Dropping Trailing None's When Calculating Emitted ExecutionPlan Outputs (#2243)

Summary:

Previously Support for outputting int/float was added in D53256808 for Seamless

This extends on this by dropping None Outputs (Which can be outputted by NanoGPT)

Differential Revision: D54328663

Author: Jack-Khuu

exir/emit/_emitter.py

@@ -1367,9 +1367,9 @@ def output(
             self.outputs.append(args_tuple.id)
         else:
             for arg in args_tuple:
-                if isinstance(arg, (int, float, bool)):
+                if isinstance(arg, (int, float, bool, type(None))):
                     arg = self._emit_evalue(self._constant_to_evalue(arg, None))
-                elif isinstance(arg, (type(None), str)):
+                elif isinstance(arg, str):
                     raise InternalError(
                         self._emit_node_specific_error(
                             self.node,

exir/emit/test/test_emit.py

@@ -213,14 +213,14 @@ def forward(
     def test_constant_output(self):
         class M(torch.nn.Module):
             def forward(self, x):
-                return [((1, 3, 1.2), True, [x + x, x * x])]
+                return [((1, 3, 1.2), True, [x + x, x * x], None)]
 
         ep = torch.export.export(M(), (torch.ones(2, 3),))
         res = ep.module()(torch.ones(2, 3))
         self.assertEqual(res[0][0], (1, 3, 1.2))
         program = to_edge(ep).to_executorch().executorch_program
         outputs = program.execution_plan[0].outputs
-        self.assertEqual(len(outputs), 6)
+        self.assertEqual(len(outputs), 7)
         self.assertEqual(program.execution_plan[0].values[outputs[0]].val.int_val, 1)
         self.assertEqual(program.execution_plan[0].values[outputs[1]].val.int_val, 3)
         self.assertEqual(
@@ -229,6 +229,7 @@ def forward(self, x):
         self.assertEqual(
             program.execution_plan[0].values[outputs[3]].val.bool_val, True
         )
+        self.assertIsInstance(program.execution_plan[0].values[outputs[6]].val, Null)
 
     def test_int_list_input(self):
         class M(torch.nn.Module):