1; REQUIRES: x86-registered-target 2 3; Test that devirtualization option -wholeprogramdevirt-check adds code to check 4; that the devirtualization decision was correct and trap or fallback if not. 5 6; The vtables have vcall_visibility metadata with hidden visibility, to enable 7; devirtualization. 8 9; Generate unsplit module with summary for ThinLTO index-based WPD. 10; RUN: opt -thinlto-bc -o %t2.o %s 11 12; Check first in trapping mode. 13; RUN: llvm-lto2 run %t2.o -save-temps -pass-remarks=. \ 14; RUN: -wholeprogramdevirt-check=trap \ 15; RUN: -o %t3 \ 16; RUN: -r=%t2.o,test,px \ 17; RUN: -r=%t2.o,_ZN1A1nEi,p \ 18; RUN: -r=%t2.o,_ZN1B1fEi,p \ 19; RUN: -r=%t2.o,_ZTV1B,px 2>&1 | FileCheck %s --check-prefix=REMARK 20; RUN: llvm-dis %t3.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK --check-prefix=TRAP 21 22; Check next in fallback mode. 23; RUN: llvm-lto2 run %t2.o -save-temps -pass-remarks=. \ 24; RUN: -wholeprogramdevirt-check=fallback \ 25; RUN: -o %t3 \ 26; RUN: -r=%t2.o,test,px \ 27; RUN: -r=%t2.o,_ZN1A1nEi,p \ 28; RUN: -r=%t2.o,_ZN1B1fEi,p \ 29; RUN: -r=%t2.o,_ZTV1B,px 2>&1 | FileCheck %s --check-prefix=REMARK 30; RUN: llvm-dis %t3.1.4.opt.bc -o - | FileCheck %s --check-prefix=CHECK --check-prefix=FALLBACK 31 32; REMARK-DAG: single-impl: devirtualized a call to _ZN1A1nEi 33 34target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" 35target triple = "x86_64-grtev4-linux-gnu" 36 37%struct.A = type { i32 (...)** } 38%struct.B = type { %struct.A } 39 40@_ZTV1B = constant { [4 x i8*] } { [4 x i8*] [i8* null, i8* undef, i8* bitcast (i32 (%struct.B*, i32)* @_ZN1B1fEi to i8*), i8* bitcast (i32 (%struct.A*, i32)* @_ZN1A1nEi to i8*)] }, !type !0, !type !1, !vcall_visibility !5 41 42 43; CHECK-LABEL: define {{(noundef )?}}i32 @test 44define i32 @test(%struct.A* %obj, i32 %a) { 45entry: 46 %0 = bitcast %struct.A* %obj to i8*** 47 %vtable = load i8**, i8*** %0 48 %1 = bitcast i8** %vtable to i8* 49 %p = call i1 @llvm.type.test(i8* %1, metadata !"_ZTS1A") 50 call void @llvm.assume(i1 %p) 51 %fptrptr = getelementptr i8*, i8** %vtable, i32 1 52 %2 = bitcast i8** %fptrptr to i32 (%struct.A*, i32)** 53 %fptr1 = load i32 (%struct.A*, i32)*, i32 (%struct.A*, i32)** %2, align 8 54 55 ; Check that the call was devirtualized, but preceeded by a check guarding 56 ; a trap if the function pointer doesn't match. 57 ; TRAP: %.not = icmp eq ptr %fptr1, @_ZN1A1nEi 58 ; Ensure !prof and !callees metadata for indirect call promotion removed. 59 ; TRAP-NOT: prof 60 ; TRAP-NOT: callees 61 ; TRAP: br i1 %.not, label %1, label %0, !prof ![[PROF:[0-9]+]] 62 ; TRAP: 0: 63 ; TRAP: tail call void @llvm.debugtrap() 64 ; TRAP: br label %1 65 ; TRAP: 1: 66 ; TRAP: tail call i32 @_ZN1A1nEi 67 ; Check that the call was devirtualized, but preceeded by a check guarding 68 ; a fallback if the function pointer doesn't match. 69 ; FALLBACK: %0 = icmp eq ptr %fptr1, @_ZN1A1nEi 70 ; FALLBACK: br i1 %0, label %if.true.direct_targ, label %if.false.orig_indirect 71 ; FALLBACK: if.true.direct_targ: 72 ; FALLBACK: tail call i32 @_ZN1A1nEi 73 ; Ensure !prof and !callees metadata for indirect call promotion removed. 74 ; FALLBACK-NOT: prof 75 ; FALLBACK-NOT: callees 76 ; FALLBACK: br label %if.end.icp 77 ; FALLBACK: if.false.orig_indirect: 78 ; FALLBACK: tail call i32 %fptr1 79 ; Ensure !prof and !callees metadata for indirect call promotion removed. 80 ; In particular, if left on the fallback indirect call ICP may perform an 81 ; additional round of promotion. 82 ; FALLBACK-NOT: prof 83 ; FALLBACK-NOT: callees 84 ; FALLBACK: br label %if.end.icp 85 %call = tail call i32 %fptr1(%struct.A* nonnull %obj, i32 %a), !prof !6, !callees !7 86 87 ret i32 %call 88} 89; CHECK-LABEL: ret i32 90; CHECK-LABEL: } 91 92; TRAP: ![[PROF]] = !{!"branch_weights", i32 1048575, i32 1} 93 94declare i1 @llvm.type.test(i8*, metadata) 95declare void @llvm.assume(i1) 96 97define i32 @_ZN1B1fEi(%struct.B* %this, i32 %a) #0 { 98 ret i32 0; 99} 100 101define i32 @_ZN1A1nEi(%struct.A* %this, i32 %a) #0 { 102 ret i32 0; 103} 104 105; Make sure we don't inline or otherwise optimize out the direct calls. 106attributes #0 = { noinline optnone } 107 108!0 = !{i64 16, !"_ZTS1A"} 109!1 = !{i64 16, !"_ZTS1B"} 110!3 = !{i64 16, !4} 111!4 = distinct !{} 112!5 = !{i64 1} 113!6 = !{!"VP", i32 0, i64 1, i64 1621563287929432257, i64 1} 114!7 = !{i32 (%struct.A*, i32)* @_ZN1A1nEi} 115