diff --git a/llvm/include/llvm/Transforms/IPO/Attributor.h b/llvm/include/llvm/Transforms/IPO/Attributor.h
--- a/llvm/include/llvm/Transforms/IPO/Attributor.h
+++ b/llvm/include/llvm/Transforms/IPO/Attributor.h
@@ -1189,6 +1189,11 @@
   ~Attributor();
 
   /// Run the analyses until a fixpoint is reached or enforced (timeout).
+  /// This method will run iteration, and manifest the changes to IR.
+  ///
+  /// This function runs all Attributor phases, for unrelated sets of
+  /// attributes, it would be more efficient to run Attributor in multiple
+  /// stages. \see runStage() and \see finalize()
   ///
   /// The attributes registered with this Attributor can be used after as long
   /// as the Attributor is not destroyed (it owns the attributes now).
@@ -1196,6 +1201,27 @@
   /// \Returns CHANGED if the IR was changed, otherwise UNCHANGED.
   ChangeStatus run();
 
+  /// Run a Attributor stage.
+  /// This method will do fixpoint iteration until fixpoint or the
+  /// maximum iteration count is reached.
+  ///
+  /// If the maximum iteration count is reached, This method will
+  /// indicate pessimistic fixpoint on attributes that transitively depend
+  /// on attributes that were scheduled for an update.
+  ///
+  /// After calling this function, you can still call `getAAFor` function to
+  /// seed more Attributes and then you can call this function again to run the
+  /// iteration one more time. Dividing the deduction process for independent
+  /// attributes helps with performance.
+  ///
+  /// The changes won't be manifested to IR until `finalize` function is called.
+  void runStage();
+
+  /// Manifest the changes to IR and clean up deleted functions.
+  ///
+  /// \Returns CHANGED if the IR was changed, otherwise UNCHANGED.
+  ChangeStatus finalize();
+
   /// Lookup an abstract attribute of type \p AAType at position \p IRP. While
   /// no abstract attribute is found equivalent positions are checked, see
   /// SubsumingPositionIterator. Thus, the returned abstract attribute
@@ -1417,6 +1443,14 @@
     return Functions.empty() || Functions.count(&Fn);
   }
 
+  /// Seed initial Attributes that should be run in the first stage.
+  /// Seeding Attributes such as `AAIsDead` before any other attribute,
+  /// helps with performance as it reduces wasted updates.
+  ///
+  /// \param F The function that is checked for attribute opportunities.
+  ///
+  void identifyFirstStageAttributes(Function &F);
+
   /// Determine opportunities to derive 'default' attributes in \p F and create
   /// abstract attribute objects for them.
   ///
@@ -1897,12 +1931,6 @@
   BumpPtrAllocator &Allocator;
 
 private:
-  /// This method will do fixpoint iteration until fixpoint or the
-  /// maximum iteration count is reached.
-  ///
-  /// If the maximum iteration count is reached, This method will
-  /// indicate pessimistic fixpoint on attributes that transitively depend
-  /// on attributes that were scheduled for an update.
   void runTillFixpoint();
 
   /// Gets called after scheduling, manifests attributes to the LLVM IR.
@@ -2008,9 +2036,6 @@
   /// Maximum number of fixedpoint iterations.
   Optional<unsigned> MaxFixpointIterations;
 
-  /// A set to remember the functions we already assume to be live and visited.
-  DenseSet<const Function *> VisitedFunctions;
-
   /// Uses we replace with a new value after manifest is done. We will remove
   /// then trivially dead instructions as well.
   DenseMap<Use *, Value *> ToBeChangedUses;
diff --git a/llvm/lib/Transforms/IPO/Attributor.cpp b/llvm/lib/Transforms/IPO/Attributor.cpp
--- a/llvm/lib/Transforms/IPO/Attributor.cpp
+++ b/llvm/lib/Transforms/IPO/Attributor.cpp
@@ -1313,15 +1313,16 @@
   return true;
 }
 
-void Attributor::runTillFixpoint() {
-  TimeTraceScope TimeScope("Attributor::runTillFixpoint");
+void Attributor::runStage() {
+  Phase = AttributorPhase::UPDATE;
+
+  TimeTraceScope TimeScope("Attributor::runIterationStage");
   LLVM_DEBUG(dbgs() << "[Attributor] Identified and initialized "
                     << DG.SyntheticRoot.Deps.size()
                     << " abstract attributes.\n");
 
   // Now that all abstract attributes are collected and initialized we start
   // the abstract analysis.
-
   unsigned IterationCounter = 1;
   unsigned MaxFixedPointIterations;
   if (MaxFixpointIterations)
@@ -1450,9 +1451,13 @@
     errs() << "\n[Attributor] Fixpoint iteration done after: "
            << IterationCounter << "/" << MaxFixedPointIterations
            << " iterations\n";
-    llvm_unreachable("The fixpoint was not reached with exactly the number of "
-                     "specified iterations!");
+    // llvm_unreachable("The fixpoint was not reached with exactly the number of
+    // "
+    //                 "specified iterations!");
   }
+
+  // We can still seed more attributes after running a fixpoint iteration.
+  Phase = AttributorPhase::SEEDING;
 }
 
 ChangeStatus Attributor::manifestAttributes() {
@@ -1817,8 +1822,7 @@
   if (PrintCallGraph)
     ACallGraph.populateAll();
 
-  Phase = AttributorPhase::UPDATE;
-  runTillFixpoint();
+  runStage();
 
   // dump graphs on demand
   if (DumpDepGraph)
@@ -1830,15 +1834,23 @@
   if (PrintDependencies)
     DG.print();
 
+  if (PrintCallGraph)
+    ACallGraph.print();
+
+  return finalize();
+}
+
+ChangeStatus Attributor::finalize() {
+  // More attributes can be seeded after the runStage() function is called.
+  assert(Phase == AttributorPhase::SEEDING &&
+         "Expected attributor to be in the seeeding stage!");
+
   Phase = AttributorPhase::MANIFEST;
   ChangeStatus ManifestChange = manifestAttributes();
 
   Phase = AttributorPhase::CLEANUP;
   ChangeStatus CleanupChange = cleanupIR();
 
-  if (PrintCallGraph)
-    ACallGraph.print();
-
   return ManifestChange | CleanupChange;
 }
 
@@ -2426,12 +2438,12 @@
   }
 }
 
-void Attributor::identifyDefaultAbstractAttributes(Function &F) {
-  if (!VisitedFunctions.insert(&F).second)
-    return;
+void Attributor::identifyFirstStageAttributes(Function &F) {
   if (F.isDeclaration())
     return;
 
+  IRPosition FPos = IRPosition::function(F);
+
   // In non-module runs we need to look at the call sites of a function to
   // determine if it is part of a must-tail call edge. This will influence what
   // attributes we can derive.
@@ -2443,13 +2455,69 @@
           FI.CalledViaMustTail = true;
   }
 
-  IRPosition FPos = IRPosition::function(F);
-
   // Check for dead BasicBlocks in every function.
   // We need dead instruction detection because we do not want to deal with
   // broken IR in which SSA rules do not apply.
   getOrCreateAAFor<AAIsDead>(FPos);
 
+  // Return attributes are only appropriate if the return type is non void.
+  Type *ReturnType = F.getReturnType();
+  if (!ReturnType->isVoidTy()) {
+    IRPosition RetPos = IRPosition::returned(F);
+
+    // Every returned value might be dead.
+    getOrCreateAAFor<AAIsDead>(RetPos);
+  }
+
+  for (Argument &Arg : F.args()) {
+    IRPosition ArgPos = IRPosition::argument(Arg);
+
+    // Every argument might be dead.
+    getOrCreateAAFor<AAIsDead>(ArgPos);
+  }
+
+  auto CallSitePred = [&](Instruction &I) -> bool {
+    auto &CB = cast<CallBase>(I);
+    IRPosition CBRetPos = IRPosition::callsite_returned(CB);
+
+    // Call sites might be dead if they do not have side effects and no live
+    // users. The return value might be dead if there are no live users.
+    getOrCreateAAFor<AAIsDead>(CBRetPos);
+
+    Function *Callee = CB.getCalledFunction();
+    // TODO: Even if the callee is not known now we might be able to simplify
+    //       the call/callee.
+    if (!Callee)
+      return true;
+
+    for (int I = 0, E = CB.getNumArgOperands(); I < E; ++I) {
+      IRPosition CBArgPos = IRPosition::callsite_argument(CB, I);
+
+      // Every call site argument might be dead.
+      getOrCreateAAFor<AAIsDead>(CBArgPos);
+    }
+
+    return true;
+  };
+
+  auto &OpcodeInstMap = InfoCache.getOpcodeInstMapForFunction(F);
+  bool Success;
+  bool UsedAssumedInformation = false;
+  Success = checkForAllInstructionsImpl(
+      nullptr, OpcodeInstMap, CallSitePred, nullptr, nullptr,
+      {(unsigned)Instruction::Invoke, (unsigned)Instruction::CallBr,
+       (unsigned)Instruction::Call},
+      UsedAssumedInformation);
+  (void)Success;
+  assert(Success && "Expected the check call to be successful!");
+}
+
+void Attributor::identifyDefaultAbstractAttributes(Function &F) {
+  if (F.isDeclaration())
+    return;
+
+  IRPosition FPos = IRPosition::function(F);
+
   // Every function might be "will-return".
   getOrCreateAAFor<AAWillReturn>(FPos);
 
@@ -2490,9 +2558,6 @@
 
     IRPosition RetPos = IRPosition::returned(F);
 
-    // Every returned value might be dead.
-    getOrCreateAAFor<AAIsDead>(RetPos);
-
     // Every function might be simplified.
     getOrCreateAAFor<AAValueSimplify>(RetPos);
 
@@ -2525,9 +2590,6 @@
     bool UsedAssumedInformation = false;
     getAssumedSimplified(ArgPos, /* AA */ nullptr, UsedAssumedInformation);
 
-    // Every argument might be dead.
-    getOrCreateAAFor<AAIsDead>(ArgPos);
-
     // Every argument might be marked noundef.
     getOrCreateAAFor<AANoUndef>(ArgPos);
 
@@ -2561,12 +2623,6 @@
 
   auto CallSitePred = [&](Instruction &I) -> bool {
     auto &CB = cast<CallBase>(I);
-    IRPosition CBRetPos = IRPosition::callsite_returned(CB);
-
-    // Call sites might be dead if they do not have side effects and no live
-    // users. The return value might be dead if there are no live users.
-    getOrCreateAAFor<AAIsDead>(CBRetPos);
-
     Function *Callee = CB.getCalledFunction();
     // TODO: Even if the callee is not known now we might be able to simplify
     //       the call/callee.
@@ -2589,9 +2645,6 @@
 
       IRPosition CBArgPos = IRPosition::callsite_argument(CB, I);
 
-      // Every call site argument might be dead.
-      getOrCreateAAFor<AAIsDead>(CBArgPos);
-
       // Call site argument might be simplified. We have to go through the
       // Attributor interface though as outside AAs can register custom
       // simplification callbacks.
@@ -2827,12 +2880,14 @@
     }
   }
 
-  for (Function *F : Functions) {
-    if (F->hasExactDefinition())
-      NumFnWithExactDefinition++;
-    else
-      NumFnWithoutExactDefinition++;
+  // We are running the Attributor in two stages.
+  // First we identify first stage attributes (like AAIsDead).
+  // We run a fixpoint iteration for those attributes
+  // before running a iteration on the rest, this helps with the performance.
 
+  // TODO: Consider running each function as a separate stage.
+
+  auto ShouldSkipFunction = [&](Function *F) {
     // We look at internal functions only on-demand but if any use is not a
     // direct call or outside the current set of analyzed functions, we have
     // to do it eagerly.
@@ -2842,15 +2897,41 @@
             return CB && CB->isCallee(&U) &&
                    Functions.count(const_cast<Function *>(CB->getCaller()));
           }))
-        continue;
+        return true;
     }
 
-    // Populate the Attributor with abstract attribute opportunities in the
-    // function and the information cache with IR information.
+    return false;
+  };
+
+  for (Function *F : Functions) {
+    if (F->hasExactDefinition())
+      NumFnWithExactDefinition++;
+    else
+      NumFnWithoutExactDefinition++;
+
+    if (ShouldSkipFunction(F))
+      continue;
+
+    // Identify AbstactAttribute oppurtunities that shuold be run on the first
+    // stage.
+    A.identifyFirstStageAttributes(*F);
+  }
+
+  // Run the first stage.
+  A.runStage();
+
+  for (Function *F : Functions) {
+    if (ShouldSkipFunction(F))
+      continue;
+
     A.identifyDefaultAbstractAttributes(*F);
   }
 
-  ChangeStatus Changed = A.run();
+  // Run the second stage.
+  A.runStage();
+
+  // Finalize the Attributor.
+  ChangeStatus Changed = A.finalize();
 
   LLVM_DEBUG(dbgs() << "[Attributor] Done with " << Functions.size()
                     << " functions, result: " << Changed << ".\n");
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/alignment.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/alignment.ll
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/alignment.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/alignment.ll
@@ -170,7 +170,7 @@
 ; IS__CGSCC_NPM-SAME: (i32 [[TMP0:%.*]]) #[[ATTR0]] {
 ; IS__CGSCC_NPM-NEXT:    [[A_PRIV:%.*]] = alloca i32, align 4
 ; IS__CGSCC_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[A_PRIV]], align 4
-; IS__CGSCC_NPM-NEXT:    [[C:%.*]] = call i32 @test(i32 undef, i64 undef) #[[ATTR1:[0-9]+]], !range [[RNG0:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[C:%.*]] = call i32 @test(i32 2, i64 undef) #[[ATTR1:[0-9]+]], !range [[RNG0:![0-9]+]]
 ; IS__CGSCC_NPM-NEXT:    ret i32 [[C]]
 ;
   %B = alloca i64
@@ -180,7 +180,7 @@
 }
 
 define i32 @callercaller() {
-; IS__TUNIT_OPM: Function Attrs: nofree nosync nounwind readnone
+; IS__TUNIT_OPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_OPM-LABEL: define {{[^@]+}}@callercaller
 ; IS__TUNIT_OPM-SAME: () #[[ATTR2:[0-9]+]] {
 ; IS__TUNIT_OPM-NEXT:    [[B:%.*]] = alloca i32, align 4
@@ -188,7 +188,7 @@
 ; IS__TUNIT_OPM-NEXT:    [[X:%.*]] = call i32 @caller(i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[B]]) #[[ATTR4:[0-9]+]]
 ; IS__TUNIT_OPM-NEXT:    ret i32 [[X]]
 ;
-; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone
+; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@callercaller
 ; IS__TUNIT_NPM-SAME: () #[[ATTR2:[0-9]+]] {
 ; IS__TUNIT_NPM-NEXT:    [[B:%.*]] = alloca i32, align 4
@@ -219,7 +219,7 @@
 ;.
 ; IS__TUNIT____: attributes #[[ATTR0:[0-9]+]] = { argmemonly nofree nosync nounwind readonly willreturn }
 ; IS__TUNIT____: attributes #[[ATTR1:[0-9]+]] = { argmemonly nofree nosync nounwind willreturn }
-; IS__TUNIT____: attributes #[[ATTR2:[0-9]+]] = { nofree nosync nounwind readnone }
+; IS__TUNIT____: attributes #[[ATTR2:[0-9]+]] = { nofree nosync nounwind readnone willreturn }
 ; IS__TUNIT____: attributes #[[ATTR3:[0-9]+]] = { nofree nosync nounwind readonly willreturn }
 ; IS__TUNIT____: attributes #[[ATTR4:[0-9]+]] = { nofree nosync nounwind willreturn }
 ;.
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/attrs.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/attrs.ll
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/attrs.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/attrs.ll
@@ -93,7 +93,7 @@
 ;
 ; IS__TUNIT_OPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_OPM-LABEL: define {{[^@]+}}@test
-; IS__TUNIT_OPM-SAME: (i32* nocapture nofree readonly align 4 [[X:%.*]]) #[[ATTR0]] {
+; IS__TUNIT_OPM-SAME: (i32* nocapture nofree readonly [[X:%.*]]) #[[ATTR0]] {
 ; IS__TUNIT_OPM-NEXT:  entry:
 ; IS__TUNIT_OPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
 ; IS__TUNIT_OPM-NEXT:    [[TMP1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 0
@@ -104,7 +104,7 @@
 ;
 ; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@test
-; IS__TUNIT_NPM-SAME: (i32* nocapture nofree readonly align 4 [[X:%.*]]) #[[ATTR0]] {
+; IS__TUNIT_NPM-SAME: (i32* nocapture nofree readonly [[X:%.*]]) #[[ATTR0]] {
 ; IS__TUNIT_NPM-NEXT:  entry:
 ; IS__TUNIT_NPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 0
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/basictest.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/basictest.ll
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/basictest.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/basictest.ll
@@ -85,7 +85,7 @@
 ; IS__CGSCC_NPM-SAME: (i32 [[TMP0:%.*]]) #[[ATTR0]] {
 ; IS__CGSCC_NPM-NEXT:    [[B_PRIV:%.*]] = alloca i32, align 4
 ; IS__CGSCC_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[B_PRIV]], align 4
-; IS__CGSCC_NPM-NEXT:    [[C:%.*]] = call i32 @test(i32 undef, i32 undef) #[[ATTR1:[0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[C:%.*]] = call i32 @test(i32 undef, i32 2) #[[ATTR1:[0-9]+]]
 ; IS__CGSCC_NPM-NEXT:    ret i32 undef
 ;
   %A = alloca i32
@@ -95,7 +95,7 @@
 }
 
 define i32 @callercaller() {
-; IS__TUNIT_OPM: Function Attrs: nofree nosync nounwind readnone
+; IS__TUNIT_OPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_OPM-LABEL: define {{[^@]+}}@callercaller
 ; IS__TUNIT_OPM-SAME: () #[[ATTR2:[0-9]+]] {
 ; IS__TUNIT_OPM-NEXT:    [[B:%.*]] = alloca i32, align 4
@@ -103,7 +103,7 @@
 ; IS__TUNIT_OPM-NEXT:    [[X:%.*]] = call i32 @caller(i32* noalias nocapture nofree noundef nonnull readonly align 4 dereferenceable(4) [[B]]) #[[ATTR4:[0-9]+]]
 ; IS__TUNIT_OPM-NEXT:    ret i32 [[X]]
 ;
-; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone
+; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@callercaller
 ; IS__TUNIT_NPM-SAME: () #[[ATTR2:[0-9]+]] {
 ; IS__TUNIT_NPM-NEXT:    [[B:%.*]] = alloca i32, align 4
@@ -134,7 +134,7 @@
 ;.
 ; IS__TUNIT____: attributes #[[ATTR0:[0-9]+]] = { argmemonly nofree nosync nounwind readonly willreturn }
 ; IS__TUNIT____: attributes #[[ATTR1:[0-9]+]] = { argmemonly nofree nosync nounwind willreturn }
-; IS__TUNIT____: attributes #[[ATTR2:[0-9]+]] = { nofree nosync nounwind readnone }
+; IS__TUNIT____: attributes #[[ATTR2:[0-9]+]] = { nofree nosync nounwind readnone willreturn }
 ; IS__TUNIT____: attributes #[[ATTR3:[0-9]+]] = { nofree nosync nounwind readonly willreturn }
 ; IS__TUNIT____: attributes #[[ATTR4:[0-9]+]] = { nofree nosync nounwind willreturn }
 ;.
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/byval-2.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/byval-2.ll
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/byval-2.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/byval-2.ll
@@ -49,7 +49,7 @@
 ;
 ; IS__TUNIT____: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@test
-; IS__TUNIT____-SAME: (i32* nocapture nofree readonly align 4 [[X:%.*]]) #[[ATTR0:[0-9]+]] {
+; IS__TUNIT____-SAME: (i32* nocapture nofree readonly [[X:%.*]]) #[[ATTR0:[0-9]+]] {
 ; IS__TUNIT____-NEXT:  entry:
 ; IS__TUNIT____-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
 ; IS__TUNIT____-NEXT:    [[TMP1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 0
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll
@@ -150,15 +150,15 @@
 ; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 0
 ; IS__TUNIT_NPM-NEXT:    store i32 1, i32* [[TMP1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[S_CAST:%.*]] = bitcast %struct.ss* [[S]] to i32*
-; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i32, i32* [[S_CAST]], align 8
-; IS__TUNIT_NPM-NEXT:    [[S_0_1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i64, i64* [[S_0_1]], align 8
-; IS__TUNIT_NPM-NEXT:    [[C0:%.*]] = call i32 @f(i32 [[TMP0]], i64 [[TMP1]]) #[[ATTR0]]
 ; IS__TUNIT_NPM-NEXT:    [[S_CAST1:%.*]] = bitcast %struct.ss* [[S]] to i32*
-; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[S_CAST1]], align 32
+; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i32, i32* [[S_CAST1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[S_0_12:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i64, i64* [[S_0_12]], align 32
+; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i64, i64* [[S_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    [[C0:%.*]] = call i32 @f(i32 [[TMP0]], i64 [[TMP1]]) #[[ATTR0]]
+; IS__TUNIT_NPM-NEXT:    [[S_CAST:%.*]] = bitcast %struct.ss* [[S]] to i32*
+; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[S_CAST]], align 32
+; IS__TUNIT_NPM-NEXT:    [[S_0_1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i64, i64* [[S_0_1]], align 32
 ; IS__TUNIT_NPM-NEXT:    [[C1:%.*]] = call i32 @g(i32 [[TMP2]], i64 [[TMP3]]) #[[ATTR0]]
 ; IS__TUNIT_NPM-NEXT:    [[A:%.*]] = add i32 [[C0]], [[C1]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[A]]
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll
@@ -15,15 +15,25 @@
 }
 
 define internal i32 @test(i32* %X, i32* %Y) {
-; NOT_CGSCC_NPM: Function Attrs: argmemonly nofree nosync nounwind willreturn writeonly
-; NOT_CGSCC_NPM-LABEL: define {{[^@]+}}@test
-; NOT_CGSCC_NPM-SAME: (i32* noalias nocapture nofree noundef writeonly align 4 [[X:%.*]]) #[[ATTR0:[0-9]+]] {
-; NOT_CGSCC_NPM-NEXT:    br i1 true, label [[LIVE:%.*]], label [[DEAD:%.*]]
-; NOT_CGSCC_NPM:       live:
-; NOT_CGSCC_NPM-NEXT:    store i32 0, i32* [[X]], align 4
-; NOT_CGSCC_NPM-NEXT:    ret i32 undef
-; NOT_CGSCC_NPM:       dead:
-; NOT_CGSCC_NPM-NEXT:    unreachable
+; IS__TUNIT____: Function Attrs: argmemonly nofree nosync nounwind willreturn writeonly
+; IS__TUNIT____-LABEL: define {{[^@]+}}@test
+; IS__TUNIT____-SAME: (i32* noalias nocapture nofree noundef writeonly align 4 [[X:%.*]], i32* noalias nocapture nofree noundef readnone align 4 [[Y:%.*]]) #[[ATTR0:[0-9]+]] {
+; IS__TUNIT____-NEXT:    br i1 true, label [[LIVE:%.*]], label [[DEAD:%.*]]
+; IS__TUNIT____:       live:
+; IS__TUNIT____-NEXT:    store i32 0, i32* [[X]], align 4
+; IS__TUNIT____-NEXT:    ret i32 undef
+; IS__TUNIT____:       dead:
+; IS__TUNIT____-NEXT:    unreachable
+;
+; IS__CGSCC_OPM: Function Attrs: argmemonly nofree nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@test
+; IS__CGSCC_OPM-SAME: (i32* noalias nocapture nofree noundef writeonly align 4 [[X:%.*]]) #[[ATTR0:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:    br i1 true, label [[LIVE:%.*]], label [[DEAD:%.*]]
+; IS__CGSCC_OPM:       live:
+; IS__CGSCC_OPM-NEXT:    store i32 0, i32* [[X]], align 4
+; IS__CGSCC_OPM-NEXT:    ret i32 undef
+; IS__CGSCC_OPM:       dead:
+; IS__CGSCC_OPM-NEXT:    unreachable
 ;
 ; IS__CGSCC_NPM: Function Attrs: argmemonly nofree norecurse nosync nounwind willreturn writeonly
 ; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@test
@@ -46,12 +56,19 @@
 }
 
 define internal i32 @caller(i32* %B) {
-; NOT_CGSCC_NPM: Function Attrs: argmemonly nofree nosync nounwind willreturn writeonly
-; NOT_CGSCC_NPM-LABEL: define {{[^@]+}}@caller
-; NOT_CGSCC_NPM-SAME: (i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR0]] {
-; NOT_CGSCC_NPM-NEXT:    [[A:%.*]] = alloca i32, align 4
-; NOT_CGSCC_NPM-NEXT:    [[C:%.*]] = call i32 @test(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B]]) #[[ATTR2:[0-9]+]]
-; NOT_CGSCC_NPM-NEXT:    ret i32 undef
+; IS__TUNIT____: Function Attrs: argmemonly nofree nosync nounwind willreturn writeonly
+; IS__TUNIT____-LABEL: define {{[^@]+}}@caller
+; IS__TUNIT____-SAME: (i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR0]] {
+; IS__TUNIT____-NEXT:    [[A:%.*]] = alloca i32, align 4
+; IS__TUNIT____-NEXT:    [[C:%.*]] = call i32 @test(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B]], i32* noalias nocapture nofree noundef nonnull readnone align 4 dereferenceable(4) [[A]]) #[[ATTR2:[0-9]+]]
+; IS__TUNIT____-NEXT:    ret i32 undef
+;
+; IS__CGSCC_OPM: Function Attrs: argmemonly nofree nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@caller
+; IS__CGSCC_OPM-SAME: (i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR0]] {
+; IS__CGSCC_OPM-NEXT:    [[A:%.*]] = alloca i32, align 4
+; IS__CGSCC_OPM-NEXT:    [[C:%.*]] = call i32 @test(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B]]) #[[ATTR2:[0-9]+]]
+; IS__CGSCC_OPM-NEXT:    ret i32 undef
 ;
 ; IS__CGSCC_NPM: Function Attrs: argmemonly nofree norecurse nosync nounwind willreturn writeonly
 ; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@caller
diff --git a/llvm/test/Transforms/Attributor/allow_list.ll b/llvm/test/Transforms/Attributor/allow_list.ll
--- a/llvm/test/Transforms/Attributor/allow_list.ll
+++ b/llvm/test/Transforms/Attributor/allow_list.ll
@@ -27,11 +27,6 @@
 ; CHECK_DISABLED_FUNCTION-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[A]], 100
 ; CHECK_DISABLED_FUNCTION-NEXT:    [[TMP2:%.*]] = zext i1 [[TMP1]] to i32
 ; CHECK_DISABLED_FUNCTION-NEXT:    ret i32 [[TMP2]]
-;
-; CHECK_ENABLED_FUNCTION: Function Attrs: noinline nounwind readnone uwtable
-; CHECK_ENABLED_FUNCTION-LABEL: define {{[^@]+}}@range_test
-; CHECK_ENABLED_FUNCTION-SAME: () #[[ATTR0:[0-9]+]] {
-; CHECK_ENABLED_FUNCTION-NEXT:    ret i32 1
 ;
   %1 = icmp sgt i32 %a, 100
   %2 = zext i1 %1 to i32
@@ -59,7 +54,7 @@
 ;
 ; CHECK_ENABLED_FUNCTION: Function Attrs: nofree noinline nosync nounwind readnone uwtable willreturn
 ; CHECK_ENABLED_FUNCTION-LABEL: define {{[^@]+}}@range_use1
-; CHECK_ENABLED_FUNCTION-SAME: () #[[ATTR1:[0-9]+]] {
+; CHECK_ENABLED_FUNCTION-SAME: () #[[ATTR0:[0-9]+]] {
 ; CHECK_ENABLED_FUNCTION-NEXT:    ret i32 1
 ;
   %1 = call i32 @range_test(i32 123)
@@ -87,9 +82,8 @@
 ;
 ; CHECK_ENABLED_FUNCTION: Function Attrs: noinline nounwind uwtable
 ; CHECK_ENABLED_FUNCTION-LABEL: define {{[^@]+}}@range_use2
-; CHECK_ENABLED_FUNCTION-SAME: () #[[ATTR2:[0-9]+]] {
-; CHECK_ENABLED_FUNCTION-NEXT:    [[TMP1:%.*]] = call i32 @range_test()
-; CHECK_ENABLED_FUNCTION-NEXT:    ret i32 [[TMP1]]
+; CHECK_ENABLED_FUNCTION-SAME: () #[[ATTR1:[0-9]+]] {
+; CHECK_ENABLED_FUNCTION-NEXT:    ret i32 1
 ;
   %1 = call i32 @range_test(i32 123)
   ret i32 %1
@@ -103,7 +97,6 @@
 ;.
 ; CHECK_DISABLED_FUNCTION: attributes #[[ATTR0]] = { noinline nounwind uwtable }
 ;.
-; CHECK_ENABLED_FUNCTION: attributes #[[ATTR0]] = { noinline nounwind readnone uwtable }
-; CHECK_ENABLED_FUNCTION: attributes #[[ATTR1]] = { nofree noinline nosync nounwind readnone uwtable willreturn }
-; CHECK_ENABLED_FUNCTION: attributes #[[ATTR2]] = { noinline nounwind uwtable }
+; CHECK_ENABLED_FUNCTION: attributes #[[ATTR0]] = { nofree noinline nosync nounwind readnone uwtable willreturn }
+; CHECK_ENABLED_FUNCTION: attributes #[[ATTR1]] = { noinline nounwind uwtable }
 ;.
diff --git a/llvm/test/Transforms/Attributor/internal-noalias.ll b/llvm/test/Transforms/Attributor/internal-noalias.ll
--- a/llvm/test/Transforms/Attributor/internal-noalias.ll
+++ b/llvm/test/Transforms/Attributor/internal-noalias.ll
@@ -112,7 +112,7 @@
 define dso_local i32 @visible_local(i32* %A) #0 {
 ; IS__TUNIT____: Function Attrs: argmemonly nofree noinline nosync nounwind uwtable willreturn
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@visible_local
-; IS__TUNIT____-SAME: (i32* nocapture nofree readonly [[A:%.*]]) #[[ATTR1:[0-9]+]] {
+; IS__TUNIT____-SAME: (i32* nocapture nofree readonly align 4 [[A:%.*]]) #[[ATTR1:[0-9]+]] {
 ; IS__TUNIT____-NEXT:  entry:
 ; IS__TUNIT____-NEXT:    [[B:%.*]] = alloca i32, align 4
 ; IS__TUNIT____-NEXT:    store i32 5, i32* [[B]], align 4
diff --git a/llvm/test/Transforms/Attributor/memory_locations.ll b/llvm/test/Transforms/Attributor/memory_locations.ll
--- a/llvm/test/Transforms/Attributor/memory_locations.ll
+++ b/llvm/test/Transforms/Attributor/memory_locations.ll
@@ -664,7 +664,7 @@
 ; IS__TUNIT____: Function Attrs: nofree nosync nounwind readnone
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@readnone_caller2
 ; IS__TUNIT____-SAME: (i1 [[C:%.*]]) #[[ATTR9]] {
-; IS__TUNIT____-NEXT:    [[R:%.*]] = call noundef i8 @recursive_readnone_internal2(i8* undef, i1 [[C]]) #[[ATTR10]], !range [[RNG0]]
+; IS__TUNIT____-NEXT:    [[R:%.*]] = call i8 @recursive_readnone_internal2(i8* undef, i1 [[C]]) #[[ATTR10]], !range [[RNG0]]
 ; IS__TUNIT____-NEXT:    ret i8 [[R]]
 ;
 ; IS__CGSCC____: Function Attrs: nofree nosync nounwind readnone
diff --git a/llvm/test/Transforms/Attributor/read_write_returned_arguments_scc.ll b/llvm/test/Transforms/Attributor/read_write_returned_arguments_scc.ll
--- a/llvm/test/Transforms/Attributor/read_write_returned_arguments_scc.ll
+++ b/llvm/test/Transforms/Attributor/read_write_returned_arguments_scc.ll
@@ -40,8 +40,8 @@
 ; IS__TUNIT____-SAME: (i32* nofree [[N0:%.*]], i32* nofree [[R0:%.*]], i32* nofree returned [[W0:%.*]]) #[[ATTR0:[0-9]+]] {
 ; IS__TUNIT____-NEXT:  entry:
 ; IS__TUNIT____-NEXT:    [[CALL:%.*]] = call i32* @internal_ret0_nw(i32* nofree [[N0]], i32* nofree [[W0]]) #[[ATTR2:[0-9]+]]
-; IS__TUNIT____-NEXT:    [[CALL1:%.*]] = call i32* @internal_ret1_rrw(i32* nofree align 4 [[R0]], i32* nofree [[R0]], i32* nofree [[W0]]) #[[ATTR2]]
-; IS__TUNIT____-NEXT:    [[CALL2:%.*]] = call i32* @external_sink_ret2_nrw(i32* nofree [[N0]], i32* nocapture nofree readonly [[R0]], i32* nofree writeonly "no-capture-maybe-returned" [[W0]]) #[[ATTR2]]
+; IS__TUNIT____-NEXT:    [[CALL1:%.*]] = call i32* @internal_ret1_rrw(i32* nofree align 4 [[R0]], i32* nofree align 4 [[R0]], i32* nofree [[W0]]) #[[ATTR2]]
+; IS__TUNIT____-NEXT:    [[CALL2:%.*]] = call i32* @external_sink_ret2_nrw(i32* nofree [[N0]], i32* nocapture nofree readonly align 4 [[R0]], i32* nofree writeonly "no-capture-maybe-returned" [[W0]]) #[[ATTR2]]
 ; IS__TUNIT____-NEXT:    [[CALL3:%.*]] = call i32* @internal_ret1_rw(i32* nofree align 4 [[R0]], i32* nofree [[W0]]) #[[ATTR2]]
 ; IS__TUNIT____-NEXT:    ret i32* [[W0]]
 ;
@@ -143,7 +143,7 @@
 define internal i32* @internal_ret1_rrw(i32* %r0, i32* %r1, i32* %w0) {
 ; IS__TUNIT____: Function Attrs: argmemonly nofree nosync nounwind
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@internal_ret1_rrw
-; IS__TUNIT____-SAME: (i32* nofree noundef nonnull align 4 dereferenceable(4) [[R0:%.*]], i32* nofree [[R1:%.*]], i32* nofree [[W0:%.*]]) #[[ATTR0]] {
+; IS__TUNIT____-SAME: (i32* nofree noundef nonnull align 4 dereferenceable(4) [[R0:%.*]], i32* nofree align 4 [[R1:%.*]], i32* nofree [[W0:%.*]]) #[[ATTR0]] {
 ; IS__TUNIT____-NEXT:  entry:
 ; IS__TUNIT____-NEXT:    [[TMP0:%.*]] = load i32, i32* [[R0]], align 4
 ; IS__TUNIT____-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP0]], 0
diff --git a/llvm/test/Transforms/Attributor/readattrs.ll b/llvm/test/Transforms/Attributor/readattrs.ll
--- a/llvm/test/Transforms/Attributor/readattrs.ll
+++ b/llvm/test/Transforms/Attributor/readattrs.ll
@@ -384,6 +384,7 @@
 ; IS__TUNIT____-SAME: (i8* nocapture readonly [[READ_ONLY:%.*]]) {
 ; IS__TUNIT____-NEXT:    call void @byval_not_readonly_1(i8* nocapture readonly byval(i8) [[READ_ONLY]]) #[[ATTR2]]
 ; IS__TUNIT____-NEXT:    call void @byval_not_readnone_1(i8* noalias nocapture readnone byval(i8) [[READ_ONLY]])
+; IS__TUNIT____-NEXT:    call void @byval_no_fnarg(i8* nocapture nofree readonly byval(i8) [[READ_ONLY]]) #[[ATTR13:[0-9]+]]
 ; IS__TUNIT____-NEXT:    ret void
 ;
 ; IS__CGSCC____: Function Attrs: readonly
@@ -486,6 +487,7 @@
 ; IS__TUNIT____: attributes #[[ATTR10]] = { willreturn writeonly }
 ; IS__TUNIT____: attributes #[[ATTR11]] = { readonly willreturn }
 ; IS__TUNIT____: attributes #[[ATTR12]] = { nounwind }
+; IS__TUNIT____: attributes #[[ATTR13]] = { nounwind writeonly }
 ;.
 ; IS__CGSCC____: attributes #[[ATTR0]] = { nofree norecurse nosync nounwind willreturn writeonly }
 ; IS__CGSCC____: attributes #[[ATTR1]] = { nofree norecurse nosync nounwind readnone willreturn }
diff --git a/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll b/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
--- a/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
+++ b/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
@@ -169,41 +169,77 @@
 ; IS__TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I12]]) #[[ATTR6]]
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____: Function Attrs: argmemonly nofree nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@local_alloca_simplifiable_1
-; IS__CGSCC____-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]]) #[[ATTR1:[0-9]+]] {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    [[S:%.*]] = alloca [[STRUCT_S]], align 4
-; IS__CGSCC____-NEXT:    [[I:%.*]] = bitcast %struct.S* [[S]] to i8*
-; IS__CGSCC____-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I]]) #[[ATTR8:[0-9]+]]
-; IS__CGSCC____-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
-; IS__CGSCC____-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
-; IS__CGSCC____-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(24) [[I1]], i32 noundef 1) #[[ATTR9:[0-9]+]]
-; IS__CGSCC____-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) [[I2]], i32 noundef 2) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(16) [[I3]], i32 noundef 3) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    [[F12:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    store float 0x3FF19999A0000000, float* [[F12]], align 4, !tbaa [[TBAA7:![0-9]+]]
-; IS__CGSCC____-NEXT:    [[MUL:%.*]] = fmul float 0x40019999A0000000, 2.000000e+00
-; IS__CGSCC____-NEXT:    [[F24:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
-; IS__CGSCC____-NEXT:    store float [[MUL]], float* [[F24]], align 4, !tbaa [[TBAA10:![0-9]+]]
-; IS__CGSCC____-NEXT:    [[ADD:%.*]] = fadd float 0x400A666660000000, 0x3FF19999A0000000
-; IS__CGSCC____-NEXT:    [[F37:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
-; IS__CGSCC____-NEXT:    store float [[ADD]], float* [[F37]], align 4, !tbaa [[TBAA11:![0-9]+]]
-; IS__CGSCC____-NEXT:    [[I19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    store i32 1, i32* [[I19]], align 4, !tbaa [[TBAA12:![0-9]+]]
-; IS__CGSCC____-NEXT:    [[MUL11:%.*]] = shl nsw i32 2, 1
-; IS__CGSCC____-NEXT:    [[I212:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
-; IS__CGSCC____-NEXT:    store i32 [[MUL11]], i32* [[I212]], align 4, !tbaa [[TBAA13:![0-9]+]]
-; IS__CGSCC____-NEXT:    [[ADD15:%.*]] = add nsw i32 3, 1
-; IS__CGSCC____-NEXT:    [[I316:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
-; IS__CGSCC____-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14:![0-9]+]]
-; IS__CGSCC____-NEXT:    [[I12:%.*]] = bitcast %struct.S* [[S]] to i8*
-; IS__CGSCC____-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I12]]) #[[ATTR8]]
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: argmemonly nofree nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@local_alloca_simplifiable_1
+; IS__CGSCC_OPM-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]]) #[[ATTR1:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_S]], align 4
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = bitcast %struct.S* [[S]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I]]) #[[ATTR11:[0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
+; IS__CGSCC_OPM-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
+; IS__CGSCC_OPM-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(24) [[I1]], i32 noundef 1) #[[ATTR12:[0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) [[I2]], i32 noundef 2) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(16) [[I3]], i32 noundef 3) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    [[F12:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    store float 0x3FF19999A0000000, float* [[F12]], align 4, !tbaa [[TBAA7:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[MUL:%.*]] = fmul float 0x40019999A0000000, 2.000000e+00
+; IS__CGSCC_OPM-NEXT:    [[F24:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
+; IS__CGSCC_OPM-NEXT:    store float [[MUL]], float* [[F24]], align 4, !tbaa [[TBAA10:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[ADD:%.*]] = fadd float 0x400A666660000000, 0x3FF19999A0000000
+; IS__CGSCC_OPM-NEXT:    [[F37:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
+; IS__CGSCC_OPM-NEXT:    store float [[ADD]], float* [[F37]], align 4, !tbaa [[TBAA11:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[I19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    store i32 1, i32* [[I19]], align 4, !tbaa [[TBAA12:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[MUL11:%.*]] = shl nsw i32 2, 1
+; IS__CGSCC_OPM-NEXT:    [[I212:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
+; IS__CGSCC_OPM-NEXT:    store i32 [[MUL11]], i32* [[I212]], align 4, !tbaa [[TBAA13:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[ADD15:%.*]] = add nsw i32 3, 1
+; IS__CGSCC_OPM-NEXT:    [[I316:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
+; IS__CGSCC_OPM-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    [[I12:%.*]] = bitcast %struct.S* [[S]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I12]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: argmemonly nofree nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@local_alloca_simplifiable_1
+; IS__CGSCC_NPM-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]]) #[[ATTR1:[0-9]+]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_S]], align 4
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = bitcast %struct.S* [[S]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I]]) #[[ATTR8:[0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
+; IS__CGSCC_NPM-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
+; IS__CGSCC_NPM-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(24) [[I1]], i32 noundef 1) #[[ATTR9:[0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) [[I2]], i32 noundef 2) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(16) [[I3]], i32 noundef 3) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    [[F12:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    store float 0x3FF19999A0000000, float* [[F12]], align 4, !tbaa [[TBAA7:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[MUL:%.*]] = fmul float 0x40019999A0000000, 2.000000e+00
+; IS__CGSCC_NPM-NEXT:    [[F24:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
+; IS__CGSCC_NPM-NEXT:    store float [[MUL]], float* [[F24]], align 4, !tbaa [[TBAA10:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[ADD:%.*]] = fadd float 0x400A666660000000, 0x3FF19999A0000000
+; IS__CGSCC_NPM-NEXT:    [[F37:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
+; IS__CGSCC_NPM-NEXT:    store float [[ADD]], float* [[F37]], align 4, !tbaa [[TBAA11:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[I19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    store i32 1, i32* [[I19]], align 4, !tbaa [[TBAA12:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[MUL11:%.*]] = shl nsw i32 2, 1
+; IS__CGSCC_NPM-NEXT:    [[I212:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
+; IS__CGSCC_NPM-NEXT:    store i32 [[MUL11]], i32* [[I212]], align 4, !tbaa [[TBAA13:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[ADD15:%.*]] = add nsw i32 3, 1
+; IS__CGSCC_NPM-NEXT:    [[I316:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
+; IS__CGSCC_NPM-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    [[I12:%.*]] = bitcast %struct.S* [[S]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 24, i8* nocapture nofree noundef nonnull align 4 dereferenceable(24) [[I12]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
 entry:
   %s = alloca %struct.S, align 4
@@ -436,7 +472,9 @@
 ; IS__TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 1024, i8* nocapture nofree noundef nonnull align 16 dereferenceable(1024) [[I24]]) #[[ATTR6]]
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@local_alloca_simplifiable_2() {
+; IS__CGSCC_OPM: Function Attrs: nofree nosync nounwind
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@local_alloca_simplifiable_2
+; IS__CGSCC_OPM-SAME: () #[[ATTR2:[0-9]+]] {
 ; IS__CGSCC_OPM-NEXT:  entry:
 ; IS__CGSCC_OPM-NEXT:    [[BYTES:%.*]] = alloca [1024 x i8], align 16
 ; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 0
@@ -451,11 +489,10 @@
 ; IS__CGSCC_OPM:       for.body:
 ; IS__CGSCC_OPM-NEXT:    [[I15:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
 ; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 [[I15]]
-; IS__CGSCC_OPM-NEXT:    store i8 0, i8* [[ARRAYIDX]], align 2, !tbaa [[TBAA15:![0-9]+]]
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC]]
 ; IS__CGSCC_OPM:       for.inc:
 ; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
 ; IS__CGSCC_OPM:       for.end:
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2:%.*]]
 ; IS__CGSCC_OPM:       for.cond2:
@@ -469,11 +506,10 @@
 ; IS__CGSCC_OPM-NEXT:    [[I17:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
 ; IS__CGSCC_OPM-NEXT:    [[I18:%.*]] = or i64 [[I17]], 1
 ; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[I16]], i64 [[I18]]
-; IS__CGSCC_OPM-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX8]], align 4, !tbaa [[TBAA18:![0-9]+]]
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC9]]
 ; IS__CGSCC_OPM:       for.inc9:
 ; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
-; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP19:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP17:![0-9]+]]
 ; IS__CGSCC_OPM:       for.end11:
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13:%.*]]
 ; IS__CGSCC_OPM:       for.cond13:
@@ -487,17 +523,15 @@
 ; IS__CGSCC_OPM-NEXT:    [[I20:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
 ; IS__CGSCC_OPM-NEXT:    [[I21:%.*]] = add nuw nsw i64 [[I20]], 2
 ; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds i64, i64* [[I19]], i64 [[I21]]
-; IS__CGSCC_OPM-NEXT:    store i64 0, i64* [[ARRAYIDX21]], align 16, !tbaa [[TBAA20:![0-9]+]]
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC22]]
 ; IS__CGSCC_OPM:       for.inc22:
 ; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
-; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP22:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP18:![0-9]+]]
 ; IS__CGSCC_OPM:       for.end24:
 ; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 1023
-; IS__CGSCC_OPM-NEXT:    store i8 0, i8* [[ARRAYIDX25]], align 1, !tbaa [[TBAA15]]
 ; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 500
 ; IS__CGSCC_OPM-NEXT:    [[I22:%.*]] = bitcast i8* [[ARRAYIDX26]] to i32*
-; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nofree noundef nonnull writeonly align 4 dereferenceable(4) [[I22]], i32 noundef 0) #[[ATTR10:[0-9]+]]
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(524) [[I22]], i32 noundef 0) #[[ATTR13:[0-9]+]]
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND28:%.*]]
 ; IS__CGSCC_OPM:       for.cond28:
 ; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC36:%.*]] ], [ 0, [[FOR_END24]] ]
@@ -506,20 +540,20 @@
 ; IS__CGSCC_OPM:       for.cond.cleanup30:
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_END38:%.*]]
 ; IS__CGSCC_OPM:       for.body31:
-; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX33:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC_OPM-NEXT:    [[I23:%.*]] = load i8, i8* [[ARRAYIDX33]], align 1, !tbaa [[TBAA15]]
 ; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX35:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC_OPM-NEXT:    store i8 [[I23]], i8* [[ARRAYIDX35]], align 1, !tbaa [[TBAA15]]
+; IS__CGSCC_OPM-NEXT:    store i8 0, i8* [[ARRAYIDX35]], align 1, !tbaa [[TBAA19:![0-9]+]]
 ; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC36]]
 ; IS__CGSCC_OPM:       for.inc36:
 ; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
-; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND28]], !llvm.loop [[LOOP23:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND28]], !llvm.loop [[LOOP20:![0-9]+]]
 ; IS__CGSCC_OPM:       for.end38:
 ; IS__CGSCC_OPM-NEXT:    [[I24:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 0
-; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 1024, i8* nofree noundef nonnull [[I24]])
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 1024, i8* nocapture nofree noundef nonnull align 16 dereferenceable(1024) [[I24]])
 ; IS__CGSCC_OPM-NEXT:    ret void
 ;
-; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@local_alloca_simplifiable_2() {
+; IS__CGSCC_NPM: Function Attrs: nofree nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@local_alloca_simplifiable_2
+; IS__CGSCC_NPM-SAME: () #[[ATTR2:[0-9]+]] {
 ; IS__CGSCC_NPM-NEXT:  entry:
 ; IS__CGSCC_NPM-NEXT:    [[BYTES:%.*]] = alloca [1024 x i8], align 16
 ; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 0
@@ -534,11 +568,10 @@
 ; IS__CGSCC_NPM:       for.body:
 ; IS__CGSCC_NPM-NEXT:    [[I15:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
 ; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 [[I15]]
-; IS__CGSCC_NPM-NEXT:    store i8 0, i8* [[ARRAYIDX]], align 2, !tbaa [[TBAA15:![0-9]+]]
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC]]
 ; IS__CGSCC_NPM:       for.inc:
 ; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP16:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
 ; IS__CGSCC_NPM:       for.end:
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2:%.*]]
 ; IS__CGSCC_NPM:       for.cond2:
@@ -552,11 +585,10 @@
 ; IS__CGSCC_NPM-NEXT:    [[I17:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
 ; IS__CGSCC_NPM-NEXT:    [[I18:%.*]] = or i64 [[I17]], 1
 ; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[I16]], i64 [[I18]]
-; IS__CGSCC_NPM-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX8]], align 4, !tbaa [[TBAA18:![0-9]+]]
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC9]]
 ; IS__CGSCC_NPM:       for.inc9:
 ; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
-; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP19:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP17:![0-9]+]]
 ; IS__CGSCC_NPM:       for.end11:
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13:%.*]]
 ; IS__CGSCC_NPM:       for.cond13:
@@ -570,17 +602,15 @@
 ; IS__CGSCC_NPM-NEXT:    [[I20:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
 ; IS__CGSCC_NPM-NEXT:    [[I21:%.*]] = add nuw nsw i64 [[I20]], 2
 ; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX21:%.*]] = getelementptr inbounds i64, i64* [[I19]], i64 [[I21]]
-; IS__CGSCC_NPM-NEXT:    store i64 0, i64* [[ARRAYIDX21]], align 16, !tbaa [[TBAA20:![0-9]+]]
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC22]]
 ; IS__CGSCC_NPM:       for.inc22:
 ; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
-; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP22:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP18:![0-9]+]]
 ; IS__CGSCC_NPM:       for.end24:
 ; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 1023
-; IS__CGSCC_NPM-NEXT:    store i8 0, i8* [[ARRAYIDX25]], align 1, !tbaa [[TBAA15]]
 ; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 500
 ; IS__CGSCC_NPM-NEXT:    [[I22:%.*]] = bitcast i8* [[ARRAYIDX26]] to i32*
-; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nofree noundef nonnull writeonly align 4 dereferenceable(4) [[I22]], i32 noundef 0) #[[ATTR10:[0-9]+]]
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(524) [[I22]], i32 noundef 0) #[[ATTR9]]
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND28:%.*]]
 ; IS__CGSCC_NPM:       for.cond28:
 ; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC36:%.*]] ], [ 0, [[FOR_END24]] ]
@@ -589,17 +619,15 @@
 ; IS__CGSCC_NPM:       for.cond.cleanup30:
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_END38:%.*]]
 ; IS__CGSCC_NPM:       for.body31:
-; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX33:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC_NPM-NEXT:    [[I23:%.*]] = load i8, i8* [[ARRAYIDX33]], align 1, !tbaa [[TBAA15]]
 ; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX35:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC_NPM-NEXT:    store i8 [[I23]], i8* [[ARRAYIDX35]], align 1, !tbaa [[TBAA15]]
+; IS__CGSCC_NPM-NEXT:    store i8 0, i8* [[ARRAYIDX35]], align 1, !tbaa [[TBAA19:![0-9]+]]
 ; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC36]]
 ; IS__CGSCC_NPM:       for.inc36:
 ; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
-; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND28]], !llvm.loop [[LOOP23:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND28]], !llvm.loop [[LOOP20:![0-9]+]]
 ; IS__CGSCC_NPM:       for.end38:
 ; IS__CGSCC_NPM-NEXT:    [[I24:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* [[BYTES]], i64 0, i64 0
-; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 1024, i8* nofree noundef nonnull [[I24]])
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 1024, i8* nocapture nofree noundef nonnull align 16 dereferenceable(1024) [[I24]]) #[[ATTR8]]
 ; IS__CGSCC_NPM-NEXT:    ret void
 ;
 entry:
@@ -726,7 +754,7 @@
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@local_alloca_simplifiable_3
-; IS__CGSCC____-SAME: () #[[ATTR2:[0-9]+]] {
+; IS__CGSCC____-SAME: () #[[ATTR3:[0-9]+]] {
 ; IS__CGSCC____-NEXT:    [[A:%.*]] = alloca i32, align 4
 ; IS__CGSCC____-NEXT:    store i32 1, i32* [[A]], align 4
 ; IS__CGSCC____-NEXT:    br label [[SPLIT:%.*]]
@@ -756,7 +784,7 @@
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@local_alloca_simplifiable_4
-; IS__CGSCC____-SAME: () #[[ATTR2]] {
+; IS__CGSCC____-SAME: () #[[ATTR3]] {
 ; IS__CGSCC____-NEXT:    ret i32 undef
 ;
   %A = alloca i32, align 4
@@ -807,23 +835,41 @@
 ; IS__TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I2]]) #[[ATTR6]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 5
 ;
-; IS__CGSCC____: Function Attrs: nofree nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@multi_obj_simplifiable_1
-; IS__CGSCC____-SAME: (i32 [[CND:%.*]]) #[[ATTR3:[0-9]+]] {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    [[L:%.*]] = alloca i32, align 4
-; IS__CGSCC____-NEXT:    [[I:%.*]] = bitcast i32* [[L]] to i8*
-; IS__CGSCC____-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8]]
-; IS__CGSCC____-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CND]], 0
-; IS__CGSCC____-NEXT:    br i1 [[TOBOOL_NOT]], label [[COND_FALSE:%.*]], label [[COND_TRUE:%.*]]
-; IS__CGSCC____:       cond.true:
-; IS__CGSCC____-NEXT:    br label [[COND_END:%.*]]
-; IS__CGSCC____:       cond.false:
-; IS__CGSCC____-NEXT:    br label [[COND_END]]
-; IS__CGSCC____:       cond.end:
-; IS__CGSCC____-NEXT:    [[I2:%.*]] = bitcast i32* [[L]] to i8*
-; IS__CGSCC____-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I2]]) #[[ATTR8]]
-; IS__CGSCC____-NEXT:    ret i32 5
+; IS__CGSCC_OPM: Function Attrs: nofree nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@multi_obj_simplifiable_1
+; IS__CGSCC_OPM-SAME: (i32 [[CND:%.*]]) #[[ATTR4:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    [[L:%.*]] = alloca i32, align 4
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CND]], 0
+; IS__CGSCC_OPM-NEXT:    br i1 [[TOBOOL_NOT]], label [[COND_FALSE:%.*]], label [[COND_TRUE:%.*]]
+; IS__CGSCC_OPM:       cond.true:
+; IS__CGSCC_OPM-NEXT:    br label [[COND_END:%.*]]
+; IS__CGSCC_OPM:       cond.false:
+; IS__CGSCC_OPM-NEXT:    br label [[COND_END]]
+; IS__CGSCC_OPM:       cond.end:
+; IS__CGSCC_OPM-NEXT:    [[I2:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I2]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    ret i32 5
+;
+; IS__CGSCC_NPM: Function Attrs: nofree nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@multi_obj_simplifiable_1
+; IS__CGSCC_NPM-SAME: (i32 [[CND:%.*]]) #[[ATTR2]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[L:%.*]] = alloca i32, align 4
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CND]], 0
+; IS__CGSCC_NPM-NEXT:    br i1 [[TOBOOL_NOT]], label [[COND_FALSE:%.*]], label [[COND_TRUE:%.*]]
+; IS__CGSCC_NPM:       cond.true:
+; IS__CGSCC_NPM-NEXT:    br label [[COND_END:%.*]]
+; IS__CGSCC_NPM:       cond.false:
+; IS__CGSCC_NPM-NEXT:    br label [[COND_END]]
+; IS__CGSCC_NPM:       cond.end:
+; IS__CGSCC_NPM-NEXT:    [[I2:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I2]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    ret i32 5
 ;
 entry:
   %L = alloca i32, align 4
@@ -893,23 +939,41 @@
 ; IS__TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR6]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 5
 ;
-; IS__CGSCC____: Function Attrs: nofree nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@multi_obj_simplifiable_2
-; IS__CGSCC____-SAME: (i32 [[CND:%.*]]) #[[ATTR3]] {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    [[L:%.*]] = alloca i32, align 4
-; IS__CGSCC____-NEXT:    [[I:%.*]] = bitcast i32* [[L]] to i8*
-; IS__CGSCC____-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8]]
-; IS__CGSCC____-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CND]], 0
-; IS__CGSCC____-NEXT:    br i1 [[TOBOOL_NOT]], label [[COND_FALSE:%.*]], label [[COND_TRUE:%.*]]
-; IS__CGSCC____:       cond.true:
-; IS__CGSCC____-NEXT:    br label [[COND_END:%.*]]
-; IS__CGSCC____:       cond.false:
-; IS__CGSCC____-NEXT:    br label [[COND_END]]
-; IS__CGSCC____:       cond.end:
-; IS__CGSCC____-NEXT:    [[I1:%.*]] = bitcast i32* [[L]] to i8*
-; IS__CGSCC____-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR8]]
-; IS__CGSCC____-NEXT:    ret i32 5
+; IS__CGSCC_OPM: Function Attrs: nofree nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@multi_obj_simplifiable_2
+; IS__CGSCC_OPM-SAME: (i32 [[CND:%.*]]) #[[ATTR4]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    [[L:%.*]] = alloca i32, align 4
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CND]], 0
+; IS__CGSCC_OPM-NEXT:    br i1 [[TOBOOL_NOT]], label [[COND_FALSE:%.*]], label [[COND_TRUE:%.*]]
+; IS__CGSCC_OPM:       cond.true:
+; IS__CGSCC_OPM-NEXT:    br label [[COND_END:%.*]]
+; IS__CGSCC_OPM:       cond.false:
+; IS__CGSCC_OPM-NEXT:    br label [[COND_END]]
+; IS__CGSCC_OPM:       cond.end:
+; IS__CGSCC_OPM-NEXT:    [[I1:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    ret i32 5
+;
+; IS__CGSCC_NPM: Function Attrs: nofree nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@multi_obj_simplifiable_2
+; IS__CGSCC_NPM-SAME: (i32 [[CND:%.*]]) #[[ATTR2]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[L:%.*]] = alloca i32, align 4
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CND]], 0
+; IS__CGSCC_NPM-NEXT:    br i1 [[TOBOOL_NOT]], label [[COND_FALSE:%.*]], label [[COND_TRUE:%.*]]
+; IS__CGSCC_NPM:       cond.true:
+; IS__CGSCC_NPM-NEXT:    br label [[COND_END:%.*]]
+; IS__CGSCC_NPM:       cond.false:
+; IS__CGSCC_NPM-NEXT:    br label [[COND_END]]
+; IS__CGSCC_NPM:       cond.end:
+; IS__CGSCC_NPM-NEXT:    [[I1:%.*]] = bitcast i32* [[L]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    ret i32 5
 ;
 entry:
   %L = alloca i32, align 4
@@ -1002,30 +1066,55 @@
 ; IS__TUNIT_NPM-NEXT:    store i32 [[ADD2]], i32* [[I3]], align 4, !tbaa [[TBAA14]]
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
-; IS__CGSCC____-LABEL: define {{[^@]+}}@static_global_simplifiable_1
-; IS__CGSCC____-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]]) #[[ATTR4:[0-9]+]] {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(24) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i32 0, i32 0), i32 noundef 1) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i64 0, i32 1), i32 noundef 2) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(16) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i64 0, i32 2), i32 noundef 3) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    store float 0x3FF19999A0000000, float* [[F1]], align 4, !tbaa [[TBAA7]]
-; IS__CGSCC____-NEXT:    [[MUL:%.*]] = fmul float 0x40019999A0000000, 2.000000e+00
-; IS__CGSCC____-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
-; IS__CGSCC____-NEXT:    store float [[MUL]], float* [[F2]], align 4, !tbaa [[TBAA10]]
-; IS__CGSCC____-NEXT:    [[ADD:%.*]] = fadd float 0x400A666660000000, 0x3FF19999A0000000
-; IS__CGSCC____-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
-; IS__CGSCC____-NEXT:    store float [[ADD]], float* [[F3]], align 4, !tbaa [[TBAA11]]
-; IS__CGSCC____-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    store i32 1, i32* [[I1]], align 4, !tbaa [[TBAA12]]
-; IS__CGSCC____-NEXT:    [[MUL1:%.*]] = shl nsw i32 2, 1
-; IS__CGSCC____-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
-; IS__CGSCC____-NEXT:    store i32 [[MUL1]], i32* [[I2]], align 4, !tbaa [[TBAA13]]
-; IS__CGSCC____-NEXT:    [[ADD2:%.*]] = add nsw i32 3, 1
-; IS__CGSCC____-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
-; IS__CGSCC____-NEXT:    store i32 [[ADD2]], i32* [[I3]], align 4, !tbaa [[TBAA14]]
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@static_global_simplifiable_1
+; IS__CGSCC_OPM-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]]) #[[ATTR5:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(24) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i32 0, i32 0), i32 noundef 1) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i64 0, i32 1), i32 noundef 2) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(16) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i64 0, i32 2), i32 noundef 3) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    store float 0x3FF19999A0000000, float* [[F1]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_OPM-NEXT:    [[MUL:%.*]] = fmul float 0x40019999A0000000, 2.000000e+00
+; IS__CGSCC_OPM-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
+; IS__CGSCC_OPM-NEXT:    store float [[MUL]], float* [[F2]], align 4, !tbaa [[TBAA10]]
+; IS__CGSCC_OPM-NEXT:    [[ADD:%.*]] = fadd float 0x400A666660000000, 0x3FF19999A0000000
+; IS__CGSCC_OPM-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
+; IS__CGSCC_OPM-NEXT:    store float [[ADD]], float* [[F3]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_OPM-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    store i32 1, i32* [[I1]], align 4, !tbaa [[TBAA12]]
+; IS__CGSCC_OPM-NEXT:    [[MUL1:%.*]] = shl nsw i32 2, 1
+; IS__CGSCC_OPM-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
+; IS__CGSCC_OPM-NEXT:    store i32 [[MUL1]], i32* [[I2]], align 4, !tbaa [[TBAA13]]
+; IS__CGSCC_OPM-NEXT:    [[ADD2:%.*]] = add nsw i32 3, 1
+; IS__CGSCC_OPM-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
+; IS__CGSCC_OPM-NEXT:    store i32 [[ADD2]], i32* [[I3]], align 4, !tbaa [[TBAA14]]
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@static_global_simplifiable_1
+; IS__CGSCC_NPM-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]]) #[[ATTR4:[0-9]+]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(24) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i32 0, i32 0), i32 noundef 1) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i64 0, i32 1), i32 noundef 2) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(16) getelementptr inbounds ([[STRUCT_S]], %struct.S* @Gs1, i64 0, i32 2), i32 noundef 3) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    store float 0x3FF19999A0000000, float* [[F1]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_NPM-NEXT:    [[MUL:%.*]] = fmul float 0x40019999A0000000, 2.000000e+00
+; IS__CGSCC_NPM-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
+; IS__CGSCC_NPM-NEXT:    store float [[MUL]], float* [[F2]], align 4, !tbaa [[TBAA10]]
+; IS__CGSCC_NPM-NEXT:    [[ADD:%.*]] = fadd float 0x400A666660000000, 0x3FF19999A0000000
+; IS__CGSCC_NPM-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
+; IS__CGSCC_NPM-NEXT:    store float [[ADD]], float* [[F3]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_NPM-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    store i32 1, i32* [[I1]], align 4, !tbaa [[TBAA12]]
+; IS__CGSCC_NPM-NEXT:    [[MUL1:%.*]] = shl nsw i32 2, 1
+; IS__CGSCC_NPM-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
+; IS__CGSCC_NPM-NEXT:    store i32 [[MUL1]], i32* [[I2]], align 4, !tbaa [[TBAA13]]
+; IS__CGSCC_NPM-NEXT:    [[ADD2:%.*]] = add nsw i32 3, 1
+; IS__CGSCC_NPM-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
+; IS__CGSCC_NPM-NEXT:    store i32 [[ADD2]], i32* [[I3]], align 4, !tbaa [[TBAA14]]
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
 entry:
   store float 0x3FF19999A0000000, float* getelementptr inbounds (%struct.S, %struct.S* @Gs1, i64 0, i32 3), align 4, !tbaa !7
@@ -1215,78 +1304,143 @@
 ; IS__TUNIT_NPM:       for.end35:
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____-LABEL: define {{[^@]+}}@static_global_simplifiable_2() {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    br label [[FOR_COND:%.*]]
-; IS__CGSCC____:       for.cond:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 0, [[ENTRY:%.*]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY:%.*]], label [[FOR_COND_CLEANUP:%.*]]
-; IS__CGSCC____:       for.cond.cleanup:
-; IS__CGSCC____-NEXT:    br label [[FOR_END:%.*]]
-; IS__CGSCC____:       for.body:
-; IS__CGSCC____-NEXT:    [[I:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
-; IS__CGSCC____-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 [[I]]
-; IS__CGSCC____-NEXT:    store i8 0, i8* [[ARRAYIDX]], align 2, !tbaa [[TBAA15:![0-9]+]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC]]
-; IS__CGSCC____:       for.inc:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]]
-; IS__CGSCC____:       for.end:
-; IS__CGSCC____-NEXT:    br label [[FOR_COND2:%.*]]
-; IS__CGSCC____:       for.cond2:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV2:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[FOR_INC9:%.*]] ], [ 0, [[FOR_END]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND6:%.*]] = icmp ne i64 [[INDVARS_IV2]], 10
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND6]], label [[FOR_BODY5:%.*]], label [[FOR_COND_CLEANUP4:%.*]]
-; IS__CGSCC____:       for.cond.cleanup4:
-; IS__CGSCC____-NEXT:    br label [[FOR_END11:%.*]]
-; IS__CGSCC____:       for.body5:
-; IS__CGSCC____-NEXT:    [[I15:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
-; IS__CGSCC____-NEXT:    [[I16:%.*]] = or i64 [[I15]], 1
-; IS__CGSCC____-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* bitcast ([1024 x i8]* @GBytes to float*), i64 [[I16]]
-; IS__CGSCC____-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX8]], align 4, !tbaa [[TBAA18:![0-9]+]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC9]]
-; IS__CGSCC____:       for.inc9:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP25:![0-9]+]]
-; IS__CGSCC____:       for.end11:
-; IS__CGSCC____-NEXT:    br label [[FOR_COND13:%.*]]
-; IS__CGSCC____:       for.cond13:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV7:%.*]] = phi i64 [ [[INDVARS_IV_NEXT8:%.*]], [[FOR_INC21:%.*]] ], [ 0, [[FOR_END11]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND11:%.*]] = icmp ne i64 [[INDVARS_IV7]], 20
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND11]], label [[FOR_BODY16:%.*]], label [[FOR_COND_CLEANUP15:%.*]]
-; IS__CGSCC____:       for.cond.cleanup15:
-; IS__CGSCC____-NEXT:    br label [[FOR_END23:%.*]]
-; IS__CGSCC____:       for.body16:
-; IS__CGSCC____-NEXT:    [[I17:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
-; IS__CGSCC____-NEXT:    [[I18:%.*]] = add nuw nsw i64 [[I17]], 2
-; IS__CGSCC____-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds i64, i64* bitcast ([1024 x i8]* @GBytes to i64*), i64 [[I18]]
-; IS__CGSCC____-NEXT:    store i64 0, i64* [[ARRAYIDX20]], align 16, !tbaa [[TBAA20:![0-9]+]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC21]]
-; IS__CGSCC____:       for.inc21:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP26:![0-9]+]]
-; IS__CGSCC____:       for.end23:
-; IS__CGSCC____-NEXT:    store i8 0, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 1023), align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nofree noundef nonnull writeonly align 4 dereferenceable(4) bitcast (i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 500) to i32*), i32 noundef 0) #[[ATTR10:[0-9]+]]
-; IS__CGSCC____-NEXT:    br label [[FOR_COND25:%.*]]
-; IS__CGSCC____:       for.cond25:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC33:%.*]] ], [ 0, [[FOR_END23]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND14:%.*]] = icmp ne i64 [[INDVARS_IV12]], 1024
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND14]], label [[FOR_BODY28:%.*]], label [[FOR_COND_CLEANUP27:%.*]]
-; IS__CGSCC____:       for.cond.cleanup27:
-; IS__CGSCC____-NEXT:    br label [[FOR_END35:%.*]]
-; IS__CGSCC____:       for.body28:
-; IS__CGSCC____-NEXT:    [[ARRAYIDX30:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC____-NEXT:    [[I19:%.*]] = load i8, i8* [[ARRAYIDX30]], align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC____-NEXT:    store i8 [[I19]], i8* [[ARRAYIDX32]], align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC33]]
-; IS__CGSCC____:       for.inc33:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND25]], !llvm.loop [[LOOP27:![0-9]+]]
-; IS__CGSCC____:       for.end35:
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@static_global_simplifiable_2
+; IS__CGSCC_OPM-SAME: () #[[ATTR6:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND:%.*]]
+; IS__CGSCC_OPM:       for.cond:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 0, [[ENTRY:%.*]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END:%.*]]
+; IS__CGSCC_OPM:       for.body:
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 [[I]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC]]
+; IS__CGSCC_OPM:       for.inc:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP21:![0-9]+]]
+; IS__CGSCC_OPM:       for.end:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2:%.*]]
+; IS__CGSCC_OPM:       for.cond2:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV2:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[FOR_INC9:%.*]] ], [ 0, [[FOR_END]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND6:%.*]] = icmp ne i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND6]], label [[FOR_BODY5:%.*]], label [[FOR_COND_CLEANUP4:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup4:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END11:%.*]]
+; IS__CGSCC_OPM:       for.body5:
+; IS__CGSCC_OPM-NEXT:    [[I15:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_OPM-NEXT:    [[I16:%.*]] = or i64 [[I15]], 1
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* bitcast ([1024 x i8]* @GBytes to float*), i64 [[I16]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC9]]
+; IS__CGSCC_OPM:       for.inc9:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP22:![0-9]+]]
+; IS__CGSCC_OPM:       for.end11:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13:%.*]]
+; IS__CGSCC_OPM:       for.cond13:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV7:%.*]] = phi i64 [ [[INDVARS_IV_NEXT8:%.*]], [[FOR_INC21:%.*]] ], [ 0, [[FOR_END11]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND11:%.*]] = icmp ne i64 [[INDVARS_IV7]], 20
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND11]], label [[FOR_BODY16:%.*]], label [[FOR_COND_CLEANUP15:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup15:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END23:%.*]]
+; IS__CGSCC_OPM:       for.body16:
+; IS__CGSCC_OPM-NEXT:    [[I17:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
+; IS__CGSCC_OPM-NEXT:    [[I18:%.*]] = add nuw nsw i64 [[I17]], 2
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds i64, i64* bitcast ([1024 x i8]* @GBytes to i64*), i64 [[I18]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC21]]
+; IS__CGSCC_OPM:       for.inc21:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP23:![0-9]+]]
+; IS__CGSCC_OPM:       for.end23:
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(524) bitcast (i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 500) to i32*), i32 noundef 0) #[[ATTR13]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND25:%.*]]
+; IS__CGSCC_OPM:       for.cond25:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC33:%.*]] ], [ 0, [[FOR_END23]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND14:%.*]] = icmp ne i64 [[INDVARS_IV12]], 1024
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND14]], label [[FOR_BODY28:%.*]], label [[FOR_COND_CLEANUP27:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup27:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END35:%.*]]
+; IS__CGSCC_OPM:       for.body28:
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
+; IS__CGSCC_OPM-NEXT:    store i8 0, i8* [[ARRAYIDX32]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC33]]
+; IS__CGSCC_OPM:       for.inc33:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND25]], !llvm.loop [[LOOP24:![0-9]+]]
+; IS__CGSCC_OPM:       for.end35:
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@static_global_simplifiable_2
+; IS__CGSCC_NPM-SAME: () #[[ATTR4]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND:%.*]]
+; IS__CGSCC_NPM:       for.cond:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 0, [[ENTRY:%.*]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END:%.*]]
+; IS__CGSCC_NPM:       for.body:
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 [[I]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC]]
+; IS__CGSCC_NPM:       for.inc:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP21:![0-9]+]]
+; IS__CGSCC_NPM:       for.end:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2:%.*]]
+; IS__CGSCC_NPM:       for.cond2:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV2:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[FOR_INC9:%.*]] ], [ 0, [[FOR_END]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND6:%.*]] = icmp ne i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND6]], label [[FOR_BODY5:%.*]], label [[FOR_COND_CLEANUP4:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup4:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END11:%.*]]
+; IS__CGSCC_NPM:       for.body5:
+; IS__CGSCC_NPM-NEXT:    [[I15:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_NPM-NEXT:    [[I16:%.*]] = or i64 [[I15]], 1
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* bitcast ([1024 x i8]* @GBytes to float*), i64 [[I16]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC9]]
+; IS__CGSCC_NPM:       for.inc9:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP22:![0-9]+]]
+; IS__CGSCC_NPM:       for.end11:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13:%.*]]
+; IS__CGSCC_NPM:       for.cond13:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV7:%.*]] = phi i64 [ [[INDVARS_IV_NEXT8:%.*]], [[FOR_INC21:%.*]] ], [ 0, [[FOR_END11]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND11:%.*]] = icmp ne i64 [[INDVARS_IV7]], 20
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND11]], label [[FOR_BODY16:%.*]], label [[FOR_COND_CLEANUP15:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup15:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END23:%.*]]
+; IS__CGSCC_NPM:       for.body16:
+; IS__CGSCC_NPM-NEXT:    [[I17:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
+; IS__CGSCC_NPM-NEXT:    [[I18:%.*]] = add nuw nsw i64 [[I17]], 2
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds i64, i64* bitcast ([1024 x i8]* @GBytes to i64*), i64 [[I18]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC21]]
+; IS__CGSCC_NPM:       for.inc21:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP23:![0-9]+]]
+; IS__CGSCC_NPM:       for.end23:
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(524) bitcast (i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @GBytes, i64 0, i64 500) to i32*), i32 noundef 0) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND25:%.*]]
+; IS__CGSCC_NPM:       for.cond25:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC33:%.*]] ], [ 0, [[FOR_END23]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND14:%.*]] = icmp ne i64 [[INDVARS_IV12]], 1024
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND14]], label [[FOR_BODY28:%.*]], label [[FOR_COND_CLEANUP27:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup27:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END35:%.*]]
+; IS__CGSCC_NPM:       for.body28:
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
+; IS__CGSCC_NPM-NEXT:    store i8 0, i8* [[ARRAYIDX32]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC33]]
+; IS__CGSCC_NPM:       for.inc33:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND25]], !llvm.loop [[LOOP24:![0-9]+]]
+; IS__CGSCC_NPM:       for.end35:
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
 entry:
   br label %for.cond
@@ -1401,12 +1555,19 @@
 ; IS__TUNIT_NPM-NEXT:    [[I:%.*]] = load i32, i32* @Flag3, align 4, !tbaa [[TBAA3]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[I]]
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@static_global_simplifiable_3
-; IS__CGSCC____-SAME: () #[[ATTR5:[0-9]+]] {
-; IS__CGSCC____-NEXT:    store i32 1, i32* @Flag3, align 4, !tbaa [[TBAA3]]
-; IS__CGSCC____-NEXT:    [[I:%.*]] = load i32, i32* @Flag3, align 4, !tbaa [[TBAA3]]
-; IS__CGSCC____-NEXT:    ret i32 [[I]]
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@static_global_simplifiable_3
+; IS__CGSCC_OPM-SAME: () #[[ATTR7:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:    store i32 1, i32* @Flag3, align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = load i32, i32* @Flag3, align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    ret i32 [[I]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@static_global_simplifiable_3
+; IS__CGSCC_NPM-SAME: () #[[ATTR5:[0-9]+]] {
+; IS__CGSCC_NPM-NEXT:    store i32 1, i32* @Flag3, align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = load i32, i32* @Flag3, align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    ret i32 [[I]]
 ;
   store i32 1, i32* @Flag3, align 4, !tbaa !3
   %i = load i32, i32* @Flag3, align 4, !tbaa !3
@@ -1531,55 +1692,105 @@
 ; IS__TUNIT_NPM-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14]]
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____: Function Attrs: argmemonly nofree norecurse nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@noalias_arg_simplifiable_1
-; IS__CGSCC____-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]], %struct.S* noalias nocapture nofree nonnull byval([[STRUCT_S]]) align 8 dereferenceable(24) [[S:%.*]]) #[[ATTR6:[0-9]+]] {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    store float 0x3FF19999A0000000, float* [[F1]], align 4, !tbaa [[TBAA7]]
-; IS__CGSCC____-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
-; IS__CGSCC____-NEXT:    store float 0x40019999A0000000, float* [[F2]], align 8, !tbaa [[TBAA10]]
-; IS__CGSCC____-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
-; IS__CGSCC____-NEXT:    store float 0x400A666660000000, float* [[F3]], align 4, !tbaa [[TBAA11]]
-; IS__CGSCC____-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 8 dereferenceable(24) [[I1]], i32 noundef 1) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) [[I2]], i32 noundef 2) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 8 dereferenceable(16) [[I3]], i32 noundef 3) #[[ATTR9]]
-; IS__CGSCC____-NEXT:    [[F11:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    [[I:%.*]] = load float, float* [[F11]], align 4, !tbaa [[TBAA7]]
-; IS__CGSCC____-NEXT:    [[F12:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    store float [[I]], float* [[F12]], align 4, !tbaa [[TBAA7]]
-; IS__CGSCC____-NEXT:    [[F23:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
-; IS__CGSCC____-NEXT:    [[I4:%.*]] = load float, float* [[F23]], align 8, !tbaa [[TBAA10]]
-; IS__CGSCC____-NEXT:    [[MUL:%.*]] = fmul float [[I4]], 2.000000e+00
-; IS__CGSCC____-NEXT:    [[F24:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
-; IS__CGSCC____-NEXT:    store float [[MUL]], float* [[F24]], align 4, !tbaa [[TBAA10]]
-; IS__CGSCC____-NEXT:    [[F35:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
-; IS__CGSCC____-NEXT:    [[I5:%.*]] = load float, float* [[F35]], align 4, !tbaa [[TBAA11]]
-; IS__CGSCC____-NEXT:    [[F16:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
-; IS__CGSCC____-NEXT:    [[I6:%.*]] = load float, float* [[F16]], align 4, !tbaa [[TBAA7]]
-; IS__CGSCC____-NEXT:    [[ADD:%.*]] = fadd float [[I5]], [[I6]]
-; IS__CGSCC____-NEXT:    [[F37:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
-; IS__CGSCC____-NEXT:    store float [[ADD]], float* [[F37]], align 4, !tbaa [[TBAA11]]
-; IS__CGSCC____-NEXT:    [[I18:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    [[I7:%.*]] = load i32, i32* [[I18]], align 8, !tbaa [[TBAA12]]
-; IS__CGSCC____-NEXT:    [[I19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    store i32 [[I7]], i32* [[I19]], align 4, !tbaa [[TBAA12]]
-; IS__CGSCC____-NEXT:    [[I210:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
-; IS__CGSCC____-NEXT:    [[I8:%.*]] = load i32, i32* [[I210]], align 4, !tbaa [[TBAA13]]
-; IS__CGSCC____-NEXT:    [[MUL11:%.*]] = shl nsw i32 [[I8]], 1
-; IS__CGSCC____-NEXT:    [[I212:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
-; IS__CGSCC____-NEXT:    store i32 [[MUL11]], i32* [[I212]], align 4, !tbaa [[TBAA13]]
-; IS__CGSCC____-NEXT:    [[I313:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
-; IS__CGSCC____-NEXT:    [[I9:%.*]] = load i32, i32* [[I313]], align 8, !tbaa [[TBAA14]]
-; IS__CGSCC____-NEXT:    [[I114:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
-; IS__CGSCC____-NEXT:    [[I10:%.*]] = load i32, i32* [[I114]], align 8, !tbaa [[TBAA12]]
-; IS__CGSCC____-NEXT:    [[ADD15:%.*]] = add nsw i32 [[I9]], [[I10]]
-; IS__CGSCC____-NEXT:    [[I316:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
-; IS__CGSCC____-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14]]
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: argmemonly nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@noalias_arg_simplifiable_1
+; IS__CGSCC_OPM-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]], %struct.S* noalias nocapture nofree nonnull byval([[STRUCT_S]]) align 8 dereferenceable(24) [[S:%.*]]) #[[ATTR8:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    store float 0x3FF19999A0000000, float* [[F1]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_OPM-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
+; IS__CGSCC_OPM-NEXT:    store float 0x40019999A0000000, float* [[F2]], align 8, !tbaa [[TBAA10]]
+; IS__CGSCC_OPM-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
+; IS__CGSCC_OPM-NEXT:    store float 0x400A666660000000, float* [[F3]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_OPM-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 8 dereferenceable(24) [[I1]], i32 noundef 1) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) [[I2]], i32 noundef 2) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 8 dereferenceable(16) [[I3]], i32 noundef 3) #[[ATTR12]]
+; IS__CGSCC_OPM-NEXT:    [[F11:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = load float, float* [[F11]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_OPM-NEXT:    [[F12:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    store float [[I]], float* [[F12]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_OPM-NEXT:    [[F23:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
+; IS__CGSCC_OPM-NEXT:    [[I4:%.*]] = load float, float* [[F23]], align 8, !tbaa [[TBAA10]]
+; IS__CGSCC_OPM-NEXT:    [[MUL:%.*]] = fmul float [[I4]], 2.000000e+00
+; IS__CGSCC_OPM-NEXT:    [[F24:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
+; IS__CGSCC_OPM-NEXT:    store float [[MUL]], float* [[F24]], align 4, !tbaa [[TBAA10]]
+; IS__CGSCC_OPM-NEXT:    [[F35:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
+; IS__CGSCC_OPM-NEXT:    [[I5:%.*]] = load float, float* [[F35]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_OPM-NEXT:    [[F16:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_OPM-NEXT:    [[I6:%.*]] = load float, float* [[F16]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_OPM-NEXT:    [[ADD:%.*]] = fadd float [[I5]], [[I6]]
+; IS__CGSCC_OPM-NEXT:    [[F37:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
+; IS__CGSCC_OPM-NEXT:    store float [[ADD]], float* [[F37]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_OPM-NEXT:    [[I18:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    [[I7:%.*]] = load i32, i32* [[I18]], align 8, !tbaa [[TBAA12]]
+; IS__CGSCC_OPM-NEXT:    [[I19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    store i32 [[I7]], i32* [[I19]], align 4, !tbaa [[TBAA12]]
+; IS__CGSCC_OPM-NEXT:    [[I210:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
+; IS__CGSCC_OPM-NEXT:    [[I8:%.*]] = load i32, i32* [[I210]], align 4, !tbaa [[TBAA13]]
+; IS__CGSCC_OPM-NEXT:    [[MUL11:%.*]] = shl nsw i32 [[I8]], 1
+; IS__CGSCC_OPM-NEXT:    [[I212:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
+; IS__CGSCC_OPM-NEXT:    store i32 [[MUL11]], i32* [[I212]], align 4, !tbaa [[TBAA13]]
+; IS__CGSCC_OPM-NEXT:    [[I313:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
+; IS__CGSCC_OPM-NEXT:    [[I9:%.*]] = load i32, i32* [[I313]], align 8, !tbaa [[TBAA14]]
+; IS__CGSCC_OPM-NEXT:    [[I114:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_OPM-NEXT:    [[I10:%.*]] = load i32, i32* [[I114]], align 8, !tbaa [[TBAA12]]
+; IS__CGSCC_OPM-NEXT:    [[ADD15:%.*]] = add nsw i32 [[I9]], [[I10]]
+; IS__CGSCC_OPM-NEXT:    [[I316:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
+; IS__CGSCC_OPM-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14]]
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: argmemonly nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@noalias_arg_simplifiable_1
+; IS__CGSCC_NPM-SAME: (%struct.S* noalias nocapture nofree nonnull writeonly sret([[STRUCT_S:%.*]]) align 4 dereferenceable(24) [[AGG_RESULT:%.*]], %struct.S* noalias nocapture nofree nonnull byval([[STRUCT_S]]) align 8 dereferenceable(24) [[S:%.*]]) #[[ATTR6:[0-9]+]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[F1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    store float 0x3FF19999A0000000, float* [[F1]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_NPM-NEXT:    [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
+; IS__CGSCC_NPM-NEXT:    store float 0x40019999A0000000, float* [[F2]], align 8, !tbaa [[TBAA10]]
+; IS__CGSCC_NPM-NEXT:    [[F3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
+; IS__CGSCC_NPM-NEXT:    store float 0x400A666660000000, float* [[F3]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_NPM-NEXT:    [[I1:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 8 dereferenceable(24) [[I1]], i32 noundef 1) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    [[I2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(20) [[I2]], i32 noundef 2) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    [[I3:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 8 dereferenceable(16) [[I3]], i32 noundef 3) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    [[F11:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = load float, float* [[F11]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_NPM-NEXT:    [[F12:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    store float [[I]], float* [[F12]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_NPM-NEXT:    [[F23:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 4
+; IS__CGSCC_NPM-NEXT:    [[I4:%.*]] = load float, float* [[F23]], align 8, !tbaa [[TBAA10]]
+; IS__CGSCC_NPM-NEXT:    [[MUL:%.*]] = fmul float [[I4]], 2.000000e+00
+; IS__CGSCC_NPM-NEXT:    [[F24:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 4
+; IS__CGSCC_NPM-NEXT:    store float [[MUL]], float* [[F24]], align 4, !tbaa [[TBAA10]]
+; IS__CGSCC_NPM-NEXT:    [[F35:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 5
+; IS__CGSCC_NPM-NEXT:    [[I5:%.*]] = load float, float* [[F35]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_NPM-NEXT:    [[F16:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 3
+; IS__CGSCC_NPM-NEXT:    [[I6:%.*]] = load float, float* [[F16]], align 4, !tbaa [[TBAA7]]
+; IS__CGSCC_NPM-NEXT:    [[ADD:%.*]] = fadd float [[I5]], [[I6]]
+; IS__CGSCC_NPM-NEXT:    [[F37:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 5
+; IS__CGSCC_NPM-NEXT:    store float [[ADD]], float* [[F37]], align 4, !tbaa [[TBAA11]]
+; IS__CGSCC_NPM-NEXT:    [[I18:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    [[I7:%.*]] = load i32, i32* [[I18]], align 8, !tbaa [[TBAA12]]
+; IS__CGSCC_NPM-NEXT:    [[I19:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    store i32 [[I7]], i32* [[I19]], align 4, !tbaa [[TBAA12]]
+; IS__CGSCC_NPM-NEXT:    [[I210:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 1
+; IS__CGSCC_NPM-NEXT:    [[I8:%.*]] = load i32, i32* [[I210]], align 4, !tbaa [[TBAA13]]
+; IS__CGSCC_NPM-NEXT:    [[MUL11:%.*]] = shl nsw i32 [[I8]], 1
+; IS__CGSCC_NPM-NEXT:    [[I212:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 1
+; IS__CGSCC_NPM-NEXT:    store i32 [[MUL11]], i32* [[I212]], align 4, !tbaa [[TBAA13]]
+; IS__CGSCC_NPM-NEXT:    [[I313:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 2
+; IS__CGSCC_NPM-NEXT:    [[I9:%.*]] = load i32, i32* [[I313]], align 8, !tbaa [[TBAA14]]
+; IS__CGSCC_NPM-NEXT:    [[I114:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[S]], i64 0, i32 0
+; IS__CGSCC_NPM-NEXT:    [[I10:%.*]] = load i32, i32* [[I114]], align 8, !tbaa [[TBAA12]]
+; IS__CGSCC_NPM-NEXT:    [[ADD15:%.*]] = add nsw i32 [[I9]], [[I10]]
+; IS__CGSCC_NPM-NEXT:    [[I316:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[AGG_RESULT]], i64 0, i32 2
+; IS__CGSCC_NPM-NEXT:    store i32 [[ADD15]], i32* [[I316]], align 4, !tbaa [[TBAA14]]
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
 entry:
   %f1 = getelementptr inbounds %struct.S, %struct.S* %s, i64 0, i32 3
@@ -1804,84 +2015,165 @@
 ; IS__TUNIT_NPM:       for.end37:
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____-LABEL: define {{[^@]+}}@noalias_arg_simplifiable_2
-; IS__CGSCC____-SAME: (i8* [[BYTES:%.*]]) {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    br label [[FOR_COND:%.*]]
-; IS__CGSCC____:       for.cond:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 0, [[ENTRY:%.*]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY:%.*]], label [[FOR_COND_CLEANUP:%.*]]
-; IS__CGSCC____:       for.cond.cleanup:
-; IS__CGSCC____-NEXT:    br label [[FOR_END:%.*]]
-; IS__CGSCC____:       for.body:
-; IS__CGSCC____-NEXT:    [[I:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
-; IS__CGSCC____-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 [[I]]
-; IS__CGSCC____-NEXT:    store i8 0, i8* [[ARRAYIDX]], align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC]]
-; IS__CGSCC____:       for.inc:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP28:![0-9]+]]
-; IS__CGSCC____:       for.end:
-; IS__CGSCC____-NEXT:    br label [[FOR_COND2:%.*]]
-; IS__CGSCC____:       for.cond2:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV2:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[FOR_INC9:%.*]] ], [ 0, [[FOR_END]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND6:%.*]] = icmp ne i64 [[INDVARS_IV2]], 10
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND6]], label [[FOR_BODY5:%.*]], label [[FOR_COND_CLEANUP4:%.*]]
-; IS__CGSCC____:       for.cond.cleanup4:
-; IS__CGSCC____-NEXT:    br label [[FOR_END11:%.*]]
-; IS__CGSCC____:       for.body5:
-; IS__CGSCC____-NEXT:    [[I15:%.*]] = bitcast i8* [[BYTES]] to float*
-; IS__CGSCC____-NEXT:    [[I16:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
-; IS__CGSCC____-NEXT:    [[I17:%.*]] = or i64 [[I16]], 1
-; IS__CGSCC____-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[I15]], i64 [[I17]]
-; IS__CGSCC____-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX8]], align 4, !tbaa [[TBAA18]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC9]]
-; IS__CGSCC____:       for.inc9:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP29:![0-9]+]]
-; IS__CGSCC____:       for.end11:
-; IS__CGSCC____-NEXT:    br label [[FOR_COND13:%.*]]
-; IS__CGSCC____:       for.cond13:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV7:%.*]] = phi i64 [ [[INDVARS_IV_NEXT8:%.*]], [[FOR_INC21:%.*]] ], [ 0, [[FOR_END11]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND11:%.*]] = icmp ne i64 [[INDVARS_IV7]], 20
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND11]], label [[FOR_BODY16:%.*]], label [[FOR_COND_CLEANUP15:%.*]]
-; IS__CGSCC____:       for.cond.cleanup15:
-; IS__CGSCC____-NEXT:    br label [[FOR_END23:%.*]]
-; IS__CGSCC____:       for.body16:
-; IS__CGSCC____-NEXT:    [[I18:%.*]] = bitcast i8* [[BYTES]] to i64*
-; IS__CGSCC____-NEXT:    [[I19:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
-; IS__CGSCC____-NEXT:    [[I20:%.*]] = add nuw nsw i64 [[I19]], 2
-; IS__CGSCC____-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds i64, i64* [[I18]], i64 [[I20]]
-; IS__CGSCC____-NEXT:    store i64 0, i64* [[ARRAYIDX20]], align 8, !tbaa [[TBAA20]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC21]]
-; IS__CGSCC____:       for.inc21:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP30:![0-9]+]]
-; IS__CGSCC____:       for.end23:
-; IS__CGSCC____-NEXT:    [[ARRAYIDX24:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 1023
-; IS__CGSCC____-NEXT:    store i8 0, i8* [[ARRAYIDX24]], align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 500
-; IS__CGSCC____-NEXT:    [[I21:%.*]] = bitcast i8* [[ARRAYIDX25]] to i32*
-; IS__CGSCC____-NEXT:    call void @write_arg(i32* nofree noundef nonnull writeonly align 4 dereferenceable(4) [[I21]], i32 noundef 0) #[[ATTR10]]
-; IS__CGSCC____-NEXT:    br label [[FOR_COND27:%.*]]
-; IS__CGSCC____:       for.cond27:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC35:%.*]] ], [ 0, [[FOR_END23]] ]
-; IS__CGSCC____-NEXT:    [[EXITCOND14:%.*]] = icmp ne i64 [[INDVARS_IV12]], 1024
-; IS__CGSCC____-NEXT:    br i1 [[EXITCOND14]], label [[FOR_BODY30:%.*]], label [[FOR_COND_CLEANUP29:%.*]]
-; IS__CGSCC____:       for.cond.cleanup29:
-; IS__CGSCC____-NEXT:    br label [[FOR_END37:%.*]]
-; IS__CGSCC____:       for.body30:
-; IS__CGSCC____-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 [[INDVARS_IV12]]
-; IS__CGSCC____-NEXT:    [[I22:%.*]] = load i8, i8* [[ARRAYIDX32]], align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    [[ARRAYIDX34:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
-; IS__CGSCC____-NEXT:    store i8 [[I22]], i8* [[ARRAYIDX34]], align 1, !tbaa [[TBAA15]]
-; IS__CGSCC____-NEXT:    br label [[FOR_INC35]]
-; IS__CGSCC____:       for.inc35:
-; IS__CGSCC____-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
-; IS__CGSCC____-NEXT:    br label [[FOR_COND27]], !llvm.loop [[LOOP31:![0-9]+]]
-; IS__CGSCC____:       for.end37:
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@noalias_arg_simplifiable_2
+; IS__CGSCC_OPM-SAME: (i8* nocapture nofree [[BYTES:%.*]]) #[[ATTR9:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND:%.*]]
+; IS__CGSCC_OPM:       for.cond:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 0, [[ENTRY:%.*]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END:%.*]]
+; IS__CGSCC_OPM:       for.body:
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 [[I]]
+; IS__CGSCC_OPM-NEXT:    store i8 0, i8* [[ARRAYIDX]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC]]
+; IS__CGSCC_OPM:       for.inc:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP25:![0-9]+]]
+; IS__CGSCC_OPM:       for.end:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2:%.*]]
+; IS__CGSCC_OPM:       for.cond2:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV2:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[FOR_INC9:%.*]] ], [ 0, [[FOR_END]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND6:%.*]] = icmp ne i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND6]], label [[FOR_BODY5:%.*]], label [[FOR_COND_CLEANUP4:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup4:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END11:%.*]]
+; IS__CGSCC_OPM:       for.body5:
+; IS__CGSCC_OPM-NEXT:    [[I15:%.*]] = bitcast i8* [[BYTES]] to float*
+; IS__CGSCC_OPM-NEXT:    [[I16:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_OPM-NEXT:    [[I17:%.*]] = or i64 [[I16]], 1
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[I15]], i64 [[I17]]
+; IS__CGSCC_OPM-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX8]], align 4, !tbaa [[TBAA26:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC9]]
+; IS__CGSCC_OPM:       for.inc9:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP27:![0-9]+]]
+; IS__CGSCC_OPM:       for.end11:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13:%.*]]
+; IS__CGSCC_OPM:       for.cond13:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV7:%.*]] = phi i64 [ [[INDVARS_IV_NEXT8:%.*]], [[FOR_INC21:%.*]] ], [ 0, [[FOR_END11]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND11:%.*]] = icmp ne i64 [[INDVARS_IV7]], 20
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND11]], label [[FOR_BODY16:%.*]], label [[FOR_COND_CLEANUP15:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup15:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END23:%.*]]
+; IS__CGSCC_OPM:       for.body16:
+; IS__CGSCC_OPM-NEXT:    [[I18:%.*]] = bitcast i8* [[BYTES]] to i64*
+; IS__CGSCC_OPM-NEXT:    [[I19:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
+; IS__CGSCC_OPM-NEXT:    [[I20:%.*]] = add nuw nsw i64 [[I19]], 2
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds i64, i64* [[I18]], i64 [[I20]]
+; IS__CGSCC_OPM-NEXT:    store i64 0, i64* [[ARRAYIDX20]], align 8, !tbaa [[TBAA28:![0-9]+]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC21]]
+; IS__CGSCC_OPM:       for.inc21:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP30:![0-9]+]]
+; IS__CGSCC_OPM:       for.end23:
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX24:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 1023
+; IS__CGSCC_OPM-NEXT:    store i8 0, i8* [[ARRAYIDX24]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 500
+; IS__CGSCC_OPM-NEXT:    [[I21:%.*]] = bitcast i8* [[ARRAYIDX25]] to i32*
+; IS__CGSCC_OPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[I21]], i32 noundef 0) #[[ATTR13]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND27:%.*]]
+; IS__CGSCC_OPM:       for.cond27:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC35:%.*]] ], [ 0, [[FOR_END23]] ]
+; IS__CGSCC_OPM-NEXT:    [[EXITCOND14:%.*]] = icmp ne i64 [[INDVARS_IV12]], 1024
+; IS__CGSCC_OPM-NEXT:    br i1 [[EXITCOND14]], label [[FOR_BODY30:%.*]], label [[FOR_COND_CLEANUP29:%.*]]
+; IS__CGSCC_OPM:       for.cond.cleanup29:
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_END37:%.*]]
+; IS__CGSCC_OPM:       for.body30:
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 [[INDVARS_IV12]]
+; IS__CGSCC_OPM-NEXT:    [[I22:%.*]] = load i8, i8* [[ARRAYIDX32]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_OPM-NEXT:    [[ARRAYIDX34:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
+; IS__CGSCC_OPM-NEXT:    store i8 [[I22]], i8* [[ARRAYIDX34]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_INC35]]
+; IS__CGSCC_OPM:       for.inc35:
+; IS__CGSCC_OPM-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
+; IS__CGSCC_OPM-NEXT:    br label [[FOR_COND27]], !llvm.loop [[LOOP31:![0-9]+]]
+; IS__CGSCC_OPM:       for.end37:
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@noalias_arg_simplifiable_2
+; IS__CGSCC_NPM-SAME: (i8* nocapture nofree [[BYTES:%.*]]) #[[ATTR5]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND:%.*]]
+; IS__CGSCC_NPM:       for.cond:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 0, [[ENTRY:%.*]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END:%.*]]
+; IS__CGSCC_NPM:       for.body:
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = mul nuw nsw i64 [[INDVARS_IV]], 10
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 [[I]]
+; IS__CGSCC_NPM-NEXT:    store i8 0, i8* [[ARRAYIDX]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC]]
+; IS__CGSCC_NPM:       for.inc:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP25:![0-9]+]]
+; IS__CGSCC_NPM:       for.end:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2:%.*]]
+; IS__CGSCC_NPM:       for.cond2:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV2:%.*]] = phi i64 [ [[INDVARS_IV_NEXT3:%.*]], [[FOR_INC9:%.*]] ], [ 0, [[FOR_END]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND6:%.*]] = icmp ne i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND6]], label [[FOR_BODY5:%.*]], label [[FOR_COND_CLEANUP4:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup4:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END11:%.*]]
+; IS__CGSCC_NPM:       for.body5:
+; IS__CGSCC_NPM-NEXT:    [[I15:%.*]] = bitcast i8* [[BYTES]] to float*
+; IS__CGSCC_NPM-NEXT:    [[I16:%.*]] = mul nuw nsw i64 [[INDVARS_IV2]], 10
+; IS__CGSCC_NPM-NEXT:    [[I17:%.*]] = or i64 [[I16]], 1
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[I15]], i64 [[I17]]
+; IS__CGSCC_NPM-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX8]], align 4, !tbaa [[TBAA26:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC9]]
+; IS__CGSCC_NPM:       for.inc9:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP27:![0-9]+]]
+; IS__CGSCC_NPM:       for.end11:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13:%.*]]
+; IS__CGSCC_NPM:       for.cond13:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV7:%.*]] = phi i64 [ [[INDVARS_IV_NEXT8:%.*]], [[FOR_INC21:%.*]] ], [ 0, [[FOR_END11]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND11:%.*]] = icmp ne i64 [[INDVARS_IV7]], 20
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND11]], label [[FOR_BODY16:%.*]], label [[FOR_COND_CLEANUP15:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup15:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END23:%.*]]
+; IS__CGSCC_NPM:       for.body16:
+; IS__CGSCC_NPM-NEXT:    [[I18:%.*]] = bitcast i8* [[BYTES]] to i64*
+; IS__CGSCC_NPM-NEXT:    [[I19:%.*]] = mul nuw nsw i64 [[INDVARS_IV7]], 10
+; IS__CGSCC_NPM-NEXT:    [[I20:%.*]] = add nuw nsw i64 [[I19]], 2
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX20:%.*]] = getelementptr inbounds i64, i64* [[I18]], i64 [[I20]]
+; IS__CGSCC_NPM-NEXT:    store i64 0, i64* [[ARRAYIDX20]], align 8, !tbaa [[TBAA28:![0-9]+]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC21]]
+; IS__CGSCC_NPM:       for.inc21:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT8]] = add nuw nsw i64 [[INDVARS_IV7]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND13]], !llvm.loop [[LOOP30:![0-9]+]]
+; IS__CGSCC_NPM:       for.end23:
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX24:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 1023
+; IS__CGSCC_NPM-NEXT:    store i8 0, i8* [[ARRAYIDX24]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX25:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 500
+; IS__CGSCC_NPM-NEXT:    [[I21:%.*]] = bitcast i8* [[ARRAYIDX25]] to i32*
+; IS__CGSCC_NPM-NEXT:    call void @write_arg(i32* nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[I21]], i32 noundef 0) #[[ATTR9]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND27:%.*]]
+; IS__CGSCC_NPM:       for.cond27:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV12:%.*]] = phi i64 [ [[INDVARS_IV_NEXT13:%.*]], [[FOR_INC35:%.*]] ], [ 0, [[FOR_END23]] ]
+; IS__CGSCC_NPM-NEXT:    [[EXITCOND14:%.*]] = icmp ne i64 [[INDVARS_IV12]], 1024
+; IS__CGSCC_NPM-NEXT:    br i1 [[EXITCOND14]], label [[FOR_BODY30:%.*]], label [[FOR_COND_CLEANUP29:%.*]]
+; IS__CGSCC_NPM:       for.cond.cleanup29:
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_END37:%.*]]
+; IS__CGSCC_NPM:       for.body30:
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX32:%.*]] = getelementptr inbounds i8, i8* [[BYTES]], i64 [[INDVARS_IV12]]
+; IS__CGSCC_NPM-NEXT:    [[I22:%.*]] = load i8, i8* [[ARRAYIDX32]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_NPM-NEXT:    [[ARRAYIDX34:%.*]] = getelementptr inbounds [1024 x i8], [1024 x i8]* @globalBytes, i64 0, i64 [[INDVARS_IV12]]
+; IS__CGSCC_NPM-NEXT:    store i8 [[I22]], i8* [[ARRAYIDX34]], align 1, !tbaa [[TBAA19]]
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_INC35]]
+; IS__CGSCC_NPM:       for.inc35:
+; IS__CGSCC_NPM-NEXT:    [[INDVARS_IV_NEXT13]] = add nuw nsw i64 [[INDVARS_IV12]], 1
+; IS__CGSCC_NPM-NEXT:    br label [[FOR_COND27]], !llvm.loop [[LOOP31:![0-9]+]]
+; IS__CGSCC_NPM:       for.end37:
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
 entry:
   br label %for.cond
@@ -1991,30 +2283,30 @@
 ;    }
 ;
 define i32 @local_alloca_not_simplifiable_1() {
-; NOT_TUNIT_NPM-LABEL: define {{[^@]+}}@local_alloca_not_simplifiable_1() {
-; NOT_TUNIT_NPM-NEXT:  entry:
-; NOT_TUNIT_NPM-NEXT:    [[X:%.*]] = alloca i32, align 4
-; NOT_TUNIT_NPM-NEXT:    [[Y:%.*]] = alloca i32, align 4
-; NOT_TUNIT_NPM-NEXT:    [[I:%.*]] = bitcast i32* [[X]] to i8*
-; NOT_TUNIT_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8:[0-9]+]]
-; NOT_TUNIT_NPM-NEXT:    [[I1:%.*]] = bitcast i32* [[Y]] to i8*
-; NOT_TUNIT_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR8]]
-; NOT_TUNIT_NPM-NEXT:    store i32 1, i32* [[Y]], align 4, !tbaa [[TBAA3]]
-; NOT_TUNIT_NPM-NEXT:    store i32 1, i32* [[X]], align 4, !tbaa [[TBAA3]]
-; NOT_TUNIT_NPM-NEXT:    [[I2:%.*]] = bitcast i32* [[X]] to i8*
-; NOT_TUNIT_NPM-NEXT:    call void @escape(i8* noundef nonnull align 4 dereferenceable(4) [[I2]])
-; NOT_TUNIT_NPM-NEXT:    call void @write_random(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[Y]])
-; NOT_TUNIT_NPM-NEXT:    [[I3:%.*]] = load i32, i32* [[X]], align 4, !tbaa [[TBAA3]]
-; NOT_TUNIT_NPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[I3]], 0
-; NOT_TUNIT_NPM-NEXT:    [[COND:%.*]] = select i1 [[TOBOOL_NOT]], i32 2, i32 1
-; NOT_TUNIT_NPM-NEXT:    [[I4:%.*]] = load i32, i32* [[Y]], align 4, !tbaa [[TBAA3]]
-; NOT_TUNIT_NPM-NEXT:    [[ADD:%.*]] = add nsw i32 [[I3]], [[I4]]
-; NOT_TUNIT_NPM-NEXT:    [[ADD1:%.*]] = add nsw i32 [[ADD]], [[COND]]
-; NOT_TUNIT_NPM-NEXT:    [[I5:%.*]] = bitcast i32* [[Y]] to i8*
-; NOT_TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I5]])
-; NOT_TUNIT_NPM-NEXT:    [[I6:%.*]] = bitcast i32* [[X]] to i8*
-; NOT_TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I6]])
-; NOT_TUNIT_NPM-NEXT:    ret i32 [[ADD1]]
+; IS__TUNIT_OPM-LABEL: define {{[^@]+}}@local_alloca_not_simplifiable_1() {
+; IS__TUNIT_OPM-NEXT:  entry:
+; IS__TUNIT_OPM-NEXT:    [[X:%.*]] = alloca i32, align 4
+; IS__TUNIT_OPM-NEXT:    [[Y:%.*]] = alloca i32, align 4
+; IS__TUNIT_OPM-NEXT:    [[I:%.*]] = bitcast i32* [[X]] to i8*
+; IS__TUNIT_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8]]
+; IS__TUNIT_OPM-NEXT:    [[I1:%.*]] = bitcast i32* [[Y]] to i8*
+; IS__TUNIT_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR8]]
+; IS__TUNIT_OPM-NEXT:    store i32 1, i32* [[Y]], align 4, !tbaa [[TBAA3]]
+; IS__TUNIT_OPM-NEXT:    store i32 1, i32* [[X]], align 4, !tbaa [[TBAA3]]
+; IS__TUNIT_OPM-NEXT:    [[I2:%.*]] = bitcast i32* [[X]] to i8*
+; IS__TUNIT_OPM-NEXT:    call void @escape(i8* noundef nonnull align 4 dereferenceable(4) [[I2]])
+; IS__TUNIT_OPM-NEXT:    call void @write_random(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[Y]])
+; IS__TUNIT_OPM-NEXT:    [[I3:%.*]] = load i32, i32* [[X]], align 4, !tbaa [[TBAA3]]
+; IS__TUNIT_OPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[I3]], 0
+; IS__TUNIT_OPM-NEXT:    [[COND:%.*]] = select i1 [[TOBOOL_NOT]], i32 2, i32 1
+; IS__TUNIT_OPM-NEXT:    [[I4:%.*]] = load i32, i32* [[Y]], align 4, !tbaa [[TBAA3]]
+; IS__TUNIT_OPM-NEXT:    [[ADD:%.*]] = add nsw i32 [[I3]], [[I4]]
+; IS__TUNIT_OPM-NEXT:    [[ADD1:%.*]] = add nsw i32 [[ADD]], [[COND]]
+; IS__TUNIT_OPM-NEXT:    [[I5:%.*]] = bitcast i32* [[Y]] to i8*
+; IS__TUNIT_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I5]])
+; IS__TUNIT_OPM-NEXT:    [[I6:%.*]] = bitcast i32* [[X]] to i8*
+; IS__TUNIT_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I6]])
+; IS__TUNIT_OPM-NEXT:    ret i32 [[ADD1]]
 ;
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@local_alloca_not_simplifiable_1() {
 ; IS__TUNIT_NPM-NEXT:  entry:
@@ -2041,6 +2333,56 @@
 ; IS__TUNIT_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I6]])
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[ADD1]]
 ;
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@local_alloca_not_simplifiable_1() {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    [[X:%.*]] = alloca i32, align 4
+; IS__CGSCC_OPM-NEXT:    [[Y:%.*]] = alloca i32, align 4
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = bitcast i32* [[X]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    [[I1:%.*]] = bitcast i32* [[Y]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR11]]
+; IS__CGSCC_OPM-NEXT:    store i32 1, i32* [[Y]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    store i32 1, i32* [[X]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    [[I2:%.*]] = bitcast i32* [[X]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @escape(i8* noundef nonnull align 4 dereferenceable(4) [[I2]])
+; IS__CGSCC_OPM-NEXT:    call void @write_random(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[Y]])
+; IS__CGSCC_OPM-NEXT:    [[I3:%.*]] = load i32, i32* [[X]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[I3]], 0
+; IS__CGSCC_OPM-NEXT:    [[COND:%.*]] = select i1 [[TOBOOL_NOT]], i32 2, i32 1
+; IS__CGSCC_OPM-NEXT:    [[I4:%.*]] = load i32, i32* [[Y]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    [[ADD:%.*]] = add nsw i32 [[I3]], [[I4]]
+; IS__CGSCC_OPM-NEXT:    [[ADD1:%.*]] = add nsw i32 [[ADD]], [[COND]]
+; IS__CGSCC_OPM-NEXT:    [[I5:%.*]] = bitcast i32* [[Y]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I5]])
+; IS__CGSCC_OPM-NEXT:    [[I6:%.*]] = bitcast i32* [[X]] to i8*
+; IS__CGSCC_OPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I6]])
+; IS__CGSCC_OPM-NEXT:    ret i32 [[ADD1]]
+;
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@local_alloca_not_simplifiable_1() {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[X:%.*]] = alloca i32, align 4
+; IS__CGSCC_NPM-NEXT:    [[Y:%.*]] = alloca i32, align 4
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = bitcast i32* [[X]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    [[I1:%.*]] = bitcast i32* [[Y]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.start.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I1]]) #[[ATTR8]]
+; IS__CGSCC_NPM-NEXT:    store i32 1, i32* [[Y]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    store i32 1, i32* [[X]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    [[I2:%.*]] = bitcast i32* [[X]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @escape(i8* noundef nonnull align 4 dereferenceable(4) [[I2]])
+; IS__CGSCC_NPM-NEXT:    call void @write_random(i32* noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[Y]])
+; IS__CGSCC_NPM-NEXT:    [[I3:%.*]] = load i32, i32* [[X]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    [[TOBOOL_NOT:%.*]] = icmp eq i32 [[I3]], 0
+; IS__CGSCC_NPM-NEXT:    [[COND:%.*]] = select i1 [[TOBOOL_NOT]], i32 2, i32 1
+; IS__CGSCC_NPM-NEXT:    [[I4:%.*]] = load i32, i32* [[Y]], align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    [[ADD:%.*]] = add nsw i32 [[I3]], [[I4]]
+; IS__CGSCC_NPM-NEXT:    [[ADD1:%.*]] = add nsw i32 [[ADD]], [[COND]]
+; IS__CGSCC_NPM-NEXT:    [[I5:%.*]] = bitcast i32* [[Y]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I5]])
+; IS__CGSCC_NPM-NEXT:    [[I6:%.*]] = bitcast i32* [[X]] to i8*
+; IS__CGSCC_NPM-NEXT:    call void @llvm.lifetime.end.p0i8(i64 noundef 4, i8* nocapture nofree noundef nonnull align 4 dereferenceable(4) [[I6]])
+; IS__CGSCC_NPM-NEXT:    ret i32 [[ADD1]]
+;
 entry:
   %X = alloca i32, align 4
   %Y = alloca i32, align 4
@@ -2089,7 +2431,7 @@
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@unknown_access_mixed_simplifiable
-; IS__CGSCC____-SAME: (i32 [[ARG1:%.*]], i32 [[ARG2:%.*]]) #[[ATTR2]] {
+; IS__CGSCC____-SAME: (i32 [[ARG1:%.*]], i32 [[ARG2:%.*]]) #[[ATTR3]] {
 ; IS__CGSCC____-NEXT:  entry:
 ; IS__CGSCC____-NEXT:    [[S:%.*]] = alloca [[STRUCT_S:%.*]], align 4
 ; IS__CGSCC____-NEXT:    [[BC:%.*]] = bitcast %struct.S* [[S]] to i32*
@@ -2152,7 +2494,7 @@
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@unknown_access_mixed_not_simplifiable
-; IS__CGSCC____-SAME: (i32 [[ARG1:%.*]], i32 [[ARG2:%.*]], i32 [[ARG3:%.*]]) #[[ATTR2]] {
+; IS__CGSCC____-SAME: (i32 [[ARG1:%.*]], i32 [[ARG2:%.*]], i32 [[ARG3:%.*]]) #[[ATTR3]] {
 ; IS__CGSCC____-NEXT:  entry:
 ; IS__CGSCC____-NEXT:    [[S:%.*]] = alloca [[STRUCT_S:%.*]], align 4
 ; IS__CGSCC____-NEXT:    [[BC:%.*]] = bitcast %struct.S* [[S]] to i32*
@@ -2220,12 +2562,19 @@
 ; IS__TUNIT_NPM-NEXT:    [[I:%.*]] = load i32, i32* @Flag0, align 4, !tbaa [[TBAA3]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[I]]
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@global_not_simplifiable_1
-; IS__CGSCC____-SAME: (i32 [[CND:%.*]]) #[[ATTR7:[0-9]+]] {
-; IS__CGSCC____-NEXT:  entry:
-; IS__CGSCC____-NEXT:    [[I:%.*]] = load i32, i32* @Flag0, align 4, !tbaa [[TBAA3]]
-; IS__CGSCC____-NEXT:    ret i32 [[I]]
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@global_not_simplifiable_1
+; IS__CGSCC_OPM-SAME: (i32 [[CND:%.*]]) #[[ATTR10:[0-9]+]] {
+; IS__CGSCC_OPM-NEXT:  entry:
+; IS__CGSCC_OPM-NEXT:    [[I:%.*]] = load i32, i32* @Flag0, align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_OPM-NEXT:    ret i32 [[I]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@global_not_simplifiable_1
+; IS__CGSCC_NPM-SAME: (i32 [[CND:%.*]]) #[[ATTR7:[0-9]+]] {
+; IS__CGSCC_NPM-NEXT:  entry:
+; IS__CGSCC_NPM-NEXT:    [[I:%.*]] = load i32, i32* @Flag0, align 4, !tbaa [[TBAA3]]
+; IS__CGSCC_NPM-NEXT:    ret i32 [[I]]
 ;
 entry:
   %i = load i32, i32* @Flag0, align 4, !tbaa !3
@@ -2323,12 +2672,19 @@
 ; IS__TUNIT_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gint1, align 4
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[L]]
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@write_read_global
-; IS__CGSCC____-SAME: () #[[ATTR5]] {
-; IS__CGSCC____-NEXT:    store i32 7, i32* @Gint1, align 4
-; IS__CGSCC____-NEXT:    [[L:%.*]] = load i32, i32* @Gint1, align 4
-; IS__CGSCC____-NEXT:    ret i32 [[L]]
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@write_read_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR7]] {
+; IS__CGSCC_OPM-NEXT:    store i32 7, i32* @Gint1, align 4
+; IS__CGSCC_OPM-NEXT:    [[L:%.*]] = load i32, i32* @Gint1, align 4
+; IS__CGSCC_OPM-NEXT:    ret i32 [[L]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@write_read_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR5]] {
+; IS__CGSCC_NPM-NEXT:    store i32 7, i32* @Gint1, align 4
+; IS__CGSCC_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gint1, align 4
+; IS__CGSCC_NPM-NEXT:    ret i32 [[L]]
 ;
   store i32 7, i32* @Gint1
   %l = load i32, i32* @Gint1
@@ -2347,11 +2703,17 @@
 ; IS__TUNIT_NPM-NEXT:    store i32 7, i32* @Gint2, align 4
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
-; IS__CGSCC____-LABEL: define {{[^@]+}}@write_global
-; IS__CGSCC____-SAME: () #[[ATTR4]] {
-; IS__CGSCC____-NEXT:    store i32 7, i32* @Gint2, align 4
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@write_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR5]] {
+; IS__CGSCC_OPM-NEXT:    store i32 7, i32* @Gint2, align 4
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@write_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR4]] {
+; IS__CGSCC_NPM-NEXT:    store i32 7, i32* @Gint2, align 4
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
   store i32 7, i32* @Gint2
   ret void
@@ -2369,11 +2731,17 @@
 ; IS__TUNIT_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gint2, align 4
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[L]]
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@read_global
-; IS__CGSCC____-SAME: () #[[ATTR7]] {
-; IS__CGSCC____-NEXT:    [[L:%.*]] = load i32, i32* @Gint2, align 4
-; IS__CGSCC____-NEXT:    ret i32 [[L]]
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@read_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR10]] {
+; IS__CGSCC_OPM-NEXT:    [[L:%.*]] = load i32, i32* @Gint2, align 4
+; IS__CGSCC_OPM-NEXT:    ret i32 [[L]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@read_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR7]] {
+; IS__CGSCC_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gint2, align 4
+; IS__CGSCC_NPM-NEXT:    ret i32 [[L]]
 ;
   %l = load i32, i32* @Gint2
   ret i32 %l
@@ -2394,12 +2762,19 @@
 ; IS__TUNIT_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int1, align 4
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[L]]
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@write_read_static_global
-; IS__CGSCC____-SAME: () #[[ATTR5]] {
-; IS__CGSCC____-NEXT:    store i32 7, i32* @Gstatic_int1, align 4
-; IS__CGSCC____-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int1, align 4
-; IS__CGSCC____-NEXT:    ret i32 [[L]]
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@write_read_static_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR7]] {
+; IS__CGSCC_OPM-NEXT:    store i32 7, i32* @Gstatic_int1, align 4
+; IS__CGSCC_OPM-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int1, align 4
+; IS__CGSCC_OPM-NEXT:    ret i32 [[L]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@write_read_static_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR5]] {
+; IS__CGSCC_NPM-NEXT:    store i32 7, i32* @Gstatic_int1, align 4
+; IS__CGSCC_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int1, align 4
+; IS__CGSCC_NPM-NEXT:    ret i32 [[L]]
 ;
   store i32 7, i32* @Gstatic_int1
   %l = load i32, i32* @Gstatic_int1
@@ -2418,11 +2793,17 @@
 ; IS__TUNIT_NPM-NEXT:    store i32 7, i32* @Gstatic_int2, align 4
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
-; IS__CGSCC____-LABEL: define {{[^@]+}}@write_static_global
-; IS__CGSCC____-SAME: () #[[ATTR4]] {
-; IS__CGSCC____-NEXT:    store i32 7, i32* @Gstatic_int2, align 4
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@write_static_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR5]] {
+; IS__CGSCC_OPM-NEXT:    store i32 7, i32* @Gstatic_int2, align 4
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@write_static_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR4]] {
+; IS__CGSCC_NPM-NEXT:    store i32 7, i32* @Gstatic_int2, align 4
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
   store i32 7, i32* @Gstatic_int2
   ret void
@@ -2440,11 +2821,17 @@
 ; IS__TUNIT_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int2, align 4
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[L]]
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
-; IS__CGSCC____-LABEL: define {{[^@]+}}@read_static_global
-; IS__CGSCC____-SAME: () #[[ATTR7]] {
-; IS__CGSCC____-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int2, align 4
-; IS__CGSCC____-NEXT:    ret i32 [[L]]
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@read_static_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR10]] {
+; IS__CGSCC_OPM-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int2, align 4
+; IS__CGSCC_OPM-NEXT:    ret i32 [[L]]
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind readonly willreturn
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@read_static_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR7]] {
+; IS__CGSCC_NPM-NEXT:    [[L:%.*]] = load i32, i32* @Gstatic_int2, align 4
+; IS__CGSCC_NPM-NEXT:    ret i32 [[L]]
 ;
   %l = load i32, i32* @Gstatic_int2
   ret i32 %l
@@ -2460,10 +2847,15 @@
 ; IS__TUNIT_NPM-SAME: () #[[ATTR4]] {
 ; IS__TUNIT_NPM-NEXT:    ret i32 7
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
-; IS__CGSCC____-LABEL: define {{[^@]+}}@write_read_static_undef_global
-; IS__CGSCC____-SAME: () #[[ATTR4]] {
-; IS__CGSCC____-NEXT:    ret i32 7
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@write_read_static_undef_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR5]] {
+; IS__CGSCC_OPM-NEXT:    ret i32 7
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@write_read_static_undef_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR4]] {
+; IS__CGSCC_NPM-NEXT:    ret i32 7
 ;
   store i32 7, i32* @Gstatic_undef_int1
   %l = load i32, i32* @Gstatic_undef_int1
@@ -2480,11 +2872,17 @@
 ; IS__TUNIT_NPM-SAME: () #[[ATTR4]] {
 ; IS__TUNIT_NPM-NEXT:    ret void
 ;
-; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
-; IS__CGSCC____-LABEL: define {{[^@]+}}@write_static_undef_global
-; IS__CGSCC____-SAME: () #[[ATTR4]] {
-; IS__CGSCC____-NEXT:    store i32 7, i32* @Gstatic_undef_int2, align 4
-; IS__CGSCC____-NEXT:    ret void
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@write_static_undef_global
+; IS__CGSCC_OPM-SAME: () #[[ATTR5]] {
+; IS__CGSCC_OPM-NEXT:    store i32 7, i32* @Gstatic_undef_int2, align 4
+; IS__CGSCC_OPM-NEXT:    ret void
+;
+; IS__CGSCC_NPM: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC_NPM-LABEL: define {{[^@]+}}@write_static_undef_global
+; IS__CGSCC_NPM-SAME: () #[[ATTR4]] {
+; IS__CGSCC_NPM-NEXT:    store i32 7, i32* @Gstatic_undef_int2, align 4
+; IS__CGSCC_NPM-NEXT:    ret void
 ;
   store i32 7, i32* @Gstatic_undef_int2
   ret void
@@ -2497,7 +2895,7 @@
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@read_static_undef_global
-; IS__CGSCC____-SAME: () #[[ATTR2]] {
+; IS__CGSCC____-SAME: () #[[ATTR3]] {
 ; IS__CGSCC____-NEXT:    ret i32 7
 ;
   %l = load i32, i32* @Gstatic_undef_int2
@@ -2512,7 +2910,7 @@
 ;
 ; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@single_read_of_static_global
-; IS__CGSCC____-SAME: () #[[ATTR2]] {
+; IS__CGSCC____-SAME: () #[[ATTR3]] {
 ; IS__CGSCC____-NEXT:    ret i32 0
 ;
   %l = load i32, i32* @Gstatic_int3
@@ -2574,81 +2972,62 @@
 ; IS__TUNIT_NPM: attributes #[[ATTR5]] = { nofree nosync nounwind readonly willreturn }
 ; IS__TUNIT_NPM: attributes #[[ATTR6]] = { willreturn }
 ;.
-; IS__CGSCC____: attributes #[[ATTR0]] = { argmemonly nofree norecurse nosync nounwind willreturn writeonly }
-; IS__CGSCC____: attributes #[[ATTR1]] = { argmemonly nofree nosync nounwind willreturn }
-; IS__CGSCC____: attributes #[[ATTR2]] = { nofree norecurse nosync nounwind readnone willreturn }
-; IS__CGSCC____: attributes #[[ATTR3]] = { nofree nosync nounwind willreturn }
-; IS__CGSCC____: attributes #[[ATTR4]] = { nofree norecurse nosync nounwind willreturn writeonly }
-; IS__CGSCC____: attributes #[[ATTR5]] = { nofree norecurse nosync nounwind willreturn }
-; IS__CGSCC____: attributes #[[ATTR6]] = { argmemonly nofree norecurse nosync nounwind willreturn }
-; IS__CGSCC____: attributes #[[ATTR7]] = { nofree norecurse nosync nounwind readonly willreturn }
-; IS__CGSCC____: attributes #[[ATTR8]] = { willreturn }
-; IS__CGSCC____: attributes #[[ATTR9]] = { nounwind willreturn writeonly }
-; IS__CGSCC____: attributes #[[ATTR10]] = { nounwind writeonly }
+; IS__CGSCC_OPM: attributes #[[ATTR0]] = { argmemonly nofree norecurse nosync nounwind willreturn writeonly }
+; IS__CGSCC_OPM: attributes #[[ATTR1]] = { argmemonly nofree nosync nounwind willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR2]] = { nofree nosync nounwind }
+; IS__CGSCC_OPM: attributes #[[ATTR3]] = { nofree norecurse nosync nounwind readnone willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR4]] = { nofree nosync nounwind willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR5]] = { nofree norecurse nosync nounwind willreturn writeonly }
+; IS__CGSCC_OPM: attributes #[[ATTR6]] = { nofree norecurse nosync nounwind writeonly }
+; IS__CGSCC_OPM: attributes #[[ATTR7]] = { nofree norecurse nosync nounwind willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR8]] = { argmemonly nofree norecurse nosync nounwind willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR9]] = { nofree norecurse nosync nounwind }
+; IS__CGSCC_OPM: attributes #[[ATTR10]] = { nofree norecurse nosync nounwind readonly willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR11]] = { willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR12]] = { nounwind willreturn writeonly }
+; IS__CGSCC_OPM: attributes #[[ATTR13]] = { nounwind writeonly }
 ;.
-; IS__TUNIT____: [[META0:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
-; IS__TUNIT____: [[META1:![0-9]+]] = !{i32 7, !"uwtable", i32 1}
-; IS__TUNIT____: [[META2:![0-9]+]] = !{!"clang version 13.0.0"}
-; IS__TUNIT____: [[TBAA3]] = !{!4, !4, i64 0}
-; IS__TUNIT____: [[META4:![0-9]+]] = !{!"int", !5, i64 0}
-; IS__TUNIT____: [[META5:![0-9]+]] = !{!"omnipotent char", !6, i64 0}
-; IS__TUNIT____: [[META6:![0-9]+]] = !{!"Simple C/C++ TBAA"}
-; IS__TUNIT____: [[META7:![0-9]+]] = !{!8, !9, i64 12}
-; IS__TUNIT____: [[META8:![0-9]+]] = !{!"S", !4, i64 0, !4, i64 4, !4, i64 8, !9, i64 12, !9, i64 16, !9, i64 20}
-; IS__TUNIT____: [[META9:![0-9]+]] = !{!"float", !5, i64 0}
-; IS__TUNIT____: [[META10:![0-9]+]] = !{!8, !9, i64 16}
-; IS__TUNIT____: [[META11:![0-9]+]] = !{!8, !9, i64 20}
-; IS__TUNIT____: [[META12:![0-9]+]] = !{!8, !4, i64 0}
-; IS__TUNIT____: [[META13:![0-9]+]] = !{!8, !4, i64 4}
-; IS__TUNIT____: [[META14:![0-9]+]] = !{!8, !4, i64 8}
-; IS__TUNIT____: [[META15:![0-9]+]] = distinct !{!15, !16}
-; IS__TUNIT____: [[META16:![0-9]+]] = !{!"llvm.loop.mustprogress"}
-; IS__TUNIT____: [[META17:![0-9]+]] = distinct !{!17, !16}
-; IS__TUNIT____: [[META18:![0-9]+]] = distinct !{!18, !16}
-; IS__TUNIT____: [[META19:![0-9]+]] = !{!5, !5, i64 0}
-; IS__TUNIT____: [[META20:![0-9]+]] = distinct !{!20, !16}
-; IS__TUNIT____: [[META21:![0-9]+]] = distinct !{!21, !16}
-; IS__TUNIT____: [[META22:![0-9]+]] = distinct !{!22, !16}
-; IS__TUNIT____: [[META23:![0-9]+]] = distinct !{!23, !16}
-; IS__TUNIT____: [[META24:![0-9]+]] = distinct !{!24, !16}
-; IS__TUNIT____: [[META25:![0-9]+]] = distinct !{!25, !16}
-; IS__TUNIT____: [[META26:![0-9]+]] = !{!9, !9, i64 0}
-; IS__TUNIT____: [[META27:![0-9]+]] = distinct !{!27, !16}
-; IS__TUNIT____: [[META28:![0-9]+]] = !{!29, !29, i64 0}
-; IS__TUNIT____: [[META29:![0-9]+]] = !{!"long long", !5, i64 0}
-; IS__TUNIT____: [[META30:![0-9]+]] = distinct !{!30, !16}
-; IS__TUNIT____: [[META31:![0-9]+]] = distinct !{!31, !16}
+; IS__CGSCC_NPM: attributes #[[ATTR0]] = { argmemonly nofree norecurse nosync nounwind willreturn writeonly }
+; IS__CGSCC_NPM: attributes #[[ATTR1]] = { argmemonly nofree nosync nounwind willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR2]] = { nofree nosync nounwind willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR3]] = { nofree norecurse nosync nounwind readnone willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR4]] = { nofree norecurse nosync nounwind willreturn writeonly }
+; IS__CGSCC_NPM: attributes #[[ATTR5]] = { nofree norecurse nosync nounwind willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR6]] = { argmemonly nofree norecurse nosync nounwind willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR7]] = { nofree norecurse nosync nounwind readonly willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR8]] = { willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR9]] = { nounwind willreturn writeonly }
 ;.
-; IS__CGSCC____: [[META0:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
-; IS__CGSCC____: [[META1:![0-9]+]] = !{i32 7, !"uwtable", i32 1}
-; IS__CGSCC____: [[META2:![0-9]+]] = !{!"clang version 13.0.0"}
-; IS__CGSCC____: [[TBAA3]] = !{!4, !4, i64 0}
-; IS__CGSCC____: [[META4:![0-9]+]] = !{!"int", !5, i64 0}
-; IS__CGSCC____: [[META5:![0-9]+]] = !{!"omnipotent char", !6, i64 0}
-; IS__CGSCC____: [[META6:![0-9]+]] = !{!"Simple C/C++ TBAA"}
-; IS__CGSCC____: [[TBAA7]] = !{!8, !9, i64 12}
-; IS__CGSCC____: [[META8:![0-9]+]] = !{!"S", !4, i64 0, !4, i64 4, !4, i64 8, !9, i64 12, !9, i64 16, !9, i64 20}
-; IS__CGSCC____: [[META9:![0-9]+]] = !{!"float", !5, i64 0}
-; IS__CGSCC____: [[TBAA10]] = !{!8, !9, i64 16}
-; IS__CGSCC____: [[TBAA11]] = !{!8, !9, i64 20}
-; IS__CGSCC____: [[TBAA12]] = !{!8, !4, i64 0}
-; IS__CGSCC____: [[TBAA13]] = !{!8, !4, i64 4}
-; IS__CGSCC____: [[TBAA14]] = !{!8, !4, i64 8}
-; IS__CGSCC____: [[TBAA15]] = !{!5, !5, i64 0}
-; IS__CGSCC____: [[META16:![0-9]+]] = distinct !{!16, !17}
-; IS__CGSCC____: [[META17:![0-9]+]] = !{!"llvm.loop.mustprogress"}
-; IS__CGSCC____: [[TBAA18]] = !{!9, !9, i64 0}
-; IS__CGSCC____: [[META19:![0-9]+]] = distinct !{!19, !17}
-; IS__CGSCC____: [[TBAA20]] = !{!21, !21, i64 0}
-; IS__CGSCC____: [[META21:![0-9]+]] = !{!"long long", !5, i64 0}
-; IS__CGSCC____: [[META22:![0-9]+]] = distinct !{!22, !17}
-; IS__CGSCC____: [[META23:![0-9]+]] = distinct !{!23, !17}
-; IS__CGSCC____: [[LOOP24]] = distinct !{!24, !17}
-; IS__CGSCC____: [[LOOP25]] = distinct !{!25, !17}
-; IS__CGSCC____: [[LOOP26]] = distinct !{!26, !17}
-; IS__CGSCC____: [[LOOP27]] = distinct !{!27, !17}
-; IS__CGSCC____: [[LOOP28]] = distinct !{!28, !17}
-; IS__CGSCC____: [[LOOP29]] = distinct !{!29, !17}
-; IS__CGSCC____: [[LOOP30]] = distinct !{!30, !17}
-; IS__CGSCC____: [[LOOP31]] = distinct !{!31, !17}
+; CHECK: [[META0:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; CHECK: [[META1:![0-9]+]] = !{i32 7, !"uwtable", i32 1}
+; CHECK: [[META2:![0-9]+]] = !{!"clang version 13.0.0"}
+; CHECK: [[TBAA3]] = !{!4, !4, i64 0}
+; CHECK: [[META4:![0-9]+]] = !{!"int", !5, i64 0}
+; CHECK: [[META5:![0-9]+]] = !{!"omnipotent char", !6, i64 0}
+; CHECK: [[META6:![0-9]+]] = !{!"Simple C/C++ TBAA"}
+; CHECK: [[META7:![0-9]+]] = !{!8, !9, i64 12}
+; CHECK: [[META8:![0-9]+]] = !{!"S", !4, i64 0, !4, i64 4, !4, i64 8, !9, i64 12, !9, i64 16, !9, i64 20}
+; CHECK: [[META9:![0-9]+]] = !{!"float", !5, i64 0}
+; CHECK: [[META10:![0-9]+]] = !{!8, !9, i64 16}
+; CHECK: [[META11:![0-9]+]] = !{!8, !9, i64 20}
+; CHECK: [[META12:![0-9]+]] = !{!8, !4, i64 0}
+; CHECK: [[META13:![0-9]+]] = !{!8, !4, i64 4}
+; CHECK: [[META14:![0-9]+]] = !{!8, !4, i64 8}
+; CHECK: [[META15:![0-9]+]] = distinct !{!15, !16}
+; CHECK: [[META16:![0-9]+]] = !{!"llvm.loop.mustprogress"}
+; CHECK: [[META17:![0-9]+]] = distinct !{!17, !16}
+; CHECK: [[META18:![0-9]+]] = distinct !{!18, !16}
+; CHECK: [[META19:![0-9]+]] = !{!5, !5, i64 0}
+; CHECK: [[META20:![0-9]+]] = distinct !{!20, !16}
+; CHECK: [[META21:![0-9]+]] = distinct !{!21, !16}
+; CHECK: [[META22:![0-9]+]] = distinct !{!22, !16}
+; CHECK: [[META23:![0-9]+]] = distinct !{!23, !16}
+; CHECK: [[META24:![0-9]+]] = distinct !{!24, !16}
+; CHECK: [[META25:![0-9]+]] = distinct !{!25, !16}
+; CHECK: [[META26:![0-9]+]] = !{!9, !9, i64 0}
+; CHECK: [[META27:![0-9]+]] = distinct !{!27, !16}
+; CHECK: [[META28:![0-9]+]] = !{!29, !29, i64 0}
+; CHECK: [[META29:![0-9]+]] = !{!"long long", !5, i64 0}
+; CHECK: [[META30:![0-9]+]] = distinct !{!30, !16}
+; CHECK: [[META31:![0-9]+]] = distinct !{!31, !16}
 ;.
diff --git a/llvm/test/Transforms/Attributor/value-simplify.ll b/llvm/test/Transforms/Attributor/value-simplify.ll
--- a/llvm/test/Transforms/Attributor/value-simplify.ll
+++ b/llvm/test/Transforms/Attributor/value-simplify.ll
@@ -1040,8 +1040,8 @@
 define void @test_callee_is_undef(void (i32)* %fn) {
 ; IS__TUNIT____-LABEL: define {{[^@]+}}@test_callee_is_undef
 ; IS__TUNIT____-SAME: (void (i32)* nocapture nofree [[FN:%.*]]) {
-; IS__TUNIT____-NEXT:    call void @callee_is_undef()
-; IS__TUNIT____-NEXT:    call void @unknown_calle_arg_is_undef(void (i32)* nocapture nofree [[FN]])
+; IS__TUNIT____-NEXT:    call void @callee_is_undef(void ()* undef)
+; IS__TUNIT____-NEXT:    call void @unknown_calle_arg_is_undef(void (i32)* nocapture nofree [[FN]], i32 undef)
 ; IS__TUNIT____-NEXT:    ret void
 ;
 ; IS__CGSCC____-LABEL: define {{[^@]+}}@test_callee_is_undef
@@ -1054,7 +1054,8 @@
 }
 define internal void @callee_is_undef(void ()* %fn) {
 ;
-; IS__TUNIT____-LABEL: define {{[^@]+}}@callee_is_undef() {
+; IS__TUNIT____-LABEL: define {{[^@]+}}@callee_is_undef
+; IS__TUNIT____-SAME: (void ()* nocapture nofree noundef nonnull [[FN:%.*]]) {
 ; IS__TUNIT____-NEXT:    call void undef()
 ; IS__TUNIT____-NEXT:    ret void
 ;
@@ -1068,10 +1069,15 @@
 }
 define internal void @unknown_calle_arg_is_undef(void (i32)* %fn, i32 %arg) {
 ;
-; CHECK-LABEL: define {{[^@]+}}@unknown_calle_arg_is_undef
-; CHECK-SAME: (void (i32)* nocapture nofree noundef nonnull [[FN:%.*]]) {
-; CHECK-NEXT:    call void [[FN]](i32 undef)
-; CHECK-NEXT:    ret void
+; IS__TUNIT____-LABEL: define {{[^@]+}}@unknown_calle_arg_is_undef
+; IS__TUNIT____-SAME: (void (i32)* nocapture nofree noundef nonnull [[FN:%.*]], i32 [[ARG:%.*]]) {
+; IS__TUNIT____-NEXT:    call void [[FN]](i32 undef)
+; IS__TUNIT____-NEXT:    ret void
+;
+; IS__CGSCC____-LABEL: define {{[^@]+}}@unknown_calle_arg_is_undef
+; IS__CGSCC____-SAME: (void (i32)* nocapture nofree noundef nonnull [[FN:%.*]]) {
+; IS__CGSCC____-NEXT:    call void [[FN]](i32 undef)
+; IS__CGSCC____-NEXT:    ret void
 ;
   call void %fn(i32 %arg)
   ret void