Merge branch 'main' into guard-no-outer-loop-option

Author: Lewuathe

.github/CODEOWNERS

@@ -7,10 +7,9 @@
 # to receive an approval from a "code owner" in particular -- any LLVM project
 # member can approve pull requests.
 #
-# Note that GitHub's concept of "code owner" is independent from LLVM's own
-# "code owner" concept, they merely happen to share terminology. See
-# https://llvm.org/docs/DeveloperPolicy.html#code-owners, as well as the
-# CODE_OWNERS.txt files in the respective subproject directories.
+# This is independent of LLVM's own "maintainer" concept.
+# See https://llvm.org/docs/DeveloperPolicy.html#maintainers as well as the
+# Maintainers.* files in the the respective subproject directories.
 
 /libcxx/ @llvm/reviewers-libcxx
 /libcxxabi/ @llvm/reviewers-libcxxabi

.github/workflows/docs.yml

@@ -112,58 +112,87 @@ jobs:
           sudo apt-get update
           # swig and graphviz are lldb specific dependencies
           sudo apt-get install -y cmake ninja-build swig graphviz
+      - name: Setup output folder
+        run: mkdir built-docs
       - name: Build LLVM docs
         if: steps.docs-changed-subprojects.outputs.llvm_any_changed == 'true'
         run: |
           cmake -B llvm-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C llvm-build docs-llvm-html docs-llvm-man
+          mkdir built-docs/llvm
+          cp -r llvm-build/docs/* built-docs/llvm/
       - name: Build Clang docs
         if: steps.docs-changed-subprojects.outputs.clang_any_changed == 'true'
         run: |
           cmake -B clang-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C clang-build docs-clang-html docs-clang-man
+          mkdir built-docs/clang
+          cp -r clang-build/docs/* built-docs/clang/
       - name: Build clang-tools-extra docs
         if: steps.docs-changed-subprojects.outputs.clang-tools-extra_any_changed == 'true'
         run: |
           cmake -B clang-tools-extra-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C clang-tools-extra-build docs-clang-tools-html docs-clang-tools-man
+          mkdir built-docs/clang-tools-extra
+          cp -r clang-tools-extra-build/docs/* built-docs/clang-tools-extra/
       - name: Build LLDB docs
         if: steps.docs-changed-subprojects.outputs.lldb_any_changed == 'true'
         run: |
           cmake -B lldb-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang;lldb" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C lldb-build docs-lldb-html docs-lldb-man
+          mkdir built-docs/lldb
+          cp -r lldb-build/docs/* built-docs/lldb/
       - name: Build libunwind docs
         if: steps.docs-changed-subprojects.outputs.libunwind_any_changed == 'true'
         run: |
           cmake -B libunwind-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_RUNTIMES="libunwind" -DLLVM_ENABLE_SPHINX=ON ./runtimes
           TZ=UTC ninja -C libunwind-build docs-libunwind-html
+          mkdir built-docs/libunwind
+          cp -r libunwind-build/docs/* built-docs/libunwind
       - name: Build libcxx docs
         if: steps.docs-changed-subprojects.outputs.libcxx_any_changed == 'true'
         run: |
           cmake -B libcxx-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_RUNTIMES="libcxxabi;libcxx;libunwind" -DLLVM_ENABLE_SPHINX=ON ./runtimes
           TZ=UTC ninja -C libcxx-build docs-libcxx-html
+          mkdir built-docs/libcxx
+          cp -r libcxx-build/docs/* built-docs/libcxx/
       - name: Build libc docs
         if: steps.docs-changed-subprojects.outputs.libc_any_changed == 'true'
         run: |
           cmake -B libc-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_RUNTIMES="libc" -DLLVM_ENABLE_SPHINX=ON ./runtimes
           TZ=UTC ninja -C libc-build docs-libc-html
+          mkdir built-docs/libc
+          cp -r libc-build/libc/docs/* built-docs/libc/
       - name: Build LLD docs
         if: steps.docs-changed-subprojects.outputs.lld_any_changed == 'true'
         run: |
           cmake -B lld-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="lld" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C lld-build docs-lld-html
+          mkdir built-docs/lld
+          cp -r lld-build/docs/* built-docs/lld/
       - name: Build OpenMP docs
         if: steps.docs-changed-subprojects.outputs.openmp_any_changed == 'true'
         run: |
           cmake -B openmp-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang;openmp" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C openmp-build docs-openmp-html
+          mkdir built-docs/openmp
+          cp -r openmp-build/docs/* built-docs/openmp/
       - name: Build Polly docs
         if: steps.docs-changed-subprojects.outputs.polly_any_changed == 'true'
         run: |
           cmake -B polly-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="polly" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C polly-build docs-polly-html docs-polly-man
+          mkdir built-docs/polly
+          cp -r polly-build/docs/* built-docs/polly/
       - name: Build Flang docs
         if: steps.docs-changed-subprojects.outputs.flang_any_changed == 'true'
         run: |
           cmake -B flang-build -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS="clang;mlir;flang" -DLLVM_ENABLE_SPHINX=ON ./llvm
           TZ=UTC ninja -C flang-build docs-flang-html
+          mkdir built-docs/flang
+          cp -r flang-build/docs/* built-docs/flang/
+      - name: Upload docs
+        uses: actions/upload-artifact@v4
+        with:
+          name: docs-output
+          path: built-docs/

bolt/CODE_OWNERS.TXT renamed to bolt/Maintainers.txt

@@ -170,6 +170,9 @@ class DataAggregator : public DataReader {
   std::string BuildIDBinaryName;
 
   /// Memory map info for a single file as recorded in perf.data
+  /// When a binary has multiple text segments, the Size is computed as the
+  /// difference of the last address of these segments from the BaseAddress.
+  /// The base addresses of all text segments must be the same.
   struct MMapInfo {
     uint64_t BaseAddress{0}; /// Base address of the mapped binary.
     uint64_t MMapAddress{0}; /// Address of the executable segment.
@@ -493,6 +496,11 @@ class DataAggregator : public DataReader {
   /// and return a file name matching a given \p FileBuildID.
   std::optional<StringRef> getFileNameForBuildID(StringRef FileBuildID);
 
+  /// Get a constant reference to the parsed binary mmap entries.
+  const std::unordered_map<uint64_t, MMapInfo> &getBinaryMMapInfo() {
+    return BinaryMMapInfo;
+  }
+
   friend class YAMLProfileWriter;
 };
 } // namespace bolt

bolt/include/bolt/Profile/DataAggregator.h

@@ -46,16 +46,17 @@ Error VeneerElimination::runOnFunctions(BinaryContext &BC) {
     if (BF.isIgnored())
       continue;
 
+    MCInst &FirstInstruction = *(BF.begin()->begin());
     const MCSymbol *VeneerTargetSymbol = 0;
     uint64_t TargetAddress;
-    if (BC.MIB->matchAbsLongVeneer(BF, TargetAddress)) {
+    if (BC.MIB->isTailCall(FirstInstruction)) {
+      VeneerTargetSymbol = BC.MIB->getTargetSymbol(FirstInstruction);
+    } else if (BC.MIB->matchAbsLongVeneer(BF, TargetAddress)) {
       if (BinaryFunction *TargetBF =
               BC.getBinaryFunctionAtAddress(TargetAddress))
         VeneerTargetSymbol = TargetBF->getSymbol();
-    } else {
-      MCInst &FirstInstruction = *(BF.begin()->begin());
-      if (BC.MIB->hasAnnotation(FirstInstruction, "AArch64Veneer"))
-        VeneerTargetSymbol = BC.MIB->getTargetSymbol(FirstInstruction, 1);
+    } else if (BC.MIB->hasAnnotation(FirstInstruction, "AArch64Veneer")) {
+      VeneerTargetSymbol = BC.MIB->getTargetSymbol(FirstInstruction, 1);
     }
 
     if (!VeneerTargetSymbol)

bolt/lib/Passes/VeneerElimination.cpp

@@ -95,6 +95,12 @@ cl::opt<bool> ReadPreAggregated(
     "pa", cl::desc("skip perf and read data from a pre-aggregated file format"),
     cl::cat(AggregatorCategory));
 
+cl::opt<std::string>
+    ReadPerfEvents("perf-script-events",
+                   cl::desc("skip perf event collection by supplying a "
+                            "perf-script output in a textual format"),
+                   cl::ReallyHidden, cl::init(""), cl::cat(AggregatorCategory));
+
 static cl::opt<bool>
 TimeAggregator("time-aggr",
   cl::desc("time BOLT aggregator"),
@@ -167,8 +173,9 @@ void DataAggregator::findPerfExecutable() {
 void DataAggregator::start() {
   outs() << "PERF2BOLT: Starting data aggregation job for " << Filename << "\n";
 
-  // Don't launch perf for pre-aggregated files
-  if (opts::ReadPreAggregated)
+  // Don't launch perf for pre-aggregated files or when perf input is specified
+  // by the user.
+  if (opts::ReadPreAggregated || !opts::ReadPerfEvents.empty())
     return;
 
   findPerfExecutable();
@@ -464,6 +471,13 @@ void DataAggregator::filterBinaryMMapInfo() {
 
 int DataAggregator::prepareToParse(StringRef Name, PerfProcessInfo &Process,
                                    PerfProcessErrorCallbackTy Callback) {
+  if (!opts::ReadPerfEvents.empty()) {
+    outs() << "PERF2BOLT: using pre-processed perf events for '" << Name
+           << "' (perf-script-events)\n";
+    ParsingBuf = opts::ReadPerfEvents;
+    return 0;
+  }
+
   std::string Error;
   outs() << "PERF2BOLT: waiting for perf " << Name
          << " collection to finish...\n";
@@ -2056,15 +2070,6 @@ std::error_code DataAggregator::parseMMapEvents() {
     if (FileMMapInfo.first == "(deleted)")
       continue;
 
-    // Consider only the first mapping of the file for any given PID
-    auto Range = GlobalMMapInfo.equal_range(FileMMapInfo.first);
-    bool PIDExists = llvm::any_of(make_range(Range), [&](const auto &MI) {
-      return MI.second.PID == FileMMapInfo.second.PID;
-    });
-
-    if (PIDExists)
-      continue;
-
     GlobalMMapInfo.insert(FileMMapInfo);
   }
 
@@ -2116,12 +2121,22 @@ std::error_code DataAggregator::parseMMapEvents() {
                << " using file offset 0x" << Twine::utohexstr(MMapInfo.Offset)
                << ". Ignoring profile data for this mapping\n";
         continue;
-      } else {
-        MMapInfo.BaseAddress = *BaseAddress;
       }
+      MMapInfo.BaseAddress = *BaseAddress;
     }
 
-    BinaryMMapInfo.insert(std::make_pair(MMapInfo.PID, MMapInfo));
+    // Try to add MMapInfo to the map and update its size. Large binaries may
+    // span to multiple text segments, so the mapping is inserted only on the
+    // first occurrence.
+    if (!BinaryMMapInfo.insert(std::make_pair(MMapInfo.PID, MMapInfo)).second)
+      assert(MMapInfo.BaseAddress == BinaryMMapInfo[MMapInfo.PID].BaseAddress &&
+             "Base address on multiple segment mappings should match");
+
+    // Update mapping size.
+    const uint64_t EndAddress = MMapInfo.MMapAddress + MMapInfo.Size;
+    const uint64_t Size = EndAddress - BinaryMMapInfo[MMapInfo.PID].BaseAddress;
+    if (Size > BinaryMMapInfo[MMapInfo.PID].Size)
+      BinaryMMapInfo[MMapInfo.PID].Size = Size;
   }
 
   if (BinaryMMapInfo.empty()) {

bolt/lib/Profile/DataAggregator.cpp

@@ -1,5 +1,5 @@
-## Check that llvm-bolt correctly recognizes long absolute thunks generated
-## by LLD.
+## Check that llvm-bolt correctly recognizes veneers/thunks for absolute code
+## generated by LLD.
 
 # 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 \
@@ -12,40 +12,63 @@
 
 .text
 .balign 4
-.global foo
-.type foo, %function
-foo:
-  adrp x1, foo
+.global far_function
+.type far_function, %function
+far_function:
   ret
-.size foo, .-foo
+.size far_function, .-far_function
+
+.global near_function
+.type near_function, %function
+near_function:
+  ret
+.size near_function, .-near_function
+
+## Force relocations against .text.
+.reloc 0, R_AARCH64_NONE
 
 .section ".mytext", "ax"
 .balign 4
 
-.global __AArch64AbsLongThunk_foo
-.type __AArch64AbsLongThunk_foo, %function
-__AArch64AbsLongThunk_foo:
+## This version of a thunk is always generated by LLD for function calls
+## spanning more than 256MB.
+.global __AArch64AbsLongThunk_far_function
+.type __AArch64AbsLongThunk_far_function, %function
+__AArch64AbsLongThunk_far_function:
   ldr x16, .L1
   br x16
-# CHECK-INPUT-LABEL: <__AArch64AbsLongThunk_foo>:
+# CHECK-INPUT-LABEL: <__AArch64AbsLongThunk_far_function>:
 # CHECK-INPUT-NEXT:    ldr
 # CHECK-INPUT-NEXT:    br
 .L1:
-  .quad foo
-.size __AArch64AbsLongThunk_foo, .-__AArch64AbsLongThunk_foo
+  .quad far_function
+.size __AArch64AbsLongThunk_far_function, .-__AArch64AbsLongThunk_far_function
+
+## If a callee is closer than 256MB away, LLD may generate a thunk with a direct
+## jump to the callee. Note, that the name might still include "AbsLong".
+.global __AArch64AbsLongThunk_near_function
+.type __AArch64AbsLongThunk_near_function, %function
+__AArch64AbsLongThunk_near_function:
+  b near_function
+# CHECK-INPUT-LABEL: <__AArch64AbsLongThunk_near_function>:
+# CHECK-INPUT-NEXT:    b {{.*}} <near_function>
+.size __AArch64AbsLongThunk_near_function, .-__AArch64AbsLongThunk_near_function
 
-## Check that the thunk was removed from .text and _start() calls foo()
+## Check that thunks were removed from .text, and _start calls functions
 ## directly.
 
-# CHECK-OUTPUT-NOT: __AArch64AbsLongThunk_foo
+# CHECK-OUTPUT-NOT: __AArch64AbsLongThunk_{{.*}}
 
 .global _start
 .type _start, %function
 _start:
 # CHECK-INPUT-LABEL:  <_start>:
 # CHECK-OUTPUT-LABEL: <_start>:
-  bl __AArch64AbsLongThunk_foo
-# CHECK-INPUT-NEXT:     bl {{.*}} <__AArch64AbsLongThunk_foo>
-# CHECK-OUTPUT-NEXT:    bl {{.*}} <foo>
+  bl __AArch64AbsLongThunk_far_function
+  bl __AArch64AbsLongThunk_near_function
+# CHECK-INPUT-NEXT:     bl {{.*}} <__AArch64AbsLongThunk_far_function>
+# CHECK-INPUT-NEXT:     bl {{.*}} <__AArch64AbsLongThunk_near_function>
+# CHECK-OUTPUT-NEXT:    bl {{.*}} <far_function>
+# CHECK-OUTPUT-NEXT:    bl {{.*}} <near_function>
   ret
 .size _start, .-_start

bolt/test/AArch64/veneer-lld-abs.s

@@ -8,6 +8,7 @@ set(LLVM_LINK_COMPONENTS
 add_bolt_unittest(CoreTests
   BinaryContext.cpp
   MCPlusBuilder.cpp
+  MemoryMaps.cpp
   DynoStats.cpp
 
   DISABLE_LLVM_LINK_LLVM_DYLIB
@@ -17,6 +18,8 @@ target_link_libraries(CoreTests
   PRIVATE
   LLVMBOLTCore
   LLVMBOLTRewrite
+  LLVMBOLTProfile
+  LLVMTestingSupport
   )
 
 foreach (tgt ${BOLT_TARGETS_TO_BUILD})