diff --git a/clang/include/clang/Basic/BuiltinsAArch64.def b/clang/include/clang/Basic/BuiltinsAArch64.def
--- a/clang/include/clang/Basic/BuiltinsAArch64.def
+++ b/clang/include/clang/Basic/BuiltinsAArch64.def
@@ -39,6 +39,8 @@
 BUILTIN(__builtin_arm_rbit64, "WUiWUi", "nc")
 BUILTIN(__builtin_arm_cls, "UiZUi", "nc")
 BUILTIN(__builtin_arm_cls64, "UiWUi", "nc")
+BUILTIN(__builtin_arm_clz, "UiZUi", "nc")
+BUILTIN(__builtin_arm_clz64, "UiWUi", "nc")
 
 // HINT
 BUILTIN(__builtin_arm_nop, "v", "")
diff --git a/clang/include/clang/Basic/BuiltinsARM.def b/clang/include/clang/Basic/BuiltinsARM.def
--- a/clang/include/clang/Basic/BuiltinsARM.def
+++ b/clang/include/clang/Basic/BuiltinsARM.def
@@ -119,6 +119,8 @@
 
 // Bit manipulation
 BUILTIN(__builtin_arm_rbit, "UiUi", "nc")
+BUILTIN(__builtin_arm_clz, "UiZUi", "nc")
+BUILTIN(__builtin_arm_clz64, "UiWUi", "nc")
 BUILTIN(__builtin_arm_cls, "UiZUi", "nc")
 BUILTIN(__builtin_arm_cls64, "UiWUi", "nc")
 
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -7948,6 +7948,17 @@
         CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
   }
 
+  if (BuiltinID == clang::ARM::BI__builtin_arm_clz ||
+      BuiltinID == clang::ARM::BI__builtin_arm_clz64) {
+    llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
+    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Arg->getType());
+    Value *Res = Builder.CreateCall(F, {Arg, Builder.getInt1(false)});
+    if (BuiltinID == clang::ARM::BI__builtin_arm_clz64)
+      Res = Builder.CreateTrunc(Res, Builder.getInt32Ty());
+    return Res;
+  }
+
+
   if (BuiltinID == clang::ARM::BI__builtin_arm_cls) {
     llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
     return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_cls), Arg, "cls");
@@ -10030,6 +10041,16 @@
         CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
   }
 
+  if (BuiltinID == clang::AArch64::BI__builtin_arm_clz ||
+      BuiltinID == clang::AArch64::BI__builtin_arm_clz64) {
+    llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
+    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Arg->getType());
+    Value *Res = Builder.CreateCall(F, {Arg, Builder.getInt1(false)});
+    if (BuiltinID == clang::AArch64::BI__builtin_arm_clz64)
+      Res = Builder.CreateTrunc(Res, Builder.getInt32Ty());
+    return Res;
+  }
+
   if (BuiltinID == clang::AArch64::BI__builtin_arm_cls) {
     llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
     return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::aarch64_cls), Arg,
diff --git a/clang/lib/Headers/arm_acle.h b/clang/lib/Headers/arm_acle.h
--- a/clang/lib/Headers/arm_acle.h
+++ b/clang/lib/Headers/arm_acle.h
@@ -140,17 +140,21 @@
 /* CLZ */
 static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
 __clz(uint32_t __t) {
-  return (uint32_t)__builtin_clz(__t);
+  return __builtin_arm_clz(__t);
 }
 
 static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
 __clzl(unsigned long __t) {
-  return (unsigned long)__builtin_clzl(__t);
+#if __SIZEOF_LONG__ == 4
+  return __builtin_arm_clz(__t);
+#else
+  return __builtin_arm_clz64(__t);
+#endif
 }
 
 static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
 __clzll(uint64_t __t) {
-  return (uint64_t)__builtin_clzll(__t);
+  return __builtin_arm_clz64(__t);
 }
 
 /* CLS */
diff --git a/clang/test/CodeGen/arm_acle.c b/clang/test/CodeGen/arm_acle.c
--- a/clang/test/CodeGen/arm_acle.c
+++ b/clang/test/CodeGen/arm_acle.c
@@ -344,8 +344,8 @@
 // AArch64-LABEL: @test_clzl(
 // AArch64-NEXT:  entry:
 // AArch64-NEXT:    [[TMP0:%.*]] = call i64 @llvm.ctlz.i64(i64 [[T:%.*]], i1 false)
-// AArch64-NEXT:    [[CAST_I:%.*]] = trunc i64 [[TMP0]] to i32
-// AArch64-NEXT:    [[CONV_I:%.*]] = sext i32 [[CAST_I]] to i64
+// AArch64-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
+// AArch64-NEXT:    [[CONV_I:%.*]] = zext i32 [[TMP1]] to i64
 // AArch64-NEXT:    ret i64 [[CONV_I]]
 //
 long test_clzl(long t) {
@@ -355,8 +355,8 @@
 // ARM-LABEL: @test_clzll(
 // ARM-NEXT:  entry:
 // ARM-NEXT:    [[TMP0:%.*]] = call i64 @llvm.ctlz.i64(i64 [[T:%.*]], i1 false)
-// ARM-NEXT:    [[CAST_I:%.*]] = trunc i64 [[TMP0]] to i32
-// ARM-NEXT:    [[CONV_I:%.*]] = sext i32 [[CAST_I]] to i64
+// ARM-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
+// ARM-NEXT:    [[CONV_I:%.*]] = zext i32 [[TMP1]] to i64
 // ARM-NEXT:    ret i64 [[CONV_I]]
 //
 uint64_t test_clzll(uint64_t t) {