Index: CodeGen/CGBuiltin.cpp
===================================================================
--- CodeGen/CGBuiltin.cpp
+++ CodeGen/CGBuiltin.cpp
@@ -268,6 +268,19 @@
   return CGF.Builder.CreateExtractValue(Result, 0);
 }
 
+static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E,
+    AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
+  assert(E->getArg(0)->getType()->isPointerType());
+
+  auto *IntTy = CGF.ConvertType(E->getType());
+  auto *Result = CGF.Builder.CreateAtomicRMW(
+                   AtomicRMWInst::Add,
+                   CGF.EmitScalarExpr(E->getArg(0)),
+                   ConstantInt::get(IntTy, 1),
+                   Ordering);
+  return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1));
+}
+
 // Emit a simple mangled intrinsic that has 1 argument and a return type
 // matching the argument type.
 static Value *emitUnaryBuiltin(CodeGenFunction &CGF,
@@ -801,6 +814,9 @@
   _InterlockedAnd_acq,
   _InterlockedAnd_rel,
   _InterlockedAnd_nf,
+  _InterlockedIncrement_acq,
+  _InterlockedIncrement_rel,
+  _InterlockedIncrement_nf,
   __fastfail,
 };
 
@@ -918,6 +934,12 @@
   case MSVCIntrin::_InterlockedAnd_nf:
     return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
                                  AtomicOrdering::Monotonic);
+  case MSVCIntrin::_InterlockedIncrement_acq:
+    return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire);
+  case MSVCIntrin::_InterlockedIncrement_rel:
+    return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release);
+  case MSVCIntrin::_InterlockedIncrement_nf:
+    return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic);
 
   case MSVCIntrin::_InterlockedDecrement: {
     llvm::Type *IntTy = ConvertType(E->getType());
@@ -928,15 +950,8 @@
       llvm::AtomicOrdering::SequentiallyConsistent);
     return Builder.CreateSub(RMWI, ConstantInt::get(IntTy, 1));
   }
-  case MSVCIntrin::_InterlockedIncrement: {
-    llvm::Type *IntTy = ConvertType(E->getType());
-    AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
-      AtomicRMWInst::Add,
-      EmitScalarExpr(E->getArg(0)),
-      ConstantInt::get(IntTy, 1),
-      llvm::AtomicOrdering::SequentiallyConsistent);
-    return Builder.CreateAdd(RMWI, ConstantInt::get(IntTy, 1));
-  }
+  case MSVCIntrin::_InterlockedIncrement:
+    return EmitAtomicIncrementValue(*this, E);
 
   case MSVCIntrin::__fastfail: {
     // Request immediate process termination from the kernel. The instruction
@@ -6291,6 +6306,18 @@
   case ARM::BI_InterlockedAnd_nf:
   case ARM::BI_InterlockedAnd64_nf:
     return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E);
+  case ARM::BI_InterlockedIncrement16_acq:
+  case ARM::BI_InterlockedIncrement_acq:
+  case ARM::BI_InterlockedIncrement64_acq:
+    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E);
+  case ARM::BI_InterlockedIncrement16_rel:
+  case ARM::BI_InterlockedIncrement_rel:
+  case ARM::BI_InterlockedIncrement64_rel:
+    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E);
+  case ARM::BI_InterlockedIncrement16_nf:
+  case ARM::BI_InterlockedIncrement_nf:
+  case ARM::BI_InterlockedIncrement64_nf:
+    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E);
   }
 
   // Get the last argument, which specifies the vector type.
@@ -8867,6 +8894,18 @@
   case AArch64::BI_InterlockedAnd_nf:
   case AArch64::BI_InterlockedAnd64_nf:
     return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E);
+  case AArch64::BI_InterlockedIncrement16_acq:
+  case AArch64::BI_InterlockedIncrement_acq:
+  case AArch64::BI_InterlockedIncrement64_acq:
+    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E);
+  case AArch64::BI_InterlockedIncrement16_rel:
+  case AArch64::BI_InterlockedIncrement_rel:
+  case AArch64::BI_InterlockedIncrement64_rel:
+    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E);
+  case AArch64::BI_InterlockedIncrement16_nf:
+  case AArch64::BI_InterlockedIncrement_nf:
+  case AArch64::BI_InterlockedIncrement64_nf:
+    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E);
 
   case AArch64::BI_InterlockedAdd: {
     Value *Arg0 = EmitScalarExpr(E->getArg(0));
Index: CodeGen/ms-intrinsics.c
===================================================================
--- CodeGen/ms-intrinsics.c
+++ CodeGen/ms-intrinsics.c
@@ -1178,6 +1178,87 @@
 // CHECK-ARM-ARM64:   [[RESULT:%[0-9]+]] = atomicrmw and i64* %value, i64 %mask monotonic
 // CHECK-ARM-ARM64:   ret i64 [[RESULT:%[0-9]+]]
 // CHECK-ARM-ARM64: }
+
+short test_InterlockedIncrement16_acq(short volatile *Addend) {
+  return _InterlockedIncrement16_acq(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i16 @test_InterlockedIncrement16_acq(i16*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i16* %Addend, i16 1 acquire
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i16 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i16 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+short test_InterlockedIncrement16_rel(short volatile *Addend) {
+  return _InterlockedIncrement16_rel(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i16 @test_InterlockedIncrement16_rel(i16*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i16* %Addend, i16 1 release
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i16 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i16 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+short test_InterlockedIncrement16_nf(short volatile *Addend) {
+  return _InterlockedIncrement16_nf(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i16 @test_InterlockedIncrement16_nf(i16*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i16* %Addend, i16 1 monotonic
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i16 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i16 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+long test_InterlockedIncrement_acq(long volatile *Addend) {
+  return _InterlockedIncrement_acq(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i32 @test_InterlockedIncrement_acq(i32*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i32* %Addend, i32 1 acquire
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i32 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i32 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+long test_InterlockedIncrement_rel(long volatile *Addend) {
+  return _InterlockedIncrement_rel(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i32 @test_InterlockedIncrement_rel(i32*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i32* %Addend, i32 1 release
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i32 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i32 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+long test_InterlockedIncrement_nf(long volatile *Addend) {
+  return _InterlockedIncrement_nf(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i32 @test_InterlockedIncrement_nf(i32*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i32* %Addend, i32 1 monotonic
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i32 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i32 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+__int64 test_InterlockedIncrement64_acq(__int64 volatile *Addend) {
+  return _InterlockedIncrement64_acq(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i64 @test_InterlockedIncrement64_acq(i64*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i64* %Addend, i64 1 acquire
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i64 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i64 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+__int64 test_InterlockedIncrement64_rel(__int64 volatile *Addend) {
+  return _InterlockedIncrement64_rel(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i64 @test_InterlockedIncrement64_rel(i64*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i64* %Addend, i64 1 release
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i64 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i64 [[RESULT]]
+// CHECK-ARM-ARM64: }
+
+__int64 test_InterlockedIncrement64_nf(__int64 volatile *Addend) {
+  return _InterlockedIncrement64_nf(Addend);
+}
+// CHECK-ARM-ARM64: define{{.*}}i64 @test_InterlockedIncrement64_nf(i64*{{[a-z_ ]*}}%Addend){{.*}}{
+// CHECK-ARM-ARM64: [[TMP:%[0-9]+]] = atomicrmw add i64* %Addend, i64 1 monotonic
+// CHECK-ARM-ARM64: [[RESULT:%[0-9]+]] = add i64 [[TMP]], 1
+// CHECK-ARM-ARM64: ret i64 [[RESULT]]
+// CHECK-ARM-ARM64: }
 #endif
 
 #if !defined(__aarch64__)
Index: Headers/intrin.h
===================================================================
--- Headers/intrin.h
+++ Headers/intrin.h
@@ -346,42 +346,15 @@
 |* Interlocked Increment
 \*----------------------------------------------------------------------------*/
 #if defined(__arm__) || defined(__aarch64__)
-static __inline__ short __DEFAULT_FN_ATTRS
-_InterlockedIncrement16_acq(short volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_ACQUIRE);
-}
-static __inline__ short __DEFAULT_FN_ATTRS
-_InterlockedIncrement16_nf(short volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_RELAXED);
-}
-static __inline__ short __DEFAULT_FN_ATTRS
-_InterlockedIncrement16_rel(short volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_RELEASE);
-}
-static __inline__ long __DEFAULT_FN_ATTRS
-_InterlockedIncrement_acq(long volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_ACQUIRE);
-}
-static __inline__ long __DEFAULT_FN_ATTRS
-_InterlockedIncrement_nf(long volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_RELAXED);
-}
-static __inline__ long __DEFAULT_FN_ATTRS
-_InterlockedIncrement_rel(long volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_RELEASE);
-}
-static __inline__ __int64 __DEFAULT_FN_ATTRS
-_InterlockedIncrement64_acq(__int64 volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_ACQUIRE);
-}
-static __inline__ __int64 __DEFAULT_FN_ATTRS
-_InterlockedIncrement64_nf(__int64 volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_RELAXED);
-}
-static __inline__ __int64 __DEFAULT_FN_ATTRS
-_InterlockedIncrement64_rel(__int64 volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, __ATOMIC_RELEASE);
-}
+short _InterlockedIncrement16_acq(short volatile *_Value);
+short _InterlockedIncrement16_nf(short volatile *_Value);
+short _InterlockedIncrement16_rel(short volatile *_Value);
+long _InterlockedIncrement_acq(long volatile *_Value);
+long _InterlockedIncrement_nf(long volatile *_Value);
+long _InterlockedIncrement_rel(long volatile *_Value);
+__int64 _InterlockedIncrement64_acq(__int64 volatile *_Value);
+__int64 _InterlockedIncrement64_nf(__int64 volatile *_Value);
+__int64 _InterlockedIncrement64_rel(__int64 volatile *_Value);
 #endif
 /*----------------------------------------------------------------------------*\
 |* Interlocked Decrement
Index: clang/Basic/BuiltinsAArch64.def
===================================================================
--- clang/Basic/BuiltinsAArch64.def
+++ clang/Basic/BuiltinsAArch64.def
@@ -182,6 +182,16 @@
 TARGET_HEADER_BUILTIN(_InterlockedAnd64_nf,  "LLiLLiD*LLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 TARGET_HEADER_BUILTIN(_InterlockedAnd64_rel, "LLiLLiD*LLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 
+TARGET_HEADER_BUILTIN(_InterlockedIncrement16_acq, "ssD*",     "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement16_nf,  "ssD*",     "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement16_rel, "ssD*",     "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement_acq,   "LiLiD*",   "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement_nf,    "LiLiD*",   "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement_rel,   "LiLiD*",   "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement64_acq, "LLiLLiD*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement64_nf,  "LLiLLiD*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement64_rel, "LLiLLiD*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+
 TARGET_HEADER_BUILTIN(_ReadWriteBarrier, "v", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 TARGET_HEADER_BUILTIN(__getReg, "ULLii", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 TARGET_HEADER_BUILTIN(_ReadStatusReg,  "ii",  "nh", "intrin.h", ALL_MS_LANGUAGES, "")
Index: clang/Basic/BuiltinsARM.def
===================================================================
--- clang/Basic/BuiltinsARM.def
+++ clang/Basic/BuiltinsARM.def
@@ -308,6 +308,16 @@
 TARGET_HEADER_BUILTIN(_InterlockedAnd64_nf,  "LLiLLiD*LLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 TARGET_HEADER_BUILTIN(_InterlockedAnd64_rel, "LLiLLiD*LLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 
+TARGET_HEADER_BUILTIN(_InterlockedIncrement16_acq, "ssD*",     "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement16_nf,  "ssD*",     "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement16_rel, "ssD*",     "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement_acq,   "LiLiD*",   "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement_nf,    "LiLiD*",   "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement_rel,   "LiLiD*",   "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement64_acq, "LLiLLiD*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement64_nf,  "LLiLLiD*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_InterlockedIncrement64_rel, "LLiLLiD*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+
 #undef BUILTIN
 #undef LANGBUILTIN
 #undef TARGET_HEADER_BUILTIN