diff --git a/clang/lib/CodeGen/CGObjC.cpp b/clang/lib/CodeGen/CGObjC.cpp
index 10aad2e26938d..93d78783cbea6 100644
--- a/clang/lib/CodeGen/CGObjC.cpp
+++ b/clang/lib/CodeGen/CGObjC.cpp
@@ -2415,9 +2415,7 @@ static llvm::Value *emitOptimizedARCReturnCall(llvm::Value *value,
   emitAutoreleasedReturnValueMarker(CGF);
 
   // Add operand bundle "clang.arc.attachedcall" to the call instead of emitting
-  // retainRV or claimRV calls in the IR. We currently do this only when the
-  // optimization level isn't -O0 since global-isel, which is currently run at
-  // -O0, doesn't know about the operand bundle.
+  // retainRV or claimRV calls in the IR.
   ObjCEntrypoints &EPs = CGF.CGM.getObjCEntrypoints();
   llvm::Function *&EP = IsRetainRV
                             ? EPs.objc_retainAutoreleasedReturnValue
@@ -2429,11 +2427,9 @@ static llvm::Value *emitOptimizedARCReturnCall(llvm::Value *value,
 
   llvm::Triple::ArchType Arch = CGF.CGM.getTriple().getArch();
 
-  // FIXME: Do this on all targets and at -O0 too. This can be enabled only if
-  // the target backend knows how to handle the operand bundle.
-  if (CGF.CGM.getCodeGenOpts().OptimizationLevel > 0 &&
-      (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_32 ||
-       Arch == llvm::Triple::x86_64)) {
+  if ((CGF.CGM.getCodeGenOpts().OptimizationLevel > 0 &&
+       Arch == llvm::Triple::x86_64) ||
+      Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_32) {
     llvm::Value *bundleArgs[] = {EP};
     llvm::OperandBundleDef OB("clang.arc.attachedcall", bundleArgs);
     auto *oldCall = cast<llvm::CallBase>(value);
diff --git a/clang/test/CodeGenObjC/arc-arm.m b/clang/test/CodeGenObjC/arc-arm.m
index 6105644c2d684..ed2ff7e51f047 100644
--- a/clang/test/CodeGenObjC/arc-arm.m
+++ b/clang/test/CodeGenObjC/arc-arm.m
@@ -1,5 +1,5 @@
 // RUN: %clang_cc1 -triple armv7-apple-darwin10 -emit-llvm -fblocks -fobjc-arc -o - %s | FileCheck %s
-// RUN: %clang_cc1 -triple arm64-apple-ios -emit-llvm -fblocks -fobjc-arc -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple arm64-apple-ios -emit-llvm -fblocks -fobjc-arc -o - %s | FileCheck %s -check-prefix=ARM64-ATTACHED
 
 // use an autorelease marker on ARM64.
 
@@ -7,6 +7,11 @@ id test0(void) {
   extern id test0_helper(void);
   // CHECK:      [[T0:%.*]] = call [[CC:(arm_aapcscc )?]]ptr @test0_helper()
   // CHECK-NEXT: ret ptr [[T0]]
+  // ARM64-ATTACHED:      %call1 = call ptr @test0_helper() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+  // ARM64-ATTACHED:      call void asm sideeffect "mov\09fp, fp\09\09// marker for objc_retainAutoreleaseReturnValue", ""()
+  // ARM64-ATTACHED:      call void (...) @llvm.objc.clang.arc.noop.use(ptr %call1) #2
+  // ARM64-ATTACHED:      %0 = tail call ptr @llvm.objc.autoreleaseReturnValue(ptr %call1) #2
+  // ARM64-ATTACHED:      ret ptr %0
   return test0_helper();
 }
 
@@ -18,6 +23,12 @@ void test1(void) {
   // CHECK-NEXT: store ptr [[T1]],
   // CHECK-NEXT: call [[CC]]void @llvm.objc.storeStrong(
   // CHECK-NEXT: ret void
+  // ARM64-ATTACHED:      %call1 = call ptr @test1_helper() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+  // ARM64-ATTACHED:      call void asm sideeffect "mov\09fp, fp\09\09// marker for objc_retainAutoreleaseReturnValue", ""()
+  // ARM64-ATTACHED:      call void (...) @llvm.objc.clang.arc.noop.use(ptr %call1) #2
+  // ARM64-ATTACHED:      store ptr %call1,
+  // ARM64-ATTACHED:      call void @llvm.objc.storeStrong(
+  // ARM64-ATTACHED:      ret void
   id x = test1_helper();
 }
 
@@ -26,6 +37,11 @@ void test1(void) {
   extern A *test2_helper(void);
   // CHECK:      [[T0:%.*]] = call [[CC]]ptr @test2_helper()
   // CHECK-NEXT: ret ptr [[T0]]
+  // ARM64-ATTACHED:      %call1 = call ptr @test2_helper() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+  // ARM64-ATTACHED:      call void asm sideeffect "mov\09fp, fp\09\09// marker for objc_retainAutoreleaseReturnValue", ""()
+  // ARM64-ATTACHED:      call void (...) @llvm.objc.clang.arc.noop.use(ptr %call1) #2
+  // ARM64-ATTACHED:      %0 = tail call ptr @llvm.objc.autoreleaseReturnValue(ptr %call1) #2
+  // ARM64-ATTACHED:      ret ptr %0
   return test2_helper();
 }
 
@@ -33,5 +49,10 @@ id test3(void) {
   extern A *test3_helper(void);
   // CHECK:      [[T0:%.*]] = call [[CC]]ptr @test3_helper()
   // CHECK-NEXT: ret ptr [[T0]]
+  // ARM64-ATTACHED:      %call1 = call ptr @test3_helper() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+  // ARM64-ATTACHED:      call void asm sideeffect "mov\09fp, fp\09\09// marker for objc_retainAutoreleaseReturnValue", ""()
+  // ARM64-ATTACHED:      call void (...) @llvm.objc.clang.arc.noop.use(ptr %call1) #2
+  // ARM64-ATTACHED:      %0 = tail call ptr @llvm.objc.autoreleaseReturnValue(ptr %call1) #2
+  // ARM64-ATTACHED:      ret ptr %0
   return test3_helper();
 }
diff --git a/clang/test/CodeGenObjC/arc-unsafeclaim.m b/clang/test/CodeGenObjC/arc-unsafeclaim.m
index 9c2b215df276f..969dd549b59ce 100644
--- a/clang/test/CodeGenObjC/arc-unsafeclaim.m
+++ b/clang/test/CodeGenObjC/arc-unsafeclaim.m
@@ -7,7 +7,7 @@
 // RUN: %clang_cc1 -triple i386-apple-darwin11 -fobjc-runtime=macosx-fragile-10.11 -fobjc-arc -emit-llvm -o - %s | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-UNOPTIMIZED -check-prefix=CHECK-MARKED -check-prefix=CALL
 
 //   Make sure it works on ARM64.
-// RUN: %clang_cc1 -triple arm64-apple-ios9 -fobjc-runtime=ios-9.0 -fobjc-arc -emit-llvm -o - %s | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-UNOPTIMIZED -check-prefix=CHECK-MARKED -check-prefix=CALL
+// RUN: %clang_cc1 -triple arm64-apple-ios9 -fobjc-runtime=ios-9.0 -fobjc-arc -emit-llvm -o - %s | FileCheck %s -check-prefix=ATTACHED-CALL
 
 //   Make sure it works on ARM.
 // RUN: %clang_cc1 -triple armv7-apple-ios9 -fobjc-runtime=ios-9.0 -fobjc-arc -emit-llvm -o - %s | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-UNOPTIMIZED -check-prefix=CHECK-MARKED -check-prefix=CALL
@@ -41,8 +41,9 @@ void test_assign(void) {
 // DISABLED-NEXT:        [[T2:%.*]] = {{.*}}call ptr @llvm.objc.retainAutoreleasedReturnValue(ptr [[T0]])
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_assign()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_assign_assign(void) {
   __unsafe_unretained id x, y;
@@ -62,8 +63,9 @@ void test_assign_assign(void) {
 // CHECK-NEXT:           ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_assign_assign()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_strong_assign_assign(void) {
   __strong id x;
@@ -88,8 +90,9 @@ void test_strong_assign_assign(void) {
 // CHECK-NEXT:           ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_strong_assign_assign()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_assign_strong_assign(void) {
   __unsafe_unretained id x;
@@ -114,8 +117,9 @@ void test_assign_strong_assign(void) {
 // CHECK-NEXT:           ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_assign_strong_assign()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_init(void) {
   __unsafe_unretained id x = makeA();
@@ -131,8 +135,9 @@ void test_init(void) {
 // CHECK-NEXT:           ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_init()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_init_assignment(void) {
   __unsafe_unretained id x;
@@ -152,8 +157,9 @@ void test_init_assignment(void) {
 // CHECK-NEXT: ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_init_assignment()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_strong_init_assignment(void) {
   __unsafe_unretained id x;
@@ -175,8 +181,9 @@ void test_strong_init_assignment(void) {
 // CHECK-NEXT: ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_strong_init_assignment()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_init_strong_assignment(void) {
   __strong id x;
@@ -200,8 +207,9 @@ void test_init_strong_assignment(void) {
 // CHECK-NEXT: ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_init_strong_assignment()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.retainAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_ignored(void) {
   makeA();
@@ -214,8 +222,9 @@ void test_ignored(void) {
 // CHECK-NEXT:        ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_ignored()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 void test_cast_to_void(void) {
   (void) makeA();
@@ -228,8 +237,9 @@ void test_cast_to_void(void) {
 // CHECK-NEXT:        ret void
 
 // ATTACHED-CALL-LABEL:      define{{.*}} void @test_cast_to_void()
-// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ],
-// ATTACHED-CALL:              call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
+// ATTACHED-CALL:              [[T0:%.*]] = call ptr @makeA() [ "clang.arc.attachedcall"(ptr @llvm.objc.unsafeClaimAutoreleasedReturnValue) ]
+// ATTACHED-CALL-NEXT:        call void asm sideeffect
+// ATTACHED-CALL-NEXT:        call void (...) @llvm.objc.clang.arc.noop.use(ptr [[T0]])
 
 
 // This is always at the end of the module.
diff --git a/llvm/lib/Target/AArch64/AArch64FastISel.cpp b/llvm/lib/Target/AArch64/AArch64FastISel.cpp
index cf344980cbaae..9bce0f19dd219 100644
--- a/llvm/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/llvm/lib/Target/AArch64/AArch64FastISel.cpp
@@ -25,6 +25,7 @@
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Analysis/BranchProbabilityInfo.h"
+#include "llvm/Analysis/ObjCARCUtil.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/FastISel.h"
 #include "llvm/CodeGen/FunctionLoweringInfo.h"
@@ -3159,6 +3160,10 @@ bool AArch64FastISel::fastLowerCall(CallLoweringInfo &CLI) {
       CLI.CB->getOperandBundle(LLVMContext::OB_kcfi))
     return false;
 
+  // Allow SelectionDAG isel to handle clang.arc.attachedcall operand bundle.
+  if (CLI.CB && objcarc::hasAttachedCallOpBundle(CLI.CB))
+    return false;
+
   // Allow SelectionDAG isel to handle tail calls.
   if (IsTailCall)
     return false;
diff --git a/llvm/test/CodeGen/AArch64/call-rv-marker.ll b/llvm/test/CodeGen/AArch64/call-rv-marker.ll
index 0bc64498da324..b9811b346175b 100644
--- a/llvm/test/CodeGen/AArch64/call-rv-marker.ll
+++ b/llvm/test/CodeGen/AArch64/call-rv-marker.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
 ; RUN: llc -o - %s | FileCheck --check-prefix=SELDAG --check-prefix=CHECK %s
 ; RUN: llc -global-isel -o - %s | FileCheck --check-prefix=GISEL --check-prefix=CHECK %s
+; RUN: llc -O0 -o - %s | FileCheck --check-prefix=O0 --check-prefix=CHECK %s
 
 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
 target triple = "arm64-apple-ios"
@@ -49,6 +50,18 @@ define dso_local ptr @rv_marker_1_retain() {
 ; GISEL-NEXT:    bl _objc_retainAutoreleasedReturnValue
 ; GISEL-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
 ; GISEL-NEXT:    ret
+;
+; O0-LABEL: rv_marker_1_retain:
+; O0:       ; %bb.0: ; %entry
+; O0-NEXT:    stp x29, x30, [sp, #-16]! ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 16
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    bl _foo1
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_retainAutoreleasedReturnValue
+; O0-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
+; O0-NEXT:    ret
 entry:
   %call = call ptr @foo1() [ "clang.arc.attachedcall"(ptr @objc_retainAutoreleasedReturnValue) ]
   ret ptr %call
@@ -78,6 +91,18 @@ define dso_local ptr @rv_marker_1_unsafeClaim() {
 ; GISEL-NEXT:    bl _objc_unsafeClaimAutoreleasedReturnValue
 ; GISEL-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
 ; GISEL-NEXT:    ret
+;
+; O0-LABEL: rv_marker_1_unsafeClaim:
+; O0:       ; %bb.0: ; %entry
+; O0-NEXT:    stp x29, x30, [sp, #-16]! ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 16
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    bl _foo1
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_unsafeClaimAutoreleasedReturnValue
+; O0-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
+; O0-NEXT:    ret
 entry:
   %call = call ptr @foo1() [ "clang.arc.attachedcall"(ptr @objc_unsafeClaimAutoreleasedReturnValue) ]
   ret ptr %call
@@ -113,6 +138,21 @@ define dso_local void @rv_marker_2_select(i32 %c) {
 ; GISEL-NEXT:    bl _objc_retainAutoreleasedReturnValue
 ; GISEL-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
 ; GISEL-NEXT:    b _foo2
+;
+; O0-LABEL: rv_marker_2_select:
+; O0:       ; %bb.0: ; %entry
+; O0-NEXT:    stp x29, x30, [sp, #-16]! ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 16
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    mov w8, #2 ; =0x2
+; O0-NEXT:    subs w9, w0, #0
+; O0-NEXT:    csinc w0, w8, wzr, eq
+; O0-NEXT:    bl _foo0
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_retainAutoreleasedReturnValue
+; O0-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
+; O0-NEXT:    b _foo2
 entry:
   %tobool.not = icmp eq i32 %c, 0
   %.sink = select i1 %tobool.not, i32 2, i32 1
@@ -247,6 +287,68 @@ define dso_local void @rv_marker_3() personality ptr @__gxx_personality_v0 {
 ; GISEL-NEXT:    .byte 0 ; On action: cleanup
 ; GISEL-NEXT:  Lcst_end0:
 ; GISEL-NEXT:    .p2align 2, 0x0
+;
+; O0-LABEL: rv_marker_3:
+; O0:       Lfunc_begin0:
+; O0-NEXT:    .cfi_startproc
+; O0-NEXT:    .cfi_personality 155, ___gxx_personality_v0
+; O0-NEXT:    .cfi_lsda 16, Lexception0
+; O0-NEXT:  ; %bb.0: ; %entry
+; O0-NEXT:    sub sp, sp, #32
+; O0-NEXT:    stp x29, x30, [sp, #16] ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 32
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    bl _foo1
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_retainAutoreleasedReturnValue
+; O0-NEXT:    str x0, [sp, #8] ; 8-byte Folded Spill
+; O0-NEXT:  Ltmp0: ; EH_LABEL
+; O0-NEXT:    bl _objc_object
+; O0-NEXT:  Ltmp1: ; EH_LABEL
+; O0-NEXT:    b LBB3_1
+; O0-NEXT:  LBB3_1: ; %invoke.cont
+; O0-NEXT:    ldr x0, [sp, #8] ; 8-byte Folded Reload
+; O0-NEXT:    adrp x1, _objc_release@GOTPAGE
+; O0-NEXT:    ldr x1, [x1, _objc_release@GOTPAGEOFF]
+; O0-NEXT:    ldp x29, x30, [sp, #16] ; 16-byte Folded Reload
+; O0-NEXT:    add sp, sp, #32
+; O0-NEXT:    br x1
+; O0-NEXT:  LBB3_2: ; %lpad
+; O0-NEXT:  Ltmp2: ; EH_LABEL
+; O0-NEXT:    mov x1, x0
+; O0-NEXT:    ldr x0, [sp, #8] ; 8-byte Folded Reload
+; O0-NEXT:    str x1, [sp] ; 8-byte Folded Spill
+; O0-NEXT:    adrp x8, _objc_release@GOTPAGE
+; O0-NEXT:    ldr x8, [x8, _objc_release@GOTPAGEOFF]
+; O0-NEXT:    blr x8
+; O0-NEXT:    ldr x0, [sp] ; 8-byte Folded Reload
+; O0-NEXT:    bl __Unwind_Resume
+; O0-NEXT:  Lfunc_end0:
+; O0-NEXT:    .cfi_endproc
+; O0-NEXT:    .section __TEXT,__gcc_except_tab
+; O0-NEXT:    .p2align 2, 0x0
+; O0-NEXT:  GCC_except_table3:
+; O0-NEXT:  Lexception0:
+; O0-NEXT:    .byte 255 ; @LPStart Encoding = omit
+; O0-NEXT:    .byte 255 ; @TType Encoding = omit
+; O0-NEXT:    .byte 1 ; Call site Encoding = uleb128
+; O0-NEXT:    .uleb128 Lcst_end0-Lcst_begin0
+; O0-NEXT:  Lcst_begin0:
+; O0-NEXT:    .uleb128 Lfunc_begin0-Lfunc_begin0 ; >> Call Site 1 <<
+; O0-NEXT:    .uleb128 Ltmp0-Lfunc_begin0 ; Call between Lfunc_begin0 and Ltmp0
+; O0-NEXT:    .byte 0 ; has no landing pad
+; O0-NEXT:    .byte 0 ; On action: cleanup
+; O0-NEXT:    .uleb128 Ltmp0-Lfunc_begin0 ; >> Call Site 2 <<
+; O0-NEXT:    .uleb128 Ltmp1-Ltmp0 ; Call between Ltmp0 and Ltmp1
+; O0-NEXT:    .uleb128 Ltmp2-Lfunc_begin0 ; jumps to Ltmp2
+; O0-NEXT:    .byte 0 ; On action: cleanup
+; O0-NEXT:    .uleb128 Ltmp1-Lfunc_begin0 ; >> Call Site 3 <<
+; O0-NEXT:    .uleb128 Lfunc_end0-Ltmp1 ; Call between Ltmp1 and Lfunc_end0
+; O0-NEXT:    .byte 0 ; has no landing pad
+; O0-NEXT:    .byte 0 ; On action: cleanup
+; O0-NEXT:  Lcst_end0:
+; O0-NEXT:    .p2align 2, 0x0
 entry:
   %call = call ptr @foo1() [ "clang.arc.attachedcall"(ptr @objc_retainAutoreleasedReturnValue) ]
   invoke void @objc_object(ptr %call) #5
@@ -419,6 +521,88 @@ define dso_local void @rv_marker_4() personality ptr @__gxx_personality_v0 {
 ; GISEL-NEXT:    .byte 0 ; On action: cleanup
 ; GISEL-NEXT:  Lcst_end1:
 ; GISEL-NEXT:    .p2align 2, 0x0
+;
+; O0-LABEL: rv_marker_4:
+; O0:       Lfunc_begin1:
+; O0-NEXT:    .cfi_startproc
+; O0-NEXT:    .cfi_personality 155, ___gxx_personality_v0
+; O0-NEXT:    .cfi_lsda 16, Lexception1
+; O0-NEXT:  ; %bb.0: ; %entry
+; O0-NEXT:    sub sp, sp, #64
+; O0-NEXT:    stp x29, x30, [sp, #48] ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 64
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:  Ltmp3: ; EH_LABEL
+; O0-NEXT:    bl _foo1
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_retainAutoreleasedReturnValue
+; O0-NEXT:    str x0, [sp, #32] ; 8-byte Folded Spill
+; O0-NEXT:  Ltmp4: ; EH_LABEL
+; O0-NEXT:    b LBB4_1
+; O0-NEXT:  LBB4_1: ; %invoke.cont
+; O0-NEXT:  Ltmp6: ; EH_LABEL
+; O0-NEXT:    ldr x0, [sp, #32] ; 8-byte Folded Reload
+; O0-NEXT:    bl _objc_object
+; O0-NEXT:  Ltmp7: ; EH_LABEL
+; O0-NEXT:    b LBB4_2
+; O0-NEXT:  LBB4_2: ; %invoke.cont2
+; O0-NEXT:    ldr x0, [sp, #32] ; 8-byte Folded Reload
+; O0-NEXT:    adrp x8, _objc_release@GOTPAGE
+; O0-NEXT:    ldr x8, [x8, _objc_release@GOTPAGEOFF]
+; O0-NEXT:    blr x8
+; O0-NEXT:    add x0, sp, #47
+; O0-NEXT:    bl __ZN1SD1Ev
+; O0-NEXT:    ldp x29, x30, [sp, #48] ; 16-byte Folded Reload
+; O0-NEXT:    add sp, sp, #64
+; O0-NEXT:    ret
+; O0-NEXT:  LBB4_3: ; %lpad
+; O0-NEXT:  Ltmp5: ; EH_LABEL
+; O0-NEXT:    str x0, [sp, #24] ; 8-byte Folded Spill
+; O0-NEXT:    b LBB4_5
+; O0-NEXT:  LBB4_4: ; %lpad1
+; O0-NEXT:  Ltmp8: ; EH_LABEL
+; O0-NEXT:    mov x1, x0
+; O0-NEXT:    ldr x0, [sp, #32] ; 8-byte Folded Reload
+; O0-NEXT:    str x1, [sp, #16] ; 8-byte Folded Spill
+; O0-NEXT:    adrp x8, _objc_release@GOTPAGE
+; O0-NEXT:    ldr x8, [x8, _objc_release@GOTPAGEOFF]
+; O0-NEXT:    blr x8
+; O0-NEXT:    ldr x0, [sp, #16] ; 8-byte Folded Reload
+; O0-NEXT:    str x0, [sp, #24] ; 8-byte Folded Spill
+; O0-NEXT:    b LBB4_5
+; O0-NEXT:  LBB4_5: ; %ehcleanup
+; O0-NEXT:    ldr x8, [sp, #24] ; 8-byte Folded Reload
+; O0-NEXT:    str x8, [sp, #8] ; 8-byte Folded Spill
+; O0-NEXT:    add x0, sp, #47
+; O0-NEXT:    bl __ZN1SD1Ev
+; O0-NEXT:    ldr x0, [sp, #8] ; 8-byte Folded Reload
+; O0-NEXT:    bl __Unwind_Resume
+; O0-NEXT:  Lfunc_end1:
+; O0-NEXT:    .cfi_endproc
+; O0-NEXT:    .section __TEXT,__gcc_except_tab
+; O0-NEXT:    .p2align 2, 0x0
+; O0-NEXT:  GCC_except_table4:
+; O0-NEXT:  Lexception1:
+; O0-NEXT:    .byte 255 ; @LPStart Encoding = omit
+; O0-NEXT:    .byte 255 ; @TType Encoding = omit
+; O0-NEXT:    .byte 1 ; Call site Encoding = uleb128
+; O0-NEXT:    .uleb128 Lcst_end1-Lcst_begin1
+; O0-NEXT:  Lcst_begin1:
+; O0-NEXT:    .uleb128 Ltmp3-Lfunc_begin1 ; >> Call Site 1 <<
+; O0-NEXT:    .uleb128 Ltmp4-Ltmp3 ; Call between Ltmp3 and Ltmp4
+; O0-NEXT:    .uleb128 Ltmp5-Lfunc_begin1 ; jumps to Ltmp5
+; O0-NEXT:    .byte 0 ; On action: cleanup
+; O0-NEXT:    .uleb128 Ltmp6-Lfunc_begin1 ; >> Call Site 2 <<
+; O0-NEXT:    .uleb128 Ltmp7-Ltmp6 ; Call between Ltmp6 and Ltmp7
+; O0-NEXT:    .uleb128 Ltmp8-Lfunc_begin1 ; jumps to Ltmp8
+; O0-NEXT:    .byte 0 ; On action: cleanup
+; O0-NEXT:    .uleb128 Ltmp7-Lfunc_begin1 ; >> Call Site 3 <<
+; O0-NEXT:    .uleb128 Lfunc_end1-Ltmp7 ; Call between Ltmp7 and Lfunc_end1
+; O0-NEXT:    .byte 0 ; has no landing pad
+; O0-NEXT:    .byte 0 ; On action: cleanup
+; O0-NEXT:  Lcst_end1:
+; O0-NEXT:    .p2align 2, 0x0
 entry:
   %s = alloca %struct.S, align 1
   call void @llvm.lifetime.start.p0(i64 1, ptr nonnull %s) #2
@@ -501,6 +685,25 @@ define dso_local ptr @rv_marker_5_indirect_call() {
 ; GISEL-NEXT:    ldp x20, x19, [sp], #32 ; 16-byte Folded Reload
 ; GISEL-NEXT:    ret
 ; GISEL-NEXT:    .loh AdrpLdr Lloh8, Lloh9
+;
+; O0-LABEL: rv_marker_5_indirect_call:
+; O0:       ; %bb.0: ; %entry
+; O0-NEXT:    sub sp, sp, #32
+; O0-NEXT:    stp x29, x30, [sp, #16] ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 32
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    adrp x8, _fptr@PAGE
+; O0-NEXT:    ldr x8, [x8, _fptr@PAGEOFF]
+; O0-NEXT:    blr x8
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_retainAutoreleasedReturnValue
+; O0-NEXT:    str x0, [sp, #8] ; 8-byte Folded Spill
+; O0-NEXT:    bl _foo2
+; O0-NEXT:    ldr x0, [sp, #8] ; 8-byte Folded Reload
+; O0-NEXT:    ldp x29, x30, [sp, #16] ; 16-byte Folded Reload
+; O0-NEXT:    add sp, sp, #32
+; O0-NEXT:    ret
 entry:
   %0 = load ptr, ptr @fptr, align 8
   %call = call ptr %0() [ "clang.arc.attachedcall"(ptr @objc_retainAutoreleasedReturnValue) ]
@@ -540,6 +743,23 @@ define dso_local void @rv_marker_multiarg(i64 %a, i64 %b, i64 %c) {
 ; GISEL-NEXT:    bl _objc_retainAutoreleasedReturnValue
 ; GISEL-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
 ; GISEL-NEXT:    ret
+;
+; O0-LABEL: rv_marker_multiarg:
+; O0:       ; %bb.0:
+; O0-NEXT:    sub sp, sp, #32
+; O0-NEXT:    stp x29, x30, [sp, #16] ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 32
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    str x0, [sp, #8] ; 8-byte Folded Spill
+; O0-NEXT:    mov x0, x2
+; O0-NEXT:    ldr x2, [sp, #8] ; 8-byte Folded Reload
+; O0-NEXT:    bl _foo
+; O0-NEXT:    mov x29, x29
+; O0-NEXT:    bl _objc_retainAutoreleasedReturnValue
+; O0-NEXT:    ldp x29, x30, [sp, #16] ; 16-byte Folded Reload
+; O0-NEXT:    add sp, sp, #32
+; O0-NEXT:    ret
   call ptr @foo(i64 %c, i64 %b, i64 %a) [ "clang.arc.attachedcall"(ptr @objc_retainAutoreleasedReturnValue) ]
   ret void
 }
@@ -566,6 +786,17 @@ define dso_local ptr @rv_marker_claim() {
 ; GISEL-NEXT:    bl _objc_claimAutoreleasedReturnValue
 ; GISEL-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
 ; GISEL-NEXT:    ret
+;
+; O0-LABEL: rv_marker_claim:
+; O0:       ; %bb.0: ; %entry
+; O0-NEXT:    stp x29, x30, [sp, #-16]! ; 16-byte Folded Spill
+; O0-NEXT:    .cfi_def_cfa_offset 16
+; O0-NEXT:    .cfi_offset w30, -8
+; O0-NEXT:    .cfi_offset w29, -16
+; O0-NEXT:    bl _foo1
+; O0-NEXT:    bl _objc_claimAutoreleasedReturnValue
+; O0-NEXT:    ldp x29, x30, [sp], #16 ; 16-byte Folded Reload
+; O0-NEXT:    ret
 entry:
   %call = call ptr @foo1() [ "clang.arc.attachedcall"(ptr @objc_claimAutoreleasedReturnValue) ]
   ret ptr %call