Merge remote-tracking branch 'origin/main' into loop-vectorize/VPReductionRecipe-RecurKind

Author: lukel97

.ci/compute-projects.sh

@@ -18,14 +18,15 @@ function compute-projects-to-test() {
   shift
   projects=${@}
   for project in ${projects}; do
+    echo "${project}"
     case ${project} in
     lld)
-      for p in lld bolt cross-project-tests; do
+      for p in bolt cross-project-tests; do
         echo $p
       done
     ;;
     llvm)
-      for p in llvm bolt clang clang-tools-extra lld lldb mlir polly; do
+      for p in bolt clang clang-tools-extra lld lldb mlir polly; do
         echo $p
       done
       # Flang is not stable in Windows CI at the moment
@@ -35,30 +36,21 @@ function compute-projects-to-test() {
     ;;
     clang)
       # lldb is temporarily removed to alleviate Linux pre-commit CI waiting times
-      for p in clang clang-tools-extra compiler-rt cross-project-tests; do
+      for p in clang-tools-extra compiler-rt cross-project-tests; do
         echo $p
       done
     ;;
     clang-tools-extra)
-      for p in clang-tools-extra libc; do
-        echo $p
-      done
+      echo libc
     ;;
     mlir)
-      echo mlir
-      # Flang is not stable in Windows CI at the moment
-      if [[ $isForWindows == 0 ]]; then
-        echo flang
-      fi
-    ;;
-    flang-rt)
       # Flang is not stable in Windows CI at the moment
       if [[ $isForWindows == 0 ]]; then
         echo flang
       fi
     ;;
     *)
-      echo "${project}"
+      # Nothing to do
     ;;
     esac
   done
@@ -73,11 +65,6 @@ function compute-runtimes-to-test() {
         echo $p
       done
     ;;
-    flang)
-      for p in flang-rt; do
-        echo $p
-      done
-    ;;
     *)
       # Nothing to do
     ;;

.ci/generate-buildkite-pipeline-premerge

@@ -73,7 +73,7 @@ fi
 # needs while letting them run on the infrastructure provided by LLVM.
 
 # Figure out which projects need to be built on each platform
-all_projects="bolt clang clang-tools-extra compiler-rt cross-project-tests flang flang-rt libc libclc lld lldb llvm mlir openmp polly pstl"
+all_projects="bolt clang clang-tools-extra compiler-rt cross-project-tests flang libc libclc lld lldb llvm mlir openmp polly pstl"
 modified_projects="$(keep-modified-projects ${all_projects})"
 
 linux_projects_to_test=$(exclude-linux $(compute-projects-to-test 0 ${modified_projects}))

.ci/metrics/metrics.py

@@ -43,7 +43,7 @@
 # This means we essentially have a list of workflows sorted by creation date,
 # and that's all we can deduce from it. So for each iteration, we'll blindly
 # process the last N workflows.
-GITHUB_WORKFLOWS_MAX_PROCESS_COUNT = 1000
+GITHUB_WORKFLOWS_MAX_PROCESS_COUNT = 2000
 # Second reason for the cut: reaching a workflow older than X.
 # This means we will miss long-tails (exceptional jobs running for more than
 # X hours), but that's also the case with the count cutoff above.

.ci/monolithic-linux.sh

@@ -65,7 +65,6 @@ cmake -S "${MONOREPO_ROOT}"/llvm -B "${BUILD_DIR}" \
       -D CMAKE_CXX_FLAGS=-gmlt \
       -D LLVM_CCACHE_BUILD=ON \
       -D MLIR_ENABLE_BINDINGS_PYTHON=ON \
-      -D FLANG_ENABLE_FLANG_RT=OFF \
       -D CMAKE_INSTALL_PREFIX="${INSTALL_DIR}"
 
 echo "--- ninja"
@@ -96,9 +95,6 @@ if [[ "${runtimes}" != "" ]]; then
   cmake -S "${MONOREPO_ROOT}/runtimes" -B "${RUNTIMES_BUILD_DIR}" -GNinja \
       -D CMAKE_C_COMPILER="${INSTALL_DIR}/bin/clang" \
       -D CMAKE_CXX_COMPILER="${INSTALL_DIR}/bin/clang++" \
-      -D CMAKE_Fortran_COMPILER="${BUILD_DIR}/bin/flang" \
-      -D CMAKE_Fortran_COMPILER_WORKS=ON \
-      -D LLVM_BINARY_DIR="${BUILD_DIR}" \
       -D LLVM_ENABLE_RUNTIMES="${runtimes}" \
       -D LIBCXX_CXX_ABI=libcxxabi \
       -D CMAKE_BUILD_TYPE=RelWithDebInfo \
@@ -117,9 +113,6 @@ if [[ "${runtimes}" != "" ]]; then
   cmake -S "${MONOREPO_ROOT}/runtimes" -B "${RUNTIMES_BUILD_DIR}" -GNinja \
       -D CMAKE_C_COMPILER="${INSTALL_DIR}/bin/clang" \
       -D CMAKE_CXX_COMPILER="${INSTALL_DIR}/bin/clang++" \
-      -D CMAKE_Fortran_COMPILER="${BUILD_DIR}/bin/flang" \
-      -D CMAKE_Fortran_COMPILER_WORKS=ON \
-      -D LLVM_BINARY_DIR="${BUILD_DIR}" \
       -D LLVM_ENABLE_RUNTIMES="${runtimes}" \
       -D LIBCXX_CXX_ABI=libcxxabi \
       -D CMAKE_BUILD_TYPE=RelWithDebInfo \
@@ -138,9 +131,6 @@ if [[ "${runtimes}" != "" ]]; then
   cmake -S "${MONOREPO_ROOT}/runtimes" -B "${RUNTIMES_BUILD_DIR}" -GNinja \
       -D CMAKE_C_COMPILER="${INSTALL_DIR}/bin/clang" \
       -D CMAKE_CXX_COMPILER="${INSTALL_DIR}/bin/clang++" \
-      -D CMAKE_Fortran_COMPILER="${BUILD_DIR}/bin/flang" \
-      -D CMAKE_Fortran_COMPILER_WORKS=ON \
-      -D LLVM_BINARY_DIR="${BUILD_DIR}" \
       -D LLVM_ENABLE_RUNTIMES="${runtimes}" \
       -D LIBCXX_CXX_ABI=libcxxabi \
       -D CMAKE_BUILD_TYPE=RelWithDebInfo \

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

@@ -37,7 +37,7 @@ jobs:
   stage1:
     if: github.repository_owner == 'llvm'
     runs-on: libcxx-self-hosted-linux
-    container: ghcr.io/llvm/libcxx-linux-builder:b319dfef21f6c7b0bc6a356d6b9f41a3b3b98ae9
+    container: ghcr.io/llvm/libcxx-linux-builder:b6bb9dc5abd7c6452c13a53fa8949cb259db459b
     continue-on-error: false
     strategy:
       fail-fast: false
@@ -75,7 +75,7 @@ jobs:
   stage2:
     if: github.repository_owner == 'llvm'
     runs-on: libcxx-self-hosted-linux
-    container: ghcr.io/llvm/libcxx-linux-builder:b319dfef21f6c7b0bc6a356d6b9f41a3b3b98ae9
+    container: ghcr.io/llvm/libcxx-linux-builder:b6bb9dc5abd7c6452c13a53fa8949cb259db459b
     needs: [ stage1 ]
     continue-on-error: false
     strategy:
@@ -167,7 +167,7 @@ jobs:
         - config: 'generic-msan'
           machine: libcxx-self-hosted-linux
     runs-on: ${{ matrix.machine }}
-    container: ghcr.io/llvm/libcxx-linux-builder:b319dfef21f6c7b0bc6a356d6b9f41a3b3b98ae9
+    container: ghcr.io/llvm/libcxx-linux-builder:b6bb9dc5abd7c6452c13a53fa8949cb259db459b
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
       - name: ${{ matrix.config }}

bolt/lib/Rewrite/RewriteInstance.cpp

@@ -2455,6 +2455,30 @@ void RewriteInstance::readDynamicRelocations(const SectionRef &Section,
     if (Symbol)
       SymbolIndex[Symbol] = getRelocationSymbol(InputFile, Rel);
 
+    const uint64_t ReferencedAddress = SymbolAddress + Addend;
+    BinaryFunction *Func =
+        BC->getBinaryFunctionContainingAddress(ReferencedAddress);
+
+    if (Relocation::isRelative(RType) && SymbolAddress == 0) {
+      if (Func) {
+        if (!Func->isInConstantIsland(ReferencedAddress)) {
+          if (const uint64_t ReferenceOffset =
+                  ReferencedAddress - Func->getAddress()) {
+            Func->addEntryPointAtOffset(ReferenceOffset);
+          }
+        } else {
+          BC->errs() << "BOLT-ERROR: referenced address at 0x"
+                     << Twine::utohexstr(ReferencedAddress)
+                     << " is in constant island of function " << *Func << "\n";
+          exit(1);
+        }
+      }
+    } else if (Relocation::isRelative(RType) && SymbolAddress != 0) {
+      BC->errs() << "BOLT-ERROR: symbol address non zero for RELATIVE "
+                    "relocation type\n";
+      exit(1);
+    }
+
     BC->addDynamicRelocation(Rel.getOffset(), Symbol, RType, Addend);
   }
 }
@@ -5696,7 +5720,7 @@ uint64_t RewriteInstance::getNewFunctionOrDataAddress(uint64_t OldAddress) {
         for (const BinaryBasicBlock &BB : *BF)
           if (BB.isEntryPoint() &&
               (BF->getAddress() + BB.getOffset()) == OldAddress)
-            return BF->getOutputAddress() + BB.getOffset();
+            return BB.getOutputStartAddress();
       }
       BC->errs() << "BOLT-ERROR: unable to get new address corresponding to "
                     "input address 0x"

bolt/lib/Target/AArch64/AArch64MCSymbolizer.cpp

@@ -125,15 +125,39 @@ AArch64MCSymbolizer::adjustRelocation(const Relocation &Rel,
     // instruction pairs and will perform necessary adjustments.
     ErrorOr<uint64_t> SymbolValue = BC.getSymbolValue(*Rel.Symbol);
     assert(SymbolValue && "Symbol value should be set");
-    (void)SymbolValue;
-
-    AdjustedRel.Symbol = BC.registerNameAtAddress("__BOLT_got_zero", 0, 0, 0);
-    AdjustedRel.Addend = Rel.Value;
+    const uint64_t SymbolPageAddr = *SymbolValue & ~0xfffULL;
+
+    // Check if defined symbol and GOT are on the same page. If they are not,
+    // disambiguate the operand.
+    if (BC.MIB->isADRP(Inst) && Rel.Addend == 0 &&
+        SymbolPageAddr == Rel.Value &&
+        !isPageAddressValidForGOT(SymbolPageAddr)) {
+      AdjustedRel.Type = ELF::R_AARCH64_ADR_PREL_PG_HI21;
+    } else {
+      AdjustedRel.Symbol = BC.registerNameAtAddress("__BOLT_got_zero", 0, 0, 0);
+      AdjustedRel.Addend = Rel.Value;
+    }
   }
 
   return AdjustedRel;
 }
 
+bool AArch64MCSymbolizer::isPageAddressValidForGOT(uint64_t PageAddress) const {
+  assert(!(PageAddress & 0xfffULL) && "Page address not aligned at 4KB");
+
+  ErrorOr<BinarySection &> GOT =
+      Function.getBinaryContext().getUniqueSectionByName(".got");
+  if (!GOT || !GOT->getSize())
+    return false;
+
+  const uint64_t GOTFirstPageAddress = GOT->getAddress() & ~0xfffULL;
+  const uint64_t GOTLastPageAddress =
+      (GOT->getAddress() + GOT->getSize() - 1) & ~0xfffULL;
+
+  return PageAddress >= GOTFirstPageAddress &&
+         PageAddress <= GOTLastPageAddress;
+}
+
 void AArch64MCSymbolizer::tryAddingPcLoadReferenceComment(raw_ostream &CStream,
                                                           int64_t Value,
                                                           uint64_t Address) {}

bolt/lib/Target/AArch64/AArch64MCSymbolizer.h

@@ -28,6 +28,9 @@ class AArch64MCSymbolizer : public MCSymbolizer {
   std::optional<Relocation> adjustRelocation(const Relocation &Rel,
                                              const MCInst &Inst) const;
 
+  /// Return true if \p PageAddress is a valid page address for .got section.
+  bool isPageAddressValidForGOT(uint64_t PageAddress) const;
+
 public:
   AArch64MCSymbolizer(BinaryFunction &Function, bool CreateNewSymbols = true)
       : MCSymbolizer(*Function.getBinaryContext().Ctx.get(), nullptr),

bolt/lib/Target/RISCV/RISCVMCPlusBuilder.cpp

@@ -244,7 +244,7 @@ class RISCVMCPlusBuilder : public MCPlusBuilder {
     Inst.clear();
     Inst.addOperand(MCOperand::createExpr(RISCVMCExpr::create(
         MCSymbolRefExpr::create(Target, MCSymbolRefExpr::VK_None, *Ctx),
-        RISCVMCExpr::VK_RISCV_CALL, *Ctx)));
+        RISCVMCExpr::VK_CALL, *Ctx)));
   }
 
   void createCall(MCInst &Inst, const MCSymbol *Target,
@@ -434,19 +434,19 @@ class RISCVMCPlusBuilder : public MCPlusBuilder {
     case ELF::R_RISCV_TLS_GOT_HI20:
       // The GOT is reused so no need to create GOT relocations
     case ELF::R_RISCV_PCREL_HI20:
-      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_RISCV_PCREL_HI, Ctx);
+      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_PCREL_HI, Ctx);
     case ELF::R_RISCV_PCREL_LO12_I:
     case ELF::R_RISCV_PCREL_LO12_S:
-      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_RISCV_PCREL_LO, Ctx);
+      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_PCREL_LO, Ctx);
     case ELF::R_RISCV_HI20:
-      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_RISCV_HI, Ctx);
+      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_HI, Ctx);
     case ELF::R_RISCV_LO12_I:
     case ELF::R_RISCV_LO12_S:
-      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_RISCV_LO, Ctx);
+      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_LO, Ctx);
     case ELF::R_RISCV_CALL:
-      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_RISCV_CALL, Ctx);
+      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_CALL, Ctx);
     case ELF::R_RISCV_CALL_PLT:
-      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_RISCV_CALL_PLT, Ctx);
+      return RISCVMCExpr::create(Expr, RISCVMCExpr::VK_CALL_PLT, Ctx);
     }
   }
 
@@ -471,8 +471,8 @@ class RISCVMCPlusBuilder : public MCPlusBuilder {
     switch (cast<RISCVMCExpr>(ImmExpr)->getKind()) {
     default:
       return false;
-    case RISCVMCExpr::VK_RISCV_CALL:
-    case RISCVMCExpr::VK_RISCV_CALL_PLT:
+    case RISCVMCExpr::VK_CALL:
+    case RISCVMCExpr::VK_CALL_PLT:
       return true;
     }
   }

bolt/test/AArch64/computed-goto.s

@@ -0,0 +1,67 @@
+// This test checks that BOLT creates entry points for addresses
+// referenced by dynamic relocations.
+// The test also checks that BOLT can map addresses inside functions.
+
+// Checks for error and entry points.
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags %t.o -o %t.exe -Wl,-q
+# RUN: llvm-bolt %t.exe -o %t.bolt 2>&1 | FileCheck %s
+# RUN: llvm-bolt %t.exe -o %t.bolt --print-cfg | FileCheck --check-prefix=CHECK-ENTRIES %s
+
+// Checks for dynamic relocations.
+# RUN: llvm-readelf -dr %t.bolt > %t.out.txt
+# RUN: llvm-objdump -j .rela.dyn -d %t.bolt >> %t.out.txt
+# RUN: FileCheck --check-prefix=CHECK-RELOCS %s --input-file=%t.out.txt
+
+// Before bolt could handle mapping addresses within moved functions, it
+// would bail out with an error of the form:
+// BOLT-ERROR: unable to get new address corresponding to input address 0x10390 in function main. Consider adding this function to --skip-funcs=...
+// These addresses arise if computed GOTO is in use.
+// Check that bolt does not emit any error.
+# CHECK-NOT: BOLT-ERROR
+
+// Check that there are dynamic relocations.
+# CHECK-RELOCS:     Dynamic section at offset {{.*}} contains {{.*}} entries:
+# CHECK-RELOCS:     Relocation section '.rela.dyn' at offset {{.*}} contains {{.*}} entries
+
+// Check that dynamic relocations were updated
+# CHECK-RELOCS: [[#%x,OFF:]] [[#%x,INFO_DYN:]] R_AARCH64_RELATIVE [[#%x,ADDR:]]
+# CHECK-RELOCS-NEXT: [[#OFF + 8]] {{0*}}[[#INFO_DYN]] R_AARCH64_RELATIVE [[#ADDR + 8]]
+# CHECK-RELOCS: [[#ADDR]] <unknown>
+# CHECK-RELOCS: [[#ADDR + 8]] <unknown>
+
+// Check that BOLT registers extra entry points for dynamic relocations.
+# CHECK-ENTRIES: Binary Function "main" after building cfg {
+# CHECK-ENTRIES:  IsMultiEntry: 1
+# CHECK-ENTRIES: .Ltmp0 {{.*}}
+# CHECK-ENTRIES-NEXT: Secondary Entry Point: {{.*}}
+# CHECK-ENTRIES: .Ltmp1 {{.*}}
+# CHECK-ENTRIES-NEXT: Secondary Entry Point: {{.*}}
+
+.globl  main
+.p2align        2
+.type   main,@function
+main:
+.cfi_startproc
+        adrp    x8, .L__const.main.ptrs+8
+        add     x8, x8, :lo12:.L__const.main.ptrs+8
+        ldr     x9, [x8], #8
+        br      x9
+
+.Label0: // Block address taken
+        ldr     x9, [x8], #8
+        br      x9
+
+.Label1: // Block address taken
+        mov     w0, #42
+        ret
+
+.Lfunc_end0:
+.size   main, .Lfunc_end0-main
+.cfi_endproc
+        .type   .L__const.main.ptrs,@object
+        .section        .data.rel.ro,"aw",@progbits
+        .p2align        3, 0x0
+.L__const.main.ptrs:
+        .xword  .Label0
+        .xword  .Label1