[LTO][GlobalDCE] Use pass parameter instead of module flag for LTO phase

D63932 added a module flag to indicate that we are executing the regular LTO post merge pipeline, so that GlobalDCE could perform more aggressive optimization for Dead Virtual Function Elimination. This caused issues trying to reuse bitcode that had already been through the LTO pipeline (see context in D139816). Instead support this by passing down a parameter flag to the GlobalDCEPass constructor, which is the more usual way for indicating this information. Most test changes are to remove incidental uses of this flag. Of the 2 real uses, llvm/test/LTO/ARM/lto-linking-metadata.ll is now obsolete and removed in this patch, and the virtual-functions-visibility-post-lto.ll test is updated to use the regular LTO default pipeline where this parameter is set to true. Differential Revision: https://reviews.llvm.org/D153655
llvm · Jun 24, 2023 · 200cc95 · 200cc95
1 parent a2a4b60
commit 200cc95
Show file tree

Hide file tree

Showing 17 changed files with 55 additions and 56 deletions.
diff --git a/llvm/include/llvm/Transforms/IPO/GlobalDCE.h b/llvm/include/llvm/Transforms/IPO/GlobalDCE.h
@@ -35,9 +35,16 @@ class Value;
 /// Pass to remove unused function declarations.
 class GlobalDCEPass : public PassInfoMixin<GlobalDCEPass> {
 public:
+  GlobalDCEPass(bool InLTOPostLink = false) : InLTOPostLink(InLTOPostLink) {}
+
   PreservedAnalyses run(Module &M, ModuleAnalysisManager &);
 
+  void printPipeline(raw_ostream &OS,
+                     function_ref<StringRef(StringRef)> MapClassName2PassName);
+
 private:
+  bool InLTOPostLink = false;
+
   SmallPtrSet<GlobalValue*, 32> AliveGlobals;
 
   /// Global -> Global that uses this global.

diff --git a/llvm/lib/LTO/LTO.cpp b/llvm/lib/LTO/LTO.cpp
@@ -1290,8 +1290,6 @@ Error LTO::runRegularLTO(AddStreamFn AddStream) {
         GV->setLinkage(GlobalValue::InternalLinkage);
     }
 
-    RegularLTO.CombinedModule->addModuleFlag(Module::Error, "LTOPostLink", 1);
-
     if (Conf.PostInternalizeModuleHook &&
         !Conf.PostInternalizeModuleHook(0, *RegularLTO.CombinedModule))
       return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));

diff --git a/llvm/lib/LTO/LTOCodeGenerator.cpp b/llvm/lib/LTO/LTOCodeGenerator.cpp
@@ -617,9 +617,6 @@ bool LTOCodeGenerator::optimize() {
   // Mark which symbols can not be internalized
   this->applyScopeRestrictions();
 
-  // Write LTOPostLink flag for passes that require all the modules.
-  MergedModule->addModuleFlag(Module::Error, "LTOPostLink", 1);
-
   // Add an appropriate DataLayout instance for this module...
   MergedModule->setDataLayout(TargetMach->createDataLayout());
 

diff --git a/llvm/lib/Passes/PassBuilder.cpp b/llvm/lib/Passes/PassBuilder.cpp
@@ -672,6 +672,10 @@ Expected<bool> parseSinglePassOption(StringRef Params, StringRef OptionName,
   return Result;
 }
 
+Expected<bool> parseGlobalDCEPassOptions(StringRef Params) {
+  return parseSinglePassOption(Params, "vfe-linkage-unit-visibility", "GlobalDCE");
+}
+
 Expected<bool> parseInlinerPassOptions(StringRef Params) {
   return parseSinglePassOption(Params, "only-mandatory", "InlinerPass");
 }

diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -1638,7 +1638,7 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
 
   // Remove unused virtual tables to improve the quality of code generated by
   // whole-program devirtualization and bitset lowering.
-  MPM.addPass(GlobalDCEPass());
+  MPM.addPass(GlobalDCEPass(/*InLTOPostLink=*/true));
 
   // Do basic inference of function attributes from known properties of system
   // libraries and other oracles.
@@ -1757,7 +1757,7 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
   MPM.addPass(OpenMPOptPass(ThinOrFullLTOPhase::FullLTOPostLink));
 
   // Garbage collect dead functions.
-  MPM.addPass(GlobalDCEPass());
+  MPM.addPass(GlobalDCEPass(/*InLTOPostLink=*/true));
 
   // If we didn't decide to inline a function, check to see if we can
   // transform it to pass arguments by value instead of by reference.
@@ -1895,7 +1895,7 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
   MPM.addPass(EliminateAvailableExternallyPass());
 
   // Now that we have optimized the program, discard unreachable functions.
-  MPM.addPass(GlobalDCEPass());
+  MPM.addPass(GlobalDCEPass(/*InLTOPostLink=*/true));
 
   if (PTO.MergeFunctions)
     MPM.addPass(MergeFunctionsPass());

diff --git a/llvm/lib/Passes/PassRegistry.def b/llvm/lib/Passes/PassRegistry.def
@@ -60,7 +60,6 @@ MODULE_PASS("elim-avail-extern", EliminateAvailableExternallyPass())
 MODULE_PASS("extract-blocks", BlockExtractorPass({}, false))
 MODULE_PASS("forceattrs", ForceFunctionAttrsPass())
 MODULE_PASS("function-import", FunctionImportPass())
-MODULE_PASS("globaldce", GlobalDCEPass())
 MODULE_PASS("globalopt", GlobalOptPass())
 MODULE_PASS("globalsplit", GlobalSplitPass())
 MODULE_PASS("hotcoldsplit", HotColdSplittingPass())
@@ -142,6 +141,13 @@ MODULE_PASS_WITH_PARAMS("loop-extract",
                         },
                         parseLoopExtractorPassOptions,
                         "single")
+MODULE_PASS_WITH_PARAMS("globaldce",
+                        "GlobalDCEPass",
+                        [](bool InLTOPostLink) {
+                          return GlobalDCEPass(InLTOPostLink);
+                        },
+                        parseGlobalDCEPassOptions,
+                        "in-lto-post-link")
 MODULE_PASS_WITH_PARAMS("hwasan",
                         "HWAddressSanitizerPass",
                         [](HWAddressSanitizerOptions Opts) {

diff --git a/llvm/lib/Transforms/IPO/GlobalDCE.cpp b/llvm/lib/Transforms/IPO/GlobalDCE.cpp
@@ -120,11 +120,6 @@ void GlobalDCEPass::ScanVTables(Module &M) {
   SmallVector<MDNode *, 2> Types;
   LLVM_DEBUG(dbgs() << "Building type info -> vtable map\n");
 
-  auto *LTOPostLinkMD =
-      cast_or_null<ConstantAsMetadata>(M.getModuleFlag("LTOPostLink"));
-  bool LTOPostLink =
-      LTOPostLinkMD && !cast<ConstantInt>(LTOPostLinkMD->getValue())->isZero();
-
   for (GlobalVariable &GV : M.globals()) {
     Types.clear();
     GV.getMetadata(LLVMContext::MD_type, Types);
@@ -151,7 +146,7 @@ void GlobalDCEPass::ScanVTables(Module &M) {
     if (auto GO = dyn_cast<GlobalObject>(&GV)) {
       GlobalObject::VCallVisibility TypeVis = GO->getVCallVisibility();
       if (TypeVis == GlobalObject::VCallVisibilityTranslationUnit ||
-          (LTOPostLink &&
+          (InLTOPostLink &&
            TypeVis == GlobalObject::VCallVisibilityLinkageUnit)) {
         LLVM_DEBUG(dbgs() << GV.getName() << " is safe for VFE\n");
         VFESafeVTables.insert(&GV);
@@ -414,3 +409,11 @@ PreservedAnalyses GlobalDCEPass::run(Module &M, ModuleAnalysisManager &MAM) {
     return PreservedAnalyses::none();
   return PreservedAnalyses::all();
 }
+
+void GlobalDCEPass::printPipeline(
+    raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
+  static_cast<PassInfoMixin<GlobalDCEPass> *>(this)->printPipeline(
+      OS, MapClassName2PassName);
+  if (InLTOPostLink)
+    OS << "<vfe-linkage-unit-visibility>";
+}
diff --git a/llvm/test/CodeGen/AMDGPU/dwarf-multi-register-use-crash.ll b/llvm/test/CodeGen/AMDGPU/dwarf-multi-register-use-crash.ll
@@ -112,7 +112,7 @@ define weak_odr void @test(i32 %0) !dbg !34 {
 attributes #0 = { nocallback nofree nosync nounwind readnone speculatable willreturn }
 
 !llvm.dbg.cu = !{!0, !25, !26}
-!llvm.module.flags = !{!27, !28, !29, !30, !31, !32, !33}
+!llvm.module.flags = !{!27, !28, !29, !30, !31, !32}
 
 !0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 15.0.0 (https://github.com/llvm/llvm-project.git 05256c8d95e0b15bcc502d595c15d902ff520f97)", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !8, imports: !20, splitDebugInlining: false, nameTableKind: None)
 !1 = !DIFile(filename: "dummy", directory: "dummy", checksumkind: CSK_MD5, checksum: "b67bec84bdce3730b4a6f2ed8d50b85c")
@@ -147,7 +147,6 @@ attributes #0 = { nocallback nofree nosync nounwind readnone speculatable willre
 !30 = !{i32 7, !"openmp", i32 50}
 !31 = !{i32 7, !"openmp-device", i32 50}
 !32 = !{i32 7, !"PIC Level", i32 2}
-!33 = !{i32 1, !"LTOPostLink", i32 1}
 !34 = distinct !DISubprogram(name: "dummy", linkageName: "dummy", scope: !35, file: !1, line: 49, type: !23, scopeLine: 288, flags: DIFlagEnumClass, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, declaration: !36, retainedNodes: !37)
 !35 = distinct !DICompositeType(tag: DW_TAG_class_type, file: !1, line: 49, size: 32, flags: DIFlagEnumClass, elements: !6, identifier: "dummy")
 !36 = !DISubprogram(name: "dummy", scope: !35, file: !1, line: 49, type: !23, scopeLine: 288, flags: DIFlagEnumClass, spFlags: DISPFlagOptimized)

diff --git a/llvm/test/DebugInfo/MIR/X86/ldv_unreachable_blocks.mir b/llvm/test/DebugInfo/MIR/X86/ldv_unreachable_blocks.mir
@@ -16,11 +16,10 @@
   }
   declare zeroext i8 @foo_len() local_unnamed_addr
   !llvm.dbg.cu = !{!0}
-  !llvm.module.flags = !{!3, !4}
+  !llvm.module.flags = !{!3}
   !0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !6, producer: "Apple clang", isOptimized: true, flags: "-fsanitize=fuzzer-no-link,address", runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2, globals: !2, splitDebugInlining: false, nameTableKind: None, sysroot: "/", sdk: "MacOSX.sdk")
   !2 = !{}
   !3 = !{i32 2, !"Debug Info Version", i32 3}
-  !4 = !{i32 1, !"LTOPostLink", i32 1}
   !5 = distinct !DISubprogram(name: "__foo_block_invoke", linkageName: "__foo_block_invoke", scope: !6, file: !6, line: 557, type: !7, scopeLine: 557, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
   !6 = !DIFile(filename: "t.c", directory: "")
   !7 = !DISubroutineType(types: !2)

diff --git a/llvm/test/DebugInfo/MIR/X86/ldv_unreachable_blocks2.mir b/llvm/test/DebugInfo/MIR/X86/ldv_unreachable_blocks2.mir
@@ -9,12 +9,11 @@
     unreachable
   }
   !llvm.dbg.cu = !{!3}
-  !llvm.module.flags = !{!5, !6}
+  !llvm.module.flags = !{!5}
   !2 = !{}
   !3 = distinct !DICompileUnit(language: DW_LANG_C99, file: !4, producer: "clang", runtimeVersion: 0, emissionKind: FullDebug)
   !4 = !DIFile(filename: "t.c", directory: "/")
   !5 = !{i32 2, !"Debug Info Version", i32 3}
-  !6 = !{i32 1, !"LTOPostLink", i32 1}
   !7 = distinct !DISubprogram(name: "__foo_block_invoke", scope: !4, file: !4, line: 573, type: !9, scopeLine: 573, flags: DIFlagPrototyped, spFlags: DISPFlagLocalToUnit | DISPFlagDefinition | DISPFlagOptimized, unit: !3, retainedNodes: !2)
   !9 = !DISubroutineType(types: !2)
   !11 = !DILocalVariable(name: ".block_descriptor", arg: 1, scope: !7, file: !4, line: 557, type: !12, flags: DIFlagArtificial)

diff --git a/llvm/test/DebugInfo/X86/subprogram-across-cus.ll b/llvm/test/DebugInfo/X86/subprogram-across-cus.ll
@@ -61,7 +61,7 @@ entry:
 
 !llvm.dbg.cu = !{!0, !9}
 !llvm.ident = !{!10, !10}
-!llvm.module.flags = !{!11, !12, !13, !14, !15, !16}
+!llvm.module.flags = !{!11, !12, !13, !14, !15}
 
 !0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 12.0.0 (git@github.com:llvm/llvm-project bc9ab9a5cd6bafc5e1293f3d5d51638f8f5cd26c)", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, globals: !3, splitDebugInlining: false, nameTableKind: None)
 !1 = !DIFile(filename: "1.cpp", directory: "/tmp/bees")
@@ -79,7 +79,6 @@ entry:
 !13 = !{i32 1, !"wchar_size", i32 4}
 !14 = !{i32 1, !"ThinLTO", i32 0}
 !15 = !{i32 1, !"EnableSplitLTOUnit", i32 1}
-!16 = !{i32 1, !"LTOPostLink", i32 1}
 !17 = distinct !DISubprogram(name: "main", scope: !8, file: !8, line: 10, type: !18, scopeLine: 10, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !9, retainedNodes: !2)
 !18 = !DISubroutineType(types: !19)
 !19 = !{!20}

diff --git a/llvm/test/LTO/ARM/lto-linking-metadata.ll b/llvm/test/LTO/ARM/lto-linking-metadata.ll
diff --git a/llvm/test/Other/new-pm-print-pipeline.ll b/llvm/test/Other/new-pm-print-pipeline.ll
@@ -114,3 +114,9 @@
 
 ; RUN: opt -disable-output -disable-verify -print-pipeline-passes -passes='function(loop(loop-rotate<no-header-duplication;no-prepare-for-lto>))' < %s | FileCheck %s --match-full-lines --check-prefixes=CHECK-33
 ; CHECK-33: function(loop(loop-rotate<no-header-duplication;no-prepare-for-lto>))
+
+; RUN: opt -disable-output -disable-verify -print-pipeline-passes -passes='globaldce' < %s | FileCheck %s --match-full-lines --check-prefixes=CHECK-34
+; CHECK-34: globaldce
+
+; RUN: opt -disable-output -disable-verify -print-pipeline-passes -passes='globaldce<vfe-linkage-unit-visibility>' < %s | FileCheck %s --match-full-lines --check-prefixes=CHECK-35
+; CHECK-35: globaldce<vfe-linkage-unit-visibility>
diff --git a/llvm/test/Transforms/GlobalDCE/virtual-functions-visibility-post-lto.ll b/llvm/test/Transforms/GlobalDCE/virtual-functions-visibility-post-lto.ll
@@ -1,8 +1,14 @@
-; RUN: opt < %s -passes=globaldce -S | FileCheck %s
+; RUN: opt < %s -passes='globaldce<vfe-linkage-unit-visibility>' -S | FileCheck %s
+; RUN: opt < %s -passes='lto<O2>' -S | FileCheck %s
 
 ; structs A, B and C have vcall_visibility of public, linkage-unit and
 ; translation-unit respectively. This test is run after LTO linking (the
-; LTOPostLink metadata is present), so B and C can be VFE'd.
+; pass parameter simulates how GlobalDCE is invoked from the regular LTO
+; pipeline), so B and C can be VFE'd.
+
+;; Try again without being in the LTO post link, we can only eliminate C.
+; RUN: opt < %s -passes='globaldce' -S | FileCheck %s --check-prefix=NO-LTO
+; RUN: opt < %s -passes='default<O2>' -S | FileCheck %s --check-prefix=NO-LTO
 
 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
 
@@ -17,6 +23,7 @@ entry:
 }
 
 ; CHECK: define {{.*}} @_ZN1A3fooEv(
+; NO-LTO: define {{.*}} @_ZN1A3fooEv(
 define internal void @_ZN1A3fooEv(ptr nocapture %this) {
 entry:
   ret void
@@ -41,6 +48,7 @@ entry:
 }
 
 ; CHECK-NOT: define {{.*}} @_ZN1B3fooEv(
+; NO-LTO: define {{.*}} @_ZN1B3fooEv(
 define internal void @_ZN1B3fooEv(ptr nocapture %this) {
 entry:
   ret void
@@ -65,6 +73,7 @@ entry:
 }
 
 ; CHECK-NOT: define {{.*}} @_ZN1C3fooEv(
+; NO-LTO-NOT: define {{.*}} @_ZN1C3fooEv(
 define internal void @_ZN1C3fooEv(ptr nocapture %this) {
 entry:
   ret void
@@ -79,12 +88,11 @@ entry:
 
 declare dso_local noalias nonnull ptr @_Znwm(i64)
 
-!llvm.module.flags = !{!5, !6}
+!llvm.module.flags = !{!6}
 
 !0 = !{i64 16, !"_ZTS1A"}
 !1 = !{i64 16, !"_ZTSM1AFvvE.virtual"}
 !2 = !{i64 0} ; public vcall visibility
 !3 = !{i64 1} ; linkage-unit vcall visibility
 !4 = !{i64 2} ; translation-unit vcall visibility
-!5 = !{i32 1, !"LTOPostLink", i32 1}
 !6 = !{i32 1, !"Virtual Function Elim", i32 1}
diff --git a/llvm/test/Transforms/GlobalDCE/vtable-rtti.ll b/llvm/test/Transforms/GlobalDCE/vtable-rtti.ll
@@ -37,10 +37,9 @@ entry:
 declare dso_local noalias nonnull ptr @_Znwm(i64)
 @_ZTVN10__cxxabiv117__class_type_infoE = external dso_local global ptr
 
-!llvm.module.flags = !{!3, !4}
+!llvm.module.flags = !{!4}
 
 !0 = !{i64 16, !"_ZTS1A"}
 !1 = !{i64 16, !"_ZTSM1AFvvE.virtual"}
 !2 = !{i64 2} ; translation-unit vcall visibility
-!3 = !{i32 1, !"LTOPostLink", i32 1}
 !4 = !{i32 1, !"Virtual Function Elim", i32 1}
diff --git a/llvm/test/Transforms/WholeProgramDevirt/devirt-single-impl2.ll b/llvm/test/Transforms/WholeProgramDevirt/devirt-single-impl2.ll
@@ -34,12 +34,11 @@ entry:
 }
 
 !llvm.ident = !{!2}
-!llvm.module.flags = !{!3, !4, !5, !6}
+!llvm.module.flags = !{!3, !4, !5}
 
 !0 = !{i64 16, !"_ZTS1A"}
 !1 = !{i64 16, !"_ZTSM1AKFivE.virtual"}
 !2 = !{!"clang version 10.0.0 (trunk 373596)"}
 !3 = !{i32 1, !"wchar_size", i32 4}
 !4 = !{i32 1, !"EnableSplitLTOUnit", i32 1}
 !5 = !{i32 1, !"ThinLTO", i32 0}
-!6 = !{i32 1, !"LTOPostLink", i32 1}
diff --git a/llvm/test/tools/llvm-reduce/reduce-module-flags.ll b/llvm/test/tools/llvm-reduce/reduce-module-flags.ll
@@ -30,11 +30,10 @@
 
 ; RESULT2: !llvm.module.flags = !{}
 
-!llvm.module.flags = !{!0, !1, !2, !3, !4, !5}
+!llvm.module.flags = !{!0, !1, !2, !3, !4}
 
 !0 = !{i32 1, !"amdgpu_code_object_version", i32 400}
 !1 = !{i32 1, !"wchar_size", i32 4}
 !2 = !{i32 7, !"openmp", i32 50}
 !3 = !{i32 7, !"openmp-device", i32 50}
 !4 = !{i32 8, !"PIC Level", i32 1}
-!5 = !{i32 1, !"LTOPostLink", i32 1}