!532 [AutoParallel]support multiple subgraphs

Merge pull request !532 from lichen/support_multiple_subgraphs_in_auto_parallel

mindspore/ccsrc/parallel/allreduce_fusion/allreduce_fusion.cc

@@ -399,7 +399,12 @@ Status AllreduceFusion::ProcessAllreduceFusion(const CNodePtr &ret) {
   ret_ = ret;
   root_graph_ = ret_->func_graph();
   MS_EXCEPTION_IF_NULL(root_graph_);
-  auto forward_graph = ForwardGraph(root_graph_);
+  auto graph_set = ForwardGraph(root_graph_);
+  if (graph_set.size() > 1) {
+    MS_LOG(WARNING) << "AllReduce fusion don't support multiple subgraphs now.";
+    return SUCCESS;
+  }
+  auto forward_graph = *(graph_set.begin());
   MS_EXCEPTION_IF_NULL(forward_graph);
   forward_ret_ = forward_graph->get_return();
   MS_EXCEPTION_IF_NULL(forward_ret_);

mindspore/ccsrc/parallel/step_parallel.h

@@ -24,6 +24,7 @@
 #include <string>
 #include <unordered_map>
 #include <utility>
+#include <set>
 
 #include "./common.h"
 #include "optimizer/opt.h"
@@ -142,13 +143,13 @@ bool StepParallel(const FuncGraphPtr &func_graph, const opt::OptimizerPtr &optim
 
 int32_t GetTupleGetItemIndex(const CNodePtr &cnode);
 
-CNodePtr FindLossCNodeFromRoot(const FuncGraphPtr &root);
+std::vector<CNodePtr> FindLossCNodeFromRoot(const FuncGraphPtr &root);
 
 Status ParallelInit();
 
 std::vector<std::string> ExtractInputsTensorName(const CNodePtr &node);
 
-FuncGraphPtr ForwardGraph(const FuncGraphPtr &root);
+std::set<FuncGraphPtr> ForwardGraph(const FuncGraphPtr &root);
 }  // namespace parallel
 }  // namespace mindspore
 

tests/ut/python/parallel/test_semi_auto_two_subgraphs.py

@@ -0,0 +1,108 @@
+# Copyright 2020 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 mindspore as ms
+from mindspore import Tensor, Parameter, ParameterTuple, context
+from mindspore import nn
+from mindspore.common.api import _executor
+from mindspore.nn.optim import Adam, FTRL
+from mindspore.ops import operations as P
+from mindspore.ops import composite as C
+from mindspore.ops import functional as F
+import numpy as np
+
+
+class Net(nn.Cell):
+    def __init__(self):
+        super(Net, self).__init__()
+        self.mul = P.Mul()
+        self.relu = P.ReLU()
+        self.param1 = Parameter(Tensor(np.ones([8, 8, 8, 8]).astype(np.float32)), name="wide")
+        self.param2 = Parameter(Tensor(np.ones([8, 8, 8, 8]).astype(np.float32)), name="deep")
+
+    def construct(self, x):
+        out = self.mul(x, self.param1)
+        out = self.mul(out, self.param2)
+        out = self.relu(out)
+        return out
+
+
+class NetWithLoss(nn.Cell):
+    def __init__(self, network):
+        super(NetWithLoss, self).__init__()
+        self.sum = P.ReduceSum(keep_dims=False).set_strategy(strategy=((4, 1, 1, 1),))
+        self.mean = P.ReduceMean(keep_dims=False).set_strategy(strategy=((8, 1, 1, 1),))
+        self.net = network
+
+    def construct(self, x):
+        net_out = self.net(x)
+        loss1 = self.sum(net_out, -1)
+        loss2 = self.mean(net_out, -1)
+        return loss1, loss2
+
+
+class IthOutputCell(nn.Cell):
+    def __init__(self, network, output_index):
+        super(IthOutputCell, self).__init__()
+        self.network = network
+        self.output_index = output_index
+
+    def construct(self, x1):
+        predict = self.network(x1)[self.output_index]
+        return predict
+
+
+class TrainStepWrap(nn.Cell):
+    def __init__(self, network, sens=1000.0):
+        super(TrainStepWrap, self).__init__()
+        self.network = network
+        self.network.set_train()
+        self.trainable_params = network.trainable_params()
+        weights_w = []
+        weights_d = []
+        for params in self.trainable_params:
+            weights_w.append(params)
+            weights_d.append(params)
+        
+        self.weights_w = ParameterTuple(weights_w)
+        self.weights_d = ParameterTuple(weights_d)
+        self.optimizer_w = FTRL(learning_rate=1e-2, params=self.weights_w,
+                                l1=1e-8, l2=1e-8, initial_accum=1.0)
+        self.optimizer_d = Adam(self.weights_d, learning_rate=3.5e-4, eps=1e-8,
+                                loss_scale=sens)
+        self.hyper_map = C.HyperMap()
+        self.grad_w = C.GradOperation('grad_w', get_by_list=True,
+                                      sens_param=True)
+        self.grad_d = C.GradOperation('grad_d', get_by_list=True,
+                                      sens_param=True)
+        self.sens = sens
+        self.loss_net_w = IthOutputCell(network, output_index=0)
+        self.loss_net_d = IthOutputCell(network, output_index=1)
+
+    def construct(self, x):
+        weights_w = self.weights_w
+        weights_d = self.weights_d
+        loss_w, loss_d = self.network(x)
+        sens_w = P.Fill()(P.DType()(loss_w), P.Shape()(loss_w), self.sens)
+        sens_d = P.Fill()(P.DType()(loss_d), P.Shape()(loss_d), self.sens)
+        grads_w = self.grad_w(self.loss_net_w, weights_w)(x, sens_w)
+        grads_d = self.grad_d(self.loss_net_d, weights_d)(x, sens_d)
+        return F.depend(loss_w, self.optimizer_w(grads_w)), F.depend(loss_d, self.optimizer_d(grads_d))
+
+
+def test_two_subgraphs():
+    context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode="semi_auto_parallel")
+    net = TrainStepWrap(NetWithLoss(Net()))
+    input_x = Tensor(np.ones([8, 8, 8, 8]), dtype=ms.float32)
+    _executor.compile(net, input_x)