Introduce -funique-source-file-names flag.

The purpose of this flag is to allow the compiler to assume that each
object file passed to the linker has been compiled using a unique
source file name. This is useful for reducing link times when doing
ThinLTO in combination with whole-program devirtualization or CFI,
as it allows modules without exported symbols to be built with ThinLTO.

Reviewers: vitalybuka, teresajohnson

Reviewed By: teresajohnson

Pull Request: https://github.com/llvm/llvm-project/pull/135728

clang/docs/ControlFlowIntegrity.rst

@@ -42,6 +42,11 @@ default visibility setting is ``-fvisibility=default``, which would disable
 CFI checks for classes without visibility attributes. Most users will want
 to specify ``-fvisibility=hidden``, which enables CFI checks for such classes.
 
+When using ``-fsanitize=cfi*`` with ``-flto=thin``, it is recommended
+to reduce link times by passing `-funique-source-file-names
+<UsersManual.html#cmdoption-f-no-unique-source-file-names>`_, provided
+that your program is compatible with it.
+
 Experimental support for :ref:`cross-DSO control flow integrity
 <cfi-cross-dso>` exists that does not require classes to have hidden LTO
 visibility. This cross-DSO support has unstable ABI at this time.

clang/docs/UsersManual.rst

@@ -2297,6 +2297,16 @@ are listed below.
    pure ThinLTO, as all split regular LTO modules are merged and LTO linked
    with regular LTO.
 
+.. option:: -f[no-]unique-source-file-names
+
+   When enabled, allows the compiler to assume that each object file
+   passed to the linker has been compiled using a unique source file
+   name. This is useful for reducing link times when doing ThinLTO
+   in combination with whole-program devirtualization or CFI.
+
+   A misuse of this flag will generally result in a duplicate symbol
+   error at link time.
+
 .. option:: -fforce-emit-vtables
 
    In order to improve devirtualization, forces emitting of vtables even in

clang/include/clang/Basic/CodeGenOptions.def

@@ -278,6 +278,8 @@ CODEGENOPT(SanitizeCfiICallNormalizeIntegers, 1, 0) ///< Normalize integer types
                                                     ///< CFI icall function signatures
 CODEGENOPT(SanitizeCfiCanonicalJumpTables, 1, 0) ///< Make jump table symbols canonical
                                                  ///< instead of creating a local jump table.
+CODEGENOPT(UniqueSourceFileNames, 1, 0) ///< Allow the compiler to assume that TUs
+                                        ///< have unique source file names at link time
 CODEGENOPT(SanitizeKcfiArity, 1, 0) ///< Embed arity in KCFI patchable function prefix
 CODEGENOPT(SanitizeCoverageType, 2, 0) ///< Type of sanitizer coverage
                                        ///< instrumentation.

clang/include/clang/Driver/Options.td

@@ -4140,6 +4140,13 @@ def ftrigraphs : Flag<["-"], "ftrigraphs">, Group<f_Group>,
 def fno_trigraphs : Flag<["-"], "fno-trigraphs">, Group<f_Group>,
   HelpText<"Do not process trigraph sequences">,
   Visibility<[ClangOption, CC1Option]>;
+defm unique_source_file_names: BoolOption<"f", "unique-source-file-names",
+  CodeGenOpts<"UniqueSourceFileNames">, DefaultFalse,
+  PosFlag<SetTrue, [], [CC1Option], "Allow">,
+  NegFlag<SetFalse, [], [], "Do not allow">,
+  BothFlags<[], [ClangOption], " the compiler to assume that each translation unit has a unique "
+                               "source file name at link time">>,
+  Group<f_clang_Group>;
 def funsigned_bitfields : Flag<["-"], "funsigned-bitfields">, Group<f_Group>;
 def funsigned_char : Flag<["-"], "funsigned-char">, Group<f_Group>;
 def fno_unsigned_char : Flag<["-"], "fno-unsigned-char">;

clang/lib/CodeGen/CodeGenModule.cpp

@@ -1144,6 +1144,10 @@ void CodeGenModule::Release() {
                               1);
   }
 
+  if (CodeGenOpts.UniqueSourceFileNames) {
+    getModule().addModuleFlag(llvm::Module::Max, "Unique Source File Names", 1);
+  }
+
   if (LangOpts.Sanitize.has(SanitizerKind::KCFI)) {
     getModule().addModuleFlag(llvm::Module::Override, "kcfi", 1);
     // KCFI assumes patchable-function-prefix is the same for all indirectly

clang/lib/Driver/ToolChains/Clang.cpp

@@ -7744,6 +7744,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   Args.addOptInFlag(CmdArgs, options::OPT_fexperimental_late_parse_attributes,
                     options::OPT_fno_experimental_late_parse_attributes);
 
+  Args.addOptInFlag(CmdArgs, options::OPT_funique_source_file_names,
+                    options::OPT_fno_unique_source_file_names);
+
   // Setup statistics file output.
   SmallString<128> StatsFile = getStatsFileName(Args, Output, Input, D);
   if (!StatsFile.empty()) {

clang/test/CodeGen/unique-source-file-names.c

@@ -0,0 +1,2 @@
+// RUN: %clang_cc1 -funique-source-file-names -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck %s
+// CHECK:  !{i32 7, !"Unique Source File Names", i32 1}

clang/test/Driver/unique-source-file-names.c

@@ -0,0 +1,5 @@
+// RUN: %clang -funique-source-file-names -### %s 2> %t
+// RUN: FileCheck < %t %s
+
+// CHECK: "-cc1"
+// CHECK: "-funique-source-file-names"

llvm/lib/Transforms/Utils/ModuleUtils.cpp

@@ -345,27 +345,25 @@ void llvm::filterDeadComdatFunctions(
 
 std::string llvm::getUniqueModuleId(Module *M) {
   MD5 Md5;
-  bool ExportsSymbols = false;
-  auto AddGlobal = [&](GlobalValue &GV) {
-    if (GV.isDeclaration() || GV.getName().starts_with("llvm.") ||
-        !GV.hasExternalLinkage() || GV.hasComdat())
-      return;
-    ExportsSymbols = true;
-    Md5.update(GV.getName());
-    Md5.update(ArrayRef<uint8_t>{0});
-  };
 
-  for (auto &F : *M)
-    AddGlobal(F);
-  for (auto &GV : M->globals())
-    AddGlobal(GV);
-  for (auto &GA : M->aliases())
-    AddGlobal(GA);
-  for (auto &IF : M->ifuncs())
-    AddGlobal(IF);
+  auto *UniqueSourceFileNames = mdconst::extract_or_null<ConstantInt>(
+      M->getModuleFlag("Unique Source File Names"));
+  if (UniqueSourceFileNames && UniqueSourceFileNames->getZExtValue()) {
+    Md5.update(M->getSourceFileName());
+  } else {
+    bool ExportsSymbols = false;
+    for (auto &GV : M->global_values()) {
+      if (GV.isDeclaration() || GV.getName().starts_with("llvm.") ||
+          !GV.hasExternalLinkage() || GV.hasComdat())
+        continue;
+      ExportsSymbols = true;
+      Md5.update(GV.getName());
+      Md5.update(ArrayRef<uint8_t>{0});
+    }
 
-  if (!ExportsSymbols)
-    return "";
+    if (!ExportsSymbols)
+      return "";
+  }
 
   MD5::MD5Result R;
   Md5.final(R);

llvm/test/Transforms/ThinLTOBitcodeWriter/unique-source-file-names.ll

@@ -0,0 +1,22 @@
+; RUN: opt -thinlto-bc -thin-link-bitcode-file=%t2 -thinlto-split-lto-unit -o %t %s
+; RUN: llvm-modextract -b -n 1 -o %t1 %t
+; RUN: llvm-dis -o - %t1 | FileCheck %s
+
+source_filename = "unique-source-file-names.c"
+
+@llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @f, ptr null }]
+
+; CHECK: @g.45934e8a5251fb7adbecfff71a4e70ed =
+@g = internal global i8 42, !type !0
+
+declare void @sink(ptr)
+
+define internal void @f() {
+  call void @sink(ptr @g)
+  ret void
+}
+
+!0 = !{i32 0, !"typeid"}
+
+!llvm.module.flags = !{!1}
+!1 = !{i32 1, !"Unique Source File Names", i32 1}