Lower lift and lift_copy ops

Author: wonjoo-wj

test/pytorch_test_base.py

@@ -192,6 +192,7 @@
         'test_random_to_xla',  # doesn't raise
         'test_copy_',  # test against complex32 which is nto supported
         'test_assertRaisesRegex_ignore_msg_non_native_device_xla',  # segfault on wheel sanity test
+        'test_index_reduce',  # takes too long
     },
 
     # test_view_ops.py

torch_xla/csrc/aten_xla_type.cpp

@@ -1467,6 +1467,19 @@ at::Tensor XLANativeFunctions::lerp(const at::Tensor& self,
       bridge::GetXlaTensor(self), bridge::GetXlaTensor(end), weight));
 }
 
+at::Tensor XLANativeFunctions::lift_fresh(const at::Tensor& self) {
+  std::cout << "WONJOO: at aten_xla_type.cpp, lift_fresh" << std::endl;
+  return at::functionalization::impl::to_functional_tensor(self);
+  // return at::functionalization::functionalize_aten_op<ATEN_OP(
+  //     lift_fresh)>::call(self);
+}
+
+at::Tensor XLANativeFunctions::lift_fresh_copy(const at::Tensor& self) {
+  std::cout << "WONJOO: at aten_xla_type.cpp, lift_fresh_copy" << std::endl;
+  return at::functionalization::functionalize_aten_op<ATEN_OP(
+      lift_fresh_copy)>::call(self);
+}
+
 at::Tensor XLANativeFunctions::linspace(const at::Scalar& start,
                                         const at::Scalar& end, int64_t steps,
                                         c10::optional<at::ScalarType> dtype,

xla_native_functions.yaml

@@ -197,6 +197,8 @@ supported:
   - leaky_relu_backward
   - lerp.Scalar
   - lerp.Tensor
+  - lift_fresh
+  - lift_fresh_copy
   - linspace
   - log
   - log1p