diff --git a/clang/lib/CodeGen/ItaniumCXXABI.cpp b/clang/lib/CodeGen/ItaniumCXXABI.cpp
--- a/clang/lib/CodeGen/ItaniumCXXABI.cpp
+++ b/clang/lib/CodeGen/ItaniumCXXABI.cpp
@@ -2485,6 +2485,21 @@
     CGF.EmitBlock(InitCheckBlock);
   }
 
+  // The semantics of dynamic initialization of variables with static or thread
+  // storage duration depends on whether they are declared at block-scope. The
+  // initialization of such variables at block-scope can be aborted with an
+  // exception and later retried (per C++20 [stmt.dcl]p4), and recursive entry
+  // to their initialization has undefined behavior (also per C++20
+  // [stmt.dcl]p4). For such variables declared at non-block scope, exceptions
+  // lead to termination (per C++20 [except.terminate]p1), and recursive
+  // references to the variables are governed only by the lifetime rules (per
+  // C++20 [class.cdtor]p2), which means such references are perfectly fine as
+  // long as they avoid touching memory. As a result, block-scope variables must
+  // not be marked as initialized until after initialization completes (unless
+  // the mark is reverted following an exception), but non-block-scope variables
+  // must be marked prior to initialization so that recursive accesses during
+  // initialization do not restart initialization.
+
   // Variables used when coping with thread-safe statics and exceptions.
   if (threadsafe) {
     // Call __cxa_guard_acquire.
@@ -2500,6 +2515,12 @@
     CGF.EHStack.pushCleanup<CallGuardAbort>(EHCleanup, guard);
 
     CGF.EmitBlock(InitBlock);
+  } else if (!D.isLocalVarDecl()) {
+    // For non-local variables, store 1 into the first byte of the guard
+    // variable before the object initialization begins so that references
+    // to the variable during initialization don't restart initialization.
+    Builder.CreateStore(llvm::ConstantInt::get(CGM.Int8Ty, 1),
+                        Builder.CreateElementBitCast(guardAddr, CGM.Int8Ty));
   }
 
   // Emit the initializer and add a global destructor if appropriate.
@@ -2512,9 +2533,10 @@
     // Call __cxa_guard_release.  This cannot throw.
     CGF.EmitNounwindRuntimeCall(getGuardReleaseFn(CGM, guardPtrTy),
                                 guardAddr.getPointer());
-  } else {
-    // Store 1 into the first byte of the guard variable after initialization is
-    // complete.
+  } else if (D.isLocalVarDecl()) {
+    // For local variables, store 1 into the first byte of the guard variable
+    // after the object initialization completes so that initialization is
+    // retried if initialization is interrupted by an exception.
     Builder.CreateStore(llvm::ConstantInt::get(CGM.Int8Ty, 1),
                         Builder.CreateElementBitCast(guardAddr, CGM.Int8Ty));
   }
diff --git a/clang/test/CodeGenCXX/aix-static-init-temp-spec-and-inline-var.cpp b/clang/test/CodeGenCXX/aix-static-init-temp-spec-and-inline-var.cpp
--- a/clang/test/CodeGenCXX/aix-static-init-temp-spec-and-inline-var.cpp
+++ b/clang/test/CodeGenCXX/aix-static-init-temp-spec-and-inline-var.cpp
@@ -127,9 +127,9 @@
 // CHECK:   br i1 %guard.uninitialized, label %init.check, label %init.end
 
 // CHECK: init.check:
+// CHECK:   store i8 1, ptr @_ZGVN5test21AIvE8instanceE, align 8
 // CHECK:   call void @_ZN5test21AIvEC1Ev(ptr {{[^,]*}} @_ZN5test21AIvE8instanceE)
 // CHECK:   %1 = call i32 @atexit(ptr @__dtor__ZN5test21AIvE8instanceE)
-// CHECK:   store i8 1, ptr @_ZGVN5test21AIvE8instanceE, align 8
 // CHECK:   br label %init.end
 
 // CHECK: init.end:
@@ -190,10 +190,10 @@
 // CHECK:   br i1 %guard.uninitialized, label %init.check, label %init.end
 
 // CHECK: init.check:
+// CHECK:   store i8 1, ptr @_ZGVN5test12t1IiEE, align 8
 // CHECK32: call void @_ZN5test15Test1C1Ei(ptr {{[^,]*}} @_ZN5test12t1IiEE, i32 noundef 2)
 // CHECK64: call void @_ZN5test15Test1C1Ei(ptr {{[^,]*}} @_ZN5test12t1IiEE, i32 noundef signext 2)
 // CHECK:   %1 = call i32 @atexit(ptr @__dtor__ZN5test12t1IiEE)
-// CHECK:   store i8 1, ptr @_ZGVN5test12t1IiEE, align 8
 // CHECK:   br label %init.end
 
 // CHECK: init.end:
diff --git a/clang/test/CodeGenCXX/cxx11-thread-local.cpp b/clang/test/CodeGenCXX/cxx11-thread-local.cpp
--- a/clang/test/CodeGenCXX/cxx11-thread-local.cpp
+++ b/clang/test/CodeGenCXX/cxx11-thread-local.cpp
@@ -193,10 +193,10 @@
 // CHECK: %[[VF_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
 // CHECK: br i1 %[[VF_M_INITIALIZED]],
 // need init:
+// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1VIfE1mE to i8*)
 // CHECK: call{{.*}} i32 @_Z1gv()
 // CHECK: [[VFM_ADDR:%.+]] = call align 4 i32* @llvm.threadlocal.address.p0i32(i32* align 4 @_ZN1VIfE1mE)
 // CHECK: store i32 %{{.*}}, i32* [[VFM_ADDR]], align 4
-// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1VIfE1mE to i8*)
 // CHECK: br label
 
 // LINUX_AIX: define internal void @[[XF_M_INIT]]()
@@ -206,11 +206,11 @@
 // CHECK: %[[XF_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
 // CHECK: br i1 %[[XF_M_INITIALIZED]],
 // need init:
+// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1XIfE1mE to i8*)
 // AIX-NOT: br
 // LINUX: call {{.*}}__cxa_thread_atexit
 // AIX: call {{.*}}__pt_atexit_np
 // DARWIN: call {{.*}}_tlv_atexit
-// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1XIfE1mE to i8*)
 // CHECK: br label
 
 // LINUX: declare i32 @__cxa_thread_atexit(void (i8*)*, i8*, i8*)
@@ -329,10 +329,10 @@
 // CHECK: %[[V_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
 // CHECK: br i1 %[[V_M_INITIALIZED]],
 // need init:
+// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1VIiE1mE to i8*)
 // CHECK: call{{.*}} i32 @_Z1gv()
 // CHECK: [[VEM_ADDR:%.+]] = call align 4 i32* @llvm.threadlocal.address.p0i32(i32* align 4 @_ZN1VIiE1mE)
 // CHECK: store i32 %{{.*}}, i32* [[VEM_ADDR]], align 4
-// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1VIiE1mE to i8*)
 // CHECK: br label
 
 // LINUX_AIX: define internal void @[[X_M_INIT]]()
@@ -342,10 +342,10 @@
 // CHECK: %[[X_M_INITIALIZED:.*]] = icmp eq i8 %{{.*}}, 0
 // CHECK: br i1 %[[X_M_INITIALIZED]],
 // need init:
+// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1XIiE1mE to i8*)
 // LINUX: call {{.*}}__cxa_thread_atexit
 // AIX: call {{.*}}__pt_atexit_np
 // DARWIN: call {{.*}}_tlv_atexit
-// CHECK: store i8 1, i8* bitcast (i64* @_ZGVN1XIiE1mE to i8*)
 // CHECK: br label
 
 // CHECK: define {{.*}}@[[GLOBAL_INIT:.*]]()
diff --git a/clang/test/CodeGenCXX/cxx1y-variable-template.cpp b/clang/test/CodeGenCXX/cxx1y-variable-template.cpp
--- a/clang/test/CodeGenCXX/cxx1y-variable-template.cpp
+++ b/clang/test/CodeGenCXX/cxx1y-variable-template.cpp
@@ -40,9 +40,9 @@
   // CHECK: load {{.*}} @_ZGVN7PR4211112_GLOBAL__N_11nILi0EEE
   // CHECK: icmp eq i8 {{.*}}, 0
   // CHECK: br i1
+  // CHECK: store i8 1, ptr @_ZGVN7PR4211112_GLOBAL__N_11nILi0EEE
   // CHECK: call noundef i32 @_ZN7PR421111fEv(
   // CHECK: [[N_ADDR:%.+]] = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @_ZN7PR4211112_GLOBAL__N_11nILi0EEE)
   // CHECK: store i32 {{.*}}, ptr [[N_ADDR]]
-  // CHECK: store i8 1, ptr @_ZGVN7PR4211112_GLOBAL__N_11nILi0EEE
   int g() { return n<> + n<>; }
 }
diff --git a/clang/test/CodeGenCXX/global-init.cpp b/clang/test/CodeGenCXX/global-init.cpp
--- a/clang/test/CodeGenCXX/global-init.cpp
+++ b/clang/test/CodeGenCXX/global-init.cpp
@@ -80,9 +80,9 @@
 
   // This needs an initialization function and guard variables.
   // CHECK: load i8, i8* bitcast (i64* @_ZGVN5test41xE to i8*)
-  // CHECK: [[CALL:%.*]] = call noundef i32 @_ZN5test43fooEv
+  // CHECK: store i8 1, i8* bitcast (i64* @_ZGVN5test41xE to i8*)
+  // CHECK-NEXT: [[CALL:%.*]] = call noundef i32 @_ZN5test43fooEv
   // CHECK-NEXT: store i32 [[CALL]], i32* @_ZN5test41xE
-  // CHECK-NEXT: store i8 1, i8* bitcast (i64* @_ZGVN5test41xE to i8*)
   __attribute__((weak)) int x = foo();
 }
 
diff --git a/clang/test/CodeGenCXX/static-data-member.cpp b/clang/test/CodeGenCXX/static-data-member.cpp
--- a/clang/test/CodeGenCXX/static-data-member.cpp
+++ b/clang/test/CodeGenCXX/static-data-member.cpp
@@ -70,9 +70,9 @@
   // CHECK:      [[GUARDBYTE:%.*]] = load i8, ptr @_ZGVN5test31AIiE1xE
   // CHECK-NEXT: [[UNINITIALIZED:%.*]] = icmp eq i8 [[GUARDBYTE]], 0
   // CHECK-NEXT: br i1 [[UNINITIALIZED]]
-  // CHECK:      [[TMP:%.*]] = call noundef i32 @_ZN5test33fooEv()
+  // CHECK:      store i8 1, ptr @_ZGVN5test31AIiE1xE
+  // CHECK-NEXT: [[TMP:%.*]] = call noundef i32 @_ZN5test33fooEv()
   // CHECK-NEXT: store i32 [[TMP]], ptr @_ZN5test31AIiE1xE, align 4
-  // CHECK-NEXT: store i8 1, ptr @_ZGVN5test31AIiE1xE
   // CHECK-NEXT: br label
   // CHECK:      ret void
 }
diff --git a/clang/test/OpenMP/threadprivate_codegen.cpp b/clang/test/OpenMP/threadprivate_codegen.cpp
--- a/clang/test/OpenMP/threadprivate_codegen.cpp
+++ b/clang/test/OpenMP/threadprivate_codegen.cpp
@@ -1585,11 +1585,11 @@
 // CHECK1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
 // CHECK1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]]
 // CHECK1:       init.check:
+// CHECK1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 // CHECK1-NEXT:    call void @__kmpc_threadprivate_register(ptr @[[GLOB1]], ptr @_ZN2STI2S4E2stE, ptr @.__kmpc_global_ctor_..9, ptr null, ptr @.__kmpc_global_dtor_..10)
 // CHECK1-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23)
 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]]
-// CHECK1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK1-NEXT:    br label [[INIT_END]]
 // CHECK1:       init.end:
 // CHECK1-NEXT:    ret void
@@ -2214,11 +2214,11 @@
 // CHECK2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
 // CHECK2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]]
 // CHECK2:       init.check:
+// CHECK2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 // CHECK2-NEXT:    call void @__kmpc_threadprivate_register(ptr @[[GLOB1]], ptr @_ZN2STI2S4E2stE, ptr @.__kmpc_global_ctor_..9, ptr null, ptr @.__kmpc_global_dtor_..10)
 // CHECK2-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23)
 // CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]]
-// CHECK2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK2-NEXT:    br label [[INIT_END]]
 // CHECK2:       init.end:
 // CHECK2-NEXT:    ret void
@@ -2697,9 +2697,9 @@
 // SIMD1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
 // SIMD1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]]
 // SIMD1:       init.check:
+// SIMD1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // SIMD1-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23)
 // SIMD1-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]]
-// SIMD1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // SIMD1-NEXT:    br label [[INIT_END]]
 // SIMD1:       init.end:
 // SIMD1-NEXT:    ret void
@@ -3167,9 +3167,9 @@
 // SIMD2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG234]]
 // SIMD2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG234]]
 // SIMD2:       init.check:
+// SIMD2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG234]]
 // SIMD2-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23), !dbg [[DBG235:![0-9]+]]
 // SIMD2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]], !dbg [[DBG234]]
-// SIMD2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG234]]
 // SIMD2-NEXT:    br label [[INIT_END]], !dbg [[DBG234]]
 // SIMD2:       init.end:
 // SIMD2-NEXT:    ret void, !dbg [[DBG237:![0-9]+]]
@@ -3767,9 +3767,9 @@
 // CHECK-TLS1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
 // CHECK-TLS1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]]
 // CHECK-TLS1:       init.check:
+// CHECK-TLS1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK-TLS1-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23)
 // CHECK-TLS1-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]]
-// CHECK-TLS1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK-TLS1-NEXT:    br label [[INIT_END]]
 // CHECK-TLS1:       init.end:
 // CHECK-TLS1-NEXT:    ret void
@@ -4302,9 +4302,9 @@
 // CHECK-TLS2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
 // CHECK-TLS2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]]
 // CHECK-TLS2:       init.check:
+// CHECK-TLS2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK-TLS2-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23)
 // CHECK-TLS2-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR4]]
-// CHECK-TLS2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // CHECK-TLS2-NEXT:    br label [[INIT_END]]
 // CHECK-TLS2:       init.end:
 // CHECK-TLS2-NEXT:    ret void
@@ -4846,9 +4846,9 @@
 // CHECK-TLS3-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG291]]
 // CHECK-TLS3-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG291]]
 // CHECK-TLS3:       init.check:
+// CHECK-TLS3-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG291]]
 // CHECK-TLS3-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23), !dbg [[DBG292:![0-9]+]]
 // CHECK-TLS3-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]], !dbg [[DBG291]]
-// CHECK-TLS3-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG291]]
 // CHECK-TLS3-NEXT:    br label [[INIT_END]], !dbg [[DBG291]]
 // CHECK-TLS3:       init.end:
 // CHECK-TLS3-NEXT:    ret void, !dbg [[DBG294:![0-9]+]]
@@ -5408,9 +5408,9 @@
 // CHECK-TLS4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG291]]
 // CHECK-TLS4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG291]]
 // CHECK-TLS4:       init.check:
+// CHECK-TLS4-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG291]]
 // CHECK-TLS4-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23), !dbg [[DBG292:![0-9]+]]
 // CHECK-TLS4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_thread_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR5]], !dbg [[DBG291]]
-// CHECK-TLS4-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG291]]
 // CHECK-TLS4-NEXT:    br label [[INIT_END]], !dbg [[DBG291]]
 // CHECK-TLS4:       init.end:
 // CHECK-TLS4-NEXT:    ret void, !dbg [[DBG294:![0-9]+]]
@@ -5810,9 +5810,9 @@
 // SIMD3-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0
 // SIMD3-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]]
 // SIMD3:       init.check:
+// SIMD3-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // SIMD3-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23)
 // SIMD3-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]]
-// SIMD3-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8
 // SIMD3-NEXT:    br label [[INIT_END]]
 // SIMD3:       init.end:
 // SIMD3-NEXT:    ret void
@@ -6280,9 +6280,9 @@
 // SIMD4-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG234]]
 // SIMD4-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG234]]
 // SIMD4:       init.check:
+// SIMD4-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG234]]
 // SIMD4-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23), !dbg [[DBG235:![0-9]+]]
 // SIMD4-NEXT:    [[TMP1:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR3]], !dbg [[DBG234]]
-// SIMD4-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG234]]
 // SIMD4-NEXT:    br label [[INIT_END]], !dbg [[DBG234]]
 // SIMD4:       init.end:
 // SIMD4-NEXT:    ret void, !dbg [[DBG237:![0-9]+]]
@@ -7076,11 +7076,11 @@
 // DEBUG1-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG334]]
 // DEBUG1-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG334]]
 // DEBUG1:       init.check:
+// DEBUG1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG334]]
 // DEBUG1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB41:[0-9]+]]), !dbg [[DBG334]]
 // DEBUG1-NEXT:    call void @__kmpc_threadprivate_register(ptr @[[GLOB41]], ptr @_ZN2STI2S4E2stE, ptr @.__kmpc_global_ctor_..9, ptr null, ptr @.__kmpc_global_dtor_..10), !dbg [[DBG334]]
 // DEBUG1-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23), !dbg [[DBG335:![0-9]+]]
 // DEBUG1-NEXT:    [[TMP2:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR4]], !dbg [[DBG334]]
-// DEBUG1-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG334]]
 // DEBUG1-NEXT:    br label [[INIT_END]], !dbg [[DBG334]]
 // DEBUG1:       init.end:
 // DEBUG1-NEXT:    ret void, !dbg [[DBG337:![0-9]+]]
@@ -7719,11 +7719,11 @@
 // DEBUG2-NEXT:    [[GUARD_UNINITIALIZED:%.*]] = icmp eq i8 [[TMP0]], 0, !dbg [[DBG285]]
 // DEBUG2-NEXT:    br i1 [[GUARD_UNINITIALIZED]], label [[INIT_CHECK:%.*]], label [[INIT_END:%.*]], !dbg [[DBG285]]
 // DEBUG2:       init.check:
+// DEBUG2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG285]]
 // DEBUG2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB41:[0-9]+]]), !dbg [[DBG285]]
 // DEBUG2-NEXT:    call void @__kmpc_threadprivate_register(ptr @[[GLOB41]], ptr @_ZN2STI2S4E2stE, ptr @.__kmpc_global_ctor_..8, ptr null, ptr @.__kmpc_global_dtor_..9), !dbg [[DBG285]]
 // DEBUG2-NEXT:    call void @_ZN2S4C1Ei(ptr noundef nonnull align 4 dereferenceable(8) @_ZN2STI2S4E2stE, i32 noundef 23), !dbg [[DBG286:![0-9]+]]
 // DEBUG2-NEXT:    [[TMP2:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2S4D1Ev, ptr @_ZN2STI2S4E2stE, ptr @__dso_handle) #[[ATTR4]], !dbg [[DBG285]]
-// DEBUG2-NEXT:    store i8 1, ptr @_ZGVN2STI2S4E2stE, align 8, !dbg [[DBG285]]
 // DEBUG2-NEXT:    br label [[INIT_END]], !dbg [[DBG285]]
 // DEBUG2:       init.end:
 // DEBUG2-NEXT:    ret void, !dbg [[DBG288:![0-9]+]]