LTO: Apply global DCE to ThinLTO modules at LTO opt level 0.

[pcc]

Summary

This is necessary because DCE is applied to full LTO modules. Without
this change, a reference from a dead ThinLTO global to a dead full
LTO global will result in an undefined reference at link time.

This problem is only observable when --gc-sections is disabled, or
when targeting COFF, as the COFF port of lld requires all symbols to
have a definition even if all references are dead (this is consistent
with link.exe).

This change also adds an EliminateAvailableExternally pass at -O0. This
is necessary to handle the situation on Windows where a non-prevailing
copy of a linkonce_odr function has an SEH filter function; any
such filters must be DCE'd because they will contain a call to the
llvm.localrecover intrinsic, passing as an argument the address of the
function that the filter belongs to, and llvm.localrecover requires
this function to be defined locally.

Fixes PR35142.

[tejohnson]

llvm/lib/LTO/LTO.cpp

1101

I think it is worth a comment about why we are doing the other summary based optimizations such as internalization at -O0, so someone doesn't go back later and make those conditional.

llvm/lib/Transforms/IPO/PassManagerBuilder.cpp

425

Just to ensure I understand correctly - needing to remove the non-prevailing linkonce_odr is in turn due to the change to do internalization/DCE at -O0?

llvm/test/LTO/Resolution/X86/dead-strip-fulllto.ll

21

add a check that we don't have @odr either?

[pcc]

• Address review comments

llvm/lib/Transforms/IPO/PassManagerBuilder.cpp

425

Yes, otherwise we can end up with a situation like in Inputs/dead-strip-fulllto.ll where odr is keeping dead3 alive. If dead3 contained references to other dead symbols, that would cause an undefined symbol error.

[tejohnson]

LGTM

[rnk]

+1, this should speed up ThinLTO -O0 in addition to being a workaround for the SEH filter function issue. Less code to generate means faster links.

llvm/lib/LTO/LTO.cpp

if (!ModuleToDefinedGVSummaries.count(Mod.first))
       ModuleToDefinedGVSummaries.try_emplace(Mod.first);
 
   StringMap<FunctionImporter::ImportMapTy> ImportLists(
       ThinLTO.ModuleMap.size());
   StringMap<FunctionImporter::ExportSetTy> ExportLists(
       ThinLTO.ModuleMap.size());
   StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR;
 
-  if (Conf.OptLevel > 0) {
+  if (Conf.OptLevel > 0)
     ComputeCrossModuleImport(ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries,
                              ImportLists, ExportLists);
 
-    std::set<GlobalValue::GUID> ExportedGUIDs;
-    for (auto &Res : GlobalResolutions) {
-      // First check if the symbol was flagged as having external references.
-      if (Res.second.Partition != GlobalResolution::External)
-        continue;
-      // IRName will be defined if we have seen the prevailing copy of
-      // this value. If not, no need to mark as exported from a ThinLTO
-      // partition (and we can't get the GUID).
-      if (Res.second.IRName.empty())
-        continue;
-      auto GUID = GlobalValue::getGUID(
-          GlobalValue::dropLLVMManglingEscape(Res.second.IRName));
-      // Mark exported unless index-based analysis determined it to be dead.
-      if (ThinLTO.CombinedIndex.isGUIDLive(GUID))
-        ExportedGUIDs.insert(GUID);
-    }
+  // Figure out which symbols need to be internalized. This also needs to happen
+  // at -O0 because summary-based DCE is implemented using internalization, and
+  // we must apply DCE consistently with the full LTO module in order to avoid
+  // undefined references during the final link.
 
-    // Any functions referenced by the jump table in the regular LTO object must
-    // be exported.
-    for (auto &Def : ThinLTO.CombinedIndex.cfiFunctionDefs())
-      ExportedGUIDs.insert(
-          GlobalValue::getGUID(GlobalValue::dropLLVMManglingEscape(Def)));
 
-    auto isExported = [&](StringRef ModuleIdentifier, GlobalValue::GUID GUID) {
-      const auto &ExportList = ExportLists.find(ModuleIdentifier);
-      return (ExportList != ExportLists.end() &&
-              ExportList->second.count(GUID)) ||
-             ExportedGUIDs.count(GUID);
-    };
-    thinLTOInternalizeAndPromoteInIndex(ThinLTO.CombinedIndex, isExported);
-  }
 
   auto isPrevailing = [&](GlobalValue::GUID GUID,
                           const GlobalValueSummary *S) {
     return ThinLTO.PrevailingModuleForGUID[GUID] == S->modulePath();
   };
   auto recordNewLinkage = [&](StringRef ModuleIdentifier,
                               GlobalValue::GUID GUID,
                               GlobalValue::LinkageTypes NewLinkage) {

llvm/lib/Transforms/IPO/PassManagerBuilder.cpp

if (OptLevel == 0) {
     // that pass manager. To prevent this we insert a no-op module pass to reset
     // the pass manager to get the same behavior as EP_OptimizerLast in non-O0
     // builds. The function merging pass is
     if (MergeFunctions)
       MPM.add(createMergeFunctionsPass());
     else if (GlobalExtensionsNotEmpty() || !Extensions.empty())
       MPM.add(createBarrierNoopPass());
 
+    if (PerformThinLTO) {
+      // Drop available_externally and unreferenced globals. This is necessary
+      // with ThinLTO in order to avoid leaving undefined references to dead
+      // globals in the object file.
+      MPM.add(createEliminateAvailableExternallyPass());
+      MPM.add(createGlobalDCEPass());
+    }
+
     addExtensionsToPM(EP_EnabledOnOptLevel0, MPM);
 
     // Rename anon globals to be able to export them in the summary.
     // This has to be done after we add the extensions to the pass manager
     // as there could be passes (e.g. Adddress sanitizer) which introduce
     // new unnamed globals.
     if (PrepareForThinLTO)
       MPM.add(createNameAnonGlobalPass());

llvm/test/LTO/Resolution/X86/Inputs/dead-strip-fulllto.ll

 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 define void @live1() {
   call void @live2()
   ret void
 }
 
 declare void @live2()
 
 define void @dead1() {
   call void @dead2()
   ret void
 }
 
 declare void @dead2()
+
+define linkonce_odr i8* @odr() {
+  ret i8* bitcast (void ()* @dead3 to i8*)
+}
+
+define internal void @dead3() {
+  ret void
+}

llvm/test/LTO/Resolution/X86/dead-strip-fulllto.ll

 ; RUN: opt -module-summary -o %t %s
 ; RUN: opt -module-summary -o %t2 %S/Inputs/dead-strip-fulllto.ll
+
 ; RUN: llvm-lto2 run %t -r %t,main,px -r %t,live1,p -r %t,live2,p -r %t,dead2,p \
-; RUN:               %t2 -r %t2,live1,p -r %t2,live2, -r %t2,dead1,p -r %t2,dead2, \
+; RUN:               %t2 -r %t2,live1,p -r %t2,live2, -r %t2,dead1,p -r %t2,dead2, -r %t2,odr, \
 ; RUN: -save-temps -o %t3
 ; RUN: llvm-nm %t3.0 | FileCheck --check-prefix=FULL %s
 ; RUN: llvm-nm %t3.1 | FileCheck --check-prefix=THIN %s
 
+; RUN: llvm-lto2 run %t -r %t,main,px -r %t,live1,p -r %t,live2,p -r %t,dead2,p \
+; RUN:               %t2 -r %t2,live1,p -r %t2,live2, -r %t2,dead1,p -r %t2,dead2, -r %t2,odr, \
+; RUN: -save-temps -o %t3 -O0
+; RUN: llvm-nm %t3.0 | FileCheck --check-prefix=FULL %s
+; RUN: llvm-nm %t3.1 | FileCheck --check-prefix=THIN %s
+
 ; FULL-NOT: dead
 ; FULL: U live1
 ; FULL: T live2
 ; FULL: T main
 
 ; THIN-NOT: dead
 ; THIN: T live1
 ; THIN: U live2
+; THIN-NOT: odr
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 define void @main() {
   call void @live1()
   ret void
 }