diff --git a/include/clang/Basic/Builtins.def b/include/clang/Basic/Builtins.def
index 3f97b78..9a4c768 100644
--- a/include/clang/Basic/Builtins.def
+++ b/include/clang/Basic/Builtins.def
@@ -577,6 +577,13 @@ BUILTIN(__sync_lock_release_4, "viD*.", "n")
 BUILTIN(__sync_lock_release_8, "vLLiD*.", "n")
 BUILTIN(__sync_lock_release_16, "vLLLiD*.", "n")
 
+BUILTIN(__sync_swap, "v.", "")
+BUILTIN(__sync_swap_1, "ccD*c.", "n")
+BUILTIN(__sync_swap_2, "ssD*s.", "n")
+BUILTIN(__sync_swap_4, "iiD*i.", "n")
+BUILTIN(__sync_swap_8, "LLiLLiD*LLi.", "n")
+BUILTIN(__sync_swap_16, "LLLiLLLiD*LLLi.", "n")
+
 
 
 // Non-overloaded atomic builtins.
diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 35507b9..af8d37a 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -721,6 +721,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
   case Builtin::BI__sync_bool_compare_and_swap:
   case Builtin::BI__sync_lock_test_and_set:
   case Builtin::BI__sync_lock_release:
+  case Builtin::BI__sync_swap:
     assert(0 && "Shouldn't make it through sema");
   case Builtin::BI__sync_fetch_and_add_1:
   case Builtin::BI__sync_fetch_and_add_2:
@@ -860,6 +861,13 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     return RValue::get(Result);
   }
 
+  case Builtin::BI__sync_swap_1:
+  case Builtin::BI__sync_swap_2:
+  case Builtin::BI__sync_swap_4:
+  case Builtin::BI__sync_swap_8:
+  case Builtin::BI__sync_swap_16:
+    return EmitBinaryAtomic(*this, Intrinsic::atomic_swap, E);
+
   case Builtin::BI__sync_lock_test_and_set_1:
   case Builtin::BI__sync_lock_test_and_set_2:
   case Builtin::BI__sync_lock_test_and_set_4:
diff --git a/lib/Sema/SemaChecking.cpp b/lib/Sema/SemaChecking.cpp
index bdadf5a..58155f4 100644
--- a/lib/Sema/SemaChecking.cpp
+++ b/lib/Sema/SemaChecking.cpp
@@ -180,6 +180,7 @@ Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case Builtin::BI__sync_bool_compare_and_swap:
   case Builtin::BI__sync_lock_test_and_set:
   case Builtin::BI__sync_lock_release:
+  case Builtin::BI__sync_swap:
     return SemaBuiltinAtomicOverloaded(move(TheCallResult));
   }
   
@@ -415,7 +416,8 @@ Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
     BUILTIN_ROW(__sync_val_compare_and_swap),
     BUILTIN_ROW(__sync_bool_compare_and_swap),
     BUILTIN_ROW(__sync_lock_test_and_set),
-    BUILTIN_ROW(__sync_lock_release)
+    BUILTIN_ROW(__sync_lock_release),
+    BUILTIN_ROW(__sync_swap)
   };
 #undef BUILTIN_ROW
 
@@ -468,6 +470,7 @@ Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
     NumFixed = 0;
     ResultType = Context.VoidTy;
     break;
+  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
   }
 
   // Now that we know how many fixed arguments we expect, first check that we
diff --git a/test/CodeGen/atomic.c b/test/CodeGen/atomic.c
index 4a7c13f..8ce2d96 100644
--- a/test/CodeGen/atomic.c
+++ b/test/CodeGen/atomic.c
@@ -44,6 +44,11 @@ int atomic(void) {
   // CHECK: call i32 @llvm.atomic.swap.i32.p0i32(i32* %val, i32 7)
   // CHECK: call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
   
+  old = __sync_swap(&val, 8);
+  // CHECK: call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
+  // CHECK: call i32 @llvm.atomic.swap.i32.p0i32(i32* %val, i32 8)
+  // CHECK: call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
+  
   old = __sync_val_compare_and_swap(&val, 4, 1976);
   // CHECK: call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
   // CHECK: call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %val, i32 4, i32 1976)