Merge branch 'main' into vp-arm-mve-transform

Author: ElvisWang123

.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 \

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/got-load-symbolization.s

@@ -0,0 +1,100 @@
+## Check that BOLT symbolizer properly handles loads from GOT, including
+## instruction sequences changed/relaxed by the linker.
+
+# RUN: split-file %s %t
+
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %t/main.s \
+# RUN:   -o %t/main.o
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %t/near.s \
+# RUN:   -o %t/near.o
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %t/far.s \
+# RUN:   -o %t/far.o
+# RUN: %clang %cflags %t/main.o %t/near.o %t/far.o -o %t/main.exe -Wl,-q -static
+# RUN: llvm-bolt %t/main.exe -o %t/main.bolt --keep-nops --print-disasm \
+# RUN:   --print-only=_start | FileCheck %s
+
+#--- main.s
+
+	.text
+	.globl _start
+	.p2align        2
+	.type _start, @function
+# CHECK-LABEL: _start
+_start:
+
+## Function address load relaxable into nop+adr.
+# CHECK: 			nop
+# CHECK-NEXT: adr x0, near
+	adrp    x0, :got:near
+	ldr     x0, [x0, :got_lo12:near]
+
+## Function address load relaxable into adrp+add.
+# CHECK-NEXT: adrp x1, far
+# CHECK-NEXT: add  x1, x1, :lo12:far
+	adrp    x1, :got:far
+	ldr     x1, [x1, :got_lo12:far]
+
+## Non-relaxable due to the instruction in-between.
+# CHECK-NEXT: adrp x2, __BOLT_got_zero
+# CHECK-NEXT: nop
+# CHECK-NEXT: ldr  x2, [x2, :lo12:__BOLT_got_zero{{.*}}]
+	adrp    x2, :got:near
+	nop
+	ldr     x2, [x2, :got_lo12:near]
+
+## Load data object with local visibility. Relaxable into adrp+add.
+# CHECK-NEXT: adrp x3, "local_far_data/1"
+# CHECK-NEXT: add  x3, x3, :lo12:"local_far_data/1"
+  adrp    x3, :got:local_far_data
+  ldr     x3, [x3, :got_lo12:local_far_data]
+
+## Global data reference relaxable into adrp+add.
+# CHECK-NEXT: adrp x4, far_data
+# CHECK-NEXT: add  x4, x4, :lo12:far_data
+  adrp    x4, :got:far_data
+  ldr     x4, [x4, :got_lo12:far_data]
+
+	ret
+	.size _start, .-_start
+
+.weak near
+.weak far
+.weak far_data
+
+## Data object separated by more than 1MB from _start.
+  .data
+  .type local_far_data, @object
+local_far_data:
+  .xword 0
+.size   local_far_data, .-local_far_data
+
+#--- near.s
+
+	.text
+	.globl near
+	.type near, @function
+near:
+  ret
+.size   near, .-near
+
+#--- far.s
+
+  .text
+
+## Insert 1MB of empty space to make objects after it unreachable by adr
+## instructions in _start.
+  .space 0x100000
+
+	.globl far
+	.type far, @function
+far:
+  ret
+.size   far, .-far
+
+  .data
+  .globl far_data
+  .type far_data, @object
+far_data:
+  .xword 0
+.size   far_data, .-far_data
+

clang/docs/ClangOffloadBundler.rst

@@ -266,15 +266,14 @@ without differentiation based on offload kind.
     The target triple of the code object. See `Target Triple
     <https://clang.llvm.org/docs/CrossCompilation.html#target-triple>`_.
 
-    The bundler accepts target triples with or without the optional environment
-    field:
+    LLVM target triples can be with or without the optional environment field:
 
     ``<arch><sub>-<vendor>-<sys>``, or
     ``<arch><sub>-<vendor>-<sys>-<env>``
 
-    However, in order to standardize outputs for tools that consume bitcode
-    bundles, bundles written by the bundler internally use only the 4-field
-    target triple:
+    However, in order to standardize outputs for tools that consume bitcode bundles
+    and to parse target ID containing dashes, the bundler only accepts target
+    triples in the 4-field format:
 
     ``<arch><sub>-<vendor>-<sys>-<env>``
 
@@ -543,4 +542,4 @@ The compressed offload bundle begins with a header followed by the compressed bi
 - **Compressed Data**:
     The actual compressed binary data follows the header. Its size can be inferred from the total size of the file minus the header size.
 
-    > **Note**: Version 3 of the format is under development. It uses 64-bit fields for Total File Size and Uncompressed Binary Size to support files larger than 4GB. To experiment with version 3, set the environment variable `COMPRESSED_BUNDLE_FORMAT_VERSION=3`. This support is experimental and not recommended for production use.
+    > **Note**: Version 3 of the format is under development. It uses 64-bit fields for Total File Size and Uncompressed Binary Size to support files larger than 4GB. To experiment with version 3, set the environment variable `COMPRESSED_BUNDLE_FORMAT_VERSION=3`. This support is experimental and not recommended for production use.

clang/docs/LanguageExtensions.rst

@@ -1652,6 +1652,7 @@ Designated initializers (N494)                                                 C
 Array & element qualification (N2607)                                          C23           C89
 Attributes (N2335)                                                             C23           C89
 ``#embed`` (N3017)                                                             C23           C89, C++
+Octal literals prefixed with ``0o`` or ``0O``                                  C2y           C89, C++
 ============================================= ================================ ============= =============
 
 Builtin type aliases
@@ -1911,6 +1912,40 @@ A simplistic usage example as might be seen in standard C++ headers follows:
   // Emulate type trait for compatibility with other compilers.
   #endif
 
+
+.. _builtin_structured_binding_size-doc:
+
+__builtin_structured_binding_size (C++)
+---------------------------------------
+
+The ``__builtin_structured_binding_size(T)`` type trait returns
+the *structured binding size* ([dcl.struct.bind]) of type ``T``
+
+This is equivalent to the size of the pack ``p`` in ``auto&& [...p] = declval<T&>();``.
+If the argument cannot be decomposed, ``__builtin_structured_binding_size(T)``
+is not a valid expression (``__builtin_structured_binding_size`` is SFINAE-friendly).
+
+builtin arrays, builtin SIMD vectors,
+builtin complex types, *tuple-like* types, and decomposable class types
+are decomposable types.
+
+A type is considered a valid *tuple-like* if ``std::tuple_size_v<T>`` is a valid expression,
+even if there is no valid ``std::tuple_element`` specialization or suitable
+``get`` function for that type.
+
+.. code-block:: c++
+
+  template<std::size_t Idx, typename T>
+  requires (Idx < __builtin_structured_binding_size(T))
+  decltype(auto) constexpr get_binding(T&& obj) {
+      auto && [...p] = std::forward<T>(obj);
+      return p...[Idx];
+  }
+  struct S { int a = 0, b = 42; };
+  static_assert(__builtin_structured_binding_size(S) == 2);
+  static_assert(get_binding<1>(S{}) == 42);
+
+
 Blocks
 ======