[Clean fluid] Add inner function _elementwise_op_with_axis

python/paddle/fluid/contrib/layers/rnn_impl.py

@@ -367,9 +367,9 @@ def get_single_direction_output(
                 new_hidden = unit_list[i](step_input, pre_hidden)
 
                 if mask:
-                    new_hidden = layers.elementwise_mul(
+                    new_hidden = paddle.tensor.math._multiply_with_axis(
                         new_hidden, step_mask, axis=0
-                    ) - layers.elementwise_mul(
+                    ) - paddle.tensor.math._multiply_with_axis(
                         pre_hidden, (step_mask - 1), axis=0
                     )
                 rnn.update_memory(pre_hidden, new_hidden)
@@ -661,14 +661,14 @@ def get_single_direction_output(
                 )
 
                 if mask:
-                    new_hidden = layers.elementwise_mul(
+                    new_hidden = paddle.tensor.math._multiply_with_axis(
                         new_hidden, step_mask, axis=0
-                    ) - layers.elementwise_mul(
+                    ) - paddle.tensor.math._multiply_with_axis(
                         pre_hidden, (step_mask - 1), axis=0
                     )
-                    new_cell = layers.elementwise_mul(
+                    new_cell = paddle.tensor.math._multiply_with_axis(
                         new_cell, step_mask, axis=0
-                    ) - layers.elementwise_mul(
+                    ) - paddle.tensor.math._multiply_with_axis(
                         pre_cell, (step_mask - 1), axis=0
                     )
 

python/paddle/fluid/layer_helper_base.py

@@ -115,8 +115,6 @@ def to_variable(self, value, name=None):
             )
 
     def _create_weight_normalize(self, attr, shape, dtype):
-        from .layers import elementwise_mul
-
         # Remove these ops when LayerHelper and layers support indicating
         # program and block.
         def __norm_op(
@@ -272,7 +270,7 @@ def __weight_normalize(g, v, dim):
             # Currently, elementwise_mul only support broadcast when the shape
             # of y is a subset of the shape of x. Thus, we reshape y to squeeze
             # to achieve the subset.
-            w = elementwise_mul(
+            w = paddle.tensor.math._multiply_with_axis(
                 x=v,
                 y=scale
                 if dim is None

python/paddle/fluid/layers/nn.py

@@ -81,9 +81,7 @@
     'lod_reset',
     'relu',
     'elementwise_add',
-    'elementwise_div',
     'elementwise_sub',
-    'elementwise_mul',
     'clip',
     'clip_by_norm',
     'mean',
@@ -2839,96 +2837,6 @@ def gen_data():
     return _elementwise_op(LayerHelper('elementwise_add', **locals()))
 
 
-@deprecated(since="2.0.0", update_to="paddle.divide")
-def elementwise_div(x, y, axis=-1, act=None, name=None):
-    """
-
-    Examples:
-
-        .. code-block:: python
-
-            import paddle.fluid as fluid
-            import numpy as np
-            import paddle
-
-            def gen_data():
-                return {
-                    "x": np.array([2, 3, 4]).astype('float32'),
-                    "y": np.array([1, 5, 2]).astype('float32')
-                }
-            paddle.enable_static()
-            x = fluid.data(name="x", shape=[3], dtype='float32')
-            y = fluid.data(name="y", shape=[3], dtype='float32')
-            z = fluid.layers.elementwise_div(x, y)
-            # z = x / y
-
-            place = fluid.CPUPlace()
-            exe = fluid.Executor(place)
-            z_value = exe.run(feed=gen_data(),
-                                fetch_list=[z.name])
-
-            print(z_value) # [2., 0.6, 2.]
-
-
-        .. code-block:: python
-
-            import paddle.fluid as fluid
-            import numpy as np
-            import paddle
-
-            def gen_data():
-                return {
-                    "x": np.ones((2, 3, 4, 5)).astype('float32'),
-                    "y": np.zeros((3, 4)).astype('float32')
-                }
-            paddle.enable_static()
-            x = fluid.data(name="x", shape=[2,3,4,5], dtype='float32')
-            y = fluid.data(name="y", shape=[3,4], dtype='float32')
-            z = fluid.layers.elementwise_div(x, y, axis=1)
-            # z = x / y
-
-            place = fluid.CPUPlace()
-            exe = fluid.Executor(place)
-
-            z_value = exe.run(feed=gen_data(),
-                                fetch_list=[z.name])
-
-            print(z_value) # z.shape=[2,3,4,5]
-
-
-        ..  code-block:: python
-
-            import paddle.fluid as fluid
-            import numpy as np
-            import paddle
-
-            def gen_data():
-                return {
-                    "x": np.random.randint(1, 5, size=[2, 3, 4, 5]).astype('float32'),
-                    "y": np.random.randint(1, 5, size=[5]).astype('float32')
-                }
-            paddle.enable_static()
-            x = fluid.data(name="x", shape=[2,3,4,5], dtype='float32')
-            y = fluid.data(name="y", shape=[5], dtype='float32')
-            z = fluid.layers.elementwise_div(x, y, axis=3)
-            # z = x / y
-
-            place = fluid.CPUPlace()
-            exe = fluid.Executor(place)
-
-            z_value = exe.run(feed=gen_data(),
-                                fetch_list=[z.name])
-            print(z_value) # z.shape=[2,3,4,5]
-
-    """
-    if _non_static_mode():
-        return _elementwise_op_in_dygraph(
-            x, y, axis=axis, act=act, op_name='elementwise_div'
-        )
-
-    return _elementwise_op(LayerHelper('elementwise_div', **locals()))
-
-
 def elementwise_sub(x, y, axis=-1, act=None, name=None):
     """
 
@@ -3018,101 +2926,9 @@ def gen_data():
     return _elementwise_op(LayerHelper('elementwise_sub', **locals()))
 
 
-@deprecated(since="2.0.0", update_to="paddle.multiply")
-def elementwise_mul(x, y, axis=-1, act=None, name=None):
-    """
-
-    Examples:
-
-        .. code-block:: python
-
-            import paddle.fluid as fluid
-            import numpy as np
-            import paddle
-
-            def gen_data():
-                return {
-                    "x": np.array([2, 3, 4]).astype('float32'),
-                    "y": np.array([1, 5, 2]).astype('float32')
-                }
-            paddle.enable_static()
-            x = fluid.data(name="x", shape=[3], dtype='float32')
-            y = fluid.data(name="y", shape=[3], dtype='float32')
-            z = fluid.layers.elementwise_mul(x, y)
-            # z = x * y
-
-            place = fluid.CPUPlace()
-            exe = fluid.Executor(place)
-            z_value = exe.run(feed=gen_data(),
-                                fetch_list=[z.name])
-
-            print(z_value) # [2., 15., 8.]
-
-
-        .. code-block:: python
-
-            import paddle.fluid as fluid
-            import numpy as np
-            import paddle
-
-            def gen_data():
-                return {
-                    "x": np.ones((2, 3, 4, 5)).astype('float32'),
-                    "y": np.zeros((3, 4)).astype('float32')
-                }
-            paddle.enable_static()
-            x = fluid.data(name="x", shape=[2,3,4,5], dtype='float32')
-            y = fluid.data(name="y", shape=[3,4], dtype='float32')
-            z = fluid.layers.elementwise_mul(x, y, axis=1)
-            # z = x * y
-
-            place = fluid.CPUPlace()
-            exe = fluid.Executor(place)
-
-            z_value = exe.run(feed=gen_data(),
-                                fetch_list=[z.name])
-
-            print(z_value) # z.shape=[2,3,4,5]
-
-
-        ..  code-block:: python
-
-            import paddle.fluid as fluid
-            import numpy as np
-            import paddle
-
-            def gen_data():
-                return {
-                    "x": np.random.randint(1, 5, size=[2, 3, 4, 5]).astype('float32'),
-                    "y": np.random.randint(1, 5, size=[5]).astype('float32')
-                }
-            paddle.enable_static()
-            x = fluid.data(name="x", shape=[2,3,4,5], dtype='float32')
-            y = fluid.data(name="y", shape=[5], dtype='float32')
-            z = fluid.layers.elementwise_mul(x, y, axis=3)
-            # z = x * y
-
-            place = fluid.CPUPlace()
-            exe = fluid.Executor(place)
-
-            z_value = exe.run(feed=gen_data(),
-                                fetch_list=[z.name])
-            print(z_value) # z.shape=[2,3,4,5]
-
-    """
-    if _non_static_mode():
-        return _elementwise_op_in_dygraph(
-            x, y, axis=axis, act=act, op_name='elementwise_mul'
-        )
-
-    return _elementwise_op(LayerHelper('elementwise_mul', **locals()))
-
-
 for func in [
     elementwise_add,
-    elementwise_div,
     elementwise_sub,
-    elementwise_mul,
 ]:
     op_proto = OpProtoHolder.instance().get_op_proto(func.__name__)
 

python/paddle/fluid/layers/rnn.py

@@ -553,9 +553,9 @@ def __getitem__(self, item):
 
 def _maybe_copy(state, new_state, step_mask):
     """update rnn state or just pass the old state through"""
-    new_state = nn.elementwise_mul(
+    new_state = paddle.tensor.math._multiply_with_axis(
         new_state, step_mask, axis=0
-    ) + nn.elementwise_mul(state, (1 - step_mask), axis=0)
+    ) + paddle.tensor.math._multiply_with_axis(state, (1 - step_mask), axis=0)
     return new_state
 
 
@@ -839,9 +839,11 @@ def _maybe_copy(state, new_state, step_mask):
             # otherwise, renamed bool gradients of would be summed up leading
             # to sum(bool) error.
             step_mask.stop_gradient = True
-        new_state = nn.elementwise_mul(
+        new_state = paddle.tensor.math._multiply_with_axis(
             state, step_mask, axis=0
-        ) - nn.elementwise_mul(new_state, (step_mask - 1), axis=0)
+        ) - paddle.tensor.math._multiply_with_axis(
+            new_state, (step_mask - 1), axis=0
+        )
         if convert_dtype(state_dtype) in ["bool"]:
             new_state = tensor.cast(new_state, dtype=state_dtype)
         return new_state
@@ -994,9 +996,11 @@ def _maybe_copy(state, new_state, step_mask):
             # otherwise, renamed bool gradients of would be summed up leading
             # to sum(bool) error.
             step_mask.stop_gradient = True
-        new_state = nn.elementwise_mul(
+        new_state = paddle.tensor.math._multiply_with_axis(
             state, step_mask, axis=0
-        ) - nn.elementwise_mul(new_state, (step_mask - 1), axis=0)
+        ) - paddle.tensor.math._multiply_with_axis(
+            new_state, (step_mask - 1), axis=0
+        )
         if convert_dtype(state_dtype) in ["bool"]:
             new_state = tensor.cast(new_state, dtype=state_dtype)
         return new_state

python/paddle/fluid/tests/unittests/collective/fleet/parallel_dygraph_se_resnext.py

@@ -139,7 +139,7 @@ def forward(self, input):
         y = paddle.nn.functional.relu(y)
         y = self._excitation(y)
         y = paddle.nn.functional.sigmoid(y)
-        y = fluid.layers.elementwise_mul(x=input, y=y, axis=0)
+        y = paddle.tensor.math._multiply_with_axis(x=input, y=y, axis=0)
         return y
 
 

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

@@ -206,7 +206,9 @@ def squeeze_excitation(self, input, num_channels, reduction_ratio):
             ),
             act='sigmoid',
         )
-        scale = fluid.layers.elementwise_mul(x=input, y=excitation, axis=0)
+        scale = paddle.tensor.math._multiply_with_axis(
+            x=input, y=excitation, axis=0
+        )
         return scale
 
 

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

@@ -1806,7 +1806,7 @@ def beam_search():
                 }
                 for cache in caches
             ]
-            pre_pos = layers.elementwise_mul(
+            pre_pos = paddle.tensor.math._multiply_with_axis(
                 x=layers.fill_constant_batch_size_like(
                     input=pre_enc_output,  # can't use pre_ids here since it has lod
                     value=1,

python/paddle/fluid/tests/unittests/dygraph_to_static/seq2seq_dygraph_model.py

@@ -194,9 +194,11 @@ def _expand_to_beam_size(self, x):
         return x
 
     def _real_state(self, state, new_state, step_mask):
-        new_state = fluid.layers.elementwise_mul(
+        new_state = paddle.tensor.math._multiply_with_axis(
             new_state, step_mask, axis=0
-        ) - fluid.layers.elementwise_mul(state, (step_mask - 1), axis=0)
+        ) - paddle.tensor.math._multiply_with_axis(
+            state, (step_mask - 1), axis=0
+        )
         return new_state
 
     def _gather(self, x, indices, batch_pos):
@@ -450,7 +452,7 @@ def beam_search(self, inputs):
                     [-1, -1, self.tar_vocab_size],
                 ),
                 noend_mask_tensor,
-            ) - fluid.layers.elementwise_mul(
+            ) - paddle.tensor.math._multiply_with_axis(
                 step_log_probs, (beam_finished - 1), axis=0
             )
             log_probs = fluid.layers.elementwise_add(
@@ -685,9 +687,11 @@ def _expand_to_beam_size(self, x):
         return x
 
     def _real_state(self, state, new_state, step_mask):
-        new_state = fluid.layers.elementwise_mul(
+        new_state = paddle.tensor.math._multiply_with_axis(
             new_state, step_mask, axis=0
-        ) - fluid.layers.elementwise_mul(state, (step_mask - 1), axis=0)
+        ) - paddle.tensor.math._multiply_with_axis(
+            state, (step_mask - 1), axis=0
+        )
         return new_state
 
     def _gather(self, x, indices, batch_pos):

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

@@ -152,7 +152,7 @@ def forward(self, input):
         y = paddle.nn.functional.relu(y)
         y = self._excitation(y)
         y = paddle.nn.functional.sigmoid(y)
-        y = fluid.layers.elementwise_mul(x=input, y=y, axis=0)
+        y = paddle.tensor.math._multiply_with_axis(x=input, y=y, axis=0)
         return y
 
 

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

@@ -756,7 +756,9 @@ def mask_probs(probs, finished, noend_mask_tensor):
                     [-1, -1, self.trg_vocab_size],
                 ),
                 noend_mask_tensor,
-            ) - layers.elementwise_mul(probs, (finished - 1), axis=0)
+            ) - paddle.tensor.math._multiply_with_axis(
+                probs, (finished - 1), axis=0
+            )
             return probs
 
         def gather(input, indices, batch_pos):

python/paddle/fluid/tests/unittests/ipu/test_elemetwise_x_op_ipu.py

@@ -35,7 +35,7 @@ def fp16_enabled(self):
             return True
 
     def set_test_op(self):
-        self.op = paddle.fluid.layers.elementwise_mul
+        self.op = paddle.tensor.math._multiply_with_axis
 
     def set_feed_attr(self):
         self.feed_shape = [x.shape for x in self.feed_fp32.values()]