Merge branch 'main' into fix/125810

Author: a-tarasyuk

.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/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/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/include/clang/Basic/arm_mve.td

@@ -1537,7 +1537,7 @@ foreach desttype = T.All in {
 let params = T.All in {
   let pnt = PNT_None in {
     def vcreateq: Intrinsic<Vector, (args u64:$a, u64:$b),
-        (vreinterpret (ielt_const (ielt_const (undef VecOf<u64>), $a, 0),
+        (vreinterpret (ielt_const (ielt_const (poison VecOf<u64>), $a, 0),
                              $b, 1), Vector)>;
     def vuninitializedq: Intrinsic<Vector, (args), (undef Vector)>;
   }

clang/include/clang/Basic/arm_mve_defs.td

@@ -102,6 +102,7 @@ def int_min: CGHelperFn<"ARMMVEConstantSplat<1,0>">;
 def int_max: CGHelperFn<"ARMMVEConstantSplat<0,1>">;
 def uint_max: CGHelperFn<"ARMMVEConstantSplat<1,1>">;
 def undef: IRFunction<"UndefValue::get">;
+def poison: IRFunction<"PoisonValue::get">;
 def icmp_eq: IRBuilder<"CreateICmpEQ">;
 def icmp_ne: IRBuilder<"CreateICmpNE">;
 def icmp_ugt: IRBuilder<"CreateICmpUGT">;

clang/include/clang/Tooling/DependencyScanning/DependencyScanningService.h

@@ -10,6 +10,7 @@
 #define LLVM_CLANG_TOOLING_DEPENDENCYSCANNING_DEPENDENCYSCANNINGSERVICE_H
 
 #include "clang/Tooling/DependencyScanning/DependencyScanningFilesystem.h"
+#include "clang/Tooling/DependencyScanning/InProcessModuleCache.h"
 #include "llvm/ADT/BitmaskEnum.h"
 
 namespace clang {
@@ -99,6 +100,8 @@ class DependencyScanningService {
     return SharedCache;
   }
 
+  ModuleCacheMutexes &getModuleCacheMutexes() { return ModCacheMutexes; }
+
 private:
   const ScanningMode Mode;
   const ScanningOutputFormat Format;
@@ -110,6 +113,8 @@ class DependencyScanningService {
   const bool TraceVFS;
   /// The global file system cache.
   DependencyScanningFilesystemSharedCache SharedCache;
+  /// The global module cache mutexes.
+  ModuleCacheMutexes ModCacheMutexes;
 };
 
 } // end namespace dependencies

clang/include/clang/Tooling/DependencyScanning/InProcessModuleCache.h

@@ -0,0 +1,31 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_TOOLING_DEPENDENCYSCANNING_INPROCESSMODULECACHE_H
+#define LLVM_CLANG_TOOLING_DEPENDENCYSCANNING_INPROCESSMODULECACHE_H
+
+#include "clang/Serialization/ModuleCache.h"
+#include "llvm/ADT/StringMap.h"
+
+#include <shared_mutex>
+
+namespace clang {
+namespace tooling {
+namespace dependencies {
+struct ModuleCacheMutexes {
+  std::mutex Mutex;
+  llvm::StringMap<std::unique_ptr<std::shared_mutex>> Map;
+};
+
+IntrusiveRefCntPtr<ModuleCache>
+makeInProcessModuleCache(ModuleCacheMutexes &Mutexes);
+} // namespace dependencies
+} // namespace tooling
+} // namespace clang
+
+#endif