diff --git a/clang/test/OpenMP/cancel_codegen.cpp b/clang/test/OpenMP/cancel_codegen.cpp
--- a/clang/test/OpenMP/cancel_codegen.cpp
+++ b/clang/test/OpenMP/cancel_codegen.cpp
@@ -1560,8 +1560,6 @@
 // CHECK3-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK3-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK3-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK3:       omp.par.outlined.exit.exitStub:
-// CHECK3-NEXT:    ret void
 // CHECK3:       omp.par.region:
 // CHECK3-NEXT:    [[TMP1:%.*]] = load float, float* @flag, align 4
 // CHECK3-NEXT:    [[TOBOOL:%.*]] = fcmp une float [[TMP1]], 0.000000e+00
@@ -1607,6 +1605,8 @@
 // CHECK3-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB]]
 // CHECK3:       .split:
 // CHECK3-NEXT:    br label [[TMP3]]
+// CHECK3:       omp.par.outlined.exit.exitStub:
+// CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@.omp_task_entry.
@@ -2187,8 +2187,6 @@
 // CHECK4-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK4-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK4-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK4:       omp.par.outlined.exit.exitStub:
-// CHECK4-NEXT:    ret void
 // CHECK4:       omp.par.region:
 // CHECK4-NEXT:    [[TMP1:%.*]] = load float, float* @flag, align 4
 // CHECK4-NEXT:    [[TOBOOL:%.*]] = fcmp une float [[TMP1]], 0.000000e+00
@@ -2234,6 +2232,8 @@
 // CHECK4-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB]]
 // CHECK4:       .split:
 // CHECK4-NEXT:    br label [[TMP3]]
+// CHECK4:       omp.par.outlined.exit.exitStub:
+// CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@.omp_task_entry.
@@ -4054,8 +4054,6 @@
 // CHECK9-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK9-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK9-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK9:       omp.par.outlined.exit.exitStub:
-// CHECK9-NEXT:    ret void
 // CHECK9:       omp.par.region:
 // CHECK9-NEXT:    [[TMP1:%.*]] = load float, float* @flag, align 4
 // CHECK9-NEXT:    [[TOBOOL:%.*]] = fcmp une float [[TMP1]], 0.000000e+00
@@ -4101,6 +4099,8 @@
 // CHECK9-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB]]
 // CHECK9:       .split:
 // CHECK9-NEXT:    br label [[TMP3]]
+// CHECK9:       omp.par.outlined.exit.exitStub:
+// CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@.omp_task_entry.
@@ -4681,8 +4681,6 @@
 // CHECK10-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK10-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK10-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK10:       omp.par.outlined.exit.exitStub:
-// CHECK10-NEXT:    ret void
 // CHECK10:       omp.par.region:
 // CHECK10-NEXT:    [[TMP1:%.*]] = load float, float* @flag, align 4
 // CHECK10-NEXT:    [[TOBOOL:%.*]] = fcmp une float [[TMP1]], 0.000000e+00
@@ -4728,6 +4726,8 @@
 // CHECK10-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB]]
 // CHECK10:       .split:
 // CHECK10-NEXT:    br label [[TMP3]]
+// CHECK10:       omp.par.outlined.exit.exitStub:
+// CHECK10-NEXT:    ret void
 //
 //
 // CHECK10-LABEL: define {{[^@]+}}@.omp_task_entry.
diff --git a/clang/test/OpenMP/parallel_codegen.cpp b/clang/test/OpenMP/parallel_codegen.cpp
--- a/clang/test/OpenMP/parallel_codegen.cpp
+++ b/clang/test/OpenMP/parallel_codegen.cpp
@@ -753,8 +753,6 @@
 // CHECK3-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK3-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK3-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK3:       omp.par.outlined.exit.exitStub:
-// CHECK3-NEXT:    ret void
 // CHECK3:       omp.par.region:
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[VLA]], i64 1
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
@@ -765,6 +763,8 @@
 // CHECK3-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 // CHECK3:       omp.par.pre_finalize:
 // CHECK3-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+// CHECK3:       omp.par.outlined.exit.exitStub:
+// CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_Z3fooIiEvT_
@@ -809,8 +809,6 @@
 // CHECK3-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTRELOADED]], align 8
 // CHECK3-NEXT:    [[VAR:%.*]] = alloca double*, align 8
 // CHECK3-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK3:       omp.par.outlined.exit.exitStub:
-// CHECK3-NEXT:    ret void
 // CHECK3:       omp.par.region:
 // CHECK3-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[ARGC_ADDR]], align 8
 // CHECK3-NEXT:    call void @_Z3fooIPPcEvT_(i8** [[TMP2]])
@@ -821,6 +819,8 @@
 // CHECK3-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 // CHECK3:       omp.par.pre_finalize:
 // CHECK3-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+// CHECK3:       omp.par.outlined.exit.exitStub:
+// CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@_Z3fooIPPcEvT_
@@ -877,8 +877,6 @@
 // CHECK4-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK4-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 // CHECK4-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK4:       omp.par.outlined.exit.exitStub:
-// CHECK4-NEXT:    ret void
 // CHECK4:       omp.par.region:
 // CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[VLA]], i64 1, !dbg [[DBG34:![0-9]+]]
 // CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX]], align 4, !dbg [[DBG34]]
@@ -889,6 +887,8 @@
 // CHECK4-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]], !dbg [[DBG34]]
 // CHECK4:       omp.par.pre_finalize:
 // CHECK4-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]], !dbg [[DBG34]]
+// CHECK4:       omp.par.outlined.exit.exitStub:
+// CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_Z3fooIiEvT_
@@ -935,8 +935,6 @@
 // CHECK4-NEXT:    [[TMP1:%.*]] = load i64, i64* [[DOTRELOADED]], align 8
 // CHECK4-NEXT:    [[VAR:%.*]] = alloca double*, align 8
 // CHECK4-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-// CHECK4:       omp.par.outlined.exit.exitStub:
-// CHECK4-NEXT:    ret void
 // CHECK4:       omp.par.region:
 // CHECK4-NEXT:    [[TMP2:%.*]] = load i8**, i8*** [[ARGC_ADDR]], align 8, !dbg [[DBG57:![0-9]+]]
 // CHECK4-NEXT:    call void @_Z3fooIPPcEvT_(i8** [[TMP2]]), !dbg [[DBG57]]
@@ -948,6 +946,8 @@
 // CHECK4-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]], !dbg [[DBG66:![0-9]+]]
 // CHECK4:       omp.par.pre_finalize:
 // CHECK4-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]], !dbg [[DBG66]]
+// CHECK4:       omp.par.outlined.exit.exitStub:
+// CHECK4-NEXT:    ret void
 //
 //
 // CHECK4-LABEL: define {{[^@]+}}@_Z3fooIPPcEvT_
@@ -958,4 +958,4 @@
 // CHECK4-NEXT:    call void @llvm.dbg.declare(metadata i8*** [[ARGC_ADDR]], metadata [[META70:![0-9]+]], metadata !DIExpression()), !dbg [[DBG71:![0-9]+]]
 // CHECK4-NEXT:    ret void, !dbg [[DBG71]]
 //
-//
\ No newline at end of file
+//
diff --git a/llvm/lib/Transforms/IPO/IROutliner.cpp b/llvm/lib/Transforms/IPO/IROutliner.cpp
--- a/llvm/lib/Transforms/IPO/IROutliner.cpp
+++ b/llvm/lib/Transforms/IPO/IROutliner.cpp
@@ -434,7 +434,6 @@
 static BasicBlock *moveFunctionData(Function &Old, Function &New) {
   Function::iterator CurrBB, NextBB, FinalBB;
   BasicBlock *NewEnd = nullptr;
-  std::vector<Instruction *> DebugInsts;
   for (CurrBB = Old.begin(), FinalBB = Old.end(); CurrBB != FinalBB;
        CurrBB = NextBB) {
     NextBB = std::next(CurrBB);
@@ -444,6 +443,8 @@
     if (isa<ReturnInst>(I))
       NewEnd = &(*CurrBB);
 
+    std::vector<Instruction *> DebugInsts;
+
     for (Instruction &Val : *CurrBB) {
       // We must handle the scoping of called functions differently than
       // other outlined instructions.
diff --git a/llvm/lib/Transforms/Utils/CodeExtractor.cpp b/llvm/lib/Transforms/Utils/CodeExtractor.cpp
--- a/llvm/lib/Transforms/Utils/CodeExtractor.cpp
+++ b/llvm/lib/Transforms/Utils/CodeExtractor.cpp
@@ -1389,12 +1389,17 @@
   Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList();
   Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList();
 
+  auto newFuncIt = newFunction->front().getIterator();
   for (BasicBlock *Block : Blocks) {
     // Delete the basic block from the old function, and the list of blocks
     oldBlocks.remove(Block);
 
     // Insert this basic block into the new function
-    newBlocks.push_back(Block);
+    // Insert the original blocks after the entry block created
+    // for the new function. The entry block may be followed
+    // by a set of exit blocks at this point, but these exit
+    // blocks better be placed at the end of the new function.
+    newFuncIt = newBlocks.insertAfter(newFuncIt, Block);
   }
 }
 
diff --git a/llvm/test/Transforms/CodeExtractor/LoopExtractor.ll b/llvm/test/Transforms/CodeExtractor/LoopExtractor.ll
--- a/llvm/test/Transforms/CodeExtractor/LoopExtractor.ll
+++ b/llvm/test/Transforms/CodeExtractor/LoopExtractor.ll
@@ -42,8 +42,6 @@
 ; CHECK-LABEL: define internal void @test3.loop.1()
 ; CHECK-NEXT:  newFuncRoot:
 ; CHECK-NEXT:    br label %loop.1
-; CHECK:       exit.exitStub:
-; CHECK-NEXT:    ret void
 ; CHECK:       loop.1:
 ; CHECK-NEXT:    %index.1 = phi i32 [ %next.1, %loop.1.loop.1_crit_edge ], [ 10, %newFuncRoot ]
 ; CHECK-NEXT:    tail call void @foo()
@@ -52,12 +50,12 @@
 ; CHECK-NEXT:    br i1 %repeat.1, label %loop.1.loop.1_crit_edge, label %exit.exitStub
 ; CHECK:       loop.1.loop.1_crit_edge:
 ; CHECK-NEXT:    br label %loop.1
+; CHECK:       exit.exitStub:
+; CHECK-NEXT:    ret void
 
 ; CHECK-LABEL: define internal void @test3.loop.0()
 ; CHECK-NEXT:  newFuncRoot:
 ; CHECK-NEXT:    br label %loop.0
-; CHECK:       loop.0.loop.1_crit_edge.exitStub:
-; CHECK-NEXT:    ret void
 ; CHECK:       loop.0:
 ; CHECK-NEXT:    %index.0 = phi i32 [ 10, %newFuncRoot ], [ %next.0, %loop.0.loop.0_crit_edge ]
 ; CHECK-NEXT:    tail call void @foo()
@@ -66,3 +64,5 @@
 ; CHECK-NEXT:    br i1 %repeat.0, label %loop.0.loop.0_crit_edge, label %loop.0.loop.1_crit_edge.exitStub
 ; CHECK:       loop.0.loop.0_crit_edge:
 ; CHECK-NEXT:    br label %loop.0
+; CHECK:       loop.0.loop.1_crit_edge.exitStub:
+; CHECK-NEXT:    ret void
diff --git a/llvm/test/Transforms/CodeExtractor/LoopExtractor_crash.ll b/llvm/test/Transforms/CodeExtractor/LoopExtractor_crash.ll
--- a/llvm/test/Transforms/CodeExtractor/LoopExtractor_crash.ll
+++ b/llvm/test/Transforms/CodeExtractor/LoopExtractor_crash.ll
@@ -37,10 +37,10 @@
 ; CHECK-LABEL: define internal void @test.loopentry()
 ; CHECK-NEXT:  newFuncRoot:
 ; CHECK-NEXT:    br label %loopentry
-; CHECK:       loopexit.exitStub:
-; CHECK-NEXT:    ret void
 ; CHECK:       loopentry:
 ; CHECK-NEXT:    br i1 false, label %loopbody, label %loopexit.exitStub
 ; CHECK:       loopbody:
 ; CHECK-NEXT:    call void @foo()
 ; CHECK-NEXT:    br label %loopentry
+; CHECK:       loopexit.exitStub:
+; CHECK-NEXT:    ret void
diff --git a/llvm/test/Transforms/CodeExtractor/LoopExtractor_infinite.ll b/llvm/test/Transforms/CodeExtractor/LoopExtractor_infinite.ll
--- a/llvm/test/Transforms/CodeExtractor/LoopExtractor_infinite.ll
+++ b/llvm/test/Transforms/CodeExtractor/LoopExtractor_infinite.ll
@@ -35,13 +35,13 @@
 ; CHECK-LABEL: define internal i1 @test.loopentry()
 ; CHECK-NEXT:  newFuncRoot:
 ; CHECK-NEXT:    br label %loopentry
-; CHECK:       exit.1.exitStub:
-; CHECK-NEXT:    ret i1 true
-; CHECK:       exit.0.exitStub:
-; CHECK-NEXT:    ret i1 false
 ; CHECK:       loopentry:
 ; CHECK-NEXT:    br i1 true, label %exit.1.exitStub, label %loopexit
 ; CHECK:       loopexit:
 ; CHECK-NEXT:    br i1 false, label %loopexit.loopentry_crit_edge, label %exit.0.exitStub
 ; CHECK:       loopexit.loopentry_crit_edge:
 ; CHECK-NEXT:    br label %loopentry
+; CHECK:       exit.1.exitStub:
+; CHECK-NEXT:    ret i1 true
+; CHECK:       exit.0.exitStub:
+; CHECK-NEXT:    ret i1 false
diff --git a/llvm/test/Transforms/CodeExtractor/live_shrink_unsafe.ll b/llvm/test/Transforms/CodeExtractor/live_shrink_unsafe.ll
--- a/llvm/test/Transforms/CodeExtractor/live_shrink_unsafe.ll
+++ b/llvm/test/Transforms/CodeExtractor/live_shrink_unsafe.ll
@@ -72,13 +72,13 @@
 ; CHECK: newFuncRoot:
 ; CHECK-NEXT: br label %bb3
 
-; CHECK: bb4.exitStub:
-; CHECK-NEXT: ret void
-
 ; CHECK: bb3:
 ; CHECK-NOT: lifetime.ed
 ; CHECK: br label %bb4.exitStub
 
+; CHECK: bb4.exitStub:
+; CHECK-NEXT: ret void
+
 
 
 !llvm.module.flags = !{!0}
diff --git a/llvm/test/Transforms/CodeExtractor/unreachable-block.ll b/llvm/test/Transforms/CodeExtractor/unreachable-block.ll
--- a/llvm/test/Transforms/CodeExtractor/unreachable-block.ll
+++ b/llvm/test/Transforms/CodeExtractor/unreachable-block.ll
@@ -9,11 +9,11 @@
 ; CHECK-LABEL: define internal void @tinkywinky.1.ontrue() {
 ; CHECK-NEXT: newFuncRoot:
 ; CHECK-NEXT:   br label %ontrue
-; CHECK: onfalse{{.*}}:
-; CHECK-NEXT:   ret void
 ; CHECK: ontrue:
 ; CHECK-NEXT:   call void @patatino()
 ; CHECK-NEXT:   br label %onfalse{{.*}}
+; CHECK: onfalse{{.*}}:
+; CHECK-NEXT:   ret void
 ; CHECK-NEXT: }
 
 declare void @patatino()
diff --git a/llvm/test/Transforms/HotColdSplit/multiple-exits.ll b/llvm/test/Transforms/HotColdSplit/multiple-exits.ll
--- a/llvm/test/Transforms/HotColdSplit/multiple-exits.ll
+++ b/llvm/test/Transforms/HotColdSplit/multiple-exits.ll
@@ -56,16 +56,16 @@
 
 ; CHECK-LABEL: define {{.*}}@foo.cold.1(
 ; CHECK: br
-; CHECK: [[exit1Stub:.*]]:
-; CHECK-NEXT: ret i1 true
-; CHECK: [[returnStub:.*]]:
-; CHECK-NEXT: ret i1 false
 ; CHECK: call {{.*}}@sink
 ; CHECK-NEXT: [[cmp:%.*]] = icmp
-; CHECK-NEXT: br i1 [[cmp]], label %[[exit1Stub]], label %exit2
+; CHECK-NEXT: br i1 [[cmp]], label %[[exit1Stub:.*]], label %exit2
 ; CHECK-LABEL: exit2:
 ; CHECK-NEXT: call {{.*}}@sideeffect(i32 2)
-; CHECK-NEXT: br label %[[returnStub]]
+; CHECK-NEXT: br label %[[returnStub:.*]]
+; CHECK: [[exit1Stub]]:
+; CHECK-NEXT: ret i1 true
+; CHECK: [[returnStub]]:
+; CHECK-NEXT: ret i1 false
 
 declare void @sink(...) cold
 
diff --git a/llvm/test/Transforms/IROutliner/outlining-different-output-blocks.ll b/llvm/test/Transforms/IROutliner/outlining-different-output-blocks.ll
--- a/llvm/test/Transforms/IROutliner/outlining-different-output-blocks.ll
+++ b/llvm/test/Transforms/IROutliner/outlining-different-output-blocks.ll
@@ -86,11 +86,6 @@
 }
 
 ; CHECK: define internal void @outlined_ir_func_0(i32* [[ARG0:%.*]], i32* [[ARG1:%.*]], i32* [[ARG2:%.*]], i32* [[ARG3:%.*]], i32* [[ARG4:%.*]], i32 [[ARG5:%.*]]) #1 {
-; CHECK: _after_outline.exitStub:
-; CHECK-NEXT:    switch i32 [[ARG5]], label [[BLOCK:%.*]] [
-; CHECK-NEXT:      i32 0, label %[[BLOCK_0:.*]]
-; CHECK-NEXT:      i32 1, label %[[BLOCK_1:.*]]
-
 ; CHECK: entry_to_outline:
 ; CHECK-NEXT:    store i32 2, i32* [[ARG0]], align 4
 ; CHECK-NEXT:    store i32 3, i32* [[ARG1]], align 4
@@ -101,6 +96,11 @@
 ; CHECK-NEXT:    store i32 [[ADD]], i32* [[ARG2]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARG2]], align 4
 
+; CHECK: _after_outline.exitStub:
+; CHECK-NEXT:    switch i32 [[ARG5]], label [[BLOCK:%.*]] [
+; CHECK-NEXT:      i32 0, label %[[BLOCK_0:.*]]
+; CHECK-NEXT:      i32 1, label %[[BLOCK_1:.*]]
+
 ; CHECK: [[BLOCK_0]]:
 ; CHECK-NEXT:    store i32 [[ADD]], i32* [[ARG3]], align 4
 ; CHECK-NEXT:    store i32 [[TMP2]], i32* [[ARG4]], align 4
diff --git a/llvm/test/Transforms/IROutliner/outlining-same-output-blocks.ll b/llvm/test/Transforms/IROutliner/outlining-same-output-blocks.ll
--- a/llvm/test/Transforms/IROutliner/outlining-same-output-blocks.ll
+++ b/llvm/test/Transforms/IROutliner/outlining-same-output-blocks.ll
@@ -84,15 +84,15 @@
 }
 
 ; CHECK: define internal void @outlined_ir_func_0(i32* [[ARG0:%.*]], i32* [[ARG1:%.*]], i32* [[ARG2:%.*]], i32* [[ARG3:%.*]], i32* [[ARG4:%.*]]) #1 {
-; CHECK: entry_after_outline.exitStub:
-; CHECK-NEXT:    store i32 [[ADD:%.*]], i32* [[ARG3]], align 4
-; CHECK-NEXT:    store i32 [[TMP2:%.*]], i32* [[ARG4]], align 4
-
 ; CHECK: entry_to_outline:
 ; CHECK-NEXT:    store i32 2, i32* [[ARG0]], align 4
 ; CHECK-NEXT:    store i32 3, i32* [[ARG1]], align 4
 ; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARG0]], align 4
 ; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[ARG1]], align 4
-; CHECK-NEXT:    [[ADD]] = add i32 [[TMP0]], [[TMP1]]
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP0]], [[TMP1]]
 ; CHECK-NEXT:    store i32 [[ADD]], i32* [[ARG2]], align 4
-; CHECK-NEXT:    [[TMP2]] = load i32, i32* [[ARG2]], align 4
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARG2]], align 4
+
+; CHECK: entry_after_outline.exitStub:
+; CHECK-NEXT:    store i32 [[ADD]], i32* [[ARG3]], align 4
+; CHECK-NEXT:    store i32 [[TMP2]], i32* [[ARG4]], align 4
diff --git a/llvm/test/Transforms/OpenMP/parallel_region_merging.ll b/llvm/test/Transforms/OpenMP/parallel_region_merging.ll
--- a/llvm/test/Transforms/OpenMP/parallel_region_merging.ll
+++ b/llvm/test/Transforms/OpenMP/parallel_region_merging.ll
@@ -4715,8 +4715,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -4731,6 +4729,8 @@
 ; CHECK1-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK1:       omp.par.pre_finalize:
 ; CHECK1-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined.
@@ -4860,8 +4860,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -4897,6 +4895,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..8
@@ -4944,8 +4944,6 @@
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TMP1:%.*]] = load float, float* [[F_RELOADED]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -4980,6 +4978,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..10
@@ -5029,8 +5029,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5069,6 +5067,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..12
@@ -5114,8 +5114,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5154,6 +5152,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..14
@@ -5206,8 +5206,6 @@
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_RELOADED]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5243,6 +5241,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..16
@@ -5291,8 +5291,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5307,6 +5305,8 @@
 ; CHECK1-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK1:       omp.par.pre_finalize:
 ; CHECK1-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..18
@@ -5368,8 +5368,6 @@
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CANCEL1_RELOADED]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5405,6 +5403,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..20
@@ -5461,8 +5461,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5480,6 +5478,8 @@
 ; CHECK1-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK1:       omp.par.pre_finalize:
 ; CHECK1-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..22
@@ -5539,8 +5539,6 @@
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_RELOADED]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5599,6 +5597,8 @@
 ; CHECK1:       omp_region.body.split:
 ; CHECK1-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK1-NEXT:    br label [[OMP_REGION_END]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..25
@@ -5764,8 +5764,6 @@
 ; CHECK1-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK1-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK1:       omp.par.outlined.exit.exitStub:
-; CHECK1-NEXT:    ret void
 ; CHECK1:       omp.par.region:
 ; CHECK1-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK1:       omp.par.merged:
@@ -5780,6 +5778,8 @@
 ; CHECK1-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK1:       omp.par.pre_finalize:
 ; CHECK1-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK1:       omp.par.outlined.exit.exitStub:
+; CHECK1-NEXT:    ret void
 ;
 ;
 ; CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..37
@@ -5832,8 +5832,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -5848,6 +5846,8 @@
 ; CHECK2-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK2:       omp.par.pre_finalize:
 ; CHECK2-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined.
@@ -5977,8 +5977,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6014,6 +6012,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..8
@@ -6061,8 +6061,6 @@
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TMP1:%.*]] = load float, float* [[F_RELOADED]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6097,6 +6095,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..10
@@ -6146,8 +6146,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6186,6 +6184,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..12
@@ -6231,8 +6231,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6271,6 +6269,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..14
@@ -6323,8 +6323,6 @@
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_RELOADED]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6360,6 +6358,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..16
@@ -6408,8 +6408,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6424,6 +6422,8 @@
 ; CHECK2-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK2:       omp.par.pre_finalize:
 ; CHECK2-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..18
@@ -6485,8 +6485,6 @@
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[CANCEL1_RELOADED]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6522,6 +6520,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..20
@@ -6578,8 +6578,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6597,6 +6595,8 @@
 ; CHECK2-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK2:       omp.par.pre_finalize:
 ; CHECK2-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..22
@@ -6656,8 +6656,6 @@
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[A_RELOADED]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6716,6 +6714,8 @@
 ; CHECK2:       omp_region.body.split:
 ; CHECK2-NEXT:    call void @__kmpc_end_master(%struct.ident_t* @[[GLOB1]], i32 [[OMP_GLOBAL_THREAD_NUM]])
 ; CHECK2-NEXT:    br label [[OMP_REGION_END]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..25
@@ -6881,8 +6881,6 @@
 ; CHECK2-NEXT:    store i32 [[TMP0]], i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    [[TID:%.*]] = load i32, i32* [[TID_ADDR_LOCAL]], align 4
 ; CHECK2-NEXT:    br label [[OMP_PAR_REGION:%.*]]
-; CHECK2:       omp.par.outlined.exit.exitStub:
-; CHECK2-NEXT:    ret void
 ; CHECK2:       omp.par.region:
 ; CHECK2-NEXT:    br label [[OMP_PAR_MERGED:%.*]]
 ; CHECK2:       omp.par.merged:
@@ -6897,6 +6895,8 @@
 ; CHECK2-NEXT:    br label [[OMP_PAR_PRE_FINALIZE:%.*]]
 ; CHECK2:       omp.par.pre_finalize:
 ; CHECK2-NEXT:    br label [[OMP_PAR_OUTLINED_EXIT_EXITSTUB:%.*]]
+; CHECK2:       omp.par.outlined.exit.exitStub:
+; CHECK2-NEXT:    ret void
 ;
 ;
 ; CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..37
diff --git a/llvm/test/tools/llvm-extract/extract-blocks-with-groups.ll b/llvm/test/tools/llvm-extract/extract-blocks-with-groups.ll
--- a/llvm/test/tools/llvm-extract/extract-blocks-with-groups.ll
+++ b/llvm/test/tools/llvm-extract/extract-blocks-with-groups.ll
@@ -8,9 +8,6 @@
 ; CHECK: newFuncRoot:
 ; CHECK:   br label %if.split
 ;
-; CHECK: end.exitStub:                                     ; preds = %end.split
-; CHECK:   ret void
-;
 ; CHECK: then:                                             ; preds = %if.split
 ; CHECK:   %tmp12 = shl i32 %arg1, 2
 ; CHECK:   %tmp13 = add nsw i32 %tmp12, %arg
@@ -32,6 +29,9 @@
 ; CHECK:   %tmp.0.ce = phi i32 [ %tmp13, %then ], [ %tmp25, %else ]
 ; CHECK:   store i32 %tmp.0.ce, i32* %tmp.0.ce.out
 ; CHECK:   br label %end.exitStub
+;
+; CHECK: end.exitStub:                                     ; preds = %end.split
+; CHECK:   ret void
 ; CHECK: }
 
 ; The second extracted function is the region composed by the blocks
@@ -40,12 +40,6 @@
 ; CHECK: newFuncRoot:
 ; CHECK:   br label %bb14
 ;
-; CHECK: bb26.exitStub:                                    ; preds = %bb14
-; CHECK:   ret i1 true
-;
-; CHECK: bb30.exitStub:                                    ; preds = %bb20
-; CHECK:   ret i1 false
-;
 ; CHECK: bb14:                                             ; preds = %newFuncRoot
 ; CHECK:   %tmp0 = and i32 %arg1, %arg
 ; CHECK:   %tmp1 = icmp slt i32 %tmp0, 0
@@ -57,6 +51,12 @@
 ; CHECK:   %tmp25 = add nsw i32 %tmp24, %tmp22
 ; CHECK:   store i32 %tmp25, i32* %tmp25.out
 ; CHECK:   br label %bb30.exitStub
+;
+; CHECK: bb26.exitStub:                                    ; preds = %bb14
+; CHECK:   ret i1 true
+;
+; CHECK: bb30.exitStub:                                    ; preds = %bb20
+; CHECK:   ret i1 false
 ; CHECK: }
 
 define i32 @foo(i32 %arg, i32 %arg1) {