[Thumb] Add test case where the machine-outliner clobbers LR.

Add ad test case where `bl OUTLINED_FUNCTION_0` clobbers LR, which in
turn is used the later call to memcpy to return to the caller.

(cherry picked from commit 20f634f)

Author: fhahn

llvm/test/CodeGen/Thumb2/outlined-fn-may-clobber-lr-in-caller.ll

@@ -0,0 +1,113 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -mtriple=thumbv7m-none-none-eabi < %s  | FileCheck %s
+
+target datalayout = "e-m:o-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+
+%struct.wibble = type { [30 x i8], i8, i32 }
+%struct.eggs = type { i32, [30 x i8], i8, i8, i8, [3 x i8] }
+
+@global = external global [3 x %struct.wibble], align 4
+@global.1 = external global [3 x %struct.wibble], align 4
+
+; Test case to make sure calling an outlined function does not clobber LR used
+; by a tail call in caller.
+; FIXME: Currently bl OUTLINED_FUNCTION_0 clobbers LR, which in turn is used
+;        by the later call to memcpy to return to the caller.
+define void @test(ptr nocapture noundef writeonly %arg, i32 noundef %arg1, i8 noundef zeroext %arg2) unnamed_addr #0 {
+; CHECK-LABEL: test:
+; CHECK:       @ %bb.0: @ %bb
+; CHECK-NEXT:    cmp r1, #2
+; CHECK-NEXT:    beq .LBB0_3
+; CHECK-NEXT:  @ %bb.1: @ %bb
+; CHECK-NEXT:    cmp r1, #1
+; CHECK-NEXT:    bne .LBB0_5
+; CHECK-NEXT:  @ %bb.2: @ %bb4
+; CHECK-NEXT:    bl OUTLINED_FUNCTION_0
+; CHECK-NEXT:    ldr r2, .LCPI0_1
+; CHECK-NEXT:    b .LBB0_4
+; CHECK-NEXT:  .LBB0_3: @ %bb14
+; CHECK-NEXT:    bl OUTLINED_FUNCTION_0
+; CHECK-NEXT:    ldr r2, .LCPI0_0
+; CHECK-NEXT:  .LBB0_4: @ %bb4
+; CHECK-NEXT:    add.w r1, r2, r1, lsl #2
+; CHECK-NEXT:    adds r0, #4
+; CHECK-NEXT:    movs r2, #30
+; CHECK-NEXT:    b __aeabi_memcpy
+; CHECK-NEXT:  .LBB0_5: @ %bb24
+; CHECK-NEXT:    .save {r7, lr}
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    bl wombat
+; CHECK-NEXT:    @APP
+; CHECK-NEXT:    @NO_APP
+; CHECK-NEXT:    pop {r7, pc}
+; CHECK-NEXT:    .p2align 2
+; CHECK-NEXT:  @ %bb.6:
+; CHECK-NEXT:  .LCPI0_0:
+; CHECK-NEXT:    .long global.1
+; CHECK-NEXT:  .LCPI0_1:
+; CHECK-NEXT:    .long global
+bb:
+  %gep = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 4
+  %zext = zext i8 %arg2 to i32
+  switch i32 %arg1, label %bb24 [
+    i32 1, label %bb4
+    i32 2, label %bb14
+  ]
+
+bb4:                                              ; preds = %bb3
+  store i8 1, ptr %gep, align 4, !tbaa !6
+  %gep5 = getelementptr inbounds [3 x %struct.wibble], ptr @global, i32 0, i32 %zext
+  %gep6 = getelementptr inbounds [3 x %struct.wibble], ptr @global, i32 0, i32 %zext, i32 2
+  %load = load i32, ptr %gep6, align 4, !tbaa !11
+  %gep7 = getelementptr inbounds [3 x %struct.wibble], ptr @global, i32 0, i32 %zext, i32 1
+  %load8 = load i8, ptr %gep7, align 2, !tbaa !13
+  %gep9 = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 3
+  %gep10 = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 2
+  store i8 30, ptr %gep10, align 2, !tbaa !16
+  %gep11 = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 1
+  tail call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(30) %gep11, ptr noundef nonnull align 4 dereferenceable(30) %gep5, i32 30, i1 false)
+  br label %bb26
+
+bb14:                                             ; preds = %bb12
+  store i8 1, ptr %gep, align 4, !tbaa !6
+  %gep16 = getelementptr inbounds [3 x %struct.wibble], ptr @global.1, i32 0, i32 %zext
+  %gep17 = getelementptr inbounds [3 x %struct.wibble], ptr @global.1, i32 0, i32 %zext, i32 2
+  %load18 = load i32, ptr %gep17, align 4, !tbaa !21
+  %gep19 = getelementptr inbounds [3 x %struct.wibble], ptr @global.1, i32 0, i32 %zext, i32 1
+  %load20 = load i8, ptr %gep19, align 2, !tbaa !23
+  %gep21 = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 3
+  %gep22 = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 2
+  store i8 30, ptr %gep22, align 2, !tbaa !16
+  %gep23 = getelementptr inbounds %struct.eggs, ptr %arg, i32 0, i32 1
+  tail call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(30) %gep23, ptr noundef nonnull align 4 dereferenceable(30) %gep16, i32 30, i1 false)
+  br label %bb26
+
+bb24:                                             ; preds = %bb
+  tail call void @wombat()
+  tail call void asm sideeffect "", ""()
+  br label %bb26
+
+bb26:                                             ; preds = %bb24, %bb14, %bb12, %bb4, %bb3
+  ret void
+}
+
+declare void @wombat()
+
+declare void @llvm.memcpy.p0.p0.i32(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i32, i1 immarg) #2
+
+attributes #0 = { minsize noimplicitfloat nounwind optsize }
+
+!6 = !{!7, !9, i64 36}
+!7 = !{!"", !8, i64 0, !9, i64 4, !9, i64 34, !9, i64 35, !9, i64 36, !9, i64 37}
+!8 = !{!"long", !9, i64 0}
+!9 = !{!"omnipotent char", !10, i64 0}
+!10 = !{!"Simple C/C++ TBAA"}
+!11 = !{!12, !8, i64 32}
+!12 = !{!"B", !9, i64 0, !9, i64 30, !8, i64 32}
+!13 = !{!12, !9, i64 30}
+!14 = !{!7, !8, i64 0}
+!15 = !{!7, !9, i64 35}
+!16 = !{!7, !9, i64 34}
+!21 = !{!22, !8, i64 32}
+!22 = !{!"A", !9, i64 0, !9, i64 30, !8, i64 32}
+!23 = !{!22, !9, i64 30}