[CodeStyle][F821] fix remaining F821 issues (#47968)

* [CodeStyle][F821] fix remained F821 issues

* refine comment

* fix _set_item

.flake8

@@ -23,7 +23,7 @@ ignore =
 
     # F, see https://flake8.pycqa.org/en/latest/user/error-codes.html
     F405,
-    F811,F821,F841,
+    F811,F841,
 
     # W, see https://pycodestyle.pycqa.org/en/latest/intro.html#error-codes
     W503
@@ -33,3 +33,7 @@ per-file-ignores =
     python/paddle/fluid/tests/unittests/collective/fleet/test_hdfs1.py:E101,W191
     # Ignore unused imports in __init__.py
     __init__.py: F401
+    # Ignore undefined variables in CMake config and some dygraph_to_static tests
+    .cmake-format.py: F821
+    python/paddle/fluid/tests/unittests/dygraph_to_static/test_loop.py: F821
+    python/paddle/fluid/tests/unittests/dygraph_to_static/test_closure_analysis.py: F821

python/paddle/distributed/auto_parallel/completion.py

@@ -1510,8 +1510,12 @@ class Completer:
                     self._dist_context.set_op_dist_attr_for_program(
                         grad_op, grad_op_dist_attr
                     )
-                    grad_op_dist_attr.impl_type = fwd_op_dist_attr.impl_type
-                    grad_op_dist_attr.impl_idx = fwd_op_dist_attr.impl_idx
+                    grad_op_dist_attr.impl_type = (
+                        fwd_op_dist_attr.impl_type  # noqa: F821
+                    )
+                    grad_op_dist_attr.impl_idx = (
+                        fwd_op_dist_attr.impl_idx  # noqa: F821
+                    )
 
                     continue
 

python/paddle/distributed/auto_parallel/cost/tensor_cost.py

@@ -100,9 +100,9 @@ class TensorCost:
 
         if dtype == paddle.float32 or dtype == paddle.int32:
             dtype_factor = 4
-        elif node.dtype == paddle.int64:
+        elif dtype == paddle.int64:
             dtype_factor = 8
-        elif node.dtype == paddle.uint8:
+        elif dtype == paddle.uint8:
             dtype_factor = 1
         else:
             dtype_factor = 2

python/paddle/distributed/auto_parallel/operators/dist_reduce_sum_p.py

@@ -141,7 +141,7 @@ class DistributedReduceSumPrimtiveImpl0(DistributedOperatorImpl):
     def backward(ctx, *args, **kwargs):
         raise RuntimeError(
             "primitive operator does NOT have backward function, op type: {}".format(
-                str(op.type)
+                str(op.type)  # noqa: F821
             )
         )
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_closure_analysis.py

@@ -299,6 +299,8 @@ class TestPushPopTrans(unittest.TestCase):
         print(paddle.jit.to_static(vlist_of_dict)(x))
 
     def test4(self):
+        import numpy as np
+
         def vlist_of_dict(x):
             a = np.array([1, 2, 3])
             for i in range(3):
@@ -310,6 +312,8 @@ class TestPushPopTrans(unittest.TestCase):
         print(paddle.jit.to_static(vlist_of_dict)(x))
 
     def test5(self):
+        import numpy as np
+
         def vlist_of_dict(x):
             a = np.array([1, 2, 3])
             for i in range(3):

python/paddle/fluid/tests/unittests/dygraph_to_static/yolov3.py

@@ -296,7 +296,9 @@ class YOLOv3(fluid.dygraph.Layer):
         blocks = self.block(inputs)
         for i, block in enumerate(blocks):
             if i > 0:
-                block = fluid.layers.concat(input=[route, block], axis=1)
+                block = fluid.layers.concat(
+                    input=[route, block], axis=1  # noqa: F821
+                )
             route, tip = self.yolo_blocks[i](block)
             block_out = self.block_outputs[i](tip)
             self.outputs.append(block_out)

python/paddle/fluid/tests/unittests/gradient_checker.py

@@ -59,7 +59,7 @@ def _get_item(t, i, np_dtype):
         raise ValueError("Not supported data type " + str(np_dtype))
 
 
-def _set_item(t, i, e, np_dtype):
+def _set_item(t, i, e, np_dtype, place):
     if np_dtype == np.float16:
         np_t = np.array(t).astype(np.float16)
         shape = np_t.shape
@@ -145,14 +145,14 @@ def _compute_numerical_jacobian(program, x, y, place, scope, delta):
     for i in range(x_size):
         orig = _get_item(x_t, i, np_type)
         x_pos = orig + delta
-        _set_item(x_t, i, x_pos, np_type)
+        _set_item(x_t, i, x_pos, np_type, place)
         y_pos = run()
 
         x_neg = orig - delta
-        _set_item(x_t, i, x_neg, np_type)
+        _set_item(x_t, i, x_neg, np_type, place)
         y_neg = run()
 
-        _set_item(x_t, i, orig, np_type)
+        _set_item(x_t, i, orig, np_type, place)
 
         for j in range(len(y)):
             jacobian[j][i, :] = (y_pos[j] - y_neg[j]) / delta / 2.0
@@ -207,7 +207,7 @@ def _compute_analytical_jacobian(program, x, y, place, scope):
     filted_idx, filted_dx = zip(*filted)
 
     for i in range(y_size):
-        _set_item(dy_t, i, 1, np_type)
+        _set_item(dy_t, i, 1, np_type, place)
 
         dx_res = exe.run(program, scope=scope, fetch_list=filted_dx)
 
@@ -220,7 +220,7 @@ def _compute_analytical_jacobian(program, x, y, place, scope):
                     dx[dx_idx].shape, dtype=np_type
                 ).flatten()
 
-        _set_item(dy_t, i, 0, np_type)
+        _set_item(dy_t, i, 0, np_type, place)
 
     return jacobian
 

python/paddle/fluid/tests/unittests/test_auto_parallel_reshard_serial.py

@@ -60,8 +60,12 @@ class MLPLayer(nn.Layer):
 
     def forward(self, input):
         if _global_parallel_strategy == "pp":
-            auto.shard_tensor(self.linear0.weight, PP_MESH_0, [None, None])
-            auto.shard_tensor(self.linear1.weight, PP_MESH_1, [None, None])
+            auto.shard_tensor(
+                self.linear0.weight, PP_MESH_0, [None, None]  # noqa: F821
+            )
+            auto.shard_tensor(
+                self.linear1.weight, PP_MESH_1, [None, None]  # noqa: F821
+            )
         else:
             auto.shard_tensor(
                 self.linear0.weight, _global_process_mesh, [None, None]
@@ -93,8 +97,8 @@ def mlp_forward(train_program, start_program):
         )
 
         if _global_parallel_strategy == "pp":
-            auto.shard_tensor(input, PP_MESH_0, [None, None])
-            auto.shard_tensor(label, PP_MESH_1, [None, None])
+            auto.shard_tensor(input, PP_MESH_0, [None, None])  # noqa: F821
+            auto.shard_tensor(label, PP_MESH_1, [None, None])  # noqa: F821
         elif _global_parallel_strategy == "dp":
             auto.shard_tensor(input, _global_process_mesh, ["x", None])
         else:

python/paddle/fluid/tests/unittests/test_reorder_lod_tensor.py

@@ -86,7 +86,9 @@ class TestReorderLoDTensor(unittest.TestCase):
                 lod_level_i = np.random.randint(
                     low=1,
                     high=5,
-                    size=self.num_seq if i == 0 else sum(lod_level_i),
+                    size=self.num_seq
+                    if i == 0
+                    else sum(lod_level_i),  # noqa: F821
                 ).tolist()
                 data_lod.append(lod_level_i)
             data_value = np.random.random(

python/paddle/fluid/tests/unittests/test_weight_normalization.py

@@ -83,7 +83,9 @@ class TestWeightNormalization(unittest.TestCase):
                 lod_level_i = np.random.randint(
                     low=1,
                     high=5,
-                    size=self.batch_size if i == 0 else sum(lod_level_i),
+                    size=self.batch_size
+                    if i == 0
+                    else sum(lod_level_i),  # noqa: F821
                 ).tolist()
                 data_lod.append(lod_level_i)
             data_value = np.random.random(