Global merge for global symbols.

This commit is to improve global merge pass and support global symbol merge.
The global symbol merge is not enabled by default. For aarch64, we need some
more back-end fix to make it really benifit ADRP CSE.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210640 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 284 insertions, 75 deletions

diff --git a/lib/Transforms/Scalar/GlobalMerge.cpp b/lib/Transforms/Scalar/GlobalMerge.cpp
index 5222712631..ecf9be861e 100644
--- a/lib/Transforms/Scalar/GlobalMerge.cpp
+++ b/lib/Transforms/Scalar/GlobalMerge.cpp
@@ -81,6 +81,13 @@ EnableGlobalMergeOnConst("global-merge-on-const", cl::Hidden,
                          cl::desc("Enable global merge pass on constants"),
                          cl::init(false));
 
+// FIXME: this could be a transitional option, and we probably need to remove
+// it if only we are sure this optimization could always benefit all targets.
+static cl::opt<bool>
+EnableGlobalMergeOnExternal("global-merge-on-external", cl::Hidden,
+     cl::desc("Enable global merge pass on external linkage"),
+     cl::init(false));
+
 STATISTIC(NumMerged      , "Number of globals merged");
 namespace {
   class GlobalMerge : public FunctionPass {
@@ -129,9 +136,19 @@ namespace {
 } // end anonymous namespace
 
 char GlobalMerge::ID = 0;
-INITIALIZE_PASS(GlobalMerge, "global-merge",
-                "Global Merge", false, false)
 
+static void *initializeGlobalMergePassOnce(PassRegistry &Registry) {
+  PassInfo *PI = new PassInfo(
+      "Merge global variables", "global-merge", &GlobalMerge::ID,
+      PassInfo::NormalCtor_t(callDefaultCtor<GlobalMerge>), false, false,
+      PassInfo::TargetMachineCtor_t(callTargetMachineCtor<GlobalMerge>));
+  Registry.registerPass(*PI, true);
+  return PI;
+}
+
+void llvm::initializeGlobalMergePass(PassRegistry &Registry) {
+  CALL_ONCE_INITIALIZATION(initializeGlobalMergePassOnce)
+}
 
 bool GlobalMerge::doMerge(SmallVectorImpl<GlobalVariable*> &Globals,
                           Module &M, bool isConst, unsigned AddrSpace) const {
@@ -154,11 +171,23 @@ bool GlobalMerge::doMerge(SmallVectorImpl<GlobalVariable*> &Globals,
 
   Type *Int32Ty = Type::getInt32Ty(M.getContext());
 
+  assert(Globals.size() > 1);
+
+  // FIXME: This simple solution merges globals all together as maximum as
+  // possible. However, with this solution it would be hard to remove dead
+  // global symbols at link-time. An alternative solution could be checking
+  // global symbols references function by function, and make the symbols
+  // being referred in the same function merged and we would probably need
+  // to introduce heuristic algorithm to solve the merge conflict from
+  // different functions.
   for (size_t i = 0, e = Globals.size(); i != e; ) {
     size_t j = 0;
     uint64_t MergedSize = 0;
     std::vector<Type*> Tys;
     std::vector<Constant*> Inits;
+
+    bool HasExternal = false;
+    GlobalVariable *TheFirstExternal = 0;
     for (j = i; j != e; ++j) {
       Type *Ty = Globals[j]->getType()->getElementType();
       MergedSize += DL->getTypeAllocSize(Ty);
@@ -167,17 +196,37 @@ bool GlobalMerge::doMerge(SmallVectorImpl<GlobalVariable*> &Globals,
       }
       Tys.push_back(Ty);
       Inits.push_back(Globals[j]->getInitializer());
+
+      if (Globals[j]->hasExternalLinkage() && !HasExternal) {
+        HasExternal = true;
+        TheFirstExternal = Globals[j];
+      }
     }
 
+    // If merged variables doesn't have external linkage, we needn't to expose
+    // the symbol after merging.
+    GlobalValue::LinkageTypes Linkage = HasExternal
+                                            ? GlobalValue::ExternalLinkage
+                                            : GlobalValue::InternalLinkage;
+
+    // If merged variables have external linkage, we use symbol name of the
+    // first variable merged as the suffix of global symbol name. This would
+    // be able to avoid the link-time naming conflict for globalm symbols.
+    Twine MergedGVName = HasExternal
+                             ? "_MergedGlobals_" + TheFirstExternal->getName()
+                             : "_MergedGlobals";
+
     StructType *MergedTy = StructType::get(M.getContext(), Tys);
     Constant *MergedInit = ConstantStruct::get(MergedTy, Inits);
-    GlobalVariable *MergedGV = new GlobalVariable(M, MergedTy, isConst,
-                                                  GlobalValue::InternalLinkage,
-                                                  MergedInit, "_MergedGlobals",
-                                                  nullptr,
-                                                  GlobalVariable::NotThreadLocal,
-                                                  AddrSpace);
+
+    GlobalVariable *MergedGV = new GlobalVariable(
+        M, MergedTy, isConst, Linkage, MergedInit, MergedGVName, nullptr,
+        GlobalVariable::NotThreadLocal, AddrSpace);
+
     for (size_t k = i; k < j; ++k) {
+      GlobalValue::LinkageTypes Linkage = Globals[k]->getLinkage();
+      std::string Name = Globals[k]->getName();
+
       Constant *Idx[2] = {
         ConstantInt::get(Int32Ty, 0),
         ConstantInt::get(Int32Ty, k-i)
@@ -185,6 +234,14 @@ bool GlobalMerge::doMerge(SmallVectorImpl<GlobalVariable*> &Globals,
       Constant *GEP = ConstantExpr::getInBoundsGetElementPtr(MergedGV, Idx);
       Globals[k]->replaceAllUsesWith(GEP);
       Globals[k]->eraseFromParent();
+
+      if (Linkage != GlobalValue::InternalLinkage) {
+        // Generate a new alias...
+        auto *PTy = cast<PointerType>(GEP->getType());
+        GlobalAlias::create(PTy->getElementType(), PTy->getAddressSpace(),
+                            Linkage, Name, GEP, &M);
+      }
+
       NumMerged++;
     }
     i = j;
@@ -245,8 +302,12 @@ bool GlobalMerge::doInitialization(Module &M) {
   // Grab all non-const globals.
   for (Module::global_iterator I = M.global_begin(),
          E = M.global_end(); I != E; ++I) {
-    // Merge is safe for "normal" internal globals only
-    if (!I->hasLocalLinkage() || I->isThreadLocal() || I->hasSection())
+    // Merge is safe for "normal" internal or external globals only
+    if (I->isDeclaration() || I->isThreadLocal() || I->hasSection())
+      continue;
+
+    if (!(EnableGlobalMergeOnExternal && I->hasExternalLinkage()) &&
+        !I->hasInternalLinkage())
       continue;
 
     PointerType *PT = dyn_cast<PointerType>(I->getType());
diff --git a/lib/Transforms/Scalar/Scalar.cpp b/lib/Transforms/Scalar/Scalar.cpp
index edf012d811..5c7db5b16e 100644
--- a/lib/Transforms/Scalar/Scalar.cpp
+++ b/lib/Transforms/Scalar/Scalar.cpp
@@ -38,6 +38,7 @@ void llvm::initializeScalarOpts(PassRegistry &Registry) {
   initializeDSEPass(Registry);
   initializeGVNPass(Registry);
   initializeEarlyCSEPass(Registry);
+  initializeGlobalMergePass(Registry);
   initializeIndVarSimplifyPass(Registry);
   initializeJumpThreadingPass(Registry);
   initializeLICMPass(Registry);
diff --git a/test/CodeGen/AArch64/global-merge.ll b/test/CodeGen/AArch64/global-merge.ll
new file mode 100644
index 0000000000..aed1dc4d1c
--- /dev/null
+++ b/test/CodeGen/AArch64/global-merge.ll
@@ -0,0 +1,30 @@
+; RUN: llc < %s -mtriple=aarch64-none-linux-gnu -O0 | FileCheck --check-prefix=NO-MERGE %s
+; RUN: llc < %s -mtriple=aarch64-none-linux-gnu -O0 -global-merge-on-external=true | FileCheck --check-prefix=NO-MERGE %s
+
+; RUN: llc < %s -mtriple=aarch64-apple-ios -O0 | FileCheck %s --check-prefix=CHECK-APPLE-IOS-NO-MERGE
+; RUN: llc < %s -mtriple=aarch64-apple-ios -O0 -global-merge-on-external=true | FileCheck %s --check-prefix=CHECK-APPLE-IOS-NO-MERGE
+
+; FIXME: add O1/O2 test for aarch64-none-linux-gnu and aarch64-apple-ios
+
+@m = internal global i32 0, align 4
+@n = internal global i32 0, align 4
+
+define void @f1(i32 %a1, i32 %a2) {
+; CHECK-LABEL: f1:
+; CHECK: adrp x{{[0-9]+}}, _MergedGlobals
+; CHECK-NOT: adrp
+
+; CHECK-APPLE-IOS-LABEL: f1:
+; CHECK-APPLE-IOS: adrp x{{[0-9]+}}, __MergedGlobals
+; CHECK-APPLE-IOS-NOT: adrp
+  store i32 %a1, i32* @m, align 4
+  store i32 %a2, i32* @n, align 4
+  ret void
+}
+
+; CHECK:        .local _MergedGlobals
+; CHECK:        .comm  _MergedGlobals,8,8
+; NO-MERGE-NOT: .local _MergedGlobals
+
+; CHECK-APPLE-IOS: .zerofill __DATA,__bss,__MergedGlobals,8,3
+; CHECK-APPLE-IOS-NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,8,3
diff --git a/test/CodeGen/ARM/global-merge-1.ll b/test/CodeGen/ARM/global-merge-1.ll
new file mode 100644
index 0000000000..341597e618
--- /dev/null
+++ b/test/CodeGen/ARM/global-merge-1.ll
@@ -0,0 +1,85 @@
+; RUN: llc %s -O0 -o - | FileCheck -check-prefix=NO-MERGE %s
+; RUN: llc %s -O0 -o - -enable-global-merge=false | FileCheck -check-prefix=NO-MERGE %s
+; RUN: llc %s -O0 -o - -enable-global-merge=true | FileCheck -check-prefix=NO-MERGE %s
+; RUN: llc %s -O1 -o - | FileCheck -check-prefix=MERGE %s
+; RUN: llc %s -O1 -o - -enable-global-merge=false | FileCheck -check-prefix=NO-MERGE %s
+; RUN: llc %s -O1 -o - -enable-global-merge=true | FileCheck -check-prefix=MERGE %s
+
+; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
+; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
+; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
+; MERGE: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
+; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
+; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
+
+; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; NO-MERGE: .zerofill __DATA,__bss,_bar,20,2
+; NO-MERGE: .zerofill __DATA,__bss,_baz,20,2
+; NO-MERGE: .zerofill __DATA,__bss,_foo,20,2
+; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
+
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
+target triple = "thumbv7-apple-ios3.0.0"
+
+@bar = internal global [5 x i32] zeroinitializer, align 4
+@baz = internal global [5 x i32] zeroinitializer, align 4
+@foo = internal global [5 x i32] zeroinitializer, align 4
+
+; Function Attrs: nounwind ssp
+define internal void @initialize() #0 {
+  %1 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %1, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 0), align 4, !tbaa !1
+  %2 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %2, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 0), align 4, !tbaa !1
+  %3 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %3, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 1), align 4, !tbaa !1
+  %4 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %4, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 1), align 4, !tbaa !1
+  %5 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %5, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 2), align 4, !tbaa !1
+  %6 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %6, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 2), align 4, !tbaa !1
+  %7 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %7, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 3), align 4, !tbaa !1
+  %8 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %8, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 3), align 4, !tbaa !1
+  %9 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %9, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 4), align 4, !tbaa !1
+  %10 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
+  store i32 %10, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 4), align 4, !tbaa !1
+  ret void
+}
+
+declare i32 @calc(...) #1
+
+; Function Attrs: nounwind ssp
+define internal void @calculate() #0 {
+  %1 = load <4 x i32>* bitcast ([5 x i32]* @bar to <4 x i32>*), align 4
+  %2 = load <4 x i32>* bitcast ([5 x i32]* @baz to <4 x i32>*), align 4
+  %3 = mul <4 x i32> %2, %1
+  store <4 x i32> %3, <4 x i32>* bitcast ([5 x i32]* @foo to <4 x i32>*), align 4
+  %4 = load i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 4), align 4, !tbaa !1
+  %5 = load i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 4), align 4, !tbaa !1
+  %6 = mul nsw i32 %5, %4
+  store i32 %6, i32* getelementptr inbounds ([5 x i32]* @foo, i32 0, i32 4), align 4, !tbaa !1
+  ret void
+}
+
+; Function Attrs: nounwind readnone ssp
+define internal i32* @returnFoo() #2 {
+  ret i32* getelementptr inbounds ([5 x i32]* @foo, i32 0, i32 0)
+}
+
+attributes #0 = { nounwind ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { nounwind readnone ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #3 = { nounwind }
+
+!llvm.ident = !{!0}
+
+!0 = metadata !{metadata !"LLVM version 3.4 "}
+!1 = metadata !{metadata !2, metadata !2, i64 0}
+!2 = metadata !{metadata !"int", metadata !3, i64 0}
+!3 = metadata !{metadata !"omnipotent char", metadata !4, i64 0}
+!4 = metadata !{metadata !"Simple C/C++ TBAA"}
diff --git a/test/Transforms/GlobalMerge/AArch64/arm64.ll b/test/Transforms/GlobalMerge/AArch64/arm64.ll
index 4715fd8de2..9174f72574 100644
--- a/test/Transforms/GlobalMerge/AArch64/arm64.ll
+++ b/test/Transforms/GlobalMerge/AArch64/arm64.ll
@@ -1,23 +1,6 @@
-; RUN: llc %s -O0 -o - | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O0 -o - -enable-global-merge=false | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O0 -o - -enable-global-merge=true | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O1 -o - | FileCheck -check-prefix=MERGE %s
-; RUN: llc %s -O1 -o - -enable-global-merge=false | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O1 -o - -enable-global-merge=true | FileCheck -check-prefix=MERGE %s
+; RUN: opt %s -mtriple=aarch64-linux-gnuabi -global-merge -S -o - | FileCheck %s
 
-; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
-; MERGE: .zerofill __DATA,__bss,__MergedGlobals,60,4
-; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
-
-; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
-; NO-MERGE: .zerofill __DATA,__bss,_bar,20,2
-; NO-MERGE: .zerofill __DATA,__bss,_baz,20,2
-; NO-MERGE: .zerofill __DATA,__bss,_foo,20,2
-; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; CHECK: @_MergedGlobals = internal global { [5 x i32], [5 x i32], [5 x i32] } zeroinitializer
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n32:64-S128"
 target triple = "arm64-apple-ios7.0.0"
diff --git a/test/Transforms/GlobalMerge/AArch64/global-merge-1.ll b/test/Transforms/GlobalMerge/AArch64/global-merge-1.ll
new file mode 100644
index 0000000000..f8854dd545
--- /dev/null
+++ b/test/Transforms/GlobalMerge/AArch64/global-merge-1.ll
@@ -0,0 +1,22 @@
+; RUN: opt %s -mtriple=aarch64-none-linux-gnu -global-merge -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-none-linux-gnu -global-merge -global-merge-on-external -S -o - | FileCheck %s
+
+; RUN: opt %s -mtriple=aarch64-linux-gnuabi -global-merge -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-linux-gnuabi -global-merge -global-merge-on-external -S -o - | FileCheck %s
+
+; RUN: opt %s -mtriple=aarch64-apple-ios -global-merge -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-apple-ios -global-merge -global-merge-on-external -S -o - | FileCheck %s
+
+@m = internal global i32 0, align 4
+@n = internal global i32 0, align 4
+
+; CHECK: @_MergedGlobals = internal global { i32, i32 } zeroinitializer
+
+define void @f1(i32 %a1, i32 %a2) {
+; CHECK-LABEL: @f1
+; CHECK: getelementptr inbounds ({ i32, i32 }* @_MergedGlobals, i32 0, i32 0)
+; CHECK: getelementptr inbounds ({ i32, i32 }* @_MergedGlobals, i32 0, i32 1)
+  store i32 %a1, i32* @m, align 4
+  store i32 %a2, i32* @n, align 4
+  ret void
+}
diff --git a/test/Transforms/GlobalMerge/AArch64/global-merge-2.ll b/test/Transforms/GlobalMerge/AArch64/global-merge-2.ll
new file mode 100644
index 0000000000..1876483882
--- /dev/null
+++ b/test/Transforms/GlobalMerge/AArch64/global-merge-2.ll
@@ -0,0 +1,30 @@
+; RUN: opt %s -mtriple=aarch64-none-linux-gnu -global-merge -global-merge-on-external -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-linux-gnuabi -global-merge -global-merge-on-external -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-apple-ios -global-merge -global-merge-on-external -S -o - | FileCheck %s
+
+@x = global i32 0, align 4
+@y = global i32 0, align 4
+@z = global i32 0, align 4
+
+; CHECK: @_MergedGlobals_x = global { i32, i32, i32 } zeroinitializer
+; CHECK: @x = alias getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 0)
+; CHECK: @y = alias getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 1)
+; CHECK: @z = alias getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 2)
+
+define void @f1(i32 %a1, i32 %a2) {
+; CHECK-LABEL: @f1
+; CHECK: getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 0)
+; CHECK: getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 1)
+  store i32 %a1, i32* @x, align 4
+  store i32 %a2, i32* @y, align 4
+  ret void
+}
+
+define void @g1(i32 %a1, i32 %a2) {
+; CHECK-LABEL: @g1
+; CHECK: getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 1)
+; CHECK: getelementptr inbounds ({ i32, i32, i32 }* @_MergedGlobals_x, i32 0, i32 2)
+  store i32 %a1, i32* @y, align 4
+  store i32 %a2, i32* @z, align 4
+  ret void
+}
diff --git a/test/Transforms/GlobalMerge/AArch64/global-merge-3.ll b/test/Transforms/GlobalMerge/AArch64/global-merge-3.ll
new file mode 100644
index 0000000000..811c3eef13
--- /dev/null
+++ b/test/Transforms/GlobalMerge/AArch64/global-merge-3.ll
@@ -0,0 +1,27 @@
+; RUN: opt %s -mtriple=aarch64-none-linux-gnu -global-merge -global-merge-on-external -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-linux-gnuabi -global-merge -global-merge-on-external -S -o - | FileCheck %s
+; RUN: opt %s -mtriple=aarch64-apple-ios -global-merge -global-merge-on-external -S -o - | FileCheck %s
+
+@x = global [1000 x i32] zeroinitializer, align 1
+@y = global [1000 x i32] zeroinitializer, align 1
+@z = internal global i32 1, align 4
+
+; CHECK: @_MergedGlobals_x = global { i32, [1000 x i32] } { i32 1, [1000 x i32] zeroinitializer }
+; CHECK: @_MergedGlobals_y = global { [1000 x i32] } zeroinitializer
+
+; CHECK: @x = alias getelementptr inbounds ({ i32, [1000 x i32] }* @_MergedGlobals_x, i32 0, i32 1)
+; CHECK: @y = alias getelementptr inbounds ({ [1000 x i32] }* @_MergedGlobals_y, i32 0, i32 0)
+
+define void @f1(i32 %a1, i32 %a2, i32 %a3) {
+; CHECK-LABEL: @f1
+; CHECK: %x3 = getelementptr inbounds [1000 x i32]* getelementptr inbounds ({ i32, [1000 x i32] }* @_MergedGlobals_x, i32 0, i32 1), i32 0, i64 3
+; CHECK: %y3 = getelementptr inbounds [1000 x i32]* getelementptr inbounds ({ [1000 x i32] }* @_MergedGlobals_y, i32 0, i32 0), i32 0, i64 3
+; CHECK: store i32 %a3, i32* getelementptr inbounds ({ i32, [1000 x i32] }* @_MergedGlobals_x, i32 0, i32 0), align 4
+
+  %x3 = getelementptr inbounds [1000 x i32]* @x, i32 0, i64 3
+  %y3 = getelementptr inbounds [1000 x i32]* @y, i32 0, i64 3
+  store i32 %a1, i32* %x3, align 4
+  store i32 %a2, i32* %y3, align 4
+  store i32 %a3, i32* @z, align 4
+  ret void
+}
diff --git a/test/Transforms/GlobalMerge/ARM/arm.ll b/test/Transforms/GlobalMerge/ARM/arm.ll
index 341597e618..e7553e9183 100644
--- a/test/Transforms/GlobalMerge/ARM/arm.ll
+++ b/test/Transforms/GlobalMerge/ARM/arm.ll
@@ -1,23 +1,4 @@
-; RUN: llc %s -O0 -o - | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O0 -o - -enable-global-merge=false | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O0 -o - -enable-global-merge=true | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O1 -o - | FileCheck -check-prefix=MERGE %s
-; RUN: llc %s -O1 -o - -enable-global-merge=false | FileCheck -check-prefix=NO-MERGE %s
-; RUN: llc %s -O1 -o - -enable-global-merge=true | FileCheck -check-prefix=MERGE %s
-
-; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
-; MERGE: .zerofill __DATA,__bss,__MergedGlobals,60,4
-; MERGE-NOT: .zerofill __DATA,__bss,_bar,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_baz,20,2
-; MERGE-NOT: .zerofill __DATA,__bss,_foo,20,2
-
-; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
-; NO-MERGE: .zerofill __DATA,__bss,_bar,20,2
-; NO-MERGE: .zerofill __DATA,__bss,_baz,20,2
-; NO-MERGE: .zerofill __DATA,__bss,_foo,20,2
-; NO-MERGE-NOT: .zerofill __DATA,__bss,__MergedGlobals,60,4
+; RUN: opt %s -mtriple=arm-linux-gnuabi -global-merge -S -o - | FileCheck %s
 
 target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
 target triple = "thumbv7-apple-ios3.0.0"
@@ -26,28 +7,30 @@ target triple = "thumbv7-apple-ios3.0.0"
 @baz = internal global [5 x i32] zeroinitializer, align 4
 @foo = internal global [5 x i32] zeroinitializer, align 4
 
+; CHECK: @_MergedGlobals = internal global { [5 x i32], [5 x i32], [5 x i32] } zeroinitializer
+
 ; Function Attrs: nounwind ssp
 define internal void @initialize() #0 {
   %1 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %1, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 0), align 4, !tbaa !1
+  store i32 %1, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 0), align 4
   %2 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %2, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 0), align 4, !tbaa !1
+  store i32 %2, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 0), align 4
   %3 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %3, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 1), align 4, !tbaa !1
+  store i32 %3, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 1), align 4
   %4 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %4, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 1), align 4, !tbaa !1
+  store i32 %4, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 1), align 4
   %5 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %5, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 2), align 4, !tbaa !1
+  store i32 %5, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 2), align 4
   %6 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %6, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 2), align 4, !tbaa !1
+  store i32 %6, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 2), align 4
   %7 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %7, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 3), align 4, !tbaa !1
+  store i32 %7, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 3), align 4
   %8 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %8, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 3), align 4, !tbaa !1
+  store i32 %8, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 3), align 4
   %9 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %9, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 4), align 4, !tbaa !1
+  store i32 %9, i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 4), align 4
   %10 = tail call i32 bitcast (i32 (...)* @calc to i32 ()*)() #3
-  store i32 %10, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 4), align 4, !tbaa !1
+  store i32 %10, i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 4), align 4
   ret void
 }
 
@@ -59,10 +42,10 @@ define internal void @calculate() #0 {
   %2 = load <4 x i32>* bitcast ([5 x i32]* @baz to <4 x i32>*), align 4
   %3 = mul <4 x i32> %2, %1
   store <4 x i32> %3, <4 x i32>* bitcast ([5 x i32]* @foo to <4 x i32>*), align 4
-  %4 = load i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 4), align 4, !tbaa !1
-  %5 = load i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 4), align 4, !tbaa !1
+  %4 = load i32* getelementptr inbounds ([5 x i32]* @bar, i32 0, i32 4), align 4
+  %5 = load i32* getelementptr inbounds ([5 x i32]* @baz, i32 0, i32 4), align 4
   %6 = mul nsw i32 %5, %4
-  store i32 %6, i32* getelementptr inbounds ([5 x i32]* @foo, i32 0, i32 4), align 4, !tbaa !1
+  store i32 %6, i32* getelementptr inbounds ([5 x i32]* @foo, i32 0, i32 4), align 4
   ret void
 }
 
@@ -70,16 +53,3 @@ define internal void @calculate() #0 {
 define internal i32* @returnFoo() #2 {
   ret i32* getelementptr inbounds ([5 x i32]* @foo, i32 0, i32 0)
 }
-
-attributes #0 = { nounwind ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
-attributes #1 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
-attributes #2 = { nounwind readnone ssp "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
-attributes #3 = { nounwind }
-
-!llvm.ident = !{!0}
-
-!0 = metadata !{metadata !"LLVM version 3.4 "}
-!1 = metadata !{metadata !2, metadata !2, i64 0}
-!2 = metadata !{metadata !"int", metadata !3, i64 0}
-!3 = metadata !{metadata !"omnipotent char", metadata !4, i64 0}
-!4 = metadata !{metadata !"Simple C/C++ TBAA"}