Implement strip.invariant.group

[Prazek]

Summary

This patch introduce new intrinsic -
strip.invariant.group that was described in the
RFC: Devirtualization v2

[Prazek]

• Add deleted tests

[Prazek]

• small fixes in tests

[Prazek]

Reformat

[Prazek]

introduced mayBeDynamicClass and added more tests

[xbolva00]

clang/include/clang/AST/DeclCXX.h

778

maybeDynamicClass?

https://github.com/llvm-mirror/llvm/search?utf8=%E2%9C%93&q=maybe&type=

[Prazek]

Changed comments

[Prazek]

After rebasing

[Prazek]

rebase

[amharc]

clang/include/clang/AST/DeclCXX.h

778

I think mayBe is more appropriate, because it grammatically consists of 'may' and 'be'. In this form it appears in multiple places in the codebase, too. A few examples:

https://github.com/llvm-mirror/llvm/blob/0b24b74655b976aaba01b2f726659b8c745a16de/include/llvm/IR/GlobalValue.h#L131
https://github.com/llvm-mirror/clang/blob/647be32c6048e24f90f470c557e850597e5526c3/lib/AST/DeclCXX.cpp#L1800
https://github.com/llvm-mirror/llvm/blob/77438c3c98c0901fa9ef56f78bdefccedc16fcc4/lib/Analysis/ValueTracking.cpp#L3659

[rjmccall]

The changes to Clang generally seem reasonable, but I think you should split them into a separate commit from the commit that adds the intrinsic itself.

clang/lib/CodeGen/CGExpr.cpp

3858

Please add a comment explaining why this is necessary. (I'm actually not sure why it is, because surely the invariant groups we generate don't contain assumptions about memory from fields, right?)

clang/lib/CodeGen/CGExprScalar.cpp

1626

Again, comment.

1767

I think assigning to IntToPtr would be clearer here. You're changing the value returned, not introducing a reason to early-exit.

And a comment, please, on all of these cases.

[Prazek]

Slitted commit into defining and using intrinsic

clang/lib/CodeGen/CGExpr.cpp

3858

Short answer: you can only make virtual calls on a dynamic pointer that carries invariant.group and you can't do anything other because it could leak the information about this pointer (which when used with comparison could break devirtualization).

[rjmccall]

clang/lib/CodeGen/CGExpr.cpp

3858

Alright, sure.

[Prazek]

friendly ping

[Prazek]

Friendly ping @hfinkel @rsmith

[rsmith]

llvm/docs/LangRef.rst

12928

does carries -> carries

clang/include/clang/AST/DeclCXX.h

static CXXRecordDecl *CreateLambda(const ASTContext &C, DeclContext *DC,
                                      bool DependentLambda, bool IsGeneric,
                                      LambdaCaptureDefault CaptureDefault);
   static CXXRecordDecl *CreateDeserialized(const ASTContext &C, unsigned ID);
 
   bool isDynamicClass() const {
     return data().Polymorphic || data().NumVBases != 0;
   }
 
+  bool mayBeDynamicClass() const {
+    return !isCompleteDefinition() || isDynamicClass();
+  }
+
   void setIsParsingBaseSpecifiers() { data().IsParsingBaseSpecifiers = true; }
 
   bool isParsingBaseSpecifiers() const {
     return data().IsParsingBaseSpecifiers;
   }
 
   unsigned getODRHash() const;
 

clang/lib/CodeGen/CGExpr.cpp

if (base.getTBAAInfo().isMayAlias() ||
 
     // Update the final access type and size.
     FieldTBAAInfo.AccessType = CGM.getTBAATypeInfo(FieldType);
     FieldTBAAInfo.Size =
         getContext().getTypeSizeInChars(FieldType).getQuantity();
   }
 
   Address addr = base.getAddress();
+  if (auto *ClassDef = dyn_cast<CXXRecordDecl>(rec)) {
+    if (CGM.getCodeGenOpts().StrictVTablePointers &&
+        ClassDef->isDynamicClass()) {
+      auto *stripped = Builder.CreateStripInvariantGroup(addr.getPointer());
+      addr = Address(stripped, addr.getAlignment());
+    }
+  }
+
   unsigned RecordCVR = base.getVRQualifiers();
   if (rec->isUnion()) {
     // For unions, there is no pointer adjustment.
     assert(!FieldType->isReferenceType() && "union has reference member");
     if (CGM.getCodeGenOpts().StrictVTablePointers &&
         hasAnyVptr(FieldType, getContext()))
       // Because unions can easily skip invariant.barriers, we need to add
       // a barrier every time CXXRecord field with vptr is referenced.

clang/lib/CodeGen/CGExprScalar.cpp

case CK_BitCast: {
     if (CGF.SanOpts.has(SanitizerKind::CFIUnrelatedCast)) {
       if (auto PT = DestTy->getAs<PointerType>())
         CGF.EmitVTablePtrCheckForCast(PT->getPointeeType(), Src,
                                       /*MayBeNull=*/true,
                                       CodeGenFunction::CFITCK_UnrelatedCast,
                                       CE->getLocStart());
     }
 
+    const CXXRecordDecl *SourceClassDecl =
+        E->getType().getTypePtr()->getPointeeCXXRecordDecl();
+    if (CGF.CGM.getCodeGenOpts().StrictVTablePointers && SourceClassDecl &&
+        SourceClassDecl->mayBeDynamicClass()) {
+      Src = Builder.CreateStripInvariantGroup(Src);
+    }
+
     return Builder.CreateBitCast(Src, DstTy);
   }
   case CK_AddressSpaceConversion: {
     Expr::EvalResult Result;
     if (E->EvaluateAsRValue(Result, CGF.getContext()) &&
         Result.Val.isNullPointer()) {
       // If E has side effect, it is emitted even if its final result is a
       // null pointer. In that case, a DCE pass should be able to
case CK_IntegralToPointer: {
     // First, convert to the correct width so that we control the kind of
     // extension.
     auto DestLLVMTy = ConvertType(DestTy);
     llvm::Type *MiddleTy = CGF.CGM.getDataLayout().getIntPtrType(DestLLVMTy);
     bool InputSigned = E->getType()->isSignedIntegerOrEnumerationType();
     llvm::Value* IntResult =
       Builder.CreateIntCast(Src, MiddleTy, InputSigned, "conv");
 
-    return Builder.CreateIntToPtr(IntResult, DestLLVMTy);
+    auto *IntToPtr = Builder.CreateIntToPtr(IntResult, DestLLVMTy);
+
+    if (CGF.CGM.getCodeGenOpts().StrictVTablePointers)
+      return Builder.CreateLaunderInvariantGroup(IntToPtr);
+
+    return IntToPtr;
   }
-  case CK_PointerToIntegral:
+  case CK_PointerToIntegral: {
     assert(!DestTy->isBooleanType() && "bool should use PointerToBool");
-    return Builder.CreatePtrToInt(Visit(E), ConvertType(DestTy));
+    auto *PtrExpr = Visit(E);
+    const CXXRecordDecl *ClassDecl =
+        E->getType().getTypePtr()->getPointeeCXXRecordDecl();
+    if (CGF.CGM.getCodeGenOpts().StrictVTablePointers && ClassDecl &&
+        ClassDecl->mayBeDynamicClass()) {
+      PtrExpr = Builder.CreateStripInvariantGroup(PtrExpr);
+    }
 
+    return Builder.CreatePtrToInt(PtrExpr, ConvertType(DestTy));
+  }
   case CK_ToVoid: {
     CGF.EmitIgnoredExpr(E);
     return nullptr;
   }
   case CK_VectorSplat: {
     llvm::Type *DstTy = ConvertType(DestTy);
     Value *Elt = Visit(const_cast<Expr*>(E));
     // Splat the element across to all elements
if (const MemberPointerType *MPT = LHSTy->getAs<MemberPointerType>()) {
     }
 
     if (LHS->getType()->isFPOrFPVectorTy()) {
       Result = Builder.CreateFCmp(FCmpOpc, LHS, RHS, "cmp");
     } else if (LHSTy->hasSignedIntegerRepresentation()) {
       Result = Builder.CreateICmp(SICmpOpc, LHS, RHS, "cmp");
     } else {
       // Unsigned integers and pointers.
+
+      if (CGF.CGM.getCodeGenOpts().StrictVTablePointers &&
+          !isa<llvm::ConstantPointerNull>(LHS) &&
+          !isa<llvm::ConstantPointerNull>(RHS)) {
+        // Based on comparisons of pointers to dynamic objects, the optimizer
+        // can replace one pointer with another, which might be incorrect in
+        // presence of invariant groups. Comparison with null is safe.
+
+        if (auto *RD = LHSTy->getPointeeCXXRecordDecl())
+          if (RD->mayBeDynamicClass())
+            LHS = Builder.CreateStripInvariantGroup(LHS);
+        if (auto *RD = RHSTy->getPointeeCXXRecordDecl())
+          if (RD->mayBeDynamicClass())
+            RHS = Builder.CreateStripInvariantGroup(RHS);
+      }
+
       Result = Builder.CreateICmp(UICmpOpc, LHS, RHS, "cmp");
     }
 
     // If this is a vector comparison, sign extend the result to the appropriate
     // vector integer type and return it (don't convert to bool).
     if (LHSTy->isVectorType())
       return Builder.CreateSExt(Result, ConvertType(E->getType()), "sext");
 

clang/test/CodeGenCXX/strict-vtable-pointers.cpp

 // RUN: %clang_cc1 %s -I%S -triple=x86_64-apple-darwin10 -fstrict-vtable-pointers -std=c++11 -disable-llvm-passes -O2 -emit-llvm -o %t.ll
 // RUN: FileCheck --check-prefix=CHECK-CTORS %s < %t.ll
 // RUN: FileCheck --check-prefix=CHECK-NEW %s < %t.ll
 // RUN: FileCheck --check-prefix=CHECK-DTORS %s < %t.ll
 // RUN: FileCheck --check-prefix=CHECK-LINK-REQ %s < %t.ll
 
 typedef __typeof__(sizeof(0)) size_t;
-void *operator new(size_t, void*) throw();
+void *operator new(size_t, void *) throw();
+using uintptr_t = unsigned long long;
 
 struct NotTrivialDtor {
   ~NotTrivialDtor();
 };
 
 struct DynamicBase1 {
   NotTrivialDtor obj;
   virtual void foo();
 };
 
 struct DynamicDerived : DynamicBase1 {
-  void foo();
+  void foo() override;
 };
 
 struct DynamicBase2 {
   virtual void bar();
   ~DynamicBase2() {
     bar();
   }
 };
 
 struct DynamicDerivedMultiple : DynamicBase1, DynamicBase2 {
-  virtual void foo();
-  virtual void bar();
+  void foo() override;
+  void bar() override;
 };
 
 struct StaticBase {
   NotTrivialDtor obj;
   void bar();
 };
 
 struct DynamicFromStatic : StaticBase {
   virtual void bar();
 };
 
 struct DynamicFromVirtualStatic1 : virtual StaticBase {
 };
 
 struct DynamicFromVirtualStatic2 : virtual StaticBase {
 };
 
-struct DynamicFrom2Virtuals :
-            DynamicFromVirtualStatic1,
+struct DynamicFrom2Virtuals : DynamicFromVirtualStatic1,
                               DynamicFromVirtualStatic2 {
 };
 
 // CHECK-NEW-LABEL: define void @_Z12LocalObjectsv()
 // CHECK-NEW-NOT: @llvm.launder.invariant.group.p0i8(
 // CHECK-NEW-LABEL: {{^}}}
 void LocalObjects() {
   DynamicBase1 DB;
   DB.foo();

 // CHECK-CTORS-NOT: @llvm.launder.invariant.group.p0i8(
 // CHECK-CTORS-LABEL: {{^}}}
 
 struct DynamicFrom2Virtuals;
 // CHECK-CTORS-LABEL: define linkonce_odr void @_ZN20DynamicFrom2VirtualsC1Ev
 // CHECK-CTORS: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-CTORS-LABEL: {{^}}}
 
-
 // CHECK-NEW-LABEL: define void @_Z9Pointers1v()
 // CHECK-NEW-NOT: @llvm.launder.invariant.group.p0i8(
 // CHECK-NEW-LABEL: call void @_ZN12DynamicBase1C1Ev(
 
 // CHECK-NEW: %[[THIS3:.*]] = call i8* @llvm.launder.invariant.group.p0i8(i8* %[[THIS2:.*]])
 // CHECK-NEW: %[[THIS4:.*]] = bitcast i8* %[[THIS3]] to %[[DynamicDerived:.*]]*
 // CHECK-NEW: call void @_ZN14DynamicDerivedC1Ev(%[[DynamicDerived:.*]]* %[[THIS4]])
 // CHECK-NEW-LABEL: {{^}}}

 }
 
 /*** Testing Constructors ***/
 struct DynamicBase1;
 // CHECK-CTORS-LABEL: define linkonce_odr void @_ZN12DynamicBase1C2Ev(
 // CHECK-CTORS-NOT: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-CTORS-LABEL: {{^}}}
 
-
 struct DynamicDerived;
 
 // CHECK-CTORS-LABEL: define linkonce_odr void @_ZN14DynamicDerivedC2Ev(
 // CHECK-CTORS: %[[THIS0:.*]] = load %[[DynamicDerived:.*]]*, %[[DynamicDerived]]** {{.*}}
 // CHECK-CTORS: %[[THIS1:.*]] = bitcast %[[DynamicDerived:.*]]* %[[THIS0]] to i8*
 // CHECK-CTORS: %[[THIS2:.*]] = call i8* @llvm.launder.invariant.group.p0i8(i8* %[[THIS1:.*]])
 // CHECK-CTORS: %[[THIS3:.*]] = bitcast i8* %[[THIS2]] to %[[DynamicDerived]]*
 // CHECK-CTORS: %[[THIS4:.*]] = bitcast %[[DynamicDerived]]* %[[THIS3]] to %[[DynamicBase:.*]]*

 // CHECK-CTORS: %[[THIS4:.*]] = bitcast %[[CLASS]]* %[[THIS3]] to %[[BASE_CLASS:.*]]*
 // CHECK-CTORS: call void @_ZN12DynamicBase1C2Ev(%[[BASE_CLASS]]* %[[THIS4]])
 
 // CHECK-CTORS: call i8* @llvm.launder.invariant.group.p0i8(
 
 // CHECK-CTORS: call void @_ZN12DynamicBase2C2Ev(
 // CHECK-CTORS-NOT: @llvm.launder.invariant.group.p0i8
 
-
 // CHECK-CTORS: %[[THIS10:.*]] = bitcast %struct.DynamicDerivedMultiple* %[[THIS0]] to i32 (...)***
 // CHECK-CTORS: store {{.*}} @_ZTV22DynamicDerivedMultiple, i32 0, inrange i32 0, i32 2) {{.*}} %[[THIS10]]
 // CHECK-CTORS: %[[THIS11:.*]] = bitcast %struct.DynamicDerivedMultiple* %[[THIS0]] to i8*
 // CHECK-CTORS: %[[THIS_ADD:.*]] = getelementptr inbounds i8, i8* %[[THIS11]], i64 16
 // CHECK-CTORS: %[[THIS12:.*]]  = bitcast i8* %[[THIS_ADD]] to i32 (...)***
 
-
 // CHECK-CTORS: store {{.*}} @_ZTV22DynamicDerivedMultiple, i32 0, inrange i32 1, i32 2) {{.*}} %[[THIS12]]
 // CHECK-CTORS-LABEL: {{^}}}
 
 struct DynamicFromStatic;
 // CHECK-CTORS-LABEL: define linkonce_odr void @_ZN17DynamicFromStaticC2Ev(
 // CHECK-CTORS-NOT: @llvm.launder.invariant.group.p0i8(
 // CHECK-CTORS-LABEL: {{^}}}
 
 struct A {
   virtual void foo();
+  int m;
 };
 struct B : A {
-  virtual void foo();
+  void foo() override;
 };
 
 union U {
   A a;
   B b;
 };
 
 void changeToB(U *u);
 void changeToA(U *u);
 
 void g2(A *a) {
   a->foo();
 }
 // We have to guard access to union fields with invariant.group, because
 // it is very easy to skip the barrier with unions. In this example the inlined
 // g2 will produce loads with the same !invariant.group metadata, and
 // u->a and u->b would use the same pointer.
 // CHECK-NEW-LABEL: define void @_Z14UnionsBarriersP1U
 void UnionsBarriers(U *u) {
   // CHECK-NEW: call void @_Z9changeToBP1U(
   changeToB(u);
   // CHECK-NEW: call i8* @llvm.launder.invariant.group.p0i8(i8*
   // CHECK-NEW: call void @_Z2g2P1A(%struct.A*
   g2(&u->b);
-  // CHECK-NEW: call void @_Z9changeToAP1U(%union.U* 
+  // CHECK-NEW: call void @_Z9changeToAP1U(%union.U*
   changeToA(u);
   // CHECK-NEW: call i8* @llvm.launder.invariant.group.p0i8(i8*
   // call void @_Z2g2P1A(%struct.A* %a)
   g2(&u->a);
   // CHECK-NEW-NOT: call i8* @llvm.launder.invariant.group.p0i8(i8*
 }
 
 struct HoldingVirtuals {
void UnionsBarrier3(U3 &u) {
   // CHECK-NEW: call void @_Z4takeR13VirtualInBase(
   take(u.v2);
 
   // CHECK-NEW: call i8* @llvm.launder.invariant.group.p0i8(i8*
   // CHECK-NEW: call void @_Z4takeR18VirtualInheritance(
   take(u.v3);
 }
 
+// CHECK-NEW-LABEL: define void @_Z7comparev()
+void compare() {
+  A *a = new A;
+  a->foo();
+  // CHECK-NEW: call i8* @llvm.launder.invariant.group.p0i8(i8*
+  A *b = new (a) B;
+
+  // CHECK-NEW: %[[a:.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8*
+  // CHECK-NEW: %[[a2:.*]] = bitcast i8* %[[a]] to %struct.A*
+  // CHECK-NEW: %[[b:.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8*
+  // CHECK-NEW: %[[b2:.*]] = bitcast i8* %[[b]] to %struct.A*
+  // CHECK-NEW: %cmp = icmp eq %struct.A* %[[a2]], %[[b2]]
+  if (a == b)
+    b->foo();
+}
+
+// CHECK-NEW-LABEL: compare2
+bool compare2(A *a, A *a2) {
+  // CHECK-NEW: %[[a:.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8*
+  // CHECK-NEW: %[[a2:.*]] = bitcast i8* %[[a]] to %struct.A*
+  // CHECK-NEW: %[[b:.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8*
+  // CHECK-NEW: %[[b2:.*]] = bitcast i8* %[[b]] to %struct.A*
+  // CHECK-NEW: %cmp = icmp ult %struct.A* %[[a2]], %[[b2]]
+  return a < a2;
+}
+// CHECK-NEW-LABEL: compareIntPointers
+bool compareIntPointers(int *a, int *b) {
+  // CHECK-NEW-NOT: call i8* @llvm.strip.invariant.group
+  return a == b;
+}
+
+struct HoldingOtherVirtuals {
+  B b;
+};
+
+// There is no need to add barriers for comparision of pointer to classes
+// that are not dynamic.
+// CHECK-NEW-LABEL: compare5
+bool compare5(HoldingOtherVirtuals *a, HoldingOtherVirtuals *b) {
+  // CHECK-NEW-NOT: call i8* @llvm.strip.invariant.group
+  return a == b;
+}
+// CHECK-NEW-LABEL: compareNull
+bool compareNull(A *a) {
+  // CHECK-NEW-NOT: call i8* @llvm.strip.invariant.group
+
+  if (a != nullptr)
+    return false;
+  if (!a)
+    return false;
+  return a == nullptr;
+}
+
+struct X;
+// We have to also introduce the barriers if comparing pointers to incomplete
+// objects
+// CHECK-NEW-LABEL: define zeroext i1 @_Z8compare4P1XS0_
+bool compare4(X *x, X *x2) {
+  // CHECK-NEW: %[[x:.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8*
+  // CHECK-NEW: %[[xp:.*]] = bitcast i8* %[[x]] to %struct.X*
+  // CHECK-NEW: %[[x2:.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8*
+  // CHECK-NEW: %[[x2p:.*]] = bitcast i8* %[[x2]] to %struct.X*
+  // CHECK-NEW: %cmp = icmp eq %struct.X* %[[xp]], %[[x2p]]
+  return x == x2;
+}
+
+// CHECK-NEW-LABEL: define void @_Z7member1P20HoldingOtherVirtuals(
+void member1(HoldingOtherVirtuals *p) {
+
+  // CHECK-NEW-NOT: call i8* @llvm.strip.invariant.group.p0i8(
+  (void)p->b;
+}
+
+// CHECK-NEW-LABEL: member2
+void member2(A *a) {
+  // CHECK-NEW: call i8* @llvm.strip.invariant.group.p0i8
+  (void)a->m;
+}
+
+// Check if from comparison of addresses of member we can't infer the equality
+// of ap and bp.
+// CHECK-NEW-LABEL: @_Z18testCompareMembersv(
+void testCompareMembers() {
+  // CHECK-NEW:    [[AP:%.*]] = alloca %struct.A*
+  // CHECK-NEW:    [[APM:%.*]] = alloca i32*
+  // CHECK-NEW:    [[BP:%.*]] = alloca %struct.B*
+  // CHECK-NEW:    [[BPM:%.*]] = alloca i32*
+
+  A *ap = new A;
+  // CHECK-NEW:   call void %{{.*}}(%struct.A* %{{.*}})
+  ap->foo();
+  // CHECK-NEW:    [[TMP7:%.*]] = load %struct.A*, %struct.A** [[AP]]
+  // CHECK-NEW:    [[TMP8:%.*]] = bitcast %struct.A* [[TMP7]] to i8*
+  // CHECK-NEW:    [[TMP9:%.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8* [[TMP8]])
+  // CHECK-NEW:    [[TMP10:%.*]] = bitcast i8* [[TMP9]] to %struct.A*
+  // CHECK-NEW:    [[M:%.*]] = getelementptr inbounds [[STRUCT_A:%.*]], %struct.A* [[TMP10]], i32 0, i32 1
+  // CHECK-NEW:    store i32* [[M]], i32** [[APM]]
+  int *const apm = &ap->m;
+
+  B *bp = new (ap) B;
+
+  // CHECK-NEW:    [[TMP20:%.*]] = load %struct.B*, %struct.B** [[BP]]
+  // CHECK-NEW:    [[TMP21:%.*]] = bitcast %struct.B* [[TMP20]] to %struct.A*
+  // CHECK-NEW:    [[TMP22:%.*]] = bitcast %struct.A* [[TMP21]] to i8*
+  // CHECK-NEW:    [[TMP23:%.*]] = call i8* @llvm.strip.invariant.group.p0i8(i8* [[TMP22]])
+  // CHECK-NEW:    [[TMP24:%.*]] = bitcast i8* [[TMP23]] to %struct.A*
+  // CHECK-NEW:    [[M4:%.*]] = getelementptr inbounds [[STRUCT_A]], %struct.A* [[TMP24]], i32 0, i32 1
+  // CHECK-NEW:    store i32* [[M4]], i32** [[BPM]]
+  int *const bpm = &bp->m;
+
+  // CHECK-NEW:    [[TMP25:%.*]] = load i32*, i32** [[APM]]
+  // CHECK-NEW:    [[TMP26:%.*]] = load i32*, i32** [[BPM]]
+  // CHECK-NEW-NOT: strip.invariant.group
+  // CHECK-NEW-NOT: launder.invariant.group
+  // CHECK-NEW:    [[CMP:%.*]] = icmp eq i32* [[TMP25]], [[TMP26]]
+  if (apm == bpm) {
+    bp->foo();
+  }
+}
+
+// CHECK-NEW-LABEL: define void @_Z9testCast1P1A(%struct.A*
+void testCast1(A *a) {
+  // Here we get rid of dynamic info
+  // CHECK-NEW: call i8* @llvm.strip.invariant.group
+  auto *v = (void *)a;
+
+  // CHECK-NEW: call i8* @llvm.strip.invariant.group
+  auto i2 = (uintptr_t)a;
+  (void)i2;
+
+  // CHECK-NEW-NOT: @llvm.strip.invariant.group
+  // CHECK-NEW-NOT: @llvm.launder.invariant.group
+
+  // The information is already stripped
+  auto i = (uintptr_t)v;
+}
+
+struct Incomplete;
+// CHECK-NEW-LABEL: define void @_Z9testCast2P10Incomplete(%struct.Incomplete*
+void testCast2(Incomplete *I) {
+  // Here we get rid of dynamic info
+  // CHECK-NEW: call i8* @llvm.strip.invariant.group
+  auto *v = (void *)I;
+
+  // CHECK-NEW: call i8* @llvm.strip.invariant.group
+  auto i2 = (uintptr_t)I;
+  (void)i2;
+
+  // CHECK-NEW-NOT: @llvm.strip.invariant.group
+  // CHECK-NEW-NOT: @llvm.launder.invariant.group
+
+  // The information is already stripped
+  auto i = (uintptr_t)v;
+}
+
+// CHECK-NEW-LABEL: define void @_Z9testCast3y(
+void testCast3(uintptr_t i) {
+  // CHECK-NEW-NOT: @llvm.strip.invariant.group
+  // CHECK-NEW: @llvm.launder.invariant.group
+  A *a3 = (A *)i;
+  (void)a3;
+
+  // CHECK-NEW: @llvm.launder.invariant.group
+  auto *v2 = (void *)i;
+  // CHECK-NEW-NOT: @llvm.launder.invariant.group
+  A *a2 = (A *)v2;
+  (void)a2;
+
+  // CHECK-NEW-LABEL: ret void
+}
+
+// CHECK-NEW-LABEL: define void @_Z9testCast4y(
+void testCast4(uintptr_t i) {
+  // CHECK-NEW-NOT: @llvm.strip.invariant.group
+  // CHECK-NEW: @llvm.launder.invariant.group
+  auto *a3 = (Incomplete *)i;
+  (void)a3;
+
+  // CHECK-NEW: @llvm.launder.invariant.group
+  auto *v2 = (void *)i;
+  // CHECK-NEW-NOT: @llvm.launder.invariant.group
+  auto *a2 = (Incomplete *)v2;
+  (void)a2;
+
+  // CHECK-NEW-LABEL: ret void
+}
+
 /** DTORS **/
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN10StaticBaseD2Ev(
 // CHECK-DTORS-NOT: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-DTORS-LABEL: {{^}}}
 
-
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN25DynamicFromVirtualStatic2D2Ev(
 // CHECK-DTORS-NOT: invariant.barrier
 // CHECK-DTORS-LABEL: {{^}}}
 
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN17DynamicFromStaticD2Ev
 // CHECK-DTORS-NOT: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-DTORS-LABEL: {{^}}}
 
-
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN22DynamicDerivedMultipleD2Ev(
 
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN12DynamicBase2D2Ev(
 // CHECK-DTORS: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-DTORS-LABEL: {{^}}}
 
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN12DynamicBase1D2Ev
 // CHECK-DTORS: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-DTORS-LABEL: {{^}}}
 
 // CHECK-DTORS-LABEL: define linkonce_odr void @_ZN14DynamicDerivedD2Ev
 // CHECK-DTORS-NOT: call i8* @llvm.launder.invariant.group.p0i8(
 // CHECK-DTORS-LABEL: {{^}}}
 
-
 // CHECK-LINK-REQ: !llvm.module.flags = !{![[FIRST:[0-9]+]], ![[SEC:[0-9]+]]{{.*}}}
 
 // CHECK-LINK-REQ: ![[FIRST]] = !{i32 1, !"StrictVTablePointers", i32 1}
 // CHECK-LINK-REQ: ![[SEC]] = !{i32 3, !"StrictVTablePointersRequirement", ![[META:.*]]}
 // CHECK-LINK-REQ: ![[META]] = !{!"StrictVTablePointers", i32 1}
-

llvm/docs/LangRef.rst

 ::
 
       declare i8* @llvm.launder.invariant.group.p0i8(i8* <ptr>)
 
 Overview:
 """""""""
 
 The '``llvm.launder.invariant.group``' intrinsic can be used when an invariant
-established by invariant.group metadata no longer holds, to obtain a new pointer
-value that does not carry the invariant information. It is an experimental
+established by ``invariant.group`` metadata no longer holds, to obtain a new pointer
+value that does carries fresh invariant group information. It is an experimental
 intrinsic, which means that its semantics might change in the future.
 
 
 Arguments:
 """"""""""
 
-The ``llvm.launder.invariant.group`` takes only one argument, which is
-the pointer to the memory for which the ``invariant.group`` no longer holds.
+The ``llvm.launder.invariant.group`` takes only one argument, which is a pointer
+to the memory.
 
 Semantics:
 """"""""""
 
 Returns another pointer that aliases its argument but which is considered different
 for the purposes of ``load``/``store`` ``invariant.group`` metadata.
 It does not read any accessible memory and the execution can be speculated.
 
+'``llvm.strip.invariant.group``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+This is an overloaded intrinsic. The memory object can belong to any address
+space. The returned pointer must belong to the same address space as the
+argument.
+
+::
+
+      declare i8* @llvm.strip.invariant.group.p0i8(i8* <ptr>)
+
+Overview:
+"""""""""
+
+The '``llvm.strip.invariant.group``' intrinsic can be used when an invariant
+established by ``invariant.group`` metadata no longer holds, to obtain a new pointer
+value that does not carry the invariant information. It is an experimental
+intrinsic, which means that its semantics might change in the future.
+
+
+Arguments:
+""""""""""
+
+The ``llvm.strip.invariant.group`` takes only one argument, which is a pointer
+to the memory.
+
+Semantics:
+""""""""""
+
+Returns another pointer that aliases its argument but which has no associated
+``invariant.group`` metadata.
+It does not read any memory and can be speculated.
+
+
+
 .. _constrainedfp:
 
 Constrained Floating-Point Intrinsics
 -------------------------------------
 
 These intrinsics are used to provide special handling of floating-point
 operations when specific rounding mode or floating-point exception behavior is
 required.  By default, LLVM optimization passes assume that the rounding mode is

llvm/include/llvm/IR/IRBuilder.h

   }
 
   /// Create a launder.invariant.group intrinsic call. If Ptr type is
   /// different from pointer to i8, it's casted to pointer to i8 in the same
   /// address space before call and casted back to Ptr type after call.
   Value *CreateLaunderInvariantGroup(Value *Ptr) {
     assert(isa<PointerType>(Ptr->getType()) &&
            "launder.invariant.group only applies to pointers.");
+    // FIXME: we could potentially avoid casts to/from i8*.
     auto *PtrType = Ptr->getType();
     auto *Int8PtrTy = getInt8PtrTy(PtrType->getPointerAddressSpace());
     if (PtrType != Int8PtrTy)
       Ptr = CreateBitCast(Ptr, Int8PtrTy);
     Module *M = BB->getParent()->getParent();
     Function *FnLaunderInvariantGroup = Intrinsic::getDeclaration(
         M, Intrinsic::launder_invariant_group, {Int8PtrTy});
 
     assert(FnLaunderInvariantGroup->getReturnType() == Int8PtrTy &&
            FnLaunderInvariantGroup->getFunctionType()->getParamType(0) ==
                Int8PtrTy &&
            "LaunderInvariantGroup should take and return the same type");
 
     CallInst *Fn = CreateCall(FnLaunderInvariantGroup, {Ptr});
 
     if (PtrType != Int8PtrTy)
       return CreateBitCast(Fn, PtrType);
     return Fn;
   }
 
+  /// \brief Create a strip.invariant.group intrinsic call. If Ptr type is
+  /// different from pointer to i8, it's casted to pointer to i8 in the same
+  /// address space before call and casted back to Ptr type after call.
+  Value *CreateStripInvariantGroup(Value *Ptr) {
+    assert(isa<PointerType>(Ptr->getType()) &&
+           "strip.invariant.group only applies to pointers.");
+
+    // FIXME: we could potentially avoid casts to/from i8*.
+    auto *PtrType = Ptr->getType();
+    auto *Int8PtrTy = getInt8PtrTy(PtrType->getPointerAddressSpace());
+    if (PtrType != Int8PtrTy)
+      Ptr = CreateBitCast(Ptr, Int8PtrTy);
+    Module *M = BB->getParent()->getParent();
+    Function *FnStripInvariantGroup = Intrinsic::getDeclaration(
+        M, Intrinsic::strip_invariant_group, {Int8PtrTy});
+
+    assert(FnStripInvariantGroup->getReturnType() == Int8PtrTy &&
+           FnStripInvariantGroup->getFunctionType()->getParamType(0) ==
+               Int8PtrTy &&
+           "StripInvariantGroup should take and return the same type");
+
+    CallInst *Fn = CreateCall(FnStripInvariantGroup, {Ptr});
+
+    if (PtrType != Int8PtrTy)
+      return CreateBitCast(Fn, PtrType);
+    return Fn;
+  }
+
   /// Return a vector value that contains \arg V broadcasted to \p
   /// NumElts elements.
   Value *CreateVectorSplat(unsigned NumElts, Value *V, const Twine &Name = "") {
     assert(NumElts > 0 && "Cannot splat to an empty vector!");
 
     // First insert it into an undef vector so we can shuffle it.
     Type *I32Ty = getInt32Ty();
     Value *Undef = UndefValue::get(VectorType::get(V->getType(), NumElts));

llvm/include/llvm/IR/Intrinsics.td

 // The argument also can't be marked with 'returned' attribute, because
 // it would remove barrier.
 // Note that it is still experimental, which means that its semantics
 // might change in the future.
 def int_launder_invariant_group : Intrinsic<[llvm_anyptr_ty],
                                             [LLVMMatchType<0>],
                                             [IntrInaccessibleMemOnly, IntrSpeculatable]>;
 
+
+def int_strip_invariant_group : Intrinsic<[llvm_anyptr_ty],
+                                          [LLVMMatchType<0>],
+                                          [IntrSpeculatable, IntrNoMem]>;
+
 //===------------------------ Stackmap Intrinsics -------------------------===//
 //
 def int_experimental_stackmap : Intrinsic<[],
                                   [llvm_i64_ty, llvm_i32_ty, llvm_vararg_ty],
                                   [Throws]>;
 def int_experimental_patchpoint_void : Intrinsic<[],
                                                  [llvm_i64_ty, llvm_i32_ty,
                                                   llvm_ptr_ty, llvm_i32_ty,

llvm/lib/Analysis/BasicAliasAnalysis.cpp

if (Op->getOpcode() == Instruction::BitCast ||
         Op->getOpcode() == Instruction::AddrSpaceCast) {
       V = Op->getOperand(0);
       continue;
     }
 
     const GEPOperator *GEPOp = dyn_cast<GEPOperator>(Op);
     if (!GEPOp) {
       if (auto CS = ImmutableCallSite(V)) {
-        // Note: getArgumentAliasingToReturnedPointer keeps CaptureTracking in
-        // sync, which is needed for correctness.
+        // CaptureTracking can know about special capturing properties of some
+        // intrinsics like launder.invariant.group, that can't be expressed with
+        // the attributes, but have properties like returning aliasing pointer.
+        // Because some analysis may assume that nocaptured pointer is not
+        // returned from some special intrinsic (because function would have to
+        // be marked with returns attribute), it is crucial to use this function
+        // because it should be in sync with CaptureTracking. Not using it may
+        // cause weird miscompilations where 2 aliasing pointers are assumed to
+        // noalias.
         if (auto *RP = getArgumentAliasingToReturnedPointer(CS)) {
           V = RP;
           continue;
         }
       }
 
       // If it's not a GEP, hand it off to SimplifyInstruction to see if it
       // can come up with something. This matches what GetUnderlyingObject does.

llvm/lib/Analysis/CaptureTracking.cpp

while (!Worklist.empty()) {
     case Instruction::Invoke: {
       CallSite CS(I);
       // Not captured if the callee is readonly, doesn't return a copy through
       // its return value and doesn't unwind (a readonly function can leak bits
       // by throwing an exception or not depending on the input value).
       if (CS.onlyReadsMemory() && CS.doesNotThrow() && I->getType()->isVoidTy())
         break;
 
-      // launder.invariant.group only captures pointer by returning it,
+      // {launder,strip}.invariant.group only captures pointer by returning it,
       // so the pointer wasn't captured if returned pointer is not captured.
       // This intrinsic is not marked as nocapture, because it would require
       // to mark the argument as returned, which would make the launder useless.
       // NOTE: CaptureTracking users should not assume that only functions
       // marked with nocapture do not capture. This means that places like
       // GetUnderlyingObject in ValueTracking or DecomposeGEPExpression
-      // in BasicAA also assume this aliasing property of the launder.
-      if (CS.getIntrinsicID() == Intrinsic::launder_invariant_group) {
+      // in BasicAA also need to know about this property. To achieve that you
+      // need to keep it in sync with getArgumentAliasingToReturnedPointer.
+      if (CS.getIntrinsicID() == Intrinsic::launder_invariant_group ||
+          CS.getIntrinsicID() == Intrinsic::strip_invariant_group) {
         AddUses(I);
         break;
       }
 
       // Volatile operations effectively capture the memory location that they
       // load and store to.
       if (auto *MI = dyn_cast<MemIntrinsic>(I))
         if (MI->isVolatile())

llvm/lib/Analysis/ConstantFolding.cpp

bool llvm::canConstantFoldCallTo(ImmutableCallSite CS, const Function *F) {
   case Intrinsic::bswap:
   case Intrinsic::ctpop:
   case Intrinsic::ctlz:
   case Intrinsic::cttz:
   case Intrinsic::fma:
   case Intrinsic::fmuladd:
   case Intrinsic::copysign:
   case Intrinsic::launder_invariant_group:
+  case Intrinsic::strip_invariant_group:
   case Intrinsic::round:
   case Intrinsic::masked_load:
   case Intrinsic::sadd_with_overflow:
   case Intrinsic::uadd_with_overflow:
   case Intrinsic::ssub_with_overflow:
   case Intrinsic::usub_with_overflow:
   case Intrinsic::smul_with_overflow:
   case Intrinsic::umul_with_overflow:
Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID, Type *Ty,
                                  const TargetLibraryInfo *TLI) {
   if (Operands.size() == 1) {
     if (isa<UndefValue>(Operands[0])) {
       // cosine(arg) is between -1 and 1. cosine(invalid arg) is NaN
       if (IntrinsicID == Intrinsic::cos)
         return Constant::getNullValue(Ty);
       if (IntrinsicID == Intrinsic::bswap ||
           IntrinsicID == Intrinsic::bitreverse ||
-          IntrinsicID == Intrinsic::launder_invariant_group)
+          IntrinsicID == Intrinsic::launder_invariant_group ||
+          IntrinsicID == Intrinsic::strip_invariant_group)
         return Operands[0];
     }
 
     if (isa<ConstantPointerNull>(Operands[0]) &&
         Operands[0]->getType()->getPointerAddressSpace() == 0) {
-      // launder(null) == null iff in addrspace 0
-      if (IntrinsicID == Intrinsic::launder_invariant_group)
+      // launder(null) == null == strip(null) iff in addrspace 0
+      if (IntrinsicID == Intrinsic::launder_invariant_group ||
+          IntrinsicID == Intrinsic::strip_invariant_group)
         return Operands[0];
       return nullptr;
     }
 
     if (auto *Op = dyn_cast<ConstantFP>(Operands[0])) {
       if (IntrinsicID == Intrinsic::convert_to_fp16) {
         APFloat Val(Op->getValueAPF());
 

llvm/lib/Analysis/ValueTracking.cpp

 }
 
 const Value *llvm::getArgumentAliasingToReturnedPointer(ImmutableCallSite CS) {
   assert(CS &&
          "getArgumentAliasingToReturnedPointer only works on nonnull CallSite");
   if (const Value *RV = CS.getReturnedArgOperand())
     return RV;
   // This can be used only as a aliasing property.
-  if (CS.getIntrinsicID() == Intrinsic::launder_invariant_group)
+  if (CS.getIntrinsicID() == Intrinsic::launder_invariant_group ||
+      CS.getIntrinsicID() == Intrinsic::strip_invariant_group)
     return CS.getArgOperand(0);
   return nullptr;
 }
 
 /// \p PN defines a loop-variant pointer to an object.  Check if the
 /// previous iteration of the loop was referring to the same object as \p PN.
 static bool isSameUnderlyingObjectInLoop(const PHINode *PN,
                                          const LoopInfo *LI) {
if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
       if (GA->isInterposable())
         return V;
       V = GA->getAliasee();
     } else if (isa<AllocaInst>(V)) {
       // An alloca can't be further simplified.
       return V;
     } else {
       if (auto CS = CallSite(V)) {
-        // Note: getArgumentAliasingToReturnedPointer keeps CaptureTracking in
-        // sync, which is needed for correctness.
+        // CaptureTracking can know about special capturing properties of some
+        // intrinsics like launder.invariant.group, that can't be expressed with
+        // the attributes, but have properties like returning aliasing pointer.
+        // Because some analysis may assume that nocaptured pointer is not
+        // returned from some special intrinsic (because function would have to
+        // be marked with returns attribute), it is crucial to use this function
+        // because it should be in sync with CaptureTracking. Not using it may
+        // cause weird miscompilations where 2 aliasing pointers are assumed to
+        // noalias.
         if (auto *RP = getArgumentAliasingToReturnedPointer(CS)) {
           V = RP;
           continue;
         }
       }
 
       // See if InstructionSimplify knows any relevant tricks.
       if (Instruction *I = dyn_cast<Instruction>(V))

llvm/lib/CodeGen/CodeGenPrepare.cpp

case Intrinsic::aarch64_stxr: {
       // Sink a zext feeding stlxr/stxr before it, so it can be folded into it.
       ExtVal->moveBefore(CI);
       // Mark this instruction as "inserted by CGP", so that other
       // optimizations don't touch it.
       InsertedInsts.insert(ExtVal);
       return true;
     }
     case Intrinsic::launder_invariant_group:
+    case Intrinsic::strip_invariant_group:
       II->replaceAllUsesWith(II->getArgOperand(0));
       II->eraseFromParent();
       return true;
 
     case Intrinsic::cttz:
     case Intrinsic::ctlz:
       // If counting zeros is expensive, try to avoid it.
       return despeculateCountZeros(II, TLI, DL, ModifiedDT);

llvm/lib/CodeGen/SelectionDAG/FastISel.cpp

case Intrinsic::objectsize: {
     Constant *ResCI = ConstantInt::get(II->getType(), Res);
     unsigned ResultReg = getRegForValue(ResCI);
     if (!ResultReg)
       return false;
     updateValueMap(II, ResultReg);
     return true;
   }
   case Intrinsic::launder_invariant_group:
+  case Intrinsic::strip_invariant_group:
   case Intrinsic::expect: {
     unsigned ResultReg = getRegForValue(II->getArgOperand(0));
     if (!ResultReg)
       return false;
     updateValueMap(II, ResultReg);
     return true;
   }
   case Intrinsic::experimental_stackmap:

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

else
       Res = DAG.getConstant(0, sdl, Ty);
 
     setValue(&I, Res);
     return nullptr;
   }
   case Intrinsic::annotation:
   case Intrinsic::ptr_annotation:
   case Intrinsic::launder_invariant_group:
+  case Intrinsic::strip_invariant_group:
     // Drop the intrinsic, but forward the value
     setValue(&I, getValue(I.getOperand(0)));
     return nullptr;
   case Intrinsic::assume:
   case Intrinsic::var_annotation:
   case Intrinsic::sideeffect:
     // Discard annotate attributes, assumptions, and artificial side-effects.
     return nullptr;

llvm/lib/IR/Value.cpp

if (auto *GEP = dyn_cast<GEPOperator>(V)) {
         if (const Value *RV = CS.getReturnedArgOperand()) {
           V = RV;
           continue;
         }
         // The result of launder.invariant.group must alias it's argument,
         // but it can't be marked with returned attribute, that's why it needs
         // special case.
         if (StripKind == PSK_ZeroIndicesAndAliasesAndInvariantGroups &&
-            CS.getIntrinsicID() == Intrinsic::launder_invariant_group) {
+            (CS.getIntrinsicID() == Intrinsic::launder_invariant_group ||
+             CS.getIntrinsicID() == Intrinsic::strip_invariant_group)) {
           V = CS.getArgOperand(0);
           continue;
         }
       }
       return V;
     }
     assert(V->getType()->isPointerTy() && "Unexpected operand type!");
   } while (Visited.insert(V).second);

llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp

static bool isCallPromotable(CallInst *CI) {
   case Intrinsic::memcpy:
   case Intrinsic::memmove:
   case Intrinsic::memset:
   case Intrinsic::lifetime_start:
   case Intrinsic::lifetime_end:
   case Intrinsic::invariant_start:
   case Intrinsic::invariant_end:
   case Intrinsic::launder_invariant_group:
+  case Intrinsic::strip_invariant_group:
   case Intrinsic::objectsize:
     return true;
   default:
     return false;
   }
 }
 
 bool AMDGPUPromoteAlloca::binaryOpIsDerivedFromSameAlloca(Value *BaseAlloca,
case Intrinsic::memset: {
                            MemSet->getLength(), MemSet->getDestAlignment(),
                            MemSet->isVolatile());
       Intr->eraseFromParent();
       continue;
     }
     case Intrinsic::invariant_start:
     case Intrinsic::invariant_end:
     case Intrinsic::launder_invariant_group:
+    case Intrinsic::strip_invariant_group:
       Intr->eraseFromParent();
       // FIXME: I think the invariant marker should still theoretically apply,
       // but the intrinsics need to be changed to accept pointers with any
       // address space.
       continue;
     case Intrinsic::objectsize: {
       Value *Src = Intr->getOperand(0);
       Type *SrcTy = Src->getType()->getPointerElementType();

llvm/test/Analysis/ValueTracking/invariant.group.ll

 ; RUN: opt -S -instsimplify -instcombine < %s | FileCheck %s
 
-; CHECK-LABEL: define void @checkNonnull()
-define void @checkNonnull() {
+; CHECK-LABEL: define void @checkNonnullLaunder()
+define void @checkNonnullLaunder() {
 ; CHECK:   %p = call i8* @llvm.launder.invariant.group.p0i8(i8* nonnull %0)
 ; CHECK:   %p2 = call i8* @llvm.launder.invariant.group.p0i8(i8* nonnull %p)
 ; CHECK:   call void @use(i8* nonnull %p2)
 entry:
   %0 = alloca i8, align 8
 
   %p = call i8* @llvm.launder.invariant.group.p0i8(i8* %0)
   %p2 = call i8* @llvm.launder.invariant.group.p0i8(i8* %p)
   call void @use(i8* %p2)
 
   ret void
 }
 
+; CHECK-LABEL: define void @checkNonnullStrip()
+define void @checkNonnullStrip() {
+; CHECK:   %p = call i8* @llvm.strip.invariant.group.p0i8(i8* nonnull %0)
+; CHECK:   %p2 = call i8* @llvm.strip.invariant.group.p0i8(i8* nonnull %p)
+; CHECK:   call void @use(i8* nonnull %p2)
+entry:
+  %0 = alloca i8, align 8
+
+  %p = call i8* @llvm.strip.invariant.group.p0i8(i8* %0)
+  %p2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %p)
+  call void @use(i8* %p2)
+
+  ret void
+}
+
 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
+
 declare void @use(i8*)

llvm/test/CodeGen/Generic/intrinsics.ll

define double @test_cos(float %F) {
         %G = call float @cosf( float %F )               ; <float> [#uses=1]
         %H = fpext float %G to double           ; <double> [#uses=1]
         %I = call double @cos( double %H )              ; <double> [#uses=1]
         ret double %I
 }
 
 declare i8* @llvm.launder.invariant.group(i8*)
 
-define i8* @barrier(i8* %p) {
+define i8* @launder(i8* %p) {
         %q = call i8* @llvm.launder.invariant.group(i8* %p)
         ret i8* %q
 }
 
+declare i8* @llvm.strip.invariant.group(i8*)
+
+define i8* @strip(i8* %p) {
+        %q = call i8* @llvm.strip.invariant.group(i8* %p)
+        ret i8* %q
+}
+
+
 ; sideeffect
 
 declare void @llvm.sideeffect()
 
 define void @test_sideeffect() {
     call void @llvm.sideeffect()
     ret void
 }

llvm/test/Other/invariant.group.ll

; CHECK: load i8, i8* %ptr3, {{.*}}!invariant.group
     ret i8 %v
 }
 
 ; This test check if optimizer is not proving equality based on mustalias
 ; CHECK-LABEL: define void @dontProveEquality(i8* %a)
 define void @dontProveEquality(i8* %a) {
   %b = call i8* @llvm.launder.invariant.group.p0i8(i8* %a)
   %r = icmp eq i8* %b, %a
-;CHECK: call void @useBool(i1 %r)
+; CHECK: call void @useBool(i1 %r)
   call void @useBool(i1 %r)
+
+  %b2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+  %r2 = icmp eq i8* %b2, %a
+; CHECK: call void @useBool(i1 %r2)
+  call void @useBool(i1 %r2)
+
   ret void
 }
 
 declare void @use(i8* readonly)
 declare void @useBool(i1)
 
 declare void @clobber(i8*)
 ; CHECK: Function Attrs: inaccessiblememonly nounwind speculatable{{$}}
 ; CHECK-NEXT: declare i8* @llvm.launder.invariant.group.p0i8(i8*)
 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
 
-!0 = !{}
+; CHECK: Function Attrs: nounwind readnone speculatable{{$}}
+; CHECK-NEXT: declare i8* @llvm.strip.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
+
 
+!0 = !{}
 No newline at end of file

llvm/test/Transforms/CodeGenPrepare/invariant.group.ll

 ; RUN: opt -codegenprepare -S < %s | FileCheck %s
 
 @tmp = global i8 0
 
 ; CHECK-LABEL: define void @foo() {
 define void @foo() {
 enter:
   ; CHECK-NOT: !invariant.group
   ; CHECK-NOT: @llvm.launder.invariant.group.p0i8(
-  ; CHECK: %val = load i8, i8* @tmp, !tbaa
-  %val = load i8, i8* @tmp, !invariant.group !0, !tbaa !{!1, !1, i64 0}
+  ; CHECK: %val = load i8, i8* @tmp{{$}}
+  %val = load i8, i8* @tmp, !invariant.group !0
   %ptr = call i8* @llvm.launder.invariant.group.p0i8(i8* @tmp)
   
   ; CHECK: store i8 42, i8* @tmp{{$}}
   store i8 42, i8* %ptr, !invariant.group !0
   
   ret void
 }
 ; CHECK-LABEL: }
 
-declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+; CHECK-LABEL: define void @foo2() {
+define void @foo2() {
+enter:
+  ; CHECK-NOT: !invariant.group
+  ; CHECK-NOT: @llvm.strip.invariant.group.p0i8(
+  ; CHECK: %val = load i8, i8* @tmp{{$}}
+  %val = load i8, i8* @tmp, !invariant.group !0
+  %ptr = call i8* @llvm.strip.invariant.group.p0i8(i8* @tmp)
+
+  ; CHECK: store i8 42, i8* @tmp{{$}}
+  store i8 42, i8* %ptr, !invariant.group !0
 
+  ret void
+}
+; CHECK-LABEL: }
+
+
+declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
 !0 = !{}
-!1 = !{!"x", !0}

llvm/test/Transforms/DeadStoreElimination/launder.invariant.group.ll

; CHECK-NOT: store i8 43
   %ptr3 = call i8* @llvm.launder.invariant.group.p0i8(i8* %ptr2)
   %ptr4 = call i8* @llvm.launder.invariant.group.p0i8(i8* %ptr3)
 
 ; CHECK: store i8 44
   store i8 44, i8* %ptr4
   ret void
 }
 
+; CHECK-LABEL: void @skip3Barriers(i8* %ptr)
+define void @skip3Barriers(i8* %ptr) {
+; CHECK-NOT: store i8 42
+  store i8 42, i8* %ptr
+; CHECK: %ptr2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr)
+  %ptr2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr)
+; CHECK-NOT: store i8 43
+  store i8 43, i8* %ptr2
+  %ptr3 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr2)
+  %ptr4 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr3)
+
+; CHECK: store i8 44
+  store i8 44, i8* %ptr4
+  ret void
+}
+
+; CHECK-LABEL: void @skip4Barriers(i8* %ptr)
+define void @skip4Barriers(i8* %ptr) {
+; CHECK-NOT: store i8 42
+  store i8 42, i8* %ptr
+; CHECK: %ptr2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr)
+  %ptr2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr)
+; CHECK-NOT: store i8 43
+  store i8 43, i8* %ptr2
+  %ptr3 = call i8* @llvm.launder.invariant.group.p0i8(i8* %ptr2)
+  %ptr4 = call i8* @llvm.strip.invariant.group.p0i8(i8* %ptr3)
+  %ptr5 = call i8* @llvm.launder.invariant.group.p0i8(i8* %ptr3)
+
+; CHECK: store i8 44
+  store i8 44, i8* %ptr5
+  ret void
+}
+
+
 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
 No newline at end of file

llvm/test/Transforms/FunctionAttrs/nocapture.ll

 @g2 = global i8* null
 ; CHECK: define void @captureLaunder(i8* %p)
 define void @captureLaunder(i8* %p) {
   %b = call i8* @llvm.launder.invariant.group.p0i8(i8* %p)
   store i8* %b, i8** @g2
   ret void
 }
 
+; CHECK: @nocaptureStrip(i8* nocapture %p)
+define void @nocaptureStrip(i8* %p) {
+entry:
+  %b = call i8* @llvm.strip.invariant.group.p0i8(i8* %p)
+  store i8 42, i8* %b
+  ret void
+}
+
+@g3 = global i8* null
+; CHECK: define void @captureStrip(i8* %p)
+define void @captureStrip(i8* %p) {
+  %b = call i8* @llvm.strip.invariant.group.p0i8(i8* %p)
+  store i8* %b, i8** @g3
+  ret void
+}
+
 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)

llvm/test/Transforms/GVN/invariant.group.ll

entry:
     
     call void @foo(i8* %ptr)
     %b = load i8, i8* %ptr, !invariant.group !0
     
 ; CHECK: ret i8 42
     ret i8 %b
 }
 
+; CHECK-LABEL: define i1 @proveEqualityForStrip(
+define i1 @proveEqualityForStrip(i8* %a) {
+; FIXME: The first call could be also removed by GVN. Right now
+; DCE removes it. The second call is CSE'd with the first one.
+; CHECK: %b1 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+  %b1 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+; CHECK-NOT: llvm.strip.invariant.group
+  %b2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+  %r = icmp eq i8* %b1, %b2
+; CHECK: ret i1 true
+  ret i1 %r
+}
 ; CHECK-LABEL: define i8 @unoptimizable1() {
 define i8 @unoptimizable1() {
 entry:
     %ptr = alloca i8
     store i8 42, i8* %ptr
     call void @foo(i8* %ptr)
     %a = load i8, i8* %ptr, !invariant.group !0
 ; CHECK: ret i8 %a

 declare void @foo2(i8*, i8)
 declare void @bar(i8)
 declare i8* @getPointer(i8*)
 declare void @_ZN1A3fooEv(%struct.A*)
 declare void @_ZN1AC1Ev(%struct.A*)
 declare void @fooBit(i1*, i1)
 
 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
+
 
-; Function Attrs: nounwind
-declare void @llvm.assume(i1 %cmp.vtables) #0
+declare void @llvm.assume(i1 %cmp.vtables)
 
 
-attributes #0 = { nounwind }
 !0 = !{}
 No newline at end of file

llvm/test/Transforms/GlobalOpt/invariant.group.ll

enter:
   call void @_optimizable()
   call void @_not_optimizable()
   ret void
 }
 
 define void @_optimizable() {
 enter:
   %valptr = alloca i32
-  
   %val = call i32 @TheAnswerToLifeTheUniverseAndEverything()
   store i32 %val, i32* @tmp
   store i32 %val, i32* %valptr
-  
   %0 = bitcast i32* %valptr to i8*
   %barr = call i8* @llvm.launder.invariant.group(i8* %0)
   %1 = bitcast i8* %barr to i32*
-  
   %val2 = load i32, i32* %1
   store i32 %val2, i32* @tmp2
   ret void
 }
 
 ; We can't step through launder.invariant.group here, because that would change
 ; this load in @usage_of_globals()
-; val = load i32, i32* %ptrVal, !invariant.group !0 
-; into 
+; val = load i32, i32* %ptrVal, !invariant.group !0
+; into
 ; %val = load i32, i32* @tmp3, !invariant.group !0
-; and then we could assume that %val and %val2 to be the same, which coud be 
+; and then we could assume that %val and %val2 to be the same, which coud be
 ; false, because @changeTmp3ValAndCallBarrierInside() may change the value
 ; of @tmp3.
 define void @_not_optimizable() {
 enter:
   store i32 13, i32* @tmp3, !invariant.group !0
-  
   %0 = bitcast i32* @tmp3 to i8*
   %barr = call i8* @llvm.launder.invariant.group(i8* %0)
   %1 = bitcast i8* %barr to i32*
-  
   store i32* %1, i32** @ptrToTmp3
   store i32 42, i32* %1, !invariant.group !0
-  
   ret void
 }
 define void @usage_of_globals() {
 entry:
   %ptrVal = load i32*, i32** @ptrToTmp3
   %val = load i32, i32* %ptrVal, !invariant.group !0
-  
   call void @changeTmp3ValAndCallBarrierInside()
   %val2 = load i32, i32* @tmp3, !invariant.group !0
   ret void;
 }
 
 declare void @changeTmp3ValAndCallBarrierInside()
 
 declare i8* @llvm.launder.invariant.group(i8*)
 
 !0 = !{}

llvm/test/Transforms/InstCombine/invariant.group.ll

 ; RUN: opt -instcombine -S < %s | FileCheck %s
 
+
 ; CHECK-LABEL: define i8* @simplifyNullLaunder()
 define i8* @simplifyNullLaunder() {
 ; CHECK-NEXT: ret i8* null
   %b2 = call i8* @llvm.launder.invariant.group.p0i8(i8* null)
   ret i8* %b2
 }
 
 ; CHECK-LABEL: define i8 addrspace(42)* @dontsimplifyNullLaunderForDifferentAddrspace()

 
 ; CHECK-LABEL: define i8 addrspace(42)* @simplifyUndefLaunder2()
 define i8 addrspace(42)* @simplifyUndefLaunder2() {
 ; CHECK-NEXT: ret i8 addrspace(42)* undef
   %b2 = call i8 addrspace(42)* @llvm.launder.invariant.group.p42i8(i8 addrspace(42)* undef)
   ret i8 addrspace(42)* %b2
 }
 
-
 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
 declare i8 addrspace(42)* @llvm.launder.invariant.group.p42i8(i8 addrspace(42)*)
+
+
+; CHECK-LABEL: define i8* @simplifyNullStrip()
+define i8* @simplifyNullStrip() {
+; CHECK-NEXT: ret i8* null
+  %b2 = call i8* @llvm.strip.invariant.group.p0i8(i8* null)
+  ret i8* %b2
+}
+
+; CHECK-LABEL: define i8 addrspace(42)* @dontsimplifyNullStripForDifferentAddrspace()
+define i8 addrspace(42)* @dontsimplifyNullStripForDifferentAddrspace() {
+; CHECK: %b2 = call i8 addrspace(42)* @llvm.strip.invariant.group.p42i8(i8 addrspace(42)* null)
+; CHECK: ret i8 addrspace(42)* %b2
+  %b2 = call i8 addrspace(42)* @llvm.strip.invariant.group.p42i8(i8 addrspace(42)* null)
+  ret i8 addrspace(42)* %b2
+}
+
+; CHECK-LABEL: define i8* @simplifyUndefStrip()
+define i8* @simplifyUndefStrip() {
+; CHECK-NEXT: ret i8* undef
+  %b2 = call i8* @llvm.strip.invariant.group.p0i8(i8* undef)
+  ret i8* %b2
+}
+
+; CHECK-LABEL: define i8 addrspace(42)* @simplifyUndefStrip2()
+define i8 addrspace(42)* @simplifyUndefStrip2() {
+; CHECK-NEXT: ret i8 addrspace(42)* undef
+  %b2 = call i8 addrspace(42)* @llvm.strip.invariant.group.p42i8(i8 addrspace(42)* undef)
+  ret i8 addrspace(42)* %b2
+}
+
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
+declare i8 addrspace(42)* @llvm.strip.invariant.group.p42i8(i8 addrspace(42)*)
+

llvm/test/Transforms/NewGVN/invariant.group.ll

entry:
     
     call void @foo(i8* %ptr)
     %b = load i8, i8* %ptr, !invariant.group !0
     
 ; CHECK: ret i8 42
     ret i8 %b
 }
 
+; CHECK-LABEL: define i1 @proveEqualityForStrip(
+define i1 @proveEqualityForStrip(i8* %a) {
+; FIXME: The first call could be also removed by GVN. Right now
+; DCE removes it. The second call is CSE'd with the first one.
+; CHECK: %b1 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+  %b1 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+; CHECK-NOT: llvm.strip.invariant.group
+  %b2 = call i8* @llvm.strip.invariant.group.p0i8(i8* %a)
+  %r = icmp eq i8* %b1, %b2
+; CHECK: ret i1 true
+  ret i1 %r
+}
+
 ; CHECK-LABEL: define i8 @unoptimizable1() {
 define i8 @unoptimizable1() {
 entry:
     %ptr = alloca i8
     store i8 42, i8* %ptr
     call void @foo(i8* %ptr)
     %a = load i8, i8* %ptr, !invariant.group !0
 ; CHECK: ret i8 %a

 declare i8* @llvm.launder.invariant.group.p0i8(i8*)
 
 ; Function Attrs: nounwind
 declare void @llvm.assume(i1 %cmp.vtables) #0
 
 
 attributes #0 = { nounwind }
 !0 = !{}
 No newline at end of file