Manually merged main:bafda89a0944d947fc4b3b5663185e07a397ac30 into amd-gfx:2e4aa61a46dc

Local branch amd-gfx 2e4aa61 Merged main:5434b04234a86273d719f96e9efa6332a53c0eba into amd-gfx:17b97598a1e4
Remote branch main bafda89 [X86] Use generic CPU tuning when tune-cpu is empty (llvm#83631)

Author: perlfu

bolt/CMakeLists.txt

@@ -1,3 +1,5 @@
+set(LLVM_SUBPROJECT_TITLE "BOLT")
+
 include(ExternalProject)
 
 set(BOLT_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR})
@@ -121,7 +123,7 @@ option(BOLT_BUILD_TOOLS
   "Build the BOLT tools. If OFF, just generate build targets." ON)
 
 add_custom_target(bolt)
-set_target_properties(bolt PROPERTIES FOLDER "BOLT")
+set_target_properties(bolt PROPERTIES FOLDER "BOLT/Metatargets")
 add_llvm_install_targets(install-bolt DEPENDS bolt COMPONENT bolt)
 
 include_directories(

bolt/cmake/modules/AddBOLT.cmake

@@ -3,7 +3,6 @@ include(LLVMDistributionSupport)
 
 macro(add_bolt_executable name)
   add_llvm_executable(${name} ${ARGN})
-  set_target_properties(${name} PROPERTIES FOLDER "BOLT")
 endmacro()
 
 macro(add_bolt_tool name)

bolt/docs/CMakeLists.txt

@@ -79,6 +79,7 @@ if (LLVM_ENABLE_DOXYGEN)
     COMMAND ${DOXYGEN_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/doxygen.cfg
     WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
     COMMENT "Generating bolt doxygen documentation." VERBATIM)
+  set_target_properties(doxygen-bolt PROPERTIES FOLDER "BOLT/Docs")
 
   if (LLVM_BUILD_DOCS)
     add_dependencies(doxygen doxygen-bolt)

bolt/include/bolt/Core/BinaryContext.h

@@ -1224,8 +1224,7 @@ class BinaryContext {
 
   /// Return a signed value of \p Size stored at \p Address. The address has
   /// to be a valid statically allocated address for the binary.
-  ErrorOr<uint64_t> getSignedValueAtAddress(uint64_t Address,
-                                            size_t Size) const;
+  ErrorOr<int64_t> getSignedValueAtAddress(uint64_t Address, size_t Size) const;
 
   /// Special case of getUnsignedValueAtAddress() that uses a pointer size.
   ErrorOr<uint64_t> getPointerAtAddress(uint64_t Address) const {

bolt/lib/Core/BinaryContext.cpp

@@ -2217,8 +2217,8 @@ ErrorOr<uint64_t> BinaryContext::getUnsignedValueAtAddress(uint64_t Address,
   return DE.getUnsigned(&ValueOffset, Size);
 }
 
-ErrorOr<uint64_t> BinaryContext::getSignedValueAtAddress(uint64_t Address,
-                                                         size_t Size) const {
+ErrorOr<int64_t> BinaryContext::getSignedValueAtAddress(uint64_t Address,
+                                                        size_t Size) const {
   const ErrorOr<const BinarySection &> Section = getSectionForAddress(Address);
   if (!Section)
     return std::make_error_code(std::errc::bad_address);

bolt/lib/Core/BinaryFunction.cpp

@@ -851,15 +851,19 @@ BinaryFunction::processIndirectBranch(MCInst &Instruction, unsigned Size,
     return IndirectBranchType::UNKNOWN;
   }
 
-  // RIP-relative addressing should be converted to symbol form by now
-  // in processed instructions (but not in jump).
-  if (DispExpr) {
+  auto getExprValue = [&](const MCExpr *Expr) {
     const MCSymbol *TargetSym;
     uint64_t TargetOffset;
-    std::tie(TargetSym, TargetOffset) = BC.MIB->getTargetSymbolInfo(DispExpr);
+    std::tie(TargetSym, TargetOffset) = BC.MIB->getTargetSymbolInfo(Expr);
     ErrorOr<uint64_t> SymValueOrError = BC.getSymbolValue(*TargetSym);
-    assert(SymValueOrError && "global symbol needs a value");
-    ArrayStart = *SymValueOrError + TargetOffset;
+    assert(SymValueOrError && "Global symbol needs a value");
+    return *SymValueOrError + TargetOffset;
+  };
+
+  // RIP-relative addressing should be converted to symbol form by now
+  // in processed instructions (but not in jump).
+  if (DispExpr) {
+    ArrayStart = getExprValue(DispExpr);
     BaseRegNum = BC.MIB->getNoRegister();
     if (BC.isAArch64()) {
       ArrayStart &= ~0xFFFULL;
@@ -3698,6 +3702,13 @@ BinaryFunction::BasicBlockListType BinaryFunction::dfs() const {
 
 size_t BinaryFunction::computeHash(bool UseDFS, HashFunction HashFunction,
                                    OperandHashFuncTy OperandHashFunc) const {
+  LLVM_DEBUG({
+    dbgs() << "BOLT-DEBUG: computeHash " << getPrintName() << ' '
+           << (UseDFS ? "dfs" : "bin") << " order "
+           << (HashFunction == HashFunction::StdHash ? "std::hash" : "xxh3")
+           << '\n';
+  });
+
   if (size() == 0)
     return 0;
 

bolt/lib/Core/DebugNames.cpp

@@ -112,8 +112,6 @@ void DWARF5AcceleratorTable::addUnit(DWARFUnit &Unit,
 // Returns true if DW_TAG_variable should be included in .debug-names based on
 // section 6.1.1.1 for DWARF5 spec.
 static bool shouldIncludeVariable(const DWARFUnit &Unit, const DIE &Die) {
-  if (Die.findAttribute(dwarf::Attribute::DW_AT_declaration))
-    return false;
   const DIEValue LocAttrInfo =
       Die.findAttribute(dwarf::Attribute::DW_AT_location);
   if (!LocAttrInfo)
@@ -148,6 +146,8 @@ static bool shouldIncludeVariable(const DWARFUnit &Unit, const DIE &Die) {
 
 bool static canProcess(const DWARFUnit &Unit, const DIE &Die,
                        std::string &NameToUse, const bool TagsOnly) {
+  if (Die.findAttribute(dwarf::Attribute::DW_AT_declaration))
+    return false;
   switch (Die.getTag()) {
   case dwarf::DW_TAG_base_type:
   case dwarf::DW_TAG_class_type:

bolt/lib/Profile/YAMLProfileReader.cpp

@@ -102,11 +102,14 @@ bool YAMLProfileReader::parseFunctionProfile(
   if (BF.empty())
     return true;
 
-  if (!opts::IgnoreHash &&
-      YamlBF.Hash != BF.computeHash(IsDFSOrder, HashFunction)) {
-    if (opts::Verbosity >= 1)
-      errs() << "BOLT-WARNING: function hash mismatch\n";
-    ProfileMatched = false;
+  if (!opts::IgnoreHash) {
+    if (!BF.getHash())
+      BF.computeHash(IsDFSOrder, HashFunction);
+    if (YamlBF.Hash != BF.getHash()) {
+      if (opts::Verbosity >= 1)
+        errs() << "BOLT-WARNING: function hash mismatch\n";
+      ProfileMatched = false;
+    }
   }
 
   if (YamlBF.NumBasicBlocks != BF.size()) {

bolt/lib/Rewrite/DWARFRewriter.cpp

@@ -73,8 +73,7 @@ static void printDie(DWARFUnit &DU, uint64_t DIEOffset) {
   DWARFDataExtractor DebugInfoData = DU.getDebugInfoExtractor();
   DWARFDebugInfoEntry DIEEntry;
   if (DIEEntry.extractFast(DU, &DIEOffset, DebugInfoData, NextCUOffset, 0)) {
-    if (const DWARFAbbreviationDeclaration *AbbrDecl =
-            DIEEntry.getAbbreviationDeclarationPtr()) {
+    if (DIEEntry.getAbbreviationDeclarationPtr()) {
       DWARFDie DDie(&DU, &DIEEntry);
       printDie(DDie);
     } else {

bolt/lib/Rewrite/LinuxKernelRewriter.cpp

@@ -393,7 +393,7 @@ void LinuxKernelRewriter::processLKKSymtab(bool IsGPL) {
 
   for (uint64_t I = 0; I < SectionSize; I += 4) {
     const uint64_t EntryAddress = SectionAddress + I;
-    ErrorOr<uint64_t> Offset = BC.getSignedValueAtAddress(EntryAddress, 4);
+    ErrorOr<int64_t> Offset = BC.getSignedValueAtAddress(EntryAddress, 4);
     assert(Offset && "Reading valid PC-relative offset for a ksymtab entry");
     const int32_t SignedOffset = *Offset;
     const uint64_t RefAddress = EntryAddress + SignedOffset;

bolt/lib/Target/X86/X86MCPlusBuilder.cpp

@@ -1932,6 +1932,19 @@ class X86MCPlusBuilder : public MCPlusBuilder {
     //    = R_X86_64_PC32(Ln) + En - JT
     //    = R_X86_64_PC32(Ln + offsetof(En))
     //
+    auto isRIPRel = [&](X86MemOperand &MO) {
+      // NB: DispExpr should be set
+      return MO.DispExpr != nullptr &&
+             MO.BaseRegNum == RegInfo->getProgramCounter() &&
+             MO.IndexRegNum == X86::NoRegister &&
+             MO.SegRegNum == X86::NoRegister;
+    };
+    auto isIndexed = [](X86MemOperand &MO, MCPhysReg R) {
+      // NB: IndexRegNum should be set.
+      return MO.IndexRegNum != X86::NoRegister && MO.BaseRegNum == R &&
+             MO.ScaleImm == 4 && MO.DispImm == 0 &&
+             MO.SegRegNum == X86::NoRegister;
+    };
     LLVM_DEBUG(dbgs() << "Checking for PIC jump table\n");
     MCInst *MemLocInstr = nullptr;
     const MCInst *MovInstr = nullptr;
@@ -1965,9 +1978,8 @@ class X86MCPlusBuilder : public MCPlusBuilder {
         std::optional<X86MemOperand> MO = evaluateX86MemoryOperand(Instr);
         if (!MO)
           break;
-        if (MO->BaseRegNum != R1 || MO->ScaleImm != 4 ||
-            MO->IndexRegNum == X86::NoRegister || MO->DispImm != 0 ||
-            MO->SegRegNum != X86::NoRegister)
+        if (!isIndexed(*MO, R1))
+          // POSSIBLE_PIC_JUMP_TABLE
           break;
         MovInstr = &Instr;
       } else {
@@ -1986,9 +1998,7 @@ class X86MCPlusBuilder : public MCPlusBuilder {
         std::optional<X86MemOperand> MO = evaluateX86MemoryOperand(Instr);
         if (!MO)
           break;
-        if (MO->BaseRegNum != RegInfo->getProgramCounter() ||
-            MO->IndexRegNum != X86::NoRegister ||
-            MO->SegRegNum != X86::NoRegister || MO->DispExpr == nullptr)
+        if (!isRIPRel(*MO))
           break;
         MemLocInstr = &Instr;
         break;
@@ -2105,13 +2115,15 @@ class X86MCPlusBuilder : public MCPlusBuilder {
       return IndirectBranchType::POSSIBLE_FIXED_BRANCH;
     }
 
-    if (Type == IndirectBranchType::POSSIBLE_PIC_JUMP_TABLE &&
-        (MO->ScaleImm != 1 || MO->BaseRegNum != RIPRegister))
-      return IndirectBranchType::UNKNOWN;
-
-    if (Type != IndirectBranchType::POSSIBLE_PIC_JUMP_TABLE &&
-        MO->ScaleImm != PtrSize)
-      return IndirectBranchType::UNKNOWN;
+    switch (Type) {
+    case IndirectBranchType::POSSIBLE_PIC_JUMP_TABLE:
+      if (MO->ScaleImm != 1 || MO->BaseRegNum != RIPRegister)
+        return IndirectBranchType::UNKNOWN;
+      break;
+    default:
+      if (MO->ScaleImm != PtrSize)
+        return IndirectBranchType::UNKNOWN;
+    }
 
     MemLocInstrOut = MemLocInstr;
 

bolt/test/CMakeLists.txt

@@ -56,15 +56,14 @@ list(APPEND BOLT_TEST_DEPS
   )
 
 add_custom_target(bolt-test-depends DEPENDS ${BOLT_TEST_DEPS})
-set_target_properties(bolt-test-depends PROPERTIES FOLDER "BOLT")
+set_target_properties(bolt-test-depends PROPERTIES FOLDER "BOLT/Tests")
 
 add_lit_testsuite(check-bolt "Running the BOLT regression tests"
   ${CMAKE_CURRENT_BINARY_DIR}
   PARAMS ${BOLT_TEST_PARAMS}
   DEPENDS ${BOLT_TEST_DEPS}
   ARGS ${BOLT_TEST_EXTRA_ARGS}
   )
-set_target_properties(check-bolt PROPERTIES FOLDER "BOLT")
 
 add_lit_testsuites(BOLT ${CMAKE_CURRENT_SOURCE_DIR}
   PARAMS ${BOLT_TEST_PARAMS}