[BOLT][DWARF][NFC] Refactor GDB Index into a new file #94405

sayhaan · 2024-06-04T22:10:16Z

Create a new class and file for functions that update GDB index.

llvmbot · 2024-06-04T22:10:44Z

@llvm/pr-subscribers-mc
@llvm/pr-subscribers-backend-x86
@llvm/pr-subscribers-clang-driver
@llvm/pr-subscribers-bolt

@llvm/pr-subscribers-lldb

Author: Sayhaan Siddiqui (sayhaan)

Changes

Create a new class and file for functions that update GDB index.

Patch is 40.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/94405.diff

24 Files Affected:

(added) bolt/include/bolt/Core/GDBIndex.h (+60)
(modified) bolt/lib/Core/BinaryEmitter.cpp (+1)
(modified) bolt/lib/Core/CMakeLists.txt (+1)
(added) bolt/lib/Core/GDBIndex.cpp (+182)
(modified) bolt/lib/Passes/ValidateMemRefs.cpp (+5-6)
(modified) clang/include/clang/Driver/Options.td (+4)
(modified) clang/lib/Driver/ToolChains/Gnu.cpp (+29)
(modified) cross-project-tests/lit.cfg.py (+13-1)
(modified) cross-project-tests/lit.site.cfg.py.in (+4)
(modified) lldb/test/API/lit.cfg.py (+5)
(modified) lldb/test/API/lit.site.cfg.py.in (+8)
(modified) lldb/test/Shell/helper/toolchain.py (+5)
(modified) lldb/test/Shell/lit.site.cfg.py.in (+9)
(modified) llvm/CMakeLists.txt (+4)
(modified) llvm/include/llvm/MC/MCFragment.h (+22)
(modified) llvm/include/llvm/MC/MCObjectStreamer.h (+2)
(modified) llvm/include/llvm/MC/MCStreamer.h (+6)
(modified) llvm/lib/MC/MCAssembler.cpp (+81-37)
(modified) llvm/lib/MC/MCExpr.cpp (+2-8)
(modified) llvm/lib/MC/MCFragment.cpp (+12)
(modified) llvm/lib/MC/MCObjectStreamer.cpp (+5)
(modified) llvm/lib/MC/MCStreamer.cpp (+2)
(modified) llvm/lib/Target/X86/AsmParser/X86AsmParser.cpp (+24)
(added) llvm/test/MC/X86/directive-avoid_end_align.s (+208)

diff --git a/bolt/include/bolt/Core/GDBIndex.h b/bolt/include/bolt/Core/GDBIndex.h
new file mode 100644
index 00000000000000..0ea588e9cba465
--- /dev/null
+++ b/bolt/include/bolt/Core/GDBIndex.h
@@ -0,0 +1,60 @@
+//===-- bolt/Core/GDBIndex.h - GDB Index support -------*- C++ -*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// This file contains declaration of classes required for generation of
+/// .gdb_index section.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef BOLT_CORE_GDB_INDEX_H
+#define BOLT_CORE_GDB_INDEX_H
+
+#include "bolt/Core/BinaryContext.h"
+#include <vector>
+
+namespace llvm {
+namespace bolt {
+
+class GDBIndex {
+public:
+  /// Contains information about TU so we can write out correct entries in GDB
+  /// index.
+  struct GDBIndexTUEntry {
+    uint64_t UnitOffset;
+    uint64_t TypeHash;
+    uint64_t TypeDIERelativeOffset;
+  };
+
+private:
+  BinaryContext &BC;
+
+  /// Entries for GDB Index Types CU List
+  using GDBIndexTUEntryType = std::vector<GDBIndexTUEntry>;
+  GDBIndexTUEntryType GDBIndexTUEntryVector;
+
+public:
+  GDBIndex(BinaryContext &BC) : BC(BC) {}
+
+  /// Adds an GDBIndexTUEntry if .gdb_index section exists.
+  void addGDBTypeUnitEntry(const GDBIndexTUEntry &Entry);
+
+  /// Rewrite .gdb_index section if present.
+  void updateGdbIndexSection(
+      CUOffsetMap &CUMap, uint32_t NumCUs,
+      std::unique_ptr<DebugARangesSectionWriter> &ARangesSectionWriter);
+
+  /// Returns all entries needed for Types CU list
+  const GDBIndexTUEntryType &getGDBIndexTUEntryVector() const {
+    return GDBIndexTUEntryVector;
+  }
+};
+
+} // namespace bolt
+} // namespace llvm
+
+#endif
diff --git a/bolt/lib/Core/BinaryEmitter.cpp b/bolt/lib/Core/BinaryEmitter.cpp
index 0b44acb0816f2f..e09b9ff92de064 100644
--- a/bolt/lib/Core/BinaryEmitter.cpp
+++ b/bolt/lib/Core/BinaryEmitter.cpp
@@ -485,6 +485,7 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         // This assumes the second instruction in the macro-op pair will get
         // assigned to its own MCRelaxableFragment. Since all JCC instructions
         // are relaxable, we should be safe.
+        Streamer.emitNeverAlignCodeAtEnd(/*Alignment to avoid=*/64, *BC.STI);
       }
 
       if (!EmitCodeOnly) {
diff --git a/bolt/lib/Core/CMakeLists.txt b/bolt/lib/Core/CMakeLists.txt
index 441df9fe084648..873cf67a56291f 100644
--- a/bolt/lib/Core/CMakeLists.txt
+++ b/bolt/lib/Core/CMakeLists.txt
@@ -25,6 +25,7 @@ add_llvm_library(LLVMBOLTCore
   DynoStats.cpp
   Exceptions.cpp
   FunctionLayout.cpp
+  GDBIndex.cpp
   HashUtilities.cpp
   JumpTable.cpp
   MCPlusBuilder.cpp
diff --git a/bolt/lib/Core/GDBIndex.cpp b/bolt/lib/Core/GDBIndex.cpp
new file mode 100644
index 00000000000000..fda19a2c8cb3ef
--- /dev/null
+++ b/bolt/lib/Core/GDBIndex.cpp
@@ -0,0 +1,182 @@
+//===- bolt/Rewrite/GDBIndex.cpp -------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "bolt/Core/GDBIndex.h"
+
+using namespace llvm::bolt;
+using namespace llvm::support::endian;
+
+void GDBIndex::addGDBTypeUnitEntry(const GDBIndexTUEntry &Entry) {
+  GDBIndexTUEntryVector.push_back(Entry);
+}
+
+void GDBIndex::updateGdbIndexSection(
+    CUOffsetMap &CUMap, uint32_t NumCUs,
+    std::unique_ptr<DebugARangesSectionWriter> &ARangesSectionWriter) {
+  if (!BC.getGdbIndexSection())
+    return;
+
+  // See https://sourceware.org/gdb/onlinedocs/gdb/Index-Section-Format.html
+  // for .gdb_index section format.
+
+  StringRef GdbIndexContents = BC.getGdbIndexSection()->getContents();
+
+  const char *Data = GdbIndexContents.data();
+
+  // Parse the header.
+  const uint32_t Version = read32le(Data);
+  if (Version != 7 && Version != 8) {
+    errs() << "BOLT-ERROR: can only process .gdb_index versions 7 and 8\n";
+    exit(1);
+  }
+
+  // Some .gdb_index generators use file offsets while others use section
+  // offsets. Hence we can only rely on offsets relative to each other,
+  // and ignore their absolute values.
+  const uint32_t CUListOffset = read32le(Data + 4);
+  const uint32_t CUTypesOffset = read32le(Data + 8);
+  const uint32_t AddressTableOffset = read32le(Data + 12);
+  const uint32_t SymbolTableOffset = read32le(Data + 16);
+  const uint32_t ConstantPoolOffset = read32le(Data + 20);
+  Data += 24;
+
+  // Map CUs offsets to indices and verify existing index table.
+  std::map<uint32_t, uint32_t> OffsetToIndexMap;
+  const uint32_t CUListSize = CUTypesOffset - CUListOffset;
+  const uint32_t TUListSize = AddressTableOffset - CUTypesOffset;
+  const unsigned NUmCUsEncoded = CUListSize / 16;
+  unsigned MaxDWARFVersion = BC.DwCtx->getMaxVersion();
+  unsigned NumDWARF5TUs =
+      getGDBIndexTUEntryVector().size() - BC.DwCtx->getNumTypeUnits();
+  bool SkipTypeUnits = false;
+  // For DWARF5 Types are in .debug_info.
+  // LLD doesn't generate Types CU List, and in CU list offset
+  // only includes CUs.
+  // GDB 11+ includes only CUs in CU list and generates Types
+  // list.
+  // GDB 9 includes CUs and TUs in CU list and generates TYpes
+  // list. The NumCUs is CUs + TUs, so need to modify the check.
+  // For split-dwarf
+  // GDB-11, DWARF5: TU units from dwo are not included.
+  // GDB-11, DWARF4: TU units from dwo are included.
+  if (MaxDWARFVersion >= 5)
+    SkipTypeUnits = !TUListSize ? true
+                                : ((NUmCUsEncoded + NumDWARF5TUs) ==
+                                   BC.DwCtx->getNumCompileUnits());
+
+  if (!((CUListSize == NumCUs * 16) ||
+        (CUListSize == (NumCUs + NumDWARF5TUs) * 16))) {
+    errs() << "BOLT-ERROR: .gdb_index: CU count mismatch\n";
+    exit(1);
+  }
+  DenseSet<uint64_t> OriginalOffsets;
+  for (unsigned Index = 0, Units = BC.DwCtx->getNumCompileUnits();
+       Index < Units; ++Index) {
+    const DWARFUnit *CU = BC.DwCtx->getUnitAtIndex(Index);
+    if (SkipTypeUnits && CU->isTypeUnit())
+      continue;
+    const uint64_t Offset = read64le(Data);
+    Data += 16;
+    if (CU->getOffset() != Offset) {
+      errs() << "BOLT-ERROR: .gdb_index CU offset mismatch\n";
+      exit(1);
+    }
+
+    OriginalOffsets.insert(Offset);
+    OffsetToIndexMap[Offset] = Index;
+  }
+
+  // Ignore old address table.
+  const uint32_t OldAddressTableSize = SymbolTableOffset - AddressTableOffset;
+  // Move Data to the beginning of symbol table.
+  Data += SymbolTableOffset - CUTypesOffset;
+
+  // Calculate the size of the new address table.
+  uint32_t NewAddressTableSize = 0;
+  for (const auto &CURangesPair : ARangesSectionWriter->getCUAddressRanges()) {
+    const SmallVector<DebugAddressRange, 2> &Ranges = CURangesPair.second;
+    NewAddressTableSize += Ranges.size() * 20;
+  }
+
+  // Difference between old and new table (and section) sizes.
+  // Could be negative.
+  int32_t Delta = NewAddressTableSize - OldAddressTableSize;
+
+  size_t NewGdbIndexSize = GdbIndexContents.size() + Delta;
+
+  // Free'd by ExecutableFileMemoryManager.
+  auto *NewGdbIndexContents = new uint8_t[NewGdbIndexSize];
+  uint8_t *Buffer = NewGdbIndexContents;
+
+  write32le(Buffer, Version);
+  write32le(Buffer + 4, CUListOffset);
+  write32le(Buffer + 8, CUTypesOffset);
+  write32le(Buffer + 12, AddressTableOffset);
+  write32le(Buffer + 16, SymbolTableOffset + Delta);
+  write32le(Buffer + 20, ConstantPoolOffset + Delta);
+  Buffer += 24;
+
+  using MapEntry = std::pair<uint32_t, CUInfo>;
+  std::vector<MapEntry> CUVector(CUMap.begin(), CUMap.end());
+  // Need to sort since we write out all of TUs in .debug_info before CUs.
+  std::sort(CUVector.begin(), CUVector.end(),
+            [](const MapEntry &E1, const MapEntry &E2) -> bool {
+              return E1.second.Offset < E2.second.Offset;
+            });
+  // Writing out CU List <Offset, Size>
+  for (auto &CUInfo : CUVector) {
+    // Skipping TU for DWARF5 when they are not included in CU list.
+    if (!OriginalOffsets.count(CUInfo.first))
+      continue;
+    write64le(Buffer, CUInfo.second.Offset);
+    // Length encoded in CU doesn't contain first 4 bytes that encode length.
+    write64le(Buffer + 8, CUInfo.second.Length + 4);
+    Buffer += 16;
+  }
+
+  // Rewrite TU CU List, since abbrevs can be different.
+  // Entry example:
+  // 0: offset = 0x00000000, type_offset = 0x0000001e, type_signature =
+  // 0x418503b8111e9a7b Spec says " triplet, the first value is the CU offset,
+  // the second value is the type offset in the CU, and the third value is the
+  // type signature" Looking at what is being generated by gdb-add-index. The
+  // first entry is TU offset, second entry is offset from it, and third entry
+  // is the type signature.
+  if (TUListSize)
+    for (const GDBIndexTUEntry &Entry : getGDBIndexTUEntryVector()) {
+      write64le(Buffer, Entry.UnitOffset);
+      write64le(Buffer + 8, Entry.TypeDIERelativeOffset);
+      write64le(Buffer + 16, Entry.TypeHash);
+      Buffer += sizeof(GDBIndexTUEntry);
+    }
+
+  // Generate new address table.
+  for (const std::pair<const uint64_t, DebugAddressRangesVector> &CURangesPair :
+       ARangesSectionWriter->getCUAddressRanges()) {
+    const uint32_t CUIndex = OffsetToIndexMap[CURangesPair.first];
+    const DebugAddressRangesVector &Ranges = CURangesPair.second;
+    for (const DebugAddressRange &Range : Ranges) {
+      write64le(Buffer, Range.LowPC);
+      write64le(Buffer + 8, Range.HighPC);
+      write32le(Buffer + 16, CUIndex);
+      Buffer += 20;
+    }
+  }
+
+  const size_t TrailingSize =
+      GdbIndexContents.data() + GdbIndexContents.size() - Data;
+  assert(Buffer + TrailingSize == NewGdbIndexContents + NewGdbIndexSize &&
+         "size calculation error");
+
+  // Copy over the rest of the original data.
+  memcpy(Buffer, Data, TrailingSize);
+
+  // Register the new section.
+  BC.registerOrUpdateNoteSection(".gdb_index", NewGdbIndexContents,
+                                 NewGdbIndexSize);
+}
diff --git a/bolt/lib/Passes/ValidateMemRefs.cpp b/bolt/lib/Passes/ValidateMemRefs.cpp
index f29a97c43f497c..1d2c230fa7106a 100644
--- a/bolt/lib/Passes/ValidateMemRefs.cpp
+++ b/bolt/lib/Passes/ValidateMemRefs.cpp
@@ -29,8 +29,7 @@ bool ValidateMemRefs::checkAndFixJTReference(BinaryFunction &BF, MCInst &Inst,
   if (!BD)
     return false;
 
-  const uint64_t TargetAddress = BD->getAddress() + Offset;
-  JumpTable *JT = BC.getJumpTableContainingAddress(TargetAddress);
+  JumpTable *JT = BC.getJumpTableContainingAddress(BD->getAddress());
   if (!JT)
     return false;
 
@@ -41,10 +40,10 @@ bool ValidateMemRefs::checkAndFixJTReference(BinaryFunction &BF, MCInst &Inst,
   // Accessing a jump table in another function. This is not a
   // legitimate jump table access, we need to replace the reference to
   // the jump table label with a regular rodata reference. Get a
-  // non-JT reference by fetching the symbol 1 byte before the JT
-  // label.
-  MCSymbol *NewSym = BC.getOrCreateGlobalSymbol(TargetAddress - 1, "DATAat");
-  BC.MIB->setOperandToSymbolRef(Inst, OperandNum, NewSym, 1, &*BC.Ctx, 0);
+  // non-JT reference by fetching the symbol 1 byte before the JT label.
+  MCSymbol *NewSym = BC.getOrCreateGlobalSymbol(BD->getAddress() - 1, "DATAat");
+  BC.MIB->setOperandToSymbolRef(Inst, OperandNum, NewSym, Offset + 1, &*BC.Ctx,
+                                0);
   LLVM_DEBUG(dbgs() << "BOLT-DEBUG: replaced reference @" << BF.getPrintName()
                     << " from " << BD->getName() << " to " << NewSym->getName()
                     << " + 1\n");
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 57f37c5023110f..7469db06839e3b 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5465,6 +5465,10 @@ def pg : Flag<["-"], "pg">, HelpText<"Enable mcount instrumentation">,
   MarshallingInfoFlag<CodeGenOpts<"InstrumentForProfiling">>;
 def pipe : Flag<["-", "--"], "pipe">,
   HelpText<"Use pipes between commands, when possible">;
+// Facebook T92898286
+def post_link_optimize : Flag<["--"], "post-link-optimize">,
+  HelpText<"Apply post-link optimizations using BOLT">;
+// End Facebook T92898286
 def prebind__all__twolevel__modules : Flag<["-"], "prebind_all_twolevel_modules">;
 def prebind : Flag<["-"], "prebind">;
 def preload : Flag<["-"], "preload">;
diff --git a/clang/lib/Driver/ToolChains/Gnu.cpp b/clang/lib/Driver/ToolChains/Gnu.cpp
index b141e5f2adfab1..f7611af5763ab7 100644
--- a/clang/lib/Driver/ToolChains/Gnu.cpp
+++ b/clang/lib/Driver/ToolChains/Gnu.cpp
@@ -672,12 +672,41 @@ void tools::gnutools::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     }
   }
 
+  // Facebook T92898286
+  if (Args.hasArg(options::OPT_post_link_optimize))
+    CmdArgs.push_back("-q");
+  // End Facebook T92898286
+
   Args.AddAllArgs(CmdArgs, options::OPT_T);
 
   const char *Exec = Args.MakeArgString(ToolChain.GetLinkerPath());
   C.addCommand(std::make_unique<Command>(JA, *this,
                                          ResponseFileSupport::AtFileCurCP(),
                                          Exec, CmdArgs, Inputs, Output));
+  // Facebook T92898286
+  if (!Args.hasArg(options::OPT_post_link_optimize) || !Output.isFilename())
+    return;
+
+  const char *MvExec = Args.MakeArgString(ToolChain.GetProgramPath("mv"));
+  ArgStringList MoveCmdArgs;
+  MoveCmdArgs.push_back(Output.getFilename());
+  const char *PreBoltBin =
+      Args.MakeArgString(Twine(Output.getFilename()) + ".pre-bolt");
+  MoveCmdArgs.push_back(PreBoltBin);
+  C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(),
+                                         MvExec, MoveCmdArgs, std::nullopt));
+
+  ArgStringList BoltCmdArgs;
+  const char *BoltExec =
+      Args.MakeArgString(ToolChain.GetProgramPath("llvm-bolt"));
+  BoltCmdArgs.push_back(PreBoltBin);
+  BoltCmdArgs.push_back("-reorder-blocks=reverse");
+  BoltCmdArgs.push_back("-update-debug-sections");
+  BoltCmdArgs.push_back("-o");
+  BoltCmdArgs.push_back(Output.getFilename());
+  C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(),
+                                         BoltExec, BoltCmdArgs, std::nullopt));
+  // End Facebook T92898286
 }
 
 void tools::gnutools::Assembler::ConstructJob(Compilation &C,
diff --git a/cross-project-tests/lit.cfg.py b/cross-project-tests/lit.cfg.py
index 774c4eaf4d976b..619634578dfe60 100644
--- a/cross-project-tests/lit.cfg.py
+++ b/cross-project-tests/lit.cfg.py
@@ -84,7 +84,13 @@ def get_required_attr(config, attr_name):
 # use_clang() and use_lld() respectively, so set them to "", if needed.
 if not hasattr(config, "clang_src_dir"):
     config.clang_src_dir = ""
-llvm_config.use_clang(required=("clang" in config.llvm_enabled_projects))
+# Facebook T92898286
+should_test_bolt = get_required_attr(config, "llvm_test_bolt")
+if should_test_bolt:
+    llvm_config.use_clang(required=("clang" in config.llvm_enabled_projects), additional_flags=["--post-link-optimize"])
+else:
+    llvm_config.use_clang(required=("clang" in config.llvm_enabled_projects))
+# End Facebook T92898286
 
 if not hasattr(config, "lld_src_dir"):
     config.lld_src_dir = ""
@@ -293,3 +299,9 @@ def get_clang_default_dwarf_version_string(triple):
 # Allow 'REQUIRES: XXX-registered-target' in tests.
 for arch in config.targets_to_build:
     config.available_features.add(arch.lower() + "-registered-target")
+
+# Facebook T92898286
+# Ensure the user's PYTHONPATH is included.
+if "PYTHONPATH" in os.environ:
+    config.environment["PYTHONPATH"] = os.environ["PYTHONPATH"]
+# End Facebook T92898286
diff --git a/cross-project-tests/lit.site.cfg.py.in b/cross-project-tests/lit.site.cfg.py.in
index 39458dfc79afd2..2d53cd377f0330 100644
--- a/cross-project-tests/lit.site.cfg.py.in
+++ b/cross-project-tests/lit.site.cfg.py.in
@@ -21,6 +21,10 @@ config.mlir_src_root = "@MLIR_SOURCE_DIR@"
 
 config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
 
+# Facebook T92898286
+config.llvm_test_bolt = lit.util.pythonize_bool("@LLVM_TEST_BOLT@")
+# End Facebook T92898286
+
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
diff --git a/lldb/test/API/lit.cfg.py b/lldb/test/API/lit.cfg.py
index d934349fe3ca3d..d4a62c51458cc1 100644
--- a/lldb/test/API/lit.cfg.py
+++ b/lldb/test/API/lit.cfg.py
@@ -248,6 +248,11 @@ def delete_module_cache(path):
 if is_configured("lldb_framework_dir"):
     dotest_cmd += ["--framework", config.lldb_framework_dir]
 
+# Facebook T92898286
+if is_configured("llvm_test_bolt"):
+    dotest_cmd += ["-E", '"--post-link-optimize"']
+# End Facebook T92898286
+
 if (
     "lldb-repro-capture" in config.available_features
     or "lldb-repro-replay" in config.available_features
diff --git a/lldb/test/API/lit.site.cfg.py.in b/lldb/test/API/lit.site.cfg.py.in
index 8b2d09ae41cd2a..602f45759e48f3 100644
--- a/lldb/test/API/lit.site.cfg.py.in
+++ b/lldb/test/API/lit.site.cfg.py.in
@@ -1,5 +1,9 @@
 @LIT_SITE_CFG_IN_HEADER@
 
+#Facebook T92898286
+import lit.util
+#End Facebook T92898286
+
 config.llvm_src_root = "@LLVM_SOURCE_DIR@"
 config.llvm_obj_root = "@LLVM_BINARY_DIR@"
 config.llvm_tools_dir = lit_config.substitute("@LLVM_TOOLS_DIR@")
@@ -39,6 +43,10 @@ config.libcxx_include_target_dir = "@LIBCXX_GENERATED_INCLUDE_TARGET_DIR@"
 config.lldb_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_LLDB@", "lldb-api")
 config.clang_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_CLANG@", "lldb-api")
 
+# Facebook T92898286
+config.llvm_test_bolt = lit.util.pythonize_bool("@LLVM_TEST_BOLT@")
+# End Facebook T92898286
+
 # Plugins
 lldb_build_intel_pt = '@LLDB_BUILD_INTEL_PT@'
 if lldb_build_intel_pt == '1':
diff --git a/lldb/test/Shell/helper/toolchain.py b/lldb/test/Shell/helper/toolchain.py
index 255955fc70d8c4..7b7be06643166d 100644
--- a/lldb/test/Shell/helper/toolchain.py
+++ b/lldb/test/Shell/helper/toolchain.py
@@ -165,6 +165,11 @@ def use_support_substitutions(config):
     if config.cmake_sysroot:
         host_flags += ["--sysroot={}".format(config.cmake_sysroot)]
 
+    # Facebook T92898286
+    if config.llvm_test_bolt:
+        host_flags += ["--post-link-optimize"]
+    # End Facebook T92898286
+
     host_flags = " ".join(host_flags)
     config.substitutions.append(("%clang_host", "%clang " + host_flags))
     config.substitutions.append(("%clangxx_host", "%clangxx " + host_flags))
diff --git a/lldb/test/Shell/lit.site.cfg.py.in b/lldb/test/Shell/lit.site.cfg.py.in
index b69e7bce1bc0be..fe8323734b7dbf 100644
--- a/lldb/test/Shell/lit.site.cfg.py.in
+++ b/lldb/test/Shell/lit.site.cfg.py.in
@@ -1,5 +1,10 @@
 @LIT_SITE_CFG_IN_HEADER@
 
+#Facebook T92898286
+import lit.util
+#End Facebook T92898286
+
+
 config.llvm_src_root = "@LLVM_SOURCE_DIR@"
 config.llvm_obj_root = "@LLVM_BINARY_DIR@"
 config.llvm_tools_dir = lit_config.substitute("@LLVM_TOOLS_DIR@")
@@ -31,6 +36,10 @@ config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
 config.lldb_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_LLDB@", "lldb-shell")
 config.clang_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_CLANG@", "lldb-shell")
 
+# Facebook T92898286
+config.llvm_test_bolt = lit.util.pythonize_bool("@LLVM_TEST_BOLT@")
+# End Facebook T92898286
+
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
diff --git a/llvm/CMakeLists.txt b/llvm/CMakeLists.txt
index 64898ab09772f4..c4ac19eaca2148 100644
--- a/llvm/CMakeLists.txt
+++ b/llvm/CMakeLists.txt
@@ -709,6 +709,10 @@ set(LLVM_LIB_FUZZING_ENGINE "" CACHE PATH
 option(LLVM_USE_SPLIT_DWARF
   "Use -gsplit-dwarf when compiling llvm and --gdb-index when linking." OFF)
 
+# Facebook T92898286
+option(LLVM_TEST_BOLT "Enable BOLT testing in non-BOLT tests that use clang" OFF)
+# End Facebook T92898286
+
 # Define an option controlling whether we should build for 32-bit on 64-bit
 # platforms, where supported.
 if( CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT (WIN32 OR ${CMAKE_SYSTEM_NAME} MATCHES "AIX"))
diff --git a/llvm/include/llv...
[truncated]

llvmbot · 2024-06-04T22:10:45Z

@llvm/pr-subscribers-clang

Author: Sayhaan Siddiqui (sayhaan)

Changes

Create a new class and file for functions that update GDB index.

Patch is 40.93 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/94405.diff

24 Files Affected:

(added) bolt/include/bolt/Core/GDBIndex.h (+60)
(modified) bolt/lib/Core/BinaryEmitter.cpp (+1)
(modified) bolt/lib/Core/CMakeLists.txt (+1)
(added) bolt/lib/Core/GDBIndex.cpp (+182)
(modified) bolt/lib/Passes/ValidateMemRefs.cpp (+5-6)
(modified) clang/include/clang/Driver/Options.td (+4)
(modified) clang/lib/Driver/ToolChains/Gnu.cpp (+29)
(modified) cross-project-tests/lit.cfg.py (+13-1)
(modified) cross-project-tests/lit.site.cfg.py.in (+4)
(modified) lldb/test/API/lit.cfg.py (+5)
(modified) lldb/test/API/lit.site.cfg.py.in (+8)
(modified) lldb/test/Shell/helper/toolchain.py (+5)
(modified) lldb/test/Shell/lit.site.cfg.py.in (+9)
(modified) llvm/CMakeLists.txt (+4)
(modified) llvm/include/llvm/MC/MCFragment.h (+22)
(modified) llvm/include/llvm/MC/MCObjectStreamer.h (+2)
(modified) llvm/include/llvm/MC/MCStreamer.h (+6)
(modified) llvm/lib/MC/MCAssembler.cpp (+81-37)
(modified) llvm/lib/MC/MCExpr.cpp (+2-8)
(modified) llvm/lib/MC/MCFragment.cpp (+12)
(modified) llvm/lib/MC/MCObjectStreamer.cpp (+5)
(modified) llvm/lib/MC/MCStreamer.cpp (+2)
(modified) llvm/lib/Target/X86/AsmParser/X86AsmParser.cpp (+24)
(added) llvm/test/MC/X86/directive-avoid_end_align.s (+208)

diff --git a/bolt/include/bolt/Core/GDBIndex.h b/bolt/include/bolt/Core/GDBIndex.h
new file mode 100644
index 00000000000000..0ea588e9cba465
--- /dev/null
+++ b/bolt/include/bolt/Core/GDBIndex.h
@@ -0,0 +1,60 @@
+//===-- bolt/Core/GDBIndex.h - GDB Index support -------*- C++ -*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// This file contains declaration of classes required for generation of
+/// .gdb_index section.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef BOLT_CORE_GDB_INDEX_H
+#define BOLT_CORE_GDB_INDEX_H
+
+#include "bolt/Core/BinaryContext.h"
+#include <vector>
+
+namespace llvm {
+namespace bolt {
+
+class GDBIndex {
+public:
+  /// Contains information about TU so we can write out correct entries in GDB
+  /// index.
+  struct GDBIndexTUEntry {
+    uint64_t UnitOffset;
+    uint64_t TypeHash;
+    uint64_t TypeDIERelativeOffset;
+  };
+
+private:
+  BinaryContext &BC;
+
+  /// Entries for GDB Index Types CU List
+  using GDBIndexTUEntryType = std::vector<GDBIndexTUEntry>;
+  GDBIndexTUEntryType GDBIndexTUEntryVector;
+
+public:
+  GDBIndex(BinaryContext &BC) : BC(BC) {}
+
+  /// Adds an GDBIndexTUEntry if .gdb_index section exists.
+  void addGDBTypeUnitEntry(const GDBIndexTUEntry &Entry);
+
+  /// Rewrite .gdb_index section if present.
+  void updateGdbIndexSection(
+      CUOffsetMap &CUMap, uint32_t NumCUs,
+      std::unique_ptr<DebugARangesSectionWriter> &ARangesSectionWriter);
+
+  /// Returns all entries needed for Types CU list
+  const GDBIndexTUEntryType &getGDBIndexTUEntryVector() const {
+    return GDBIndexTUEntryVector;
+  }
+};
+
+} // namespace bolt
+} // namespace llvm
+
+#endif
diff --git a/bolt/lib/Core/BinaryEmitter.cpp b/bolt/lib/Core/BinaryEmitter.cpp
index 0b44acb0816f2f..e09b9ff92de064 100644
--- a/bolt/lib/Core/BinaryEmitter.cpp
+++ b/bolt/lib/Core/BinaryEmitter.cpp
@@ -485,6 +485,7 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
         // This assumes the second instruction in the macro-op pair will get
         // assigned to its own MCRelaxableFragment. Since all JCC instructions
         // are relaxable, we should be safe.
+        Streamer.emitNeverAlignCodeAtEnd(/*Alignment to avoid=*/64, *BC.STI);
       }
 
       if (!EmitCodeOnly) {
diff --git a/bolt/lib/Core/CMakeLists.txt b/bolt/lib/Core/CMakeLists.txt
index 441df9fe084648..873cf67a56291f 100644
--- a/bolt/lib/Core/CMakeLists.txt
+++ b/bolt/lib/Core/CMakeLists.txt
@@ -25,6 +25,7 @@ add_llvm_library(LLVMBOLTCore
   DynoStats.cpp
   Exceptions.cpp
   FunctionLayout.cpp
+  GDBIndex.cpp
   HashUtilities.cpp
   JumpTable.cpp
   MCPlusBuilder.cpp
diff --git a/bolt/lib/Core/GDBIndex.cpp b/bolt/lib/Core/GDBIndex.cpp
new file mode 100644
index 00000000000000..fda19a2c8cb3ef
--- /dev/null
+++ b/bolt/lib/Core/GDBIndex.cpp
@@ -0,0 +1,182 @@
+//===- bolt/Rewrite/GDBIndex.cpp -------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "bolt/Core/GDBIndex.h"
+
+using namespace llvm::bolt;
+using namespace llvm::support::endian;
+
+void GDBIndex::addGDBTypeUnitEntry(const GDBIndexTUEntry &Entry) {
+  GDBIndexTUEntryVector.push_back(Entry);
+}
+
+void GDBIndex::updateGdbIndexSection(
+    CUOffsetMap &CUMap, uint32_t NumCUs,
+    std::unique_ptr<DebugARangesSectionWriter> &ARangesSectionWriter) {
+  if (!BC.getGdbIndexSection())
+    return;
+
+  // See https://sourceware.org/gdb/onlinedocs/gdb/Index-Section-Format.html
+  // for .gdb_index section format.
+
+  StringRef GdbIndexContents = BC.getGdbIndexSection()->getContents();
+
+  const char *Data = GdbIndexContents.data();
+
+  // Parse the header.
+  const uint32_t Version = read32le(Data);
+  if (Version != 7 && Version != 8) {
+    errs() << "BOLT-ERROR: can only process .gdb_index versions 7 and 8\n";
+    exit(1);
+  }
+
+  // Some .gdb_index generators use file offsets while others use section
+  // offsets. Hence we can only rely on offsets relative to each other,
+  // and ignore their absolute values.
+  const uint32_t CUListOffset = read32le(Data + 4);
+  const uint32_t CUTypesOffset = read32le(Data + 8);
+  const uint32_t AddressTableOffset = read32le(Data + 12);
+  const uint32_t SymbolTableOffset = read32le(Data + 16);
+  const uint32_t ConstantPoolOffset = read32le(Data + 20);
+  Data += 24;
+
+  // Map CUs offsets to indices and verify existing index table.
+  std::map<uint32_t, uint32_t> OffsetToIndexMap;
+  const uint32_t CUListSize = CUTypesOffset - CUListOffset;
+  const uint32_t TUListSize = AddressTableOffset - CUTypesOffset;
+  const unsigned NUmCUsEncoded = CUListSize / 16;
+  unsigned MaxDWARFVersion = BC.DwCtx->getMaxVersion();
+  unsigned NumDWARF5TUs =
+      getGDBIndexTUEntryVector().size() - BC.DwCtx->getNumTypeUnits();
+  bool SkipTypeUnits = false;
+  // For DWARF5 Types are in .debug_info.
+  // LLD doesn't generate Types CU List, and in CU list offset
+  // only includes CUs.
+  // GDB 11+ includes only CUs in CU list and generates Types
+  // list.
+  // GDB 9 includes CUs and TUs in CU list and generates TYpes
+  // list. The NumCUs is CUs + TUs, so need to modify the check.
+  // For split-dwarf
+  // GDB-11, DWARF5: TU units from dwo are not included.
+  // GDB-11, DWARF4: TU units from dwo are included.
+  if (MaxDWARFVersion >= 5)
+    SkipTypeUnits = !TUListSize ? true
+                                : ((NUmCUsEncoded + NumDWARF5TUs) ==
+                                   BC.DwCtx->getNumCompileUnits());
+
+  if (!((CUListSize == NumCUs * 16) ||
+        (CUListSize == (NumCUs + NumDWARF5TUs) * 16))) {
+    errs() << "BOLT-ERROR: .gdb_index: CU count mismatch\n";
+    exit(1);
+  }
+  DenseSet<uint64_t> OriginalOffsets;
+  for (unsigned Index = 0, Units = BC.DwCtx->getNumCompileUnits();
+       Index < Units; ++Index) {
+    const DWARFUnit *CU = BC.DwCtx->getUnitAtIndex(Index);
+    if (SkipTypeUnits && CU->isTypeUnit())
+      continue;
+    const uint64_t Offset = read64le(Data);
+    Data += 16;
+    if (CU->getOffset() != Offset) {
+      errs() << "BOLT-ERROR: .gdb_index CU offset mismatch\n";
+      exit(1);
+    }
+
+    OriginalOffsets.insert(Offset);
+    OffsetToIndexMap[Offset] = Index;
+  }
+
+  // Ignore old address table.
+  const uint32_t OldAddressTableSize = SymbolTableOffset - AddressTableOffset;
+  // Move Data to the beginning of symbol table.
+  Data += SymbolTableOffset - CUTypesOffset;
+
+  // Calculate the size of the new address table.
+  uint32_t NewAddressTableSize = 0;
+  for (const auto &CURangesPair : ARangesSectionWriter->getCUAddressRanges()) {
+    const SmallVector<DebugAddressRange, 2> &Ranges = CURangesPair.second;
+    NewAddressTableSize += Ranges.size() * 20;
+  }
+
+  // Difference between old and new table (and section) sizes.
+  // Could be negative.
+  int32_t Delta = NewAddressTableSize - OldAddressTableSize;
+
+  size_t NewGdbIndexSize = GdbIndexContents.size() + Delta;
+
+  // Free'd by ExecutableFileMemoryManager.
+  auto *NewGdbIndexContents = new uint8_t[NewGdbIndexSize];
+  uint8_t *Buffer = NewGdbIndexContents;
+
+  write32le(Buffer, Version);
+  write32le(Buffer + 4, CUListOffset);
+  write32le(Buffer + 8, CUTypesOffset);
+  write32le(Buffer + 12, AddressTableOffset);
+  write32le(Buffer + 16, SymbolTableOffset + Delta);
+  write32le(Buffer + 20, ConstantPoolOffset + Delta);
+  Buffer += 24;
+
+  using MapEntry = std::pair<uint32_t, CUInfo>;
+  std::vector<MapEntry> CUVector(CUMap.begin(), CUMap.end());
+  // Need to sort since we write out all of TUs in .debug_info before CUs.
+  std::sort(CUVector.begin(), CUVector.end(),
+            [](const MapEntry &E1, const MapEntry &E2) -> bool {
+              return E1.second.Offset < E2.second.Offset;
+            });
+  // Writing out CU List <Offset, Size>
+  for (auto &CUInfo : CUVector) {
+    // Skipping TU for DWARF5 when they are not included in CU list.
+    if (!OriginalOffsets.count(CUInfo.first))
+      continue;
+    write64le(Buffer, CUInfo.second.Offset);
+    // Length encoded in CU doesn't contain first 4 bytes that encode length.
+    write64le(Buffer + 8, CUInfo.second.Length + 4);
+    Buffer += 16;
+  }
+
+  // Rewrite TU CU List, since abbrevs can be different.
+  // Entry example:
+  // 0: offset = 0x00000000, type_offset = 0x0000001e, type_signature =
+  // 0x418503b8111e9a7b Spec says " triplet, the first value is the CU offset,
+  // the second value is the type offset in the CU, and the third value is the
+  // type signature" Looking at what is being generated by gdb-add-index. The
+  // first entry is TU offset, second entry is offset from it, and third entry
+  // is the type signature.
+  if (TUListSize)
+    for (const GDBIndexTUEntry &Entry : getGDBIndexTUEntryVector()) {
+      write64le(Buffer, Entry.UnitOffset);
+      write64le(Buffer + 8, Entry.TypeDIERelativeOffset);
+      write64le(Buffer + 16, Entry.TypeHash);
+      Buffer += sizeof(GDBIndexTUEntry);
+    }
+
+  // Generate new address table.
+  for (const std::pair<const uint64_t, DebugAddressRangesVector> &CURangesPair :
+       ARangesSectionWriter->getCUAddressRanges()) {
+    const uint32_t CUIndex = OffsetToIndexMap[CURangesPair.first];
+    const DebugAddressRangesVector &Ranges = CURangesPair.second;
+    for (const DebugAddressRange &Range : Ranges) {
+      write64le(Buffer, Range.LowPC);
+      write64le(Buffer + 8, Range.HighPC);
+      write32le(Buffer + 16, CUIndex);
+      Buffer += 20;
+    }
+  }
+
+  const size_t TrailingSize =
+      GdbIndexContents.data() + GdbIndexContents.size() - Data;
+  assert(Buffer + TrailingSize == NewGdbIndexContents + NewGdbIndexSize &&
+         "size calculation error");
+
+  // Copy over the rest of the original data.
+  memcpy(Buffer, Data, TrailingSize);
+
+  // Register the new section.
+  BC.registerOrUpdateNoteSection(".gdb_index", NewGdbIndexContents,
+                                 NewGdbIndexSize);
+}
diff --git a/bolt/lib/Passes/ValidateMemRefs.cpp b/bolt/lib/Passes/ValidateMemRefs.cpp
index f29a97c43f497c..1d2c230fa7106a 100644
--- a/bolt/lib/Passes/ValidateMemRefs.cpp
+++ b/bolt/lib/Passes/ValidateMemRefs.cpp
@@ -29,8 +29,7 @@ bool ValidateMemRefs::checkAndFixJTReference(BinaryFunction &BF, MCInst &Inst,
   if (!BD)
     return false;
 
-  const uint64_t TargetAddress = BD->getAddress() + Offset;
-  JumpTable *JT = BC.getJumpTableContainingAddress(TargetAddress);
+  JumpTable *JT = BC.getJumpTableContainingAddress(BD->getAddress());
   if (!JT)
     return false;
 
@@ -41,10 +40,10 @@ bool ValidateMemRefs::checkAndFixJTReference(BinaryFunction &BF, MCInst &Inst,
   // Accessing a jump table in another function. This is not a
   // legitimate jump table access, we need to replace the reference to
   // the jump table label with a regular rodata reference. Get a
-  // non-JT reference by fetching the symbol 1 byte before the JT
-  // label.
-  MCSymbol *NewSym = BC.getOrCreateGlobalSymbol(TargetAddress - 1, "DATAat");
-  BC.MIB->setOperandToSymbolRef(Inst, OperandNum, NewSym, 1, &*BC.Ctx, 0);
+  // non-JT reference by fetching the symbol 1 byte before the JT label.
+  MCSymbol *NewSym = BC.getOrCreateGlobalSymbol(BD->getAddress() - 1, "DATAat");
+  BC.MIB->setOperandToSymbolRef(Inst, OperandNum, NewSym, Offset + 1, &*BC.Ctx,
+                                0);
   LLVM_DEBUG(dbgs() << "BOLT-DEBUG: replaced reference @" << BF.getPrintName()
                     << " from " << BD->getName() << " to " << NewSym->getName()
                     << " + 1\n");
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index 57f37c5023110f..7469db06839e3b 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -5465,6 +5465,10 @@ def pg : Flag<["-"], "pg">, HelpText<"Enable mcount instrumentation">,
   MarshallingInfoFlag<CodeGenOpts<"InstrumentForProfiling">>;
 def pipe : Flag<["-", "--"], "pipe">,
   HelpText<"Use pipes between commands, when possible">;
+// Facebook T92898286
+def post_link_optimize : Flag<["--"], "post-link-optimize">,
+  HelpText<"Apply post-link optimizations using BOLT">;
+// End Facebook T92898286
 def prebind__all__twolevel__modules : Flag<["-"], "prebind_all_twolevel_modules">;
 def prebind : Flag<["-"], "prebind">;
 def preload : Flag<["-"], "preload">;
diff --git a/clang/lib/Driver/ToolChains/Gnu.cpp b/clang/lib/Driver/ToolChains/Gnu.cpp
index b141e5f2adfab1..f7611af5763ab7 100644
--- a/clang/lib/Driver/ToolChains/Gnu.cpp
+++ b/clang/lib/Driver/ToolChains/Gnu.cpp
@@ -672,12 +672,41 @@ void tools::gnutools::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     }
   }
 
+  // Facebook T92898286
+  if (Args.hasArg(options::OPT_post_link_optimize))
+    CmdArgs.push_back("-q");
+  // End Facebook T92898286
+
   Args.AddAllArgs(CmdArgs, options::OPT_T);
 
   const char *Exec = Args.MakeArgString(ToolChain.GetLinkerPath());
   C.addCommand(std::make_unique<Command>(JA, *this,
                                          ResponseFileSupport::AtFileCurCP(),
                                          Exec, CmdArgs, Inputs, Output));
+  // Facebook T92898286
+  if (!Args.hasArg(options::OPT_post_link_optimize) || !Output.isFilename())
+    return;
+
+  const char *MvExec = Args.MakeArgString(ToolChain.GetProgramPath("mv"));
+  ArgStringList MoveCmdArgs;
+  MoveCmdArgs.push_back(Output.getFilename());
+  const char *PreBoltBin =
+      Args.MakeArgString(Twine(Output.getFilename()) + ".pre-bolt");
+  MoveCmdArgs.push_back(PreBoltBin);
+  C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(),
+                                         MvExec, MoveCmdArgs, std::nullopt));
+
+  ArgStringList BoltCmdArgs;
+  const char *BoltExec =
+      Args.MakeArgString(ToolChain.GetProgramPath("llvm-bolt"));
+  BoltCmdArgs.push_back(PreBoltBin);
+  BoltCmdArgs.push_back("-reorder-blocks=reverse");
+  BoltCmdArgs.push_back("-update-debug-sections");
+  BoltCmdArgs.push_back("-o");
+  BoltCmdArgs.push_back(Output.getFilename());
+  C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(),
+                                         BoltExec, BoltCmdArgs, std::nullopt));
+  // End Facebook T92898286
 }
 
 void tools::gnutools::Assembler::ConstructJob(Compilation &C,
diff --git a/cross-project-tests/lit.cfg.py b/cross-project-tests/lit.cfg.py
index 774c4eaf4d976b..619634578dfe60 100644
--- a/cross-project-tests/lit.cfg.py
+++ b/cross-project-tests/lit.cfg.py
@@ -84,7 +84,13 @@ def get_required_attr(config, attr_name):
 # use_clang() and use_lld() respectively, so set them to "", if needed.
 if not hasattr(config, "clang_src_dir"):
     config.clang_src_dir = ""
-llvm_config.use_clang(required=("clang" in config.llvm_enabled_projects))
+# Facebook T92898286
+should_test_bolt = get_required_attr(config, "llvm_test_bolt")
+if should_test_bolt:
+    llvm_config.use_clang(required=("clang" in config.llvm_enabled_projects), additional_flags=["--post-link-optimize"])
+else:
+    llvm_config.use_clang(required=("clang" in config.llvm_enabled_projects))
+# End Facebook T92898286
 
 if not hasattr(config, "lld_src_dir"):
     config.lld_src_dir = ""
@@ -293,3 +299,9 @@ def get_clang_default_dwarf_version_string(triple):
 # Allow 'REQUIRES: XXX-registered-target' in tests.
 for arch in config.targets_to_build:
     config.available_features.add(arch.lower() + "-registered-target")
+
+# Facebook T92898286
+# Ensure the user's PYTHONPATH is included.
+if "PYTHONPATH" in os.environ:
+    config.environment["PYTHONPATH"] = os.environ["PYTHONPATH"]
+# End Facebook T92898286
diff --git a/cross-project-tests/lit.site.cfg.py.in b/cross-project-tests/lit.site.cfg.py.in
index 39458dfc79afd2..2d53cd377f0330 100644
--- a/cross-project-tests/lit.site.cfg.py.in
+++ b/cross-project-tests/lit.site.cfg.py.in
@@ -21,6 +21,10 @@ config.mlir_src_root = "@MLIR_SOURCE_DIR@"
 
 config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
 
+# Facebook T92898286
+config.llvm_test_bolt = lit.util.pythonize_bool("@LLVM_TEST_BOLT@")
+# End Facebook T92898286
+
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
diff --git a/lldb/test/API/lit.cfg.py b/lldb/test/API/lit.cfg.py
index d934349fe3ca3d..d4a62c51458cc1 100644
--- a/lldb/test/API/lit.cfg.py
+++ b/lldb/test/API/lit.cfg.py
@@ -248,6 +248,11 @@ def delete_module_cache(path):
 if is_configured("lldb_framework_dir"):
     dotest_cmd += ["--framework", config.lldb_framework_dir]
 
+# Facebook T92898286
+if is_configured("llvm_test_bolt"):
+    dotest_cmd += ["-E", '"--post-link-optimize"']
+# End Facebook T92898286
+
 if (
     "lldb-repro-capture" in config.available_features
     or "lldb-repro-replay" in config.available_features
diff --git a/lldb/test/API/lit.site.cfg.py.in b/lldb/test/API/lit.site.cfg.py.in
index 8b2d09ae41cd2a..602f45759e48f3 100644
--- a/lldb/test/API/lit.site.cfg.py.in
+++ b/lldb/test/API/lit.site.cfg.py.in
@@ -1,5 +1,9 @@
 @LIT_SITE_CFG_IN_HEADER@
 
+#Facebook T92898286
+import lit.util
+#End Facebook T92898286
+
 config.llvm_src_root = "@LLVM_SOURCE_DIR@"
 config.llvm_obj_root = "@LLVM_BINARY_DIR@"
 config.llvm_tools_dir = lit_config.substitute("@LLVM_TOOLS_DIR@")
@@ -39,6 +43,10 @@ config.libcxx_include_target_dir = "@LIBCXX_GENERATED_INCLUDE_TARGET_DIR@"
 config.lldb_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_LLDB@", "lldb-api")
 config.clang_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_CLANG@", "lldb-api")
 
+# Facebook T92898286
+config.llvm_test_bolt = lit.util.pythonize_bool("@LLVM_TEST_BOLT@")
+# End Facebook T92898286
+
 # Plugins
 lldb_build_intel_pt = '@LLDB_BUILD_INTEL_PT@'
 if lldb_build_intel_pt == '1':
diff --git a/lldb/test/Shell/helper/toolchain.py b/lldb/test/Shell/helper/toolchain.py
index 255955fc70d8c4..7b7be06643166d 100644
--- a/lldb/test/Shell/helper/toolchain.py
+++ b/lldb/test/Shell/helper/toolchain.py
@@ -165,6 +165,11 @@ def use_support_substitutions(config):
     if config.cmake_sysroot:
         host_flags += ["--sysroot={}".format(config.cmake_sysroot)]
 
+    # Facebook T92898286
+    if config.llvm_test_bolt:
+        host_flags += ["--post-link-optimize"]
+    # End Facebook T92898286
+
     host_flags = " ".join(host_flags)
     config.substitutions.append(("%clang_host", "%clang " + host_flags))
     config.substitutions.append(("%clangxx_host", "%clangxx " + host_flags))
diff --git a/lldb/test/Shell/lit.site.cfg.py.in b/lldb/test/Shell/lit.site.cfg.py.in
index b69e7bce1bc0be..fe8323734b7dbf 100644
--- a/lldb/test/Shell/lit.site.cfg.py.in
+++ b/lldb/test/Shell/lit.site.cfg.py.in
@@ -1,5 +1,10 @@
 @LIT_SITE_CFG_IN_HEADER@
 
+#Facebook T92898286
+import lit.util
+#End Facebook T92898286
+
+
 config.llvm_src_root = "@LLVM_SOURCE_DIR@"
 config.llvm_obj_root = "@LLVM_BINARY_DIR@"
 config.llvm_tools_dir = lit_config.substitute("@LLVM_TOOLS_DIR@")
@@ -31,6 +36,10 @@ config.llvm_use_sanitizer = "@LLVM_USE_SANITIZER@"
 config.lldb_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_LLDB@", "lldb-shell")
 config.clang_module_cache = os.path.join("@LLDB_TEST_MODULE_CACHE_CLANG@", "lldb-shell")
 
+# Facebook T92898286
+config.llvm_test_bolt = lit.util.pythonize_bool("@LLVM_TEST_BOLT@")
+# End Facebook T92898286
+
 import lit.llvm
 lit.llvm.initialize(lit_config, config)
 
diff --git a/llvm/CMakeLists.txt b/llvm/CMakeLists.txt
index 64898ab09772f4..c4ac19eaca2148 100644
--- a/llvm/CMakeLists.txt
+++ b/llvm/CMakeLists.txt
@@ -709,6 +709,10 @@ set(LLVM_LIB_FUZZING_ENGINE "" CACHE PATH
 option(LLVM_USE_SPLIT_DWARF
   "Use -gsplit-dwarf when compiling llvm and --gdb-index when linking." OFF)
 
+# Facebook T92898286
+option(LLVM_TEST_BOLT "Enable BOLT testing in non-BOLT tests that use clang" OFF)
+# End Facebook T92898286
+
 # Define an option controlling whether we should build for 32-bit on 64-bit
 # platforms, where supported.
 if( CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT (WIN32 OR ${CMAKE_SYSTEM_NAME} MATCHES "AIX"))
diff --git a/llvm/include/llv...
[truncated]

Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D58156977

ayermolo · 2024-06-04T22:30:57Z

Add [BOLT][DWARF] to title

maksfb · 2024-06-04T23:14:13Z

Add [BOLT][DWARF] to title

And NFC somewhere.

bolt/include/bolt/Core/GDBIndex.h

bolt/lib/Core/GDBIndex.cpp

Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D58193743

github-actions · 2024-06-05T16:39:57Z

✅ With the latest revision this PR passed the C/C++ code formatter.

Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags:

bolt/include/bolt/Core/GDBIndex.h

Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags:

bolt/lib/Core/GDBIndex.cpp

Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags:

bolt/lib/Core/GDBIndex.cpp

ayermolo · 2024-06-05T21:38:38Z

LGTM

sayhaan requested review from aaupov, maksfb, rafaelauler, ayermolo, dcci and JDevlieghere as code owners June 4, 2024 22:10

llvmbot added clang Clang issues not falling into any other category lldb backend:X86 clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' mc Machine (object) code BOLT labels Jun 4, 2024

sayhaan force-pushed the gdb-index-file branch from 37c9d34 to fb913e9 Compare June 4, 2024 22:11

Refactor GDBIndex into new file

8223e75

Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D58156977

sayhaan force-pushed the gdb-index-file branch from fb913e9 to 8223e75 Compare June 4, 2024 22:12

sayhaan changed the title ~~Refactor GDB Index into a new file~~ [BOLT][DWARF] Refactor GDB Index into a new file Jun 4, 2024

sayhaan changed the title ~~[BOLT][DWARF] Refactor GDB Index into a new file~~ [BOLT][DWARF][NFC] Refactor GDB Index into a new file Jun 4, 2024