case ELF::R_RISCV_TPREL_ADD:
+  case ELF::R_RISCV_TPREL_LO12_I:
+  case ELF::R_RISCV_TPREL_LO12_S:
+  case ELFReserved::R_RISCV_TPREL_I:
+  case ELFReserved::R_RISCV_TPREL_S:
     return true;
   }
 }
@@ -214,6 +228,7 @@ static size_t getSizeForTypeRISCV(uint64_t Type) {
     return 4;
   case ELF::R_RISCV_64:
   case ELF::R_RISCV_GOT_HI20:
+  case ELF::R_RISCV_TLS_GOT_HI20:
     // See extractValueRISCV for why this is necessary.
     return 8;
   }
@@ -532,6 +547,7 @@ static uint64_t extractValueRISCV(uint64_t Type, uint64_t Contents,
   case ELF::R_RISCV_BRANCH:
     return extractBImmRISCV(Contents);
   case ELF::R_RISCV_GOT_HI20:
+  case ELF::R_RISCV_TLS_GOT_HI20:
     // We need to know the exact address of the GOT entry so we extract the
     // value from both the AUIPC and L[D|W]. We cannot rely on the symbol in the
     // relocation for this since it simply refers to the object that is stored
@@ -600,6 +616,7 @@ static bool isGOTRISCV(uint64_t Type) {
   default:
     return false;
   case ELF::R_RISCV_GOT_HI20:
+  case ELF::R_RISCV_TLS_GOT_HI20:
     return true;
   }
 }
@@ -636,6 +653,14 @@ static bool isTLSRISCV(uint64_t Type) {
   switch (Type) {
   default:
     return false;
+  case ELF::R_RISCV_TLS_GOT_HI20:
+  case ELF::R_RISCV_TPREL_HI20:
+  case ELF::R_RISCV_TPREL_ADD:
+  case ELF::R_RISCV_TPREL_LO12_I:
+  case ELF::R_RISCV_TPREL_LO12_S:
+  case ELFReserved::R_RISCV_TPREL_I:
+  case ELFReserved::R_RISCV_TPREL_S:
+    return true;
   }
 }
 
@@ -733,6 +758,7 @@ static bool isPCRelativeRISCV(uint64_t Type) {
   case ELF::R_RISCV_RVC_JUMP:
   case ELF::R_RISCV_RVC_BRANCH:
   case ELF::R_RISCV_32_PCREL:
+  case ELF::R_RISCV_TLS_GOT_HI20:
     return true;
   }
 }
@@ -832,6 +858,19 @@ bool Relocation::isTLS(uint64_t Type) {
   return isTLSX86(Type);
 }
 
+bool Relocation::isInstructionReference(uint64_t Type) {
+  if (Arch != Triple::riscv64)
+    return false;
+
+  switch (Type) {
+  default:
+    return false;
+  case ELF::R_RISCV_PCREL_LO12_I:
+  case ELF::R_RISCV_PCREL_LO12_S:
+    return true;
+  }
+}
+
 uint64_t Relocation::getNone() {
   if (Arch == Triple::aarch64)
     return ELF::R_AARCH64_NONE;

bolt/lib/Passes/BinaryPasses.cpp

@@ -575,6 +575,7 @@ bool CheckLargeFunctions::shouldOptimize(const BinaryFunction &BF) const {
 
 void LowerAnnotations::runOnFunctions(BinaryContext &BC) {
   std::vector<std::pair<MCInst *, uint32_t>> PreservedOffsetAnnotations;
+  std::vector<std::pair<MCInst *, MCSymbol *>> PreservedLabelAnnotations;
 
   for (auto &It : BC.getBinaryFunctions()) {
     BinaryFunction &BF = It.second;
@@ -609,6 +610,8 @@ void LowerAnnotations::runOnFunctions(BinaryContext &BC) {
           if (BF.requiresAddressTranslation() && BC.MIB->getOffset(*II))
             PreservedOffsetAnnotations.emplace_back(&(*II),
                                                     *BC.MIB->getOffset(*II));
+          if (auto Label = BC.MIB->getLabel(*II))
+            PreservedLabelAnnotations.emplace_back(&*II, *Label);
           BC.MIB->stripAnnotations(*II);
         }
       }
@@ -625,6 +628,8 @@ void LowerAnnotations::runOnFunctions(BinaryContext &BC) {
   // Reinsert preserved annotations we need during code emission.
   for (const std::pair<MCInst *, uint32_t> &Item : PreservedOffsetAnnotations)
     BC.MIB->setOffset(*Item.first, Item.second);
+  for (auto [Instr, Label] : PreservedLabelAnnotations)
+    BC.MIB->setLabel(*Instr, Label);
 }
 
 // Check for dirty state in MCSymbol objects that might be a consequence

bolt/lib/Passes/FixRISCVCallsPass.cpp

@@ -43,7 +43,12 @@ void FixRISCVCallsPass::runOnFunction(BinaryFunction &BF) {
 
         MCInst OldCall = *NextII;
         auto L = BC.scopeLock();
-        MIB->createCall(*II, Target, Ctx);
+
+        if (MIB->isTailCall(*NextII))
+          MIB->createTailCall(*II, Target, Ctx);
+        else
+          MIB->createCall(*II, Target, Ctx);
+
         MIB->moveAnnotations(std::move(OldCall), *II);
 
         // The original offset was set on the jalr of the auipc+jalr pair. Since