CodeGen: Fix address space of global variable

[yaxunl]

Summary

Certain targets (e.g. amdgcn) require global variable to stay in global or constant address
space. In C or C++ global variables are emitted in the default (generic) address space.
This patch introduces virtual functions TargetCodeGenInfo::getGlobalVarAddressSpace
and TargetInfo::getConstantAddressSpace to handle this in a general approach.

It only affects IR generated for amdgcn target.

[rjmccall]

lib/CodeGen/CGExpr.cpp

360

This is not an appropriate place to stick target-specific code. You should add a target hook that returns the address space into which to allocate constant temporaries, with the requirement that the target must not return an address space which cannot be lifted into LangAS::Default. For some configurations this may not be possible (e.g. OpenCL C++?), so the hook should return an Optional<unsigned> so that a None return will disable the optimization.

And you should go ahead and lift the pointer here if necessary.

You may have to make createReferenceTemporary return whether it constant-emitted the temporary initializer.

[yaxunl]

Add TargetInfo::getTargetConstantAddressSpace.

[rjmccall]

include/clang/Basic/TargetInfo.h

959 ↗

(On Diff #102204)

"Return an AST address space which can be used opportunistically for constant global memory. It must be possible to convert pointers into this address space to LangAS::Default. If no such address space exists, this may return None, and such optimizations will be disabled."

lib/CodeGen/CGExpr.cpp

357

This isn't the right logic. Part of what I was saying before is that this optimization isn't obviously supportable on all targets. You should call getTargetConstantAddressSpace() and then only try to do this optimization if it doesn't return None.

For parity, the default implementation should return 0.

452

Every time you're tempted to use getPointerCast, you should instead be using the target hook to lift the pointer into the right address space.

[yaxunl]

include/clang/Basic/TargetInfo.h

959 ↗

(On Diff #102204)

I will change it. Also I think getConstantAddressSpace may be a better name since we will return AST addr space.

lib/CodeGen/CGExpr.cpp

452

I am going to fix it. However this seems not the right place to do the addr space cast. I am going to do it in createReferenceTemporary instead.

[yaxunl]

Revised by John's comments.

[rjmccall]

include/clang/Basic/TargetInfo.h

969 ↗

(On Diff #103613)

I'm not sure this really needs to exist.

959 ↗

(On Diff #102204)

Sorry, I typo'ed my original suggestion. "pointers in this address space".

lib/CodeGen/CGBlocks.cpp

1144 ↗

(On Diff #103613)

You used VoidPtrTy above.

lib/CodeGen/CGExpr.cpp

344

Why are you passing the TargetCodeGenInfo down separately when it can be easily reacquired from the CodeGenFunction?

Oh, it looks like getTargetHooks() is private; there's no reason for that, just make it public.

383

GV->getValueType()->getPointerTo(...)

lib/CodeGen/CodeGenModule.cpp

2549

This does not seem like a very useful assertion.

2554

I would expect this to be the general rule for all language modes. Why is OpenCL an exception?

It looks to me like GetGlobalVarAddressSpace is only ever called with a possibly-null D by GetOrCreateLLVMGlobal, which is only called with a null D by CreateRuntimeVariable. I expect that every call to CreateRuntimeVariable will probably need to be audited if it's going to suddenly start returning pointers that aren't in the default address space, and we'll probably end up needing some very different behavior at that point. So for now, let's just move the LanguageExpected case out to the one call site that uses it.

I do like the simplification of just taking the declaration instead of taking that and an address space.

2571

It's impossible for K to be LanguageExpected here, you already checked it above.

2578

Okay, I think I see what you're trying to do here: when you compile normal (i.e. non-OpenCL) code for your target, you want to implicitly change where global variables are allocated by default, then translate the pointer into LangAS::Default. It's essentially the global-variable equivalent of what we're already doing with stack pointers.

The right way to do this is to make a hook on the TargetCodeGenInfo like getASTAllocaAddressSpace() which allows the target to override the address space for a global on a case-by-case basis. You can then do whatever target-specific logic you need there, and you won't need to introduce getGlobalAddressSpace() on TargetInfo. The default implementation, of course, will just return D->getType().getAddressSpace().

lib/CodeGen/TargetInfo.cpp

7418

You have a couple things in this patch that just touch your target and are basically unrelated to the rest of your logic. Please split those into a separate patch.

[yaxunl]

include/clang/Basic/TargetInfo.h

969 ↗

(On Diff #103613)

Removed.

lib/CodeGen/CGBlocks.cpp

1144 ↗

(On Diff #103613)

They are different fields with different types.

lib/CodeGen/CodeGenModule.cpp

2549

This checks global variable has valid address space. If a global variable having private or generic address space is passed to backend, it may cause issues difficult to debug. This assertion helps to detect the issue early.

2554

For OpenCL, TargetFavored and LanguageExpected are the same, i.e. global. For C++, TargetFavored is globa, but LanguageExpected is Default.

Done.

2578

Done.

I am wondering if I should just move the whole function to TargetCodeGenInfo::getGlobalVarAddressSpace, which seems cleaner.

lib/CodeGen/TargetInfo.cpp

7418

Done.

[yaxunl]

Revised by John's comments.

[rjmccall]

Looks great, thanks! Just a few minor changes.

lib/CodeGen/CGBlocks.cpp

1144 ↗

(On Diff #103613)

They're both the isa field of a block object. The difference is just between allocating the block globally and allocating it on the stack.

lib/CodeGen/TargetInfo.h

237 ↗

(On Diff #105310)

The CGM is conventionally the first argument.

262 ↗

(On Diff #105310)

This should probably take the CGM for completeness.

[yaxunl]

lib/CodeGen/CGBlocks.cpp

1144 ↗

(On Diff #103613)

You are right, actually they should both be VoidPtrPtrTy to match the type of getNSConcreteGlobalBlock() and getNSConcreteStackBlock().

Fixed.

lib/CodeGen/TargetInfo.h

237 ↗

(On Diff #105310)

done

262 ↗

(On Diff #105310)

done.

[yaxunl]

Revised by John's comments.

[rjmccall]

Great, thanks! LGTM.

[mehdi_amini]

cfe/trunk/lib/CodeGen/CodeGenModule.cpp

3759 ↗

(On Diff #105770)

Coverity reports:

>>>     CID 1377397:  Null pointer dereferences  (REVERSE_INULL)
>>>     Null-checking "VD" suggests that it may be null, but it has already been dereferenced on all paths leading to the check.

Can you look into this?

lib/Basic/Targets.cpp

     auto IsGenericZero = isGenericZero(Triple);
     resetDataLayout(getTriple().getArch() == llvm::Triple::amdgcn ?
                     (IsGenericZero ? DataLayoutStringSIGenericIsZero :
                         DataLayoutStringSIPrivateIsZero)
                     : DataLayoutStringR600);
     assert(DataLayout->getAllocaAddrSpace() == AS.Private);
 
     UseAddrSpaceMapMangling = true;
+
+    if (getMaxPointerWidth() == 64) {
+      LongWidth = LongAlign = 64;
+      SizeType = UnsignedLong;
+      PtrDiffType = SignedLong;
+      IntPtrType = SignedLong;
+    }
   }
 
   void adjust(LangOptions &Opts) override {
     TargetInfo::adjust(Opts);
     if (isGenericZero(getTriple())) {
       AddrSpaceMap = Opts.OpenCL ? &AMDGPUOpenCLGenericIsZeroMap
                                  : &AMDGPUNonOpenCLGenericIsZeroMap;
     } else {
       AddrSpaceMap = Opts.OpenCL ? &AMDGPUOpenCLPrivateIsZeroMap
                                  : &AMDGPUNonOpenCLPrivateIsZeroMap;
     }
+    // Set default pointer width and alignment.
+    PointerWidth = getPointerWidthV(Opts.OpenCL ? AS.Private : AS.Generic);
+    PointerAlign = PointerWidth;
   }
 
   uint64_t getPointerWidthV(unsigned AddrSpace) const override {
     if (GPU <= GK_CAYMAN)
       return 32;
 
     if (AddrSpace == AS.Private || AddrSpace == AS.Local) {
       return 32;

lib/CodeGen/CGExpr.cpp

pushTemporaryCleanup(CodeGenFunction &CGF, const MaterializeTemporaryExpr *M,
   case SD_Dynamic:
     llvm_unreachable("temporary cannot have dynamic storage duration");
   }
 }
 
 static Address
 createReferenceTemporary(CodeGenFunction &CGF,
                          const MaterializeTemporaryExpr *M, const Expr *Inner) {
   switch (M->getStorageDuration()) {
   case SD_FullExpression:
   case SD_Automatic: {
     // If we have a constant temporary array or record try to promote it into a
     // constant global under the same rules a normal constant would've been
     // promoted. This is easier on the optimizer and generally emits fewer
     // instructions.
     QualType Ty = Inner->getType();
     if (CGF.CGM.getCodeGenOpts().MergeAllConstants &&
         (Ty->isArrayType() || Ty->isRecordType()) &&
         CGF.CGM.isTypeConstant(Ty, true))
       if (llvm::Constant *Init = CGF.CGM.EmitConstantExpr(Inner, Ty, &CGF)) {
+        unsigned AddrSpace = 0;
+        if (CGF.CGM.getTriple().getArch() == llvm::Triple::amdgcn) {
+          AddrSpace =
+              CGF.getContext().getTargetAddressSpace(LangAS::opencl_constant);
+        }
         auto *GV = new llvm::GlobalVariable(
             CGF.CGM.getModule(), Init->getType(), /*isConstant=*/true,
-            llvm::GlobalValue::PrivateLinkage, Init, ".ref.tmp");
+            llvm::GlobalValue::PrivateLinkage, Init, ".ref.tmp", nullptr,
+            llvm::GlobalValue::NotThreadLocal, AddrSpace);
         CharUnits alignment = CGF.getContext().getTypeAlignInChars(Ty);
         GV->setAlignment(alignment.getQuantity());
         // FIXME: Should we put the new global into a COMDAT?
         return Address(GV, alignment);
       }
     return CGF.CreateMemTemp(Ty, "ref.tmp");
   }
   case SD_Thread:
   case SD_Static:
     return CGF.CGM.GetAddrOfGlobalTemporary(M, Inner);
 
   case SD_Dynamic:
     llvm_unreachable("temporary can't have dynamic storage duration");
   }
   llvm_unreachable("unknown storage duration");
 }
 
 LValue CodeGenFunction::
 EmitMaterializeTemporaryExpr(const MaterializeTemporaryExpr *M) {
   const Expr *E = M->GetTemporaryExpr();
 
     // FIXME: ideally this would use EmitAnyExprToMem, however, we cannot do so
     // as that will cause the lifetime adjustment to be lost for ARC
   auto ownership = M->getType().getObjCLifetime();
   if (ownership != Qualifiers::OCL_None &&
       ownership != Qualifiers::OCL_ExplicitNone) {
     Address Object = createReferenceTemporary(*this, M, E);
if (opaque->getType()->isRecordType()) {
       assert(Adjustments.empty());
       return EmitOpaqueValueLValue(opaque);
     }
   }
 
   // Create and initialize the reference temporary.
   Address Object = createReferenceTemporary(*this, M, E);
   if (auto *Var = dyn_cast<llvm::GlobalVariable>(Object.getPointer())) {
-    Object = Address(llvm::ConstantExpr::getBitCast(
+    Object = Address(llvm::ConstantExpr::getPointerCast(
                          Var, ConvertTypeForMem(E->getType())->getPointerTo()),
                      Object.getAlignment());
     // If the temporary is a global and has a constant initializer or is a
     // constant temporary that we promoted to a global, we may have already
     // initialized it.
     if (!Var->hasInitializer()) {
       Var->setInitializer(CGM.EmitNullConstant(E->getType()));
       EmitAnyExprToMem(E, Object, Qualifiers(), /*IsInit*/true);
     }
   } else {

lib/CodeGen/CodeGenModule.cpp

 
 CharUnits CodeGenModule::GetTargetTypeStoreSize(llvm::Type *Ty) const {
   return Context.toCharUnitsFromBits(
       getDataLayout().getTypeStoreSizeInBits(Ty));
 }
 
 unsigned CodeGenModule::GetGlobalVarAddressSpace(const VarDecl *D,
                                                  unsigned AddrSpace) {
-  if (D && LangOpts.CUDA && LangOpts.CUDAIsDevice) {
-    if (D->hasAttr<CUDAConstantAttr>())
-      AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_constant);
-    else if (D->hasAttr<CUDASharedAttr>())
-      AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_shared);
-    else
-      AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_device);
+  if (D) {
+    if (LangOpts.CUDA && LangOpts.CUDAIsDevice) {
+      if (D->hasAttr<CUDAConstantAttr>())
+        AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_constant);
+      else if (D->hasAttr<CUDASharedAttr>())
+        AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_shared);
+      else
+        AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_device);
+    } else if (getTriple().getArch() == llvm::Triple::amdgcn) {
+      auto LangAddr = D->getType().getAddressSpace();
+      if (LangOpts.OpenCL) {
+        assert(LangAddr == LangAS::opencl_global ||
+               LangAddr == LangAS::opencl_constant ||
+               LangAddr == LangAS::opencl_local ||
+               LangAddr >= LangAS::FirstTargetAddressSpace);
+      } else {
+        assert(LangAddr == LangAS::Default ||
+               LangAddr >= LangAS::FirstTargetAddressSpace);
+      }
+      if (!LangOpts.OpenCL && LangAddr == LangAS::Default) {
+        if (isTypeConstant(D->getType(), false)) {
+          LangAddr = LangAS::opencl_constant;
+        } else {
+          LangAddr = LangAS::opencl_global;
+        }
+        AddrSpace = getContext().getTargetAddressSpace(LangAddr);
+      }
+    }
   }
 
   return AddrSpace;
 }
 
 template<typename SomeDecl>
 void CodeGenModule::MaybeHandleStaticInExternC(const SomeDecl *D,
                                                llvm::GlobalValue *GV) {

   if (!CGM.supportsCOMDAT())
     return false;
 
   if (D.hasAttr<SelectAnyAttr>())
     return true;
 
   GVALinkage Linkage;
   if (auto *VD = dyn_cast<VarDecl>(&D))
     Linkage = CGM.getContext().GetGVALinkageForVariable(VD);
   else
     Linkage = CGM.getContext().GetGVALinkageForFunction(cast<FunctionDecl>(&D));
 
   switch (Linkage) {
   case GVA_Internal:
   case GVA_AvailableExternally:
   case GVA_StrongExternal:
     return false;
   case GVA_DiscardableODR:
   case GVA_StrongODR:
     return true;
   }
   llvm_unreachable("No such linkage");
 }
 
 void CodeGenModule::maybeSetTrivialComdat(const Decl &D,
                                           llvm::GlobalObject &GO) {
   if (!shouldBeInCOMDAT(*this, D))
     return;
   GO.setComdat(TheModule.getOrInsertComdat(GO.getName()));
 }
 
 /// Pass IsTentative as true if you want to create a tentative definition.
 void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D,
                                             bool IsTentative) {
   // OpenCL global variables of sampler type are translated to function calls,
   // therefore no need to be translated.
   QualType ASTTy = D->getType();
   if (getLangOpts().OpenCL && ASTTy->isSamplerT())
     return;
 
   llvm::Constant *Init = nullptr;
   CXXRecordDecl *RD = ASTTy->getBaseElementTypeUnsafe()->getAsCXXRecordDecl();
   bool NeedsGlobalCtor = false;
   bool NeedsGlobalDtor = RD && !RD->hasTrivialDestructor();
 
   const VarDecl *InitDecl;

lib/CodeGen/TargetInfo.cpp

 
   if (const auto *Attr = FD->getAttr<AMDGPUNumVGPRAttr>()) {
     uint32_t NumVGPR = Attr->getNumVGPR();
 
     if (NumVGPR != 0)
       F->addFnAttr("amdgpu-num-vgpr", llvm::utostr(NumVGPR));
   }
 
-  appendOpenCLVersionMD(M);
+  if (M.getLangOpts().OpenCL)
+    appendOpenCLVersionMD(M);
 }
 
 unsigned AMDGPUTargetCodeGenInfo::getOpenCLKernelCallingConv() const {
   return llvm::CallingConv::AMDGPU_KERNEL;
 }
 
 // Currently LLVM assumes null pointers always have value 0,
 // which results in incorrectly transformed IR. Therefore, instead of

 ABIArgInfo
 SparcV8ABIInfo::classifyReturnType(QualType Ty) const {
   if (Ty->isAnyComplexType()) {
     return ABIArgInfo::getDirect();
   }
   else {
     return DefaultABIInfo::classifyReturnType(Ty);
   }
 }
 
 void SparcV8ABIInfo::computeInfo(CGFunctionInfo &FI) const {
 
   FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
   for (auto &Arg : FI.arguments())
     Arg.info = classifyArgumentType(Arg.type);
 }
 

test/CodeGen/address-space.c

 // CHECK: load i32, i32 addrspace(1)*
 // CHECK-NEXT: ret i32
 int test2(int i) { return ban[i]; }
 
 // Both A and B point into addrspace(2).
 __attribute__((address_space(2))) int *A, *B;
 
 // CHECK-LABEL: define void @test3()
-// GIZ: load i32 addrspace(2)*, i32 addrspace(2)** @B
-// PIZ: load i32 addrspace(2)*, i32 addrspace(2)* addrspace(4)* @B
+// X86: load i32 addrspace(2)*, i32 addrspace(2)** @B
+// AMDGIZ: load i32 addrspace(2)*, i32 addrspace(2)** addrspacecast (i32 addrspace(2)* addrspace(1)* @B to i32 addrspace(2)**)
+// PIZ: load i32 addrspace(2)*, i32 addrspace(2)* addrspace(4)* addrspacecast (i32 addrspace(2)* addrspace(1)* @B to i32 addrspace(2)* addrspace(4)*)
 // CHECK: load i32, i32 addrspace(2)*
-// GIZ: load i32 addrspace(2)*, i32 addrspace(2)** @A
-// PIZ: load i32 addrspace(2)*, i32 addrspace(2)* addrspace(4)* @A
+// X86: load i32 addrspace(2)*, i32 addrspace(2)** @A
+// AMDGIZ: load i32 addrspace(2)*, i32 addrspace(2)** addrspacecast (i32 addrspace(2)* addrspace(1)* @A to i32 addrspace(2)**)
+// PIZ: load i32 addrspace(2)*, i32 addrspace(2)* addrspace(4)* addrspacecast (i32 addrspace(2)* addrspace(1)* @A to i32 addrspace(2)* addrspace(4)*)
 // CHECK: store i32 {{.*}}, i32 addrspace(2)*
 void test3() {
   *A = *B;
 }
 
 // PR7437
 typedef struct {
   float aData[1];

test/CodeGen/default-address-space.c

 // RUN: %clang_cc1 -triple amdgcn -emit-llvm < %s | FileCheck -check-prefixes=PIZ,COM %s
 // RUN: %clang_cc1 -triple amdgcn---amdgiz -emit-llvm < %s | FileCheck -check-prefixes=CHECK,COM %s
 
-// PIZ-DAG: @foo = common addrspace(4) global i32 0
-// CHECK-DAG: @foo = common global i32 0
+// PIZ-DAG: @foo = common addrspace(1) global i32 0
+// CHECK-DAG: @foo = common addrspace(1) global i32 0
 int foo;
 
-// PIZ-DAG: @ban = common addrspace(4) global [10 x i32] zeroinitializer
-// CHECK-DAG: @ban = common global [10 x i32] zeroinitializer
+// PIZ-DAG: @ban = common addrspace(1) global [10 x i32] zeroinitializer
+// CHECK-DAG: @ban = common addrspace(1) global [10 x i32] zeroinitializer
 int ban[10];
 
-// PIZ-DAG: @A = common addrspace(4) global i32 addrspace(4)* null
-// PIZ-DAG: @B = common addrspace(4) global i32 addrspace(4)* null
-// CHECK-DAG: @A = common global i32* null
-// CHECK-DAG: @B = common global i32* null
+// PIZ-DAG: @A = common addrspace(1) global i32 addrspace(4)* null
+// PIZ-DAG: @B = common addrspace(1) global i32 addrspace(4)* null
+// CHECK-DAG: @A = common addrspace(1) global i32* null
+// CHECK-DAG: @B = common addrspace(1) global i32* null
 int *A;
 int *B;
 
 // COM-LABEL: define i32 @test1()
-// PIZ: load i32, i32 addrspace(4)* @foo
-// CHECK: load i32, i32* @foo
+// PIZ: load i32, i32 addrspace(4)* addrspacecast{{[^@]+}} @foo
+// CHECK: load i32, i32* addrspacecast{{[^@]+}} @foo
 int test1() { return foo; }
 
 // COM-LABEL: define i32 @test2(i32 %i)
 // COM: %[[addr:.*]] = getelementptr
 // PIZ: load i32, i32 addrspace(4)* %[[addr]]
 // PIZ-NEXT: ret i32
 // CHECK: load i32, i32* %[[addr]]
 // CHECK-NEXT: ret i32
 int test2(int i) { return ban[i]; }
 
 // COM-LABEL: define void @test3()
-// PIZ: load i32 addrspace(4)*, i32 addrspace(4)* addrspace(4)* @B
+// PIZ: load i32 addrspace(4)*, i32 addrspace(4)* addrspace(4)* addrspacecast{{[^@]+}} @B
 // PIZ: load i32, i32 addrspace(4)*
-// PIZ: load i32 addrspace(4)*, i32 addrspace(4)* addrspace(4)* @A
+// PIZ: load i32 addrspace(4)*, i32 addrspace(4)* addrspace(4)* addrspacecast{{[^@]+}} @A
 // PIZ: store i32 {{.*}}, i32 addrspace(4)*
-// CHECK: load i32*, i32** @B
+// CHECK: load i32*, i32** addrspacecast{{.*}} @B
 // CHECK: load i32, i32*
-// CHECK: load i32*, i32** @A
+// CHECK: load i32*, i32** addrspacecast{{.*}} @A
 // CHECK: store i32 {{.*}}, i32*
 void test3() {
   *A = *B;
 }
 
 // PIZ-LABEL: define void @test4(i32 addrspace(4)* %a)
 // PIZ: %[[alloca:.*]] = alloca i32 addrspace(4)*
 // PIZ: %[[a_addr:.*]] = addrspacecast{{.*}} %[[alloca]] to i32 addrspace(4)* addrspace(4)*
 // PIZ: store i32 addrspace(4)* %a, i32 addrspace(4)* addrspace(4)* %[[a_addr]]
 // PIZ: %[[r0:.*]] = load i32 addrspace(4)*, i32 addrspace(4)* addrspace(4)* %[[a_addr]]
 // PIZ: %[[arrayidx:.*]] = getelementptr inbounds i32, i32 addrspace(4)* %[[r0]]
 // PIZ: store i32 0, i32 addrspace(4)* %[[arrayidx]]
 // CHECK-LABEL: define void @test4(i32* %a)
-// CHECK: %[[alloca:.*]] = alloca i32*, align 4, addrspace(5)
+// CHECK: %[[alloca:.*]] = alloca i32*, align 8, addrspace(5)
 // CHECK: %[[a_addr:.*]] = addrspacecast{{.*}} %[[alloca]] to i32**
 // CHECK: store i32* %a, i32** %[[a_addr]]
 // CHECK: %[[r0:.*]] = load i32*, i32** %[[a_addr]]
 // CHECK: %[[arrayidx:.*]] = getelementptr inbounds i32, i32* %[[r0]]
 // CHECK: store i32 0, i32* %[[arrayidx]]
 void test4(int *a) {
   a[0] = 0;
 }

test/CodeGenCXX/amdgcn-automatic-variable.cpp

void func1(int *x) {
   *x = 1;
 }
 
 // CHECK-LABEL: define void @_Z5func2v()
 void func2(void) {
   // CHECK: %lv1 = alloca i32, align 4, addrspace(5)
   // CHECK: %lv2 = alloca i32, align 4, addrspace(5)
   // CHECK: %la = alloca [100 x i32], align 4, addrspace(5)
-  // CHECK: %lp1 = alloca i32*, align 4, addrspace(5)
-  // CHECK: %lp2 = alloca i32*, align 4, addrspace(5)
+  // CHECK: %lp1 = alloca i32*, align 8, addrspace(5)
+  // CHECK: %lp2 = alloca i32*, align 8, addrspace(5)
   // CHECK: %lvc = alloca i32, align 4, addrspace(5)
 
   // CHECK: %[[r0:.*]] = addrspacecast i32 addrspace(5)* %lv1 to i32*
   // CHECK: store i32 1, i32* %[[r0]]
   int lv1;
   lv1 = 1;
   // CHECK: %[[r1:.*]] = addrspacecast i32 addrspace(5)* %lv2 to i32*
   // CHECK: store i32 2, i32* %[[r1]]
   int lv2 = 2;
 
   // CHECK: %[[r2:.*]] = addrspacecast [100 x i32] addrspace(5)* %la to [100 x i32]*
   // CHECK: %[[arrayidx:.*]] = getelementptr inbounds [100 x i32], [100 x i32]* %[[r2]], i64 0, i64 0
   // CHECK: store i32 3, i32* %[[arrayidx]], align 4
   int la[100];
   la[0] = 3;
 
   // CHECK: %[[r3:.*]] = addrspacecast i32* addrspace(5)* %lp1 to i32**
-  // CHECK: store i32* %[[r0]], i32** %[[r3]], align 4
+  // CHECK: store i32* %[[r0]], i32** %[[r3]], align 8
   int *lp1 = &lv1;
 
   // CHECK: %[[r4:.*]] = addrspacecast i32* addrspace(5)* %lp2 to i32**
   // CHECK: %[[arraydecay:.*]] = getelementptr inbounds [100 x i32], [100 x i32]* %[[r2]], i32 0, i32 0
-  // CHECK: store i32* %[[arraydecay]], i32** %[[r4]], align 4
+  // CHECK: store i32* %[[arraydecay]], i32** %[[r4]], align 8
   int *lp2 = la;
 
   // CHECK: call void @_Z5func1Pi(i32* %[[r0]])
   func1(&lv1);
 
   // CHECK: %[[r5:.*]] = addrspacecast i32 addrspace(5)* %lvc to i32*
   // CHECK: store i32 4, i32* %[[r5]]
   // CHECK: store i32 4, i32* %[[r0]]

 // CHECK-LABEL: define void @_Z5func4i
 void func4(int x) {
   // CHECK: %[[x_addr:.*]] = alloca i32, align 4, addrspace(5)
   // CHECK: %[[r0:.*]] = addrspacecast i32 addrspace(5)* %[[x_addr]] to i32*
   // CHECK: store i32 %x, i32* %[[r0]], align 4
   // CHECK: call void @_Z5func1Pi(i32* %[[r0]])
   func1(&x);
 }
+
+// CHECK-NOT: !opencl.ocl.version 

test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

-// RUN: %clang_cc1 -std=c++11 -triple x86_64-none-linux-gnu -emit-llvm -o - %s | FileCheck %s
+// RUN1: %clang_cc1 -std=c++11 -triple x86_64-none-linux-gnu -emit-llvm -o - %s | FileCheck -check-prefixes=X86,CHECK %s
+// RUN: %clang_cc1 -std=c++11 -triple amdgcn-amd-amdhsa-amdgiz -DNO_TLS -emit-llvm -o - %s | FileCheck -check-prefixes=AMD,CHECK %s
 
 namespace std {
   typedef decltype(sizeof(int)) size_t;
 
   // libc++'s implementation
   template <class _E>
   class initializer_list
   {

 struct witharg1 {
   witharg1(const destroyme1&);
   ~witharg1();
 };
 struct wantslist1 {
   wantslist1(std::initializer_list<destroyme1>);
   ~wantslist1();
 };
-
-// CHECK: @_ZGR15globalInitList1_ = internal constant [3 x i32] [i32 1, i32 2, i32 3]
-// CHECK: @globalInitList1 = global %{{[^ ]+}} { i32* getelementptr inbounds ([3 x i32], [3 x i32]* @_ZGR15globalInitList1_, i32 0, i32 0), i{{32|64}} 3 }
+// X86: @_ZGR15globalInitList1_ = internal constant [3 x i32] [i32 1, i32 2, i32 3]
+// X86: @globalInitList1 = global %{{[^ ]+}} { i32* getelementptr inbounds ([3 x i32], [3 x i32]* @_ZGR15globalInitList1_, i32 0, i32 0), i{{32|64}} 3 }
+// AMD: @_ZGR15globalInitList1_ = internal addrspace(1) constant [3 x i32] [i32 1, i32 2, i32 3]
+// AMD: @globalInitList1 = addrspace(1) global %{{[^ ]+}} { i32* addrspacecast (i32 addrspace(1)* getelementptr inbounds ([3 x i32], [3 x i32] addrspace(1)* @_ZGR15globalInitList1_, i32 0, i32 0) to i32*), i{{32|64}} 3 }
 std::initializer_list<int> globalInitList1 = {1, 2, 3};
 
+#ifndef NO_TLS
 namespace thread_local_global_array {
   // FIXME: We should be able to constant-evaluate this even though the
   // initializer is not a constant expression (pointers to thread_local
   // objects aren't really a problem).
   //
-  // CHECK: @_ZN25thread_local_global_array1xE = thread_local global
-  // CHECK: @_ZGRN25thread_local_global_array1xE_ = internal thread_local constant [4 x i32] [i32 1, i32 2, i32 3, i32 4]
+  // X86: @_ZN25thread_local_global_array1xE = thread_local global
+  // X86: @_ZGRN25thread_local_global_array1xE_ = internal thread_local constant [4 x i32] [i32 1, i32 2, i32 3, i32 4]
   std::initializer_list<int> thread_local x = { 1, 2, 3, 4 };
 }
+#endif
 
-// CHECK: @globalInitList2 = global %{{[^ ]+}} zeroinitializer
-// CHECK: @_ZGR15globalInitList2_ = internal global [2 x %[[WITHARG:[^ ]*]]] zeroinitializer
-
-// CHECK: @_ZN15partly_constant1kE = global i32 0, align 4
-// CHECK: @_ZN15partly_constant2ilE = global {{.*}} null, align 8
-// CHECK: @[[PARTLY_CONSTANT_OUTER:_ZGRN15partly_constant2ilE.*]] = internal global {{.*}} zeroinitializer, align 8
-// CHECK: @[[PARTLY_CONSTANT_INNER:_ZGRN15partly_constant2ilE.*]] = internal global [3 x {{.*}}] zeroinitializer, align 8
-// CHECK: @[[PARTLY_CONSTANT_FIRST:_ZGRN15partly_constant2ilE.*]] = internal constant [3 x i32] [i32 1, i32 2, i32 3], align 4
-// CHECK: @[[PARTLY_CONSTANT_SECOND:_ZGRN15partly_constant2ilE.*]] = internal global [2 x i32] zeroinitializer, align 4
-// CHECK: @[[PARTLY_CONSTANT_THIRD:_ZGRN15partly_constant2ilE.*]] = internal constant [4 x i32] [i32 5, i32 6, i32 7, i32 8], align 4
-
-// CHECK: @[[REFTMP1:.*]] = private constant [2 x i32] [i32 42, i32 43], align 4
-// CHECK: @[[REFTMP2:.*]] = private constant [3 x %{{.*}}] [%{{.*}} { i32 1 }, %{{.*}} { i32 2 }, %{{.*}} { i32 3 }], align 4
+// X86: @globalInitList2 = global %{{[^ ]+}} zeroinitializer
+// X86: @_ZGR15globalInitList2_ = internal global [2 x %[[WITHARG:[^ ]*]]] zeroinitializer
+// AMD: @globalInitList2 = addrspace(1) global %{{[^ ]+}} zeroinitializer
+// AMD: @_ZGR15globalInitList2_ = internal addrspace(1) global [2 x %[[WITHARG:[^ ]*]]] zeroinitializer
+
+// X86: @_ZN15partly_constant1kE = global i32 0, align 4
+// X86: @_ZN15partly_constant2ilE = global {{.*}} null, align 8
+// X86: @[[PARTLY_CONSTANT_OUTER:_ZGRN15partly_constant2ilE.*]] = internal global {{.*}} zeroinitializer, align 8
+// X86: @[[PARTLY_CONSTANT_INNER:_ZGRN15partly_constant2ilE.*]] = internal global [3 x {{.*}}] zeroinitializer, align 8
+// X86: @[[PARTLY_CONSTANT_FIRST:_ZGRN15partly_constant2ilE.*]] = internal constant [3 x i32] [i32 1, i32 2, i32 3], align 4
+// X86: @[[PARTLY_CONSTANT_SECOND:_ZGRN15partly_constant2ilE.*]] = internal global [2 x i32] zeroinitializer, align 4
+// X86: @[[PARTLY_CONSTANT_THIRD:_ZGRN15partly_constant2ilE.*]] = internal constant [4 x i32] [i32 5, i32 6, i32 7, i32 8], align 4
+// AMD: @_ZN15partly_constant1kE = addrspace(1) global i32 0, align 4
+// AMD: @_ZN15partly_constant2ilE = addrspace(2) global {{.*}} null, align 8
+// AMD: @[[PARTLY_CONSTANT_OUTER:_ZGRN15partly_constant2ilE.*]] = internal addrspace(2) global {{.*}} zeroinitializer, align 8
+// AMD: @[[PARTLY_CONSTANT_INNER:_ZGRN15partly_constant2ilE.*]] = internal addrspace(2) global [3 x {{.*}}] zeroinitializer, align 8
+// AMD: @[[PARTLY_CONSTANT_FIRST:_ZGRN15partly_constant2ilE.*]] = internal addrspace(2) constant [3 x i32] [i32 1, i32 2, i32 3], align 4
+// AMD: @[[PARTLY_CONSTANT_SECOND:_ZGRN15partly_constant2ilE.*]] = internal addrspace(2) global [2 x i32] zeroinitializer, align 4
+// AMD: @[[PARTLY_CONSTANT_THIRD:_ZGRN15partly_constant2ilE.*]] = internal addrspace(2) constant [4 x i32] [i32 5, i32 6, i32 7, i32 8], align 4
+
+// X86: @[[REFTMP1:.*]] = private constant [2 x i32] [i32 42, i32 43], align 4
+// X86: @[[REFTMP2:.*]] = private constant [3 x %{{.*}}] [%{{.*}} { i32 1 }, %{{.*}} { i32 2 }, %{{.*}} { i32 3 }], align 4
+// AMD: @[[REFTMP1:.*]] = private addrspace(2) constant [2 x i32] [i32 42, i32 43], align 4
+// AMD: @[[REFTMP2:.*]] = private addrspace(2) constant [3 x %{{.*}}] [%{{.*}} { i32 1 }, %{{.*}} { i32 2 }, %{{.*}} { i32 3 }], align 4
 
 // CHECK: appending global
 
-
 // thread_local initializer:
-// CHECK-LABEL: define internal void
-// CHECK: store i32* getelementptr inbounds ([4 x i32], [4 x i32]* @_ZGRN25thread_local_global_array1xE_, i64 0, i64 0),
-// CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* @_ZN25thread_local_global_array1xE, i32 0, i32 0), align 8
-// CHECK: store i64 4, i64* getelementptr inbounds ({{.*}}, {{.*}}* @_ZN25thread_local_global_array1xE, i32 0, i32 1), align 8
-
-
-// CHECK-LABEL: define internal void
-// CHECK: call void @_ZN8witharg1C1ERK10destroyme1(%[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* @_ZGR15globalInitList2_, i{{32|64}} 0, i{{32|64}} 0
-// CHECK: call void @_ZN8witharg1C1ERK10destroyme1(%[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* @_ZGR15globalInitList2_, i{{32|64}} 0, i{{32|64}} 1
-// CHECK: __cxa_atexit
-// CHECK: store %[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* @_ZGR15globalInitList2_, i64 0, i64 0),
-// CHECK:       %[[WITHARG]]** getelementptr inbounds (%{{.*}}, %{{.*}}* @globalInitList2, i32 0, i32 0), align 8
-// CHECK: store i64 2, i64* getelementptr inbounds (%{{.*}}, %{{.*}}* @globalInitList2, i32 0, i32 1), align 8
-// CHECK: call void @_ZN10destroyme1D1Ev
+// X86-LABEL: define internal void @__cxx_global_var_init
+// X86: store i32* getelementptr inbounds ([4 x i32], [4 x i32]* @_ZGRN25thread_local_global_array1xE_, i64 0, i64 0),
+// X86:       i32** getelementptr inbounds ({{.*}}, {{.*}}* @_ZN25thread_local_global_array1xE, i32 0, i32 0), align 8
+// X86: store i64 4, i64* getelementptr inbounds ({{.*}}, {{.*}}* @_ZN25thread_local_global_array1xE, i32 0, i32 1), align 8
+
+// CHECK-LABEL: define internal void @__cxx_global_var_init
+// X86: call void @_ZN8witharg1C1ERK10destroyme1(%[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* @_ZGR15globalInitList2_, i{{32|64}} 0, i{{32|64}} 0
+// X86: call void @_ZN8witharg1C1ERK10destroyme1(%[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* @_ZGR15globalInitList2_, i{{32|64}} 0, i{{32|64}} 1
+// AMD: call void @_ZN8witharg1C1ERK10destroyme1(%[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* addrspacecast ({{[^@]+}} @_ZGR15globalInitList2_ {{[^)]+}}), i{{32|64}} 0, i{{32|64}} 0
+// AMD: call void @_ZN8witharg1C1ERK10destroyme1(%[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* addrspacecast ({{[^@]+}} @_ZGR15globalInitList2_ {{[^)]+}}), i{{32|64}} 0, i{{32|64}} 1
+// CHECK: call i32 @__cxa_atexit
+// X86: store %[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* @_ZGR15globalInitList2_, i64 0, i64 0),
+// X86:       %[[WITHARG]]** getelementptr inbounds (%{{.*}}, %{{.*}}* @globalInitList2, i32 0, i32 0), align 8
+// X86: store i64 2, i64* getelementptr inbounds (%{{.*}}, %{{.*}}* @globalInitList2, i32 0, i32 1), align 8
+// AMD: store %[[WITHARG]]* getelementptr inbounds ([2 x %[[WITHARG]]], [2 x %[[WITHARG]]]* addrspacecast ({{[^@]+}} @_ZGR15globalInitList2_ {{[^)]+}}), i64 0, i64 0),
+// AMD:       %[[WITHARG]]** getelementptr inbounds (%{{.*}}, %{{.*}}* addrspacecast ({{[^@]+}} @globalInitList2 {{[^)]+}}), i32 0, i32 0), align 8
+// AMD: store i64 2, i64* getelementptr inbounds (%{{.*}}, %{{.*}}* addrspacecast ({{[^@]+}} @globalInitList2 {{[^)]+}}), i32 0, i32 1), align 8
 // CHECK: call void @_ZN10destroyme1D1Ev
+// CHECK-NEXT: call void @_ZN10destroyme1D1Ev
+// CHECK-NEXT: ret void
 std::initializer_list<witharg1> globalInitList2 = {
   witharg1(destroyme1()), witharg1(destroyme1())
 };
 
 void fn1(int i) {
   // CHECK-LABEL: define void @_Z3fn1i
   // temporary array
-  // CHECK: [[array:%[^ ]+]] = alloca [3 x i32]
+  // X86: [[array:%[^ ]+]] = alloca [3 x i32]
+  // AMD: [[alloca:%[^ ]+]] = alloca [3 x i32], align 4, addrspace(5)
+  // AMD: [[array:%[^ ]+]] = addrspacecast [3 x i32] addrspace(5)* [[alloca]] to [3 x i32]*
   // CHECK: getelementptr inbounds [3 x i32], [3 x i32]* [[array]], i{{32|64}} 0
   // CHECK-NEXT: store i32 1, i32*
   // CHECK-NEXT: getelementptr
   // CHECK-NEXT: store
   // CHECK-NEXT: getelementptr
   // CHECK-NEXT: load
   // CHECK-NEXT: store
   // init the list
void fn6() {
   // CHECK: call void @_ZN10wantslist1C1ESt16initializer_listI10destroyme1E
   // CHECK: call void @_Z6targetRK10wantslist1
   target({ destroyme1(), destroyme1() });
   // CHECK: call void @_ZN10wantslist1D1Ev
   // CHECK: call void @_ZN10destroyme1D1Ev
   destroyme2 dm2;
   // CHECK: call void @_ZN10destroyme2D1Ev
 }
-
 void fn7() {
   // CHECK-LABEL: define void @_Z3fn7v
   // temps should be destroyed before dm2
   // object should be destroyed after dm2
   // CHECK: call void @_ZN10wantslist1C1ESt16initializer_listI10destroyme1E
   wantslist1 wl = { destroyme1(), destroyme1() };
   // CHECK: call void @_ZN10destroyme1D1Ev
   destroyme2 dm2;
namespace dtors {
   }
 }
 
 namespace partly_constant {
   int k;
   std::initializer_list<std::initializer_list<int>> &&il = { { 1, 2, 3 }, { 4, k }, { 5, 6, 7, 8 } };
   // First init list.
   // CHECK-NOT: @[[PARTLY_CONSTANT_FIRST]],
-  // CHECK: store i32* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_FIRST]], i64 0, i64 0),
-  // CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_INNER]], i64 0, i64 0, i32 0)
-  // CHECK: store i64 3, i64* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_INNER]], i64 0, i64 0, i32 1)
+  // CHECK: store i32* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_FIRST]]{{.*}}, i64 0, i64 0),
+  // CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 0, i64 0, i32 0)
+  // CHECK: store i64 3, i64* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 0, i64 0, i32 1)
   // CHECK-NOT: @[[PARTLY_CONSTANT_FIRST]],
   //
   // Second init list array (non-constant).
-  // CHECK: store i32 4, i32* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_SECOND]], i64 0, i64 0)
-  // CHECK: load i32, i32* @_ZN15partly_constant1kE
-  // CHECK: store i32 {{.*}}, i32* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_SECOND]], i64 0, i64 1)
+  // CHECK: store i32 4, i32* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_SECOND]]{{.*}}, i64 0, i64 0)
+  // CHECK: load i32, i32* {{.*}}@_ZN15partly_constant1kE
+  // CHECK: store i32 {{.*}}, i32* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_SECOND]]{{.*}}, i64 0, i64 1)
   //
   // Second init list.
-  // CHECK: store i32* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_SECOND]], i64 0, i64 0),
-  // CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_INNER]], i64 0, i64 1, i32 0)
-  // CHECK: store i64 2, i64* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_INNER]], i64 0, i64 1, i32 1)
+  // CHECK: store i32* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_SECOND]]{{.*}}, i64 0, i64 0),
+  // CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 0, i64 1, i32 0)
+  // CHECK: store i64 2, i64* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 0, i64 1, i32 1)
   //
   // Third init list.
   // CHECK-NOT: @[[PARTLY_CONSTANT_THIRD]],
-  // CHECK: store i32* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_THIRD]], i64 0, i64 0),
-  // CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_INNER]], i64 0, i64 2, i32 0)
-  // CHECK: store i64 4, i64* getelementptr inbounds ({{.*}}, {{.*}}* @_ZGRN15partly_constant2ilE4_, i64 0, i64 2, i32 1)
+  // CHECK: store i32* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_THIRD]]{{.*}}, i64 0, i64 0),
+  // CHECK:       i32** getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 0, i64 2, i32 0)
+  // CHECK: store i64 4, i64* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@_ZGRN15partly_constant2ilE4_{{.*}}, i64 0, i64 2, i32 1)
   // CHECK-NOT: @[[PARTLY_CONSTANT_THIRD]],
   //
   // Outer init list.
-  // CHECK: store {{.*}}* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_INNER]], i64 0, i64 0),
-  // CHECK:       {{.*}}** getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_OUTER]], i32 0, i32 0)
-  // CHECK: store i64 3, i64* getelementptr inbounds ({{.*}}, {{.*}}* @[[PARTLY_CONSTANT_OUTER]], i32 0, i32 1)
+  // CHECK: store {{.*}}* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_INNER]]{{.*}}, i64 0, i64 0),
+  // CHECK:       {{.*}}** getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_OUTER]]{{.*}}, i32 0, i32 0)
+  // CHECK: store i64 3, i64* getelementptr inbounds ({{.*}}, {{.*}}* {{.*}}@[[PARTLY_CONSTANT_OUTER]]{{.*}}, i32 0, i32 1)
   //
   // 'il' reference.
-  // CHECK: store {{.*}}* @[[PARTLY_CONSTANT_OUTER]], {{.*}}** @_ZN15partly_constant2ilE, align 8
+  // CHECK: store {{.*}}* {{.*}}@[[PARTLY_CONSTANT_OUTER]]{{.*}}, {{.*}}** {{.*}}@_ZN15partly_constant2ilE{{.*}}, align 8
 }
-
 namespace nested {
   struct A { A(); ~A(); };
   struct B { const A &a; ~B(); };
   struct C { std::initializer_list<B> b; ~C(); };
   void f();
   // CHECK-LABEL: define void @_ZN6nested1gEv(
   void g() {
     // CHECK: call void @_ZN6nested1AC1Ev(

 }
 
 namespace PR20445 {
   struct vector { vector(std::initializer_list<int>); };
   struct MyClass { explicit MyClass(const vector &v); };
   template<int x> void f() { new MyClass({42, 43}); }
   template void f<0>();
   // CHECK-LABEL: define {{.*}} @_ZN7PR204451fILi0EEEvv(
-  // CHECK: store i32* getelementptr inbounds ([2 x i32], [2 x i32]* @[[REFTMP1]], i64 0, i64 0)
+  // CHECK: store i32* getelementptr inbounds ([2 x i32], [2 x i32]* {{.*}}@[[REFTMP1]]{{.*}}, i64 0, i64 0)
   // CHECK: call void @_ZN7PR204456vectorC1ESt16initializer_listIiE(
   // CHECK: call void @_ZN7PR204457MyClassC1ERKNS_6vectorE(
 }
 
 namespace ConstExpr {
   class C {
     int x;
   public:
     constexpr C(int x) : x(x) {}
   };
   void f(std::initializer_list<C>);
   void g() {
     // CHECK-LABEL: _ZN9ConstExpr1gEv
-// CHECK: store %"class.ConstExpr::C"* getelementptr inbounds ([3 x %"class.ConstExpr::C"], [3 x %"class.ConstExpr::C"]* @[[REFTMP2]], i64 0, i64 0)
+    // CHECK: store %"class.ConstExpr::C"* getelementptr inbounds ([3 x %"class.ConstExpr::C"], [3 x %"class.ConstExpr::C"]* {{.*}}@[[REFTMP2]]{{.*}}, i64 0, i64 0)
     // CHECK: call void @_ZN9ConstExpr1fESt16initializer_listINS_1CEE
     f({C(1), C(2), C(3)});
   }
 }
 
 namespace B19773010 {
   template <class T1, class T2> struct pair {
     T1 first;
     T2 second;
     constexpr pair() : first(), second() {}
     constexpr pair(T1 a, T2 b) : first(a), second(b) {}
   };
 
   enum E { ENUM_CONSTANT };
   struct testcase {
     testcase(std::initializer_list<pair<const char *, E>>);
   };
   void f1() {
     // CHECK-LABEL: @_ZN9B197730102f1Ev
     testcase a{{"", ENUM_CONSTANT}};
-    // CHECK: store %"struct.B19773010::pair"* getelementptr inbounds ([1 x %"struct.B19773010::pair"], [1 x %"struct.B19773010::pair"]* bitcast ([1 x { i8*, i32 }]* @.ref.tmp{{.*}} to [1 x %"struct.B19773010::pair"]*), i64 0, i64 0), %"struct.B19773010::pair"** %{{.*}}, align 8
+    // X86: store %"struct.B19773010::pair"* getelementptr inbounds ([1 x %"struct.B19773010::pair"], [1 x %"struct.B19773010::pair"]* bitcast ([1 x { i8*, i32 }]* @.ref.tmp{{.*}} to [1 x %"struct.B19773010::pair"]*), i64 0, i64 0), %"struct.B19773010::pair"** %{{.*}}, align 8
+    // AMD: store %"struct.B19773010::pair"* getelementptr inbounds ([1 x %"struct.B19773010::pair"], [1 x %"struct.B19773010::pair"]* addrspacecast{{.*}} bitcast ([1 x { i8*, i32 }] addrspace(2)* @.ref.tmp{{.*}} to [1 x %"struct.B19773010::pair"] addrspace(2)*){{.*}}, i64 0, i64 0), %"struct.B19773010::pair"** %{{.*}}, align 8
   }
   void f2() {
     // CHECK-LABEL: @_ZN9B197730102f2Ev
-    // CHECK: store %"struct.B19773010::pair"* getelementptr inbounds ([1 x %"struct.B19773010::pair"], [1 x %"struct.B19773010::pair"]* bitcast ([1 x { i8*, i32 }]* @_ZGRZN9B197730102f2EvE1p_ to [1 x %"struct.B19773010::pair"]*), i64 0, i64 0), %"struct.B19773010::pair"** getelementptr inbounds ([2 x %"class.std::initializer_list.10"], [2 x %"class.std::initializer_list.10"]* @_ZZN9B197730102f2EvE1p, i64 0, i64 1, i32 0), align 16
+    // X86: store %"struct.B19773010::pair"* getelementptr inbounds ([1 x %"struct.B19773010::pair"], [1 x %"struct.B19773010::pair"]* bitcast ([1 x { i8*, i32 }]* @_ZGRZN9B197730102f2EvE1p_ to [1 x %"struct.B19773010::pair"]*), i64 0, i64 0), %"struct.B19773010::pair"** getelementptr inbounds ([2 x %"class.std::initializer_list.10"], [2 x %"class.std::initializer_list.10"]* @_ZZN9B197730102f2EvE1p, i64 0, i64 1, i32 0), align 16
+    // AMD: store %"struct.B19773010::pair"* getelementptr inbounds ([1 x %"struct.B19773010::pair"], [1 x %"struct.B19773010::pair"]* addrspacecast{{.*}} bitcast ([1 x { i8*, i32 }] addrspace(1)* @_ZGRZN9B197730102f2EvE1p_ to [1 x %"struct.B19773010::pair"] addrspace(1)*){{.*}}, i64 0, i64 0), %"struct.B19773010::pair"** getelementptr inbounds ([2 x %"class.std::initializer_list.10"], [2 x %"class.std::initializer_list.10"]* addrspacecast{{.*}}@_ZZN9B197730102f2EvE1p{{.*}}, i64 0, i64 1, i32 0), align 8
     static std::initializer_list<pair<const char *, E>> a, p[2] =
         {a, {{"", ENUM_CONSTANT}}};
   }
 
   void PR22940_helper(const pair<void*, int>&) { }
   void PR22940() {
     // CHECK-LABEL: @_ZN9B197730107PR22940Ev
     // CHECK: call {{.*}} @_ZN9B197730104pairIPviEC{{.}}Ev(
     // CHECK: call {{.*}} @_ZN9B1977301014PR22940_helperERKNS_4pairIPviEE(
     PR22940_helper(pair<void*, int>());
   }
 }