Dynamic shape testing for softmax

Differential Revision: D68586985

Pull Request resolved: pytorch#7914

Author: digantdesai

backends/xnnpack/test/ops/test_softmax.py

@@ -24,17 +24,22 @@ def _test_softmax(self, inputs):
         # as xnnpack only supports softmax on the last dimension.
         valid_dims = [len(inputs[0]) - 1, -1]
 
+        dynamic_shape = {}
+        for i in range(len(inputs[0].shape)):
+            dynamic_shape[i] = torch.export.Dim(f"dynamic_dim{i}", min=1, max=100)
+        dynamic_shape = (dynamic_shape,)
+
         for dim in valid_dims:
             (
-                Tester(self.Softmax(dim), inputs)
+                Tester(self.Softmax(dim), inputs, dynamic_shapes=dynamic_shape)
                 .export()
                 .check_count({"torch.ops.aten.softmax": 1})
                 .to_edge_transform_and_lower()
                 .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
                 .check_not(["executorch_exir_dialects_edge__ops_aten__softmax_default"])
                 .to_executorch()
                 .serialize()
-                .run_method_and_compare_outputs()
+                .run_method_and_compare_outputs(num_runs=5)
             )
 
     def test_fp16_softmax(self):