Merge branch 'upstream' into sse41-scalar-blendv

Author: RKSimon

.github/workflows/libc-fullbuild-tests.yml

@@ -15,6 +15,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
+        build_type: [Debug, Release, MinSizeRel]
         include:
           - os: ubuntu-24.04
             ccache-variant: sccache
@@ -68,7 +69,7 @@ jobs:
         cmake -B ${{ steps.strings.outputs.build-output-dir }}
         -DCMAKE_CXX_COMPILER=${{ matrix.cpp_compiler }}
         -DCMAKE_C_COMPILER=${{ matrix.c_compiler }}
-        -DCMAKE_BUILD_TYPE=MinSizeRel
+        -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
         -DCMAKE_C_COMPILER_LAUNCHER=${{ matrix.ccache-variant }}
         -DCMAKE_CXX_COMPILER_LAUNCHER=${{ matrix.ccache-variant }}
         -DCMAKE_INSTALL_PREFIX=${{ steps.strings.outputs.build-install-dir }}

.github/workflows/libc-overlay-tests.yml

@@ -16,6 +16,7 @@ jobs:
       # Set fail-fast to false to ensure that feedback is delivered for all matrix combinations.
       fail-fast: false
       matrix:
+        build_type: [Debug, Release, MinSizeRel]
         include:
           # TODO: add linux gcc when it is fixed
           - os: ubuntu-24.04
@@ -95,7 +96,7 @@ jobs:
         cmake -B ${{ steps.strings.outputs.build-output-dir }}
         -DCMAKE_CXX_COMPILER=${{ matrix.compiler.cpp_compiler }}
         -DCMAKE_C_COMPILER=${{ matrix.compiler.c_compiler }}
-        -DCMAKE_BUILD_TYPE=MinSizeRel
+        -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
         -DCMAKE_C_COMPILER_LAUNCHER=${{ matrix.ccache-variant }}
         -DCMAKE_CXX_COMPILER_LAUNCHER=${{ matrix.ccache-variant }}
         -DCMAKE_POLICY_DEFAULT_CMP0141=NEW

.github/workflows/premerge.yaml

@@ -21,15 +21,16 @@ on:
       - 'main'
       - 'release/**'
 
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}
+  cancel-in-progress: true
+
 jobs:
   premerge-checks-linux:
     if: >-
         github.repository_owner == 'llvm' &&
         (github.event_name != 'pull_request' || github.event.action != 'closed')
     runs-on: llvm-premerge-linux-runners
-    concurrency:
-      group: ${{ github.workflow }}-linux-${{ github.event.pull_request.number || github.sha }}
-      cancel-in-progress: true
     steps:
       - name: Checkout LLVM
         uses: actions/checkout@v4
@@ -88,9 +89,6 @@ jobs:
         github.repository_owner == 'llvm' &&
         (github.event_name != 'pull_request' || github.event.action != 'closed')
     runs-on: llvm-premerge-windows-runners
-    concurrency:
-      group: ${{ github.workflow }}-windows-${{ github.event.pull_request.number || github.sha }}
-      cancel-in-progress: true
     defaults:
       run:
         shell: bash
@@ -148,9 +146,6 @@ jobs:
 
   permerge-check-macos:
     runs-on: macos-14
-    concurrency:
-      group: ${{ github.workflow }}-macos-${{ github.event.pull_request.number || github.sha }}
-      cancel-in-progress: true
     if: >-
       github.repository_owner == 'llvm' &&
       (startswith(github.ref_name, 'release/') ||

bolt/include/bolt/Core/BinaryContext.h

@@ -1435,6 +1435,17 @@ class BinaryContext {
                         bool PrintRelocations = false,
                         StringRef Endl = "\n") const;
 
+  /// Print data when embedded in the instruction stream keeping the format
+  /// similar to printInstruction().
+  void printData(raw_ostream &OS, ArrayRef<uint8_t> Data,
+                 uint64_t Offset) const;
+
+  /// Extract data from the binary corresponding to [Address, Address + Size)
+  /// range. Return an empty ArrayRef if the address range does not belong to
+  /// any section in the binary, crosses a section boundary, or falls into a
+  /// virtual section.
+  ArrayRef<uint8_t> extractData(uint64_t Address, uint64_t Size) const;
+
   /// Print a range of instructions.
   template <typename Itr>
   uint64_t

bolt/include/bolt/Core/BinaryFunction.h

@@ -2060,6 +2060,11 @@ class BinaryFunction {
     return Islands ? Islands->getAlignment() : 1;
   }
 
+  /// If there is a constant island in the range [StartOffset, EndOffset),
+  /// return its address.
+  std::optional<uint64_t> getIslandInRange(uint64_t StartOffset,
+                                           uint64_t EndOffset) const;
+
   uint64_t
   estimateConstantIslandSize(const BinaryFunction *OnBehalfOf = nullptr) const {
     if (!Islands)

bolt/lib/Core/BinaryContext.cpp

@@ -1759,7 +1759,11 @@ void BinaryContext::preprocessDebugInfo() {
           dwarf::toString(CU->getUnitDIE().find(dwarf::DW_AT_name), nullptr);
       if (std::optional<uint64_t> DWOID = CU->getDWOId()) {
         auto Iter = DWOCUs.find(*DWOID);
-        assert(Iter != DWOCUs.end() && "DWO CU was not found.");
+        if (Iter == DWOCUs.end()) {
+          this->errs() << "BOLT-ERROR: DWO CU was not found for " << Name
+                       << '\n';
+          exit(1);
+        }
         Name = dwarf::toString(
             Iter->second->getUnitDIE().find(dwarf::DW_AT_name), nullptr);
       }
@@ -1942,6 +1946,43 @@ static void printDebugInfo(raw_ostream &OS, const MCInst &Instruction,
     OS << " discriminator:" << Row.Discriminator;
 }
 
+ArrayRef<uint8_t> BinaryContext::extractData(uint64_t Address,
+                                             uint64_t Size) const {
+  ArrayRef<uint8_t> Res;
+
+  const ErrorOr<const BinarySection &> Section = getSectionForAddress(Address);
+  if (!Section || Section->isVirtual())
+    return Res;
+
+  if (!Section->containsRange(Address, Size))
+    return Res;
+
+  auto *Bytes =
+      reinterpret_cast<const uint8_t *>(Section->getContents().data());
+  return ArrayRef<uint8_t>(Bytes + Address - Section->getAddress(), Size);
+}
+
+void BinaryContext::printData(raw_ostream &OS, ArrayRef<uint8_t> Data,
+                              uint64_t Offset) const {
+  DataExtractor DE(Data, AsmInfo->isLittleEndian(),
+                   AsmInfo->getCodePointerSize());
+  uint64_t DataOffset = 0;
+  while (DataOffset + 4 <= Data.size()) {
+    OS << format("    %08" PRIx64 ": \t.word\t0x", Offset + DataOffset);
+    const auto Word = DE.getUnsigned(&DataOffset, 4);
+    OS << Twine::utohexstr(Word) << '\n';
+  }
+  if (DataOffset + 2 <= Data.size()) {
+    OS << format("    %08" PRIx64 ": \t.short\t0x", Offset + DataOffset);
+    const auto Short = DE.getUnsigned(&DataOffset, 2);
+    OS << Twine::utohexstr(Short) << '\n';
+  }
+  if (DataOffset + 1 == Data.size()) {
+    OS << format("    %08" PRIx64 ": \t.byte\t0x%x\n", Offset + DataOffset,
+                 Data[DataOffset]);
+  }
+}
+
 void BinaryContext::printInstruction(raw_ostream &OS, const MCInst &Instruction,
                                      uint64_t Offset,
                                      const BinaryFunction *Function,

bolt/lib/Core/BinaryFunction.cpp

@@ -491,11 +491,27 @@ void BinaryFunction::print(raw_ostream &OS, std::string Annotation) {
   // Offset of the instruction in function.
   uint64_t Offset = 0;
 
+  auto printConstantIslandInRange = [&](uint64_t Start, uint64_t End) {
+    assert(Start <= End && "Invalid range");
+    std::optional<uint64_t> IslandOffset = getIslandInRange(Start, End);
+
+    if (!IslandOffset)
+      return;
+
+    const size_t IslandSize = getSizeOfDataInCodeAt(*IslandOffset);
+    BC.printData(OS, BC.extractData(getAddress() + *IslandOffset, IslandSize),
+                 *IslandOffset);
+  };
+
   if (BasicBlocks.empty() && !Instructions.empty()) {
     // Print before CFG was built.
+    uint64_t PrevOffset = 0;
     for (const std::pair<const uint32_t, MCInst> &II : Instructions) {
       Offset = II.first;
 
+      // Print any constant islands inbeetween the instructions.
+      printConstantIslandInRange(PrevOffset, Offset);
+
       // Print label if exists at this offset.
       auto LI = Labels.find(Offset);
       if (LI != Labels.end()) {
@@ -506,7 +522,12 @@ void BinaryFunction::print(raw_ostream &OS, std::string Annotation) {
       }
 
       BC.printInstruction(OS, II.second, Offset, this);
+
+      PrevOffset = Offset;
     }
+
+    // Print any data at the end of the function.
+    printConstantIslandInRange(PrevOffset, getMaxSize());
   }
 
   StringRef SplitPointMsg = "";
@@ -1048,6 +1069,19 @@ size_t BinaryFunction::getSizeOfDataInCodeAt(uint64_t Offset) const {
   return getSize() - Offset;
 }
 
+std::optional<uint64_t>
+BinaryFunction::getIslandInRange(uint64_t StartOffset,
+                                 uint64_t EndOffset) const {
+  if (!Islands)
+    return std::nullopt;
+
+  auto Iter = llvm::lower_bound(Islands->DataOffsets, StartOffset);
+  if (Iter != Islands->DataOffsets.end() && *Iter < EndOffset)
+    return *Iter;
+
+  return std::nullopt;
+}
+
 bool BinaryFunction::isZeroPaddingAt(uint64_t Offset) const {
   ArrayRef<uint8_t> FunctionData = *getData();
   uint64_t EndOfCode = getSize();

bolt/lib/RuntimeLibs/RuntimeLibrary.cpp

@@ -18,6 +18,7 @@
 #include "llvm/Object/Archive.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Support/Path.h"
+#include "llvm/Support/Program.h"
 
 #define DEBUG_TYPE "bolt-rtlib"
 
@@ -38,6 +39,23 @@ std::string RuntimeLibrary::getLibPathByToolPath(StringRef ToolPath,
     llvm::sys::path::append(LibPath, "lib" LLVM_LIBDIR_SUFFIX);
   }
   llvm::sys::path::append(LibPath, LibFileName);
+  if (!llvm::sys::fs::exists(LibPath)) {
+    // If it is a symlink, check the directory that the symlink points to.
+    if (llvm::sys::fs::is_symlink_file(ToolPath)) {
+      SmallString<256> RealPath;
+      llvm::sys::fs::real_path(ToolPath, RealPath);
+      if (llvm::ErrorOr<std::string> P =
+              llvm::sys::findProgramByName(RealPath)) {
+        outs() << "BOLT-INFO: library not found: " << LibPath << "\n"
+               << "BOLT-INFO: " << ToolPath << " is a symlink; will look up "
+               << LibFileName
+               << " at the target directory that the symlink points to\n";
+        return getLibPath(*P, LibFileName);
+      }
+    }
+    errs() << "BOLT-ERROR: library not found: " << LibPath << "\n";
+    exit(1);
+  }
   return std::string(LibPath);
 }
 

bolt/test/AArch64/data-in-code.s

@@ -0,0 +1,31 @@
+## Check that llvm-bolt prints data embedded in code.
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags -fno-PIC -no-pie %t.o -o %t.exe -nostdlib \
+# RUN:    -fuse-ld=lld -Wl,-q
+
+## Check disassembly of BOLT input.
+# RUN: llvm-objdump %t.exe -d | FileCheck %s
+
+# RUN: llvm-bolt %t.exe -o %t.bolt --print-disasm | FileCheck %s
+
+.text
+.balign 4
+
+.global _start
+.type _start, %function
+_start:
+  mov x0, #0x0
+  .word 0x4f82e010
+  ret
+  .byte 0x0, 0xff, 0x42
+# CHECK-LABEL: _start
+# CHECK:        mov x0, #0x0
+# CHECK-NEXT:   .word 0x4f82e010
+# CHECK-NEXT:   ret
+# CHECK-NEXT:   .short 0xff00
+# CHECK-NEXT:   .byte 0x42
+.size _start, .-_start
+
+## Force relocation mode.
+  .reloc 0, R_AARCH64_NONE

bolt/test/AArch64/exceptions-plt.cpp

@@ -2,7 +2,9 @@
 
 // REQUIRES: system-linux
 
-// RUN: %clangxx %cxxflags -O1 -Wl,-q,-znow %s -o %t.exe
+// RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+// Link against a DSO to ensure PLT entries.
+// RUN: %clangxx %cxxflags -O1 -Wl,-q,-znow %s %t.so -o %t.exe
 // RUN: llvm-bolt %t.exe -o %t.bolt.exe --plt=all --print-only=.*main.* \
 // RUN:   --print-finalized 2>&1 | FileCheck %s