[LTO] Ignore unreachable virtual functions in WPD in thin LTO

[mingmingl]

Summary

For thin LTO, ignore unreachable functions (by reading from index) when finding virtual functions.

• This helps to de-virtualize. Unit test are added in test case in llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll
• Rename devirt_hybrid_after_filtering_unreachable.ll to devirt_after_filtering_unreachable.ll, and Inputs/devirt_hybrid... to Inputs/devirt...

[mingmingl]

Add unit test for thinlto.

[mingmingl]

Rebase to main

[tejohnson]

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

64

This stuff is checked in other tests. Unless you are checking something specific to this test, you can remove all of this. Actually, it would be good to test here and in the earlier hybrid case that the summary for _ZN4BaseD0Ev says it is unreachable. But that's probably all you need to test.

Ditto for %t4.o

127

Unless you need to share only some of the check results with another invocation, just use a single check prefix.

[mingmingl]

Change test IR by removing irrelevant lines and using one prefix per FileCheck.

[mingmingl]

thanks for review!

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

64

Done by

1. checking the bit mustBeUnreachable for thin LTO and hybrid LTO, and remove irrelevant lines
2. Use one prefix per FileCheck.

I spent some time trying to figure out why the regular LTO split modules (files with ".1" suffix in mod-extract output) don't have function summaries.

Didn't figure out from code perspective (starting from Index->enableSplitLTOUnit() in https://github.com/llvm/llvm-project/blob/main/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp#L3926), but it's probably intended in the sense that regular LTO module won't need index (which is generated for thin).

[tejohnson]

lgtm with one more test simplification below.

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

11

Since we don't ever use variable BaseD0, there isn't a need to capture a value there. In fact, you can ignore it completely and then change all 4 llvm-dis instances in this test to check the same pattern like:

; FLAG: gv: (name: "_ZN4BaseD0Ev", {{.*}}, funcFlags: ({{.*}} mustBeUnreachable: 1

(and no need for the -DAG handling since there is only one pattern being checked per llvm-dis invocation).

[mingmingl]

Simplify test by using non-DAG prefix (for one pattern), and avoid capturing values since they are not used.

[tejohnson]

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

11

You can combine all of these (ENABLESPLITFLAG, LIBENABLESPLITFLAG, etc) now into a single filecheck pattern.

[mingmingl]

thanks for review!

Also in the previous reply, I forgot to mention I removed the the line assert ((VTP.FuncVI) && "VTP.FuncVI must be nullptr after finding existing references doesn't guarantee that [1], and mustBeUnreachableFunction will return false if an empty VI is passed in anyway.

I'll run a ninja check (may take a while) and then push this.

[1] For field VirtFuncOffset::FuncVI, there are if and assertions about empty value (e.g., https://github.com/llvm/llvm-project/blob/main/llvm/lib/AsmParser/LLParser.cpp#L7834)

[mingmingl]

Simplify test by sharing one pattern across two filecheck lines.

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

11

(Saw this after previous reply) Done by sharing the pattern between two filecheck lines.

[tejohnson]

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

16

I think it can be shared with the equivalent checks in the unsplit case as well. And probably with a check-prefix name that reflects what is being checked (e.g. UNREACHABLEFLAG), or maybe something simpler (e.g. FLAG).

[mingmingl]

Simplify filecheck by re-using one prefix to test the unreachable bit.

[mingmingl]

thanks for the inputs!

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

16

Use UNREACHABLEFLAG and share it across split and unsplit case.

[mingmingl]

Simplify test case by using -O3 to generate IR.

[mingmingl]

In D115648#3196142, @luna wrote:

Simplify test case by using -O3 to generate IR.

Probably because of the diff from ThinLTO/X86/devirt_hybrid_after_filtering_unreachable.ll to ThinLTO/X86/devirt_after_filtering_unreachable.ll is big, git regard a git mv as git add and git mv here (this stack overflow) question echoes.

So I'm going to keep it (without trying to make it look like a mv), wait for the pre-merge test and the local ninja check-llvm-lto, and then land this.

llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp

for (const TypeIdOffsetVtableInfo &P : TIdInfo) {
     if (!VS)
       return false;
     if (!VS->isLive())
       continue;
     for (auto VTP : VS->vTableFuncs()) {
       if (VTP.VTableOffset != P.AddressPointOffset + ByteOffset)
         continue;
 
+      if (mustBeUnreachableFunction(VTP.FuncVI))
+        continue;
+
       TargetsForSlot.push_back(VTP.FuncVI);
     }
   }
 
   // Give up if we couldn't find any targets.
   return !TargetsForSlot.empty();
 }
 

llvm/test/ThinLTO/X86/Inputs/devirt_after_filtering_unreachable_lib.ll

 $_ZN4BaseD0Ev = comdat any
 
 $_ZTS4Base = comdat any
 
 $_ZTI4Base = comdat any
 
 $_ZTV4Base = comdat any
 
-@.str = private unnamed_addr constant [12 x i8] c"Derived::x\0A\00", align 1
 @_ZTV7Derived = hidden unnamed_addr constant { [5 x i8*] } { [5 x i8*] [i8* null, i8* bitcast ({ i8*, i8*, i8* }* @_ZTI7Derived to i8*), i8* bitcast (void (%class.Derived*)* @_ZN7DerivedD2Ev to i8*), i8* bitcast (void (%class.Derived*)* @_ZN7DerivedD0Ev to i8*), i8* bitcast (void (%class.Derived*)* @_ZN7Derived1xEv to i8*)] }, align 8, !type !0, !type !1, !type !2, !type !3, !vcall_visibility !4
 @_ZTVN10__cxxabiv120__si_class_type_infoE = external dso_local global i8*
 @_ZTS7Derived = hidden constant [9 x i8] c"7Derived\00", align 1
 @_ZTVN10__cxxabiv117__class_type_infoE = external dso_local global i8*
 @_ZTS4Base = linkonce_odr hidden constant [6 x i8] c"4Base\00", comdat, align 1
 @_ZTI4Base = linkonce_odr hidden constant { i8*, i8* } { i8* bitcast (i8** getelementptr inbounds (i8*, i8** @_ZTVN10__cxxabiv117__class_type_infoE, i64 2) to i8*), i8* getelementptr inbounds ([6 x i8], [6 x i8]* @_ZTS4Base, i32 0, i32 0) }, comdat, align 8
 @_ZTI7Derived = hidden constant { i8*, i8*, i8* } { i8* bitcast (i8** getelementptr inbounds (i8*, i8** @_ZTVN10__cxxabiv120__si_class_type_infoE, i64 2) to i8*), i8* getelementptr inbounds ([9 x i8], [9 x i8]* @_ZTS7Derived, i32 0, i32 0), i8* bitcast ({ i8*, i8* }* @_ZTI4Base to i8*) }, align 8
-@.str.1 = private unnamed_addr constant [24 x i8] c"In Derived::~Derived()\0A\00", align 1
 @_ZTV4Base = linkonce_odr hidden unnamed_addr constant { [5 x i8*] } { [5 x i8*] [i8* null, i8* bitcast ({ i8*, i8* }* @_ZTI4Base to i8*), i8* bitcast (void (%class.Base*)* @_ZN4BaseD2Ev to i8*), i8* bitcast (void (%class.Base*)* @_ZN4BaseD0Ev to i8*), i8* bitcast (void ()* @__cxa_pure_virtual to i8*)] }, comdat, align 8, !type !0, !type !1, !vcall_visibility !4
-@.str.2 = private unnamed_addr constant [18 x i8] c"In Base::~Base()\0A\00", align 1
+@str = private unnamed_addr constant [11 x i8] c"Derived::x\00", align 1
+@str.3 = private unnamed_addr constant [23 x i8] c"In Derived::~Derived()\00", align 1
+@str.4 = private unnamed_addr constant [17 x i8] c"In Base::~Base()\00", align 1
 
-define hidden void @_ZN7Derived1xEv(%class.Derived* nonnull align 8 dereferenceable(8) %this) unnamed_addr align 2 {
+; Function Attrs: mustprogress nofree nounwind uwtable
+define hidden void @_ZN7Derived1xEv(%class.Derived* nocapture nonnull readnone align 8 %this) unnamed_addr align 2 {
 entry:
-  %call = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([12 x i8], [12 x i8]* @.str, i64 0, i64 0))
+  %puts = tail call i32 @puts(i8* nonnull dereferenceable(1) getelementptr inbounds ([11 x i8], [11 x i8]* @str, i64 0, i64 0))
   ret void
 }
 
-declare dso_local i32 @printf(i8*, ...)
-
-define hidden void @_Z3fooP4Base(%class.Base* %b) {
+; Function Attrs: mustprogress nounwind uwtable
+define hidden void @_Z3fooP4Base(%class.Base* %b) local_unnamed_addr {
 entry:
-  %b.addr = alloca %class.Base*, align 8
-  store %class.Base* %b, %class.Base** %b.addr, align 8
-  %0 = load %class.Base*, %class.Base** %b.addr, align 8
-  %isnull = icmp eq %class.Base* %0, null
+  %isnull = icmp eq %class.Base* %b, null
   br i1 %isnull, label %delete.end, label %delete.notnull
 
 delete.notnull:                                   ; preds = %entry
-  %1 = bitcast %class.Base* %0 to void (%class.Base*)***
-  %vtable = load void (%class.Base*)**, void (%class.Base*)*** %1, align 8
-  %2 = bitcast void (%class.Base*)** %vtable to i8*
-  %3 = call i1 @llvm.type.test(i8* %2, metadata !"_ZTS4Base")
-  call void @llvm.assume(i1 %3)
+  %0 = bitcast %class.Base* %b to void (%class.Base*)***
+  %vtable = load void (%class.Base*)**, void (%class.Base*)*** %0, align 8, !tbaa !9
+  %1 = bitcast void (%class.Base*)** %vtable to i8*
+  %2 = tail call i1 @llvm.type.test(i8* %1, metadata !"_ZTS4Base")
+  tail call void @llvm.assume(i1 %2)
   %vfn = getelementptr inbounds void (%class.Base*)*, void (%class.Base*)** %vtable, i64 1
-  %4 = load void (%class.Base*)*, void (%class.Base*)** %vfn, align 8
-  call void %4(%class.Base* nonnull align 8 dereferenceable(8) %0)
+  %3 = load void (%class.Base*)*, void (%class.Base*)** %vfn, align 8
+  tail call void %3(%class.Base* nonnull align 8 dereferenceable(8) %b)
   br label %delete.end
 
 delete.end:                                       ; preds = %delete.notnull, %entry
   ret void
 }
 
 declare i1 @llvm.type.test(i8*, metadata)
 
 declare void @llvm.assume(i1 noundef)
 
 define linkonce_odr hidden void @_ZN7DerivedD2Ev(%class.Derived* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
 entry:
-  %0 = bitcast %class.Derived* %this to i32 (...)***
-  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV7Derived, i32 0, inrange i32 0, i32 2) to i32 (...)**), i32 (...)*** %0, align 8
-  %call = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([24 x i8], [24 x i8]* @.str.1, i64 0, i64 0))
-  %1 = bitcast %class.Derived* %this to %class.Base*
-  call void @_ZN4BaseD2Ev(%class.Base* nonnull align 8 dereferenceable(8) %1)
+  %0 = getelementptr inbounds %class.Derived, %class.Derived* %this, i64 0, i32 0, i32 0
+  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV7Derived, i64 0, inrange i32 0, i64 2) to i32 (...)**), i32 (...)*** %0, align 8, !tbaa !9
+  %puts = tail call i32 @puts(i8* nonnull dereferenceable(1) getelementptr inbounds ([23 x i8], [23 x i8]* @str.3, i64 0, i64 0))
+  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV4Base, i64 0, inrange i32 0, i64 2) to i32 (...)**), i32 (...)*** %0, align 8, !tbaa !9
+  %puts.i = tail call i32 @puts(i8* nonnull dereferenceable(1) getelementptr inbounds ([17 x i8], [17 x i8]* @str.4, i64 0, i64 0))
   ret void
 }
 
+; Function Attrs: nounwind uwtable
 define linkonce_odr hidden void @_ZN7DerivedD0Ev(%class.Derived* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
 entry:
-  call void @_ZN7DerivedD2Ev(%class.Derived* nonnull align 8 dereferenceable(8) %this)
-  %0 = bitcast %class.Derived* %this to i8*
-  call void @_ZdlPv(i8* %0)
+  %0 = getelementptr inbounds %class.Derived, %class.Derived* %this, i64 0, i32 0, i32 0
+  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV7Derived, i64 0, inrange i32 0, i64 2) to i32 (...)**), i32 (...)*** %0, align 8, !tbaa !9
+  %puts.i = tail call i32 @puts(i8* nonnull dereferenceable(1) getelementptr inbounds ([23 x i8], [23 x i8]* @str.3, i64 0, i64 0))
+  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV4Base, i64 0, inrange i32 0, i64 2) to i32 (...)**), i32 (...)*** %0, align 8, !tbaa !9
+  %puts.i.i = tail call i32 @puts(i8* nonnull dereferenceable(1) getelementptr inbounds ([17 x i8], [17 x i8]* @str.4, i64 0, i64 0))
+  %1 = bitcast %class.Derived* %this to i8*
+  tail call void @_ZdlPv(i8* nonnull %1)
   ret void
 }
 
+; Function Attrs: nounwind uwtable
 define linkonce_odr hidden void @_ZN4BaseD2Ev(%class.Base* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
 entry:
-  %0 = bitcast %class.Base* %this to i32 (...)***
-  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV4Base, i32 0, inrange i32 0, i32 2) to i32 (...)**), i32 (...)*** %0, align 8
-  %call = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([18 x i8], [18 x i8]* @.str.2, i64 0, i64 0))
+  %0 = getelementptr inbounds %class.Base, %class.Base* %this, i64 0, i32 0
+  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV4Base, i64 0, inrange i32 0, i64 2) to i32 (...)**), i32 (...)*** %0, align 8, !tbaa !9
+  %puts = tail call i32 @puts(i8* nonnull dereferenceable(1) getelementptr inbounds ([17 x i8], [17 x i8]* @str.4, i64 0, i64 0))
   ret void
 }
 
+; Function Attrs: nounwind uwtable
 define linkonce_odr hidden void @_ZN4BaseD0Ev(%class.Base* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
 entry:
-  call void @llvm.trap()
+  tail call void @llvm.trap()
   unreachable
 }
 
 declare dso_local void @__cxa_pure_virtual() unnamed_addr
 
 declare void @llvm.trap()
 
-declare dso_local void @_ZdlPv(i8*)
+declare dso_local void @_ZdlPv(i8*) local_unnamed_addr
+
+declare noundef i32 @puts(i8* nocapture noundef readonly) local_unnamed_addr
+
+!llvm.module.flags = !{!5, !6, !7}
 
-!llvm.module.flags = !{!5}
 !0 = !{i64 16, !"_ZTS4Base"}
 !1 = !{i64 32, !"_ZTSM4BaseFvvE.virtual"}
 !2 = !{i64 16, !"_ZTS7Derived"}
 !3 = !{i64 32, !"_ZTSM7DerivedFvvE.virtual"}
 !4 = !{i64 1}
-!5 = !{i32 1, !"Virtual Function Elim", i32 0}
+!5 = !{i32 1, !"wchar_size", i32 4}
+!6 = !{i32 1, !"Virtual Function Elim", i32 0}
+!7 = !{i32 7, !"uwtable", i32 1}
+!9 = !{!10, !10, i64 0}
+!10 = !{!"vtable pointer", !11, i64 0}
+!11 = !{!"Simple C++ TBAA"}

llvm/test/ThinLTO/X86/devirt_after_filtering_unreachable.ll

+; Test that unreachable functions are ignored by WPD in hybrid LTO, and thin LTO.
+; In this test case, the unreachable function is the virtual deleting destructor of an abstract class.
+
+; Generate split module with summary for hybrid Regular LTO WPD
+; RUN: opt -thinlto-bc -thinlto-split-lto-unit %s -o %t-main.bc
+; RUN: opt -thinlto-bc -thinlto-split-lto-unit %p/Inputs/devirt_after_filtering_unreachable_lib.ll -o %t-foo.bc
+
+; Check that deleting destructor of pure virtual class is unreachable.
+
+; Check that deleting destructor of pure virtual class is unreachable.
+; RUN: llvm-modextract -b -n=0 %t-foo.bc -o %t-foo.bc.0
+; RUN: llvm-dis -o - %t-foo.bc.0 | FileCheck %s --check-prefix=UNREACHABLEFLAG
+
+; Tests that devirtualization happens.
+; RUN: llvm-lto2 run -save-temps %t-main.bc %t-foo.bc -pass-remarks=. -o %t \
+; RUN:   -whole-program-visibility \
+; RUN:   -r=%t-foo.bc,_ZN7Derived1xEv,pl \
+; RUN:   -r=%t-foo.bc,puts, \
+; RUN:   -r=%t-foo.bc,_Z3fooP4Base,pl \
+; RUN:   -r=%t-foo.bc,_ZN7DerivedD2Ev,pl \
+; RUN:   -r=%t-foo.bc,_ZN7DerivedD0Ev,pl \
+; RUN:   -r=%t-foo.bc,_ZN4BaseD2Ev,pl \
+; RUN:   -r=%t-foo.bc,_ZN4BaseD0Ev,pl \
+; RUN:   -r=%t-foo.bc,__cxa_pure_virtual, \
+; RUN:   -r=%t-foo.bc,_ZdlPv, \
+; RUN:   -r=%t-foo.bc,_ZTV7Derived,l \
+; RUN:   -r=%t-foo.bc,_ZTVN10__cxxabiv120__si_class_type_infoE, \
+; RUN:   -r=%t-foo.bc,_ZTS7Derived,pl \
+; RUN:   -r=%t-foo.bc,_ZTVN10__cxxabiv117__class_type_infoE, \
+; RUN:   -r=%t-foo.bc,_ZTS4Base,pl \
+; RUN:   -r=%t-foo.bc,_ZTI4Base,pl \
+; RUN:   -r=%t-foo.bc,_ZTI7Derived,pl \
+; RUN:   -r=%t-foo.bc,_ZTV4Base,l \
+; RUN:   -r=%t-foo.bc,__cxa_pure_virtual, \
+; RUN:   -r=%t-foo.bc,_ZN7Derived1xEv, \
+; RUN:   -r=%t-foo.bc,_ZN7DerivedD2Ev, \
+; RUN:   -r=%t-foo.bc,_ZN7DerivedD0Ev, \
+; RUN:   -r=%t-foo.bc,_ZN4BaseD2Ev, \
+; RUN:   -r=%t-foo.bc,_ZN4BaseD0Ev, \
+; RUN:   -r=%t-foo.bc,_ZTV7Derived,pl \
+; RUN:   -r=%t-foo.bc,_ZTI4Base, \
+; RUN:   -r=%t-foo.bc,_ZTI7Derived, \
+; RUN:   -r=%t-foo.bc,_ZTV4Base,pl \
+; RUN:   -r=%t-main.bc,main,plx \
+; RUN:   -r=%t-main.bc,_Znwm,pl \
+; RUN:   -r=%t-main.bc,_Z3fooP4Base, \
+; RUN:   -r=%t-main.bc,_ZTV7Derived, 2>&1 | FileCheck %s --check-prefix=REMARK
+ 
+; REMARK-COUNT-1: single-impl: devirtualized a call to _ZN7DerivedD0Ev
+
+; Generate unsplit module with summary for ThinLTO index-based WPD.
+; RUN: opt -thinlto-bc -o %t3.o %s
+
+; Check that deleting destructor of pure virtual class is unreachable.
+; RUN: opt -thinlto-bc -o %t4.o %p/Inputs/devirt_after_filtering_unreachable_lib.ll
+; RUN: llvm-dis -o - %t4.o | FileCheck %s --check-prefix=UNREACHABLEFLAG
+
+; UNREACHABLEFLAG: gv: (name: "_ZN4BaseD0Ev", {{.*}}, funcFlags: ({{.*}} mustBeUnreachable: 1
+
+; Test that devirtualized happen in index based WPD
+; RUN: llvm-lto2 run %t4.o %t3.o -save-temps -pass-remarks=. \
+; RUN:   -whole-program-visibility \
+; RUN:   -wholeprogramdevirt-print-index-based \
+; RUN:   -o %t5 \
+; RUN:   -r=%t4.o,_ZN7Derived1xEv,pl \
+; RUN:   -r=%t4.o,puts, \
+; RUN:   -r=%t4.o,_Z3fooP4Base,pl \
+; RUN:   -r=%t4.o,_ZN7DerivedD2Ev,pl \
+; RUN:   -r=%t4.o,_ZN7DerivedD0Ev,pl \
+; RUN:   -r=%t4.o,_ZN4BaseD2Ev,pl \
+; RUN:   -r=%t4.o,_ZN4BaseD0Ev,pl \
+; RUN:   -r=%t4.o,__cxa_pure_virtual, \
+; RUN:   -r=%t4.o,_ZdlPv, \
+; RUN:   -r=%t4.o,_ZTV7Derived,pl \
+; RUN:   -r=%t4.o,_ZTVN10__cxxabiv120__si_class_type_infoE, \
+; RUN:   -r=%t4.o,_ZTS7Derived,pl \
+; RUN:   -r=%t4.o,_ZTVN10__cxxabiv117__class_type_infoE, \
+; RUN:   -r=%t4.o,_ZTS4Base,pl \
+; RUN:   -r=%t4.o,_ZTI4Base,pl \
+; RUN:   -r=%t4.o,_ZTI7Derived,pl \
+; RUN:   -r=%t4.o,_ZTV4Base,pl \
+; RUN:   -r=%t3.o,main,plx \
+; RUN:   -r=%t3.o,_Znwm, \
+; RUN:   -r=%t3.o,_Z3fooP4Base, \
+; RUN:   -r=%t3.o,_ZTV7Derived,  2>&1 | FileCheck %s --check-prefix=THINREMARK
+
+; THINREMARK: Devirtualized call to {{.*}} (_ZN7DerivedD0Ev)
+; THINREMARK: single-impl: devirtualized a call to _ZN7DerivedD0Ev
+; THINREMARK: single-impl: devirtualized a call to _ZN7DerivedD0Ev
+
+; ModuleID = 'tmp.cc'
+source_filename = "tmp.cc"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%class.Derived = type { %class.Base }
+%class.Base = type { i32 (...)** }
+
+@_ZTV7Derived = external dso_local unnamed_addr constant { [5 x i8*] }, align 8
+
+define hidden i32 @main() local_unnamed_addr {
+entry:
+  %call = tail call noalias nonnull dereferenceable(8) i8* @_Znwm(i64 8)
+  %0 = bitcast i8* %call to %class.Derived*
+  %1 = getelementptr inbounds %class.Derived, %class.Derived* %0, i64 0, i32 0, i32 0
+  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV7Derived, i64 0, inrange i32 0, i64 2) to i32 (...)**), i32 (...)*** %1, align 8, !tbaa !4
+  %2 = getelementptr %class.Derived, %class.Derived* %0, i64 0, i32 0
+  tail call void @_Z3fooP4Base(%class.Base* nonnull %2)
+  ret i32 0
+}
+
+declare dso_local nonnull i8* @_Znwm(i64) local_unnamed_addr
+
+declare dso_local void @_Z3fooP4Base(%class.Base*) local_unnamed_addr
+
+!llvm.module.flags = !{!0, !1, !2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, !"Virtual Function Elim", i32 0}
+!2 = !{i32 7, !"uwtable", i32 1}
+!4 = !{!5, !5, i64 0}
+!5 = !{!"vtable pointer", !6, i64 0}
+!6 = !{!"Simple C++ TBAA"}
 No newline at end of file

llvm/test/ThinLTO/X86/devirt_hybrid_after_filtering_unreachable.ll

-; Test that unreachable functions are ignored by WPD in hybrid LTO.
-; In this test case, the unreachable function is the virtual deleting destructor of an abstract class.
-
-; Generate split module with summary for hybrid Regular LTO WPD
-; RUN: opt -thinlto-bc -thinlto-split-lto-unit %s -o %t-main.bc
-; RUN: opt -thinlto-bc -thinlto-split-lto-unit %p/Inputs/devirt_hybrid_after_filtering_unreachable_lib.ll -o %t-foo.bc
-; Tests that devirtualization happens.
-; RUN: llvm-lto2 run -save-temps %t-main.bc %t-foo.bc -pass-remarks=. -o %t \
-; RUN:   -whole-program-visibility \
-; RUN:   -r=%t-foo.bc,_ZN7Derived1xEv,pl \
-; RUN:   -r=%t-foo.bc,printf, \
-; RUN:   -r=%t-foo.bc,_Z3fooP4Base,pl \
-; RUN:   -r=%t-foo.bc,_ZN7DerivedD2Ev,pl \
-; RUN:   -r=%t-foo.bc,_ZN7DerivedD0Ev,pl \
-; RUN:   -r=%t-foo.bc,_ZN4BaseD2Ev,pl \
-; RUN:   -r=%t-foo.bc,_ZN4BaseD0Ev,pl \
-; RUN:   -r=%t-foo.bc,__cxa_pure_virtual, \
-; RUN:   -r=%t-foo.bc,_ZdlPv, \
-; RUN:   -r=%t-foo.bc,_ZTV7Derived,l \
-; RUN:   -r=%t-foo.bc,_ZTVN10__cxxabiv120__si_class_type_infoE, \
-; RUN:   -r=%t-foo.bc,_ZTS7Derived,pl \
-; RUN:   -r=%t-foo.bc,_ZTVN10__cxxabiv117__class_type_infoE, \
-; RUN:   -r=%t-foo.bc,_ZTS4Base,pl \
-; RUN:   -r=%t-foo.bc,_ZTI4Base,pl \
-; RUN:   -r=%t-foo.bc,_ZTI7Derived,pl \
-; RUN:   -r=%t-foo.bc,_ZTV4Base,l \
-; RUN:   -r=%t-foo.bc,__cxa_pure_virtual, \
-; RUN:   -r=%t-foo.bc,_ZN7Derived1xEv, \
-; RUN:   -r=%t-foo.bc,_ZN7DerivedD2Ev, \
-; RUN:   -r=%t-foo.bc,_ZN7DerivedD0Ev, \
-; RUN:   -r=%t-foo.bc,_ZN4BaseD2Ev, \
-; RUN:   -r=%t-foo.bc,_ZN4BaseD0Ev, \
-; RUN:   -r=%t-foo.bc,_ZTV7Derived,pl \
-; RUN:   -r=%t-foo.bc,_ZTI4Base, \
-; RUN:   -r=%t-foo.bc,_ZTI7Derived, \
-; RUN:   -r=%t-foo.bc,_ZTV4Base,pl \
-; RUN:   -r=%t-main.bc,main,plx \
-; RUN:   -r=%t-main.bc,_Znwm,pl \
-; RUN:   -r=%t-main.bc,_ZN7DerivedC2Ev,pl \
-; RUN:   -r=%t-main.bc,_Z3fooP4Base, \
-; RUN:   -r=%t-main.bc,_ZN4BaseC2Ev,pl \
-; RUN:   -r=%t-main.bc,_ZN4BaseD2Ev, \
-; RUN:   -r=%t-main.bc,_ZN4BaseD0Ev, \
-; RUN:   -r=%t-main.bc,__cxa_pure_virtual, \
-; RUN:   -r=%t-main.bc,printf, \
-; RUN:   -r=%t-main.bc,_ZTV7Derived, \
-; RUN:   -r=%t-main.bc,_ZTV4Base, \
-; RUN:   -r=%t-main.bc,_ZTVN10__cxxabiv117__class_type_infoE, \
-; RUN:   -r=%t-main.bc,_ZTS4Base, \
-; RUN:   -r=%t-main.bc,_ZTI4Base, \
-; RUN:   -r=%t-main.bc,__cxa_pure_virtual, \
-; RUN:   -r=%t-main.bc,_ZN4BaseD2Ev, \
-; RUN:   -r=%t-main.bc,_ZN4BaseD0Ev, \
-; RUN:   -r=%t-main.bc,_ZTV4Base, \
-; RUN:   -r=%t-main.bc,_ZTI4Base, 2>&1 | FileCheck %s --check-prefix=REMARK
- 
-; REMARK-COUNT-1: single-impl: devirtualized a call to _ZN7DerivedD0Ev
-
-; ModuleID = 'tmp.cc'
-source_filename = "tmp.cc"
-target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64-unknown-linux-gnu"
-
-%class.Derived = type { %class.Base }
-%class.Base = type { i32 (...)** }
-
-$_ZN7DerivedC2Ev = comdat any
-
-$_ZN4BaseC2Ev = comdat any
-
-$_ZN4BaseD2Ev = comdat any
-
-$_ZN4BaseD0Ev = comdat any
-
-$_ZTV4Base = comdat any
-
-$_ZTS4Base = comdat any
-
-$_ZTI4Base = comdat any
-
-@_ZTV7Derived = external dso_local unnamed_addr constant { [5 x i8*] }, align 8
-@_ZTV4Base = linkonce_odr hidden unnamed_addr constant { [5 x i8*] } { [5 x i8*] [i8* null, i8* bitcast ({ i8*, i8* }* @_ZTI4Base to i8*), i8* bitcast (void (%class.Base*)* @_ZN4BaseD2Ev to i8*), i8* bitcast (void (%class.Base*)* @_ZN4BaseD0Ev to i8*), i8* bitcast (void ()* @__cxa_pure_virtual to i8*)] }, comdat, align 8, !type !0, !type !1, !vcall_visibility !2
-@_ZTVN10__cxxabiv117__class_type_infoE = external dso_local global i8*
-@_ZTS4Base = linkonce_odr hidden constant [6 x i8] c"4Base\00", comdat, align 1
-@_ZTI4Base = linkonce_odr hidden constant { i8*, i8* } { i8* bitcast (i8** getelementptr inbounds (i8*, i8** @_ZTVN10__cxxabiv117__class_type_infoE, i64 2) to i8*), i8* getelementptr inbounds ([6 x i8], [6 x i8]* @_ZTS4Base, i32 0, i32 0) }, comdat, align 8
-@.str = private unnamed_addr constant [18 x i8] c"In Base::~Base()\0A\00", align 1
-
-define hidden i32 @main() {
-entry:
-  %d = alloca %class.Derived*, align 8
-  %call = call noalias nonnull i8* @_Znwm(i64 8)
-  %0 = bitcast i8* %call to %class.Derived*
-  call void @_ZN7DerivedC2Ev(%class.Derived* nonnull align 8 dereferenceable(8) %0)
-  store %class.Derived* %0, %class.Derived** %d, align 8
-  %1 = load %class.Derived*, %class.Derived** %d, align 8
-  %2 = bitcast %class.Derived* %1 to %class.Base*
-  call void @_Z3fooP4Base(%class.Base* %2)
-  ret i32 0
-}
-
-declare dso_local nonnull i8* @_Znwm(i64)
-
-define linkonce_odr hidden void @_ZN7DerivedC2Ev(%class.Derived* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
-entry:
-  %0 = bitcast %class.Derived* %this to %class.Base*
-  call void @_ZN4BaseC2Ev(%class.Base* nonnull align 8 dereferenceable(8) %0)
-  %1 = bitcast %class.Derived* %this to i32 (...)***
-  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV7Derived, i32 0, inrange i32 0, i32 2) to i32 (...)**), i32 (...)*** %1, align 8
-  ret void
-}
-
-declare dso_local void @_Z3fooP4Base(%class.Base*)
-
-define linkonce_odr hidden void @_ZN4BaseC2Ev(%class.Base* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
-entry:
-  %0 = bitcast %class.Base* %this to i32 (...)***
-  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV4Base, i32 0, inrange i32 0, i32 2) to i32 (...)**), i32 (...)*** %0, align 8
-  ret void
-}
-
-define linkonce_odr hidden void @_ZN4BaseD2Ev(%class.Base* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
-entry:
-  %0 = bitcast %class.Base* %this to i32 (...)***
-  store i32 (...)** bitcast (i8** getelementptr inbounds ({ [5 x i8*] }, { [5 x i8*] }* @_ZTV4Base, i32 0, inrange i32 0, i32 2) to i32 (...)**), i32 (...)*** %0, align 8
-  %call = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([18 x i8], [18 x i8]* @.str, i64 0, i64 0))
-  ret void
-}
-
-define linkonce_odr hidden void @_ZN4BaseD0Ev(%class.Base* nonnull align 8 dereferenceable(8) %this) unnamed_addr comdat align 2 {
-entry:
-  call void @llvm.trap()
-  unreachable
-}
-
-declare dso_local void @__cxa_pure_virtual() unnamed_addr
-
-declare dso_local i32 @printf(i8*, ...)
-
-declare void @llvm.trap()
-
-!llvm.module.flags = !{!3}
-
-!0 = !{i64 16, !"_ZTS4Base"}
-!1 = !{i64 32, !"_ZTSM4BaseFvvE.virtual"}
-!2 = !{i64 1}
-!3 = !{i32 1, !"Virtual Function Elim", i32 0}