diff --git a/src/device/intrinsics/output.jl b/src/device/intrinsics/output.jl
index 2b2ba00a..53ece0f9 100644
--- a/src/device/intrinsics/output.jl
+++ b/src/device/intrinsics/output.jl
@@ -75,10 +75,16 @@ end
         T_pint8 = LLVM.PointerType(LLVM.Int8Type())
         T_pint8a2 = LLVM.PointerType(LLVM.Int8Type(), 2)
 
+
+
         # create functions
         param_types = LLVMType[convert(LLVMType, typ) for typ in arg_types]
-        llvm_f, llvm_ft = create_function(T_void, LLVMType[]; vararg=true)
-        mod = LLVM.parent(llvm_f)
+        wrapper_f, wrapper_ft = create_function(T_void, param_types)
+        mod = LLVM.parent(wrapper_f)
+
+        llvm_ft = LLVM.FunctionType(T_void, LLVMType[]; vararg=true)
+        llvm_f = LLVM.Function(mod, "vararg", llvm_ft)
+        push!(function_attributes(llvm_f), EnumAttribute("alwaysinline", 0))
 
         # generate IR
         @dispose builder=IRBuilder() begin
@@ -86,6 +92,9 @@ end
             position!(builder, entry)
 
             str = globalstring_ptr!(builder, String(fmt), addrspace=2)
+            subsystem_str = globalstring_ptr!(builder, MTLLOG_SUBSYSTEM, addrspace=2)
+            category_str = globalstring_ptr!(builder, MTLLOG_CATEGRORY, addrspace=2)
+            log_type = LLVM.ConstantInt(T_int32, __METAL_OS_LOG_TYPE_DEBUG__)
 
             # compute argsize
             argtypes = LLVM.StructType(param_types)
@@ -103,16 +112,11 @@ end
             va_start_fty = LLVM.FunctionType(T_void, [T_pint8])
             va_start = LLVM.Function(mod, "llvm.va_start", va_start_fty)
             call!(builder, va_start_fty, va_start, [buffer])
-
-            # invoke @air.os_log and return
-            subsystem_str = globalstring_ptr!(builder, MTLLOG_SUBSYSTEM, addrspace=2)
-            category_str = globalstring_ptr!(builder, MTLLOG_CATEGRORY, addrspace=2)
-            log_type = LLVM.ConstantInt(T_int32, __METAL_OS_LOG_TYPE_DEBUG__)
-            os_log_fty = LLVM.FunctionType(T_void, [T_pint8a2, T_pint8a2, T_int32, T_pint8a2, T_pint8, T_int64])
-            os_log = LLVM.Function(mod, "air.os_log", os_log_fty)
             
             arg_ptr = load!(builder, T_pint8, alloc)
             
+            os_log_fty = LLVM.FunctionType(T_void, [T_pint8a2, T_pint8a2, T_int32, T_pint8a2, T_pint8, T_int64])
+            os_log = LLVM.Function(mod, "air.os_log", os_log_fty)
             call!(builder, os_log_fty, os_log, [subsystem_str, category_str, log_type, str, arg_ptr, arg_size])
 
             va_end_fty = LLVM.FunctionType(T_void, [T_pint8])
@@ -126,19 +130,17 @@ end
             ret!(builder)
         end
 
-        wrapper_f, wrapper_ft = create_function(T_void, param_types)
-
-        @dispose builder=IRBuilder() begin
-            entry = BasicBlock(wrapper_f, "entry")
-            position!(builder, entry)
+        # @dispose builder=IRBuilder() begin
+        #     entry = BasicBlock(wrapper_f, "entry")
+        #     position!(builder, entry)
             
-            # call!(builder, llvm_ft, llvm_f, collect(parameters(wrapper_f)))
+        #     call!(builder, llvm_ft, llvm_f, collect(parameters(wrapper_f)))
             
-            ret!(builder)
-        end
+        #     ret!(builder)
+        # end
 
 
-        call_function(wrapper_f, Nothing, Tuple{arg_types...}, arg_exprs...)
+        call_function(llvm_f, Nothing, Tuple{arg_types...}, arg_exprs...)
     end
 end