merge parameter slice compile graph only once

mindspore/nn/cell.py

@@ -90,6 +90,7 @@ class Cell(Cell_):
         self._phase = 'train'
         self._parameter_layout_dict = {}
         self._parallel_parameter_name_list = ()
+        self._parallel_parameter_merge_net_dict = {}
         self._create_time = int(time.time() * 1e9)
         self.phase_prefix = ""
         self.parameter_broadcast_done = False
@@ -224,6 +225,16 @@ class Cell(Cell_):
             raise TypeError("'parallel_parameter_name_list' must be list type.")
         self._parallel_parameter_name_list = value
 
+    @property
+    def parallel_parameter_merge_net_dict(self):
+        return self._parallel_parameter_merge_net_dict
+
+    @parallel_parameter_merge_net_dict.setter
+    def parallel_parameter_merge_net_dict(self, value):
+        if not isinstance(value, dict):
+            raise TypeError("'parallel_parameter_merge_net_dict' must be dict type.")
+        self._parallel_parameter_merge_net_dict = value
+
     def get_func_graph_proto(self):
         """Return graph binary proto."""
         return _executor._get_func_graph_proto(self, self.phase + "." + str(self.create_time), "anf_ir", True)
@@ -382,46 +393,63 @@ class Cell(Cell_):
             self._add_attr(key, value)
         self._attr_synced = True
 
-    def __setattr__(self, name, value):
+    def _set_attr_for_parameter(self, name, value):
+        """Set attr for parameter."""
         cells = self.__dict__.get('_cells')
         params = self.__dict__.get('_params')
+        if params is None:
+            raise AttributeError("Can not assign params before Cell.__init__() call.")
+        if name in self.__dict__:
+            if self.__dict__[name] is not None:
+                raise TypeError("The type of value should not be Parameter or Cell, but got Parameter.")
+            del self.__dict__[name]
+        if cells and name in cells:
+            raise TypeError("The type of value should be Cell, but got Parameter.")
+        self.insert_param_to_cell(name, value)
+
+    def _set_attr_for_parameter_tuple(self, name, value):
+        """Set attr for parameter tuple."""
+        params = self.__dict__.get('_params')
         params_list = self.__dict__.get('_params_list')
-        tensor_list = self.__dict__.get('_tensor_list')
-        if isinstance(value, Parameter):
-            if params is None:
-                raise AttributeError("Can not assign params before Cell.__init__() call.")
+        if params is None:
+            raise AttributeError("Can not assign params before Cell.__init__() call.")
+        for item in value:
+            self.insert_param_to_cell(item.name, item, check_name=False)
+        if context.get_context("mode") == context.PYNATIVE_MODE:
             if name in self.__dict__:
-                if self.__dict__[name] is not None:
-                    raise TypeError("The type of value should not be Parameter or Cell, but got Parameter.")
                 del self.__dict__[name]
-            if cells and name in cells:
-                raise TypeError("The type of value should be Cell, but got Parameter.")
-            self.insert_param_to_cell(name, value)
+            if name in params:
+                del params[name]
+            params_list[name] = value
+        else:
+            object.__setattr__(self, name, value)
+
+    def _set_attr_for_cell(self, name, value):
+        """Set attr for cell."""
+        cells = self.__dict__.get('_cells')
+        params = self.__dict__.get('_params')
+        if cells is None:
+            raise AttributeError("Can not assign cells before Cell.__init__() call.")
+        if name in self.__dict__:
+            del self.__dict__[name]
+        if params and name in params:
+            raise TypeError("The type of value should be Parameter, but got Cell.")
+        if self._auto_prefix:
+            value.update_parameters_name(name + '.')
+        cells[name] = value
+        if hasattr(self, '_cell_init_args'):
+            self.cell_init_args += str({name: value})
+
+    def __setattr__(self, name, value):
+        cells = self.__dict__.get('_cells')
+        params = self.__dict__.get('_params')
+        tensor_list = self.__dict__.get('_tensor_list')
+        if isinstance(value, Parameter):
+            self._set_attr_for_parameter(name, value)
         elif isinstance(value, ParameterTuple):
-            if params is None:
-                raise AttributeError("Can not assign params before Cell.__init__() call.")
-            for item in value:
-                self.insert_param_to_cell(item.name, item, check_name=False)
-            if context.get_context("mode") == context.PYNATIVE_MODE:
-                if name in self.__dict__:
-                    del self.__dict__[name]
-                if name in params:
-                    del params[name]
-                params_list[name] = value
-            else:
-                object.__setattr__(self, name, value)
+            self._set_attr_for_parameter_tuple(name, value)
         elif isinstance(value, Cell):
-            if cells is None:
-                raise AttributeError("Can not assign cells before Cell.__init__() call.")
-            if name in self.__dict__:
-                del self.__dict__[name]
-            if params and name in params:
-                raise TypeError("The type of value should be Parameter, but got Cell.")
-            if self._auto_prefix:
-                value.update_parameters_name(name + '.')
-            cells[name] = value
-            if hasattr(self, '_cell_init_args'):
-                self.cell_init_args += str({name: value})
+            self._set_attr_for_cell(name, value)
         elif params and name in params:
             if isinstance(value, Tensor) and self._params[name] is not None:
                 self._params[name].set_data(value)

mindspore/train/serialization.py

@@ -455,6 +455,12 @@ def _get_merged_param_data(net, param_name, param_data, integrated_save):
     uniform_split = layout[4]
     opt_shard_group = layout[5]
 
+    allgather_net = None
+    if param_name in net.parallel_parameter_merge_net_dict:
+        allgather_net = net.parallel_parameter_merge_net_dict[param_name]
+    else:
+        logger.info("need to create allgather net for %s", param_name)
+
     if integrated_save:
         if uniform_split == 0:
             raise RuntimeError("Integrated save checkpoint only support uniform split tensor now.")
@@ -462,19 +468,25 @@ def _get_merged_param_data(net, param_name, param_data, integrated_save):
         # pipeline parallel need to be supported here later
         for dim in tensor_map:
             if dim != -1:
-                if opt_shard_group:
-                    allgather_net = get_allgather_cell(opt_shard_group, True)
-                else:
-                    allgather_net = get_allgather_cell(opt_shard_group, False)
+                if allgather_net is None:
+                    if opt_shard_group:
+                        allgather_net = get_allgather_cell(opt_shard_group, True)
+                    else:
+                        allgather_net = get_allgather_cell(opt_shard_group, False)
+                    net.parallel_parameter_merge_net_dict[param_name] = allgather_net
                 param_data = allgather_net(param_data)
                 if field_size:
                     return _reshape_param_data_with_weight(param_data, dev_mat, field_size)
                 return _reshape_param_data(param_data, dev_mat, tensor_map)
         if opt_shard_group:
-            allgather_net = get_allgather_cell(opt_shard_group, False)
+            if allgather_net is None:
+                allgather_net = get_allgather_cell(opt_shard_group, False)
+                net.parallel_parameter_merge_net_dict[param_name] = allgather_net
             param_data = allgather_net(param_data)
     elif opt_shard_group:
-        allgather_net = get_allgather_cell(opt_shard_group, False)
+        if allgather_net is None:
+            allgather_net = get_allgather_cell(opt_shard_group, False)
+            net.parallel_parameter_merge_net_dict[param_name] = allgather_net
         param_data = allgather_net(param_data)
     return param_data
 

tests/ut/python/parallel/test_parameter_merge.py

@@ -0,0 +1,105 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+import os
+import numpy as np
+
+import mindspore as ms
+from mindspore import context, Tensor, Parameter
+from mindspore.nn import Cell, Momentum
+from mindspore.ops import operations as P
+from mindspore.train import Model
+from mindspore.train.callback import CheckpointConfig, ModelCheckpoint
+from tests.dataset_mock import MindData
+
+
+class Dataset(MindData):
+    def __init__(self, predict, label, length=3):
+        super(Dataset, self).__init__(size=length)
+        self.predict = predict
+        self.label = label
+        self.index = 0
+        self.length = length
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        if self.index >= self.length:
+            raise StopIteration
+        self.index += 1
+        return self.predict, self.label
+
+    def reset(self):
+        self.index = 0
+
+
+class Net(Cell):
+    def __init__(self, weight, w2, begin, end, strides, strategy1=None, strategy2=None, mask=0):
+        super().__init__()
+        self.mul = P.Mul().shard(strategy1)
+        self.strided_slice = P.StridedSlice(begin_mask=mask).shard(strategy2)
+        self.weight = Parameter(weight, "w1")
+        self.mul2 = P.Mul()
+        self.weight2 = Parameter(w2, "w2")
+        self.begin = begin
+        self.end = end
+        self.strides = strides
+
+    def construct(self, x, b):
+        out = self.strided_slice(
+            self.weight, self.begin, self.end, self.strides)
+        out = self.mul(x, out)
+        out = self.mul2(out, self.weight2)
+        return out
+
+
+_x = Tensor(np.ones([16, 64, 1]), dtype=ms.float32)
+_b = Tensor(np.ones([16, 64, 32]), dtype=ms.float32)
+_w1 = Tensor(np.ones([256, 64, 32]), dtype=ms.float32)
+_w2 = Tensor(np.ones([128, 64, 1]), dtype=ms.float32)
+
+
+def clean_all_ckpt_files(folder_path):
+    if os.path.exists(folder_path):
+        for file_name in os.listdir(folder_path):
+            if file_name.endswith('.ckpt') or file_name.endswith('.meta'):
+                os.remove(os.path.join(folder_path, file_name))
+
+
+def compile_net(net):
+    context.set_context(save_graphs=True)
+    learning_rate = 0.1
+    momentum = 0.9
+    epoch_size = 2
+    dataset = Dataset(_x, _b)
+    opt = Momentum(net.trainable_params(), learning_rate, momentum)
+    model = Model(net, optimizer=opt)
+    ckpt_config = CheckpointConfig(keep_checkpoint_max=1)
+    ckpt_path = "./parallel_ckpt"
+    ckpt_cb = ModelCheckpoint(prefix="parallel", directory=ckpt_path, config=ckpt_config)
+    model.train(epoch_size, dataset, dataset_sink_mode=False, callbacks=[ckpt_cb])
+    assert len(model._train_network.parallel_parameter_merge_net_dict) == 4
+    clean_all_ckpt_files(ckpt_path)
+    context.reset_auto_parallel_context()
+
+
+def test_stridedslice_parameter():
+    context.set_auto_parallel_context(
+        parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
+    strategy1 = ((1, 4, 1), (1, 4, 2))
+    strategy2 = ((1, 4, 2),)
+    net = Net(_w1, _w2, (0, 0, 0), (128, 64, 32), (1, 1, 1),
+              strategy1, strategy2)
+    compile_net(net)