!1219 Add an new output to FusedMulApplyMomentum

Merge pull request !1219 from YuJianfeng/master

mindspore/ccsrc/pre_activate/ascend/ir_fusion/momentum_lossscale_fusion.cc

@@ -23,6 +23,7 @@
 namespace mindspore {
 namespace opt {
 namespace {
+constexpr size_t kAccumIndex = 1;
 bool CheckValueNodeInputOfMul(const AnfNodePtr &node) {
   MS_EXCEPTION_IF_NULL(node);
   if (!node->isa<ValueNode>()) {
@@ -79,9 +80,19 @@ const AnfNodePtr MomentumLossscaleFusion::Process(const FuncGraphPtr &func_graph
   input_names_value[3] = "x1";
   input_names_value.emplace_back("x2");
   AnfAlgo::SetNodeAttr(kAttrInputNames, MakeValue(input_names_value), new_node);
-  new_node->set_abstract(node->abstract());
+  auto node_to_output = cnode->input(kAccumIndex + 1);
+  MS_EXCEPTION_IF_NULL(node_to_output);
+  AbstractBasePtrList abstract_list{node->abstract(), node_to_output->abstract()};
+  auto abstract_tuple = std::make_shared<abstract::AbstractTuple>(abstract_list);
+  new_node->set_abstract(abstract_tuple);
   new_node->set_scope(node->scope());
-  return new_node;
+  // Create Output
+  std::vector<AnfNodePtr> new_outputs;
+  CreateMultipleOutputsOfAnfNode(func_graph, new_node, kFusedMulApplyMomentumOutputNum, &new_outputs);
+  if (new_outputs.size() != kFusedMulApplyMomentumOutputNum) {
+    MS_LOG(EXCEPTION) << "Failed to create outputs of " << new_node->DebugString();
+  }
+  return new_outputs[0];
 }
 }  // namespace opt
 }  // namespace mindspore

mindspore/ccsrc/pre_activate/common/helper.h

@@ -92,6 +92,7 @@ constexpr size_t kApplyMomentumInputNum = 6;
 constexpr size_t kBiasAddInputNum = 3;
 constexpr size_t kTopkInputNum = 3;
 constexpr size_t kLarsV2InputNum = 5;
+constexpr size_t kFusedMulApplyMomentumOutputNum = 2;
 
 enum FusedBatchNormInput {
   kX = 1,

tests/ut/cpp/python_input/gtest_input/pre_activate/momentum_lossscale_fusion_test.py

@@ -47,6 +47,6 @@ def test_momentum_lossscale_fusion(tag):
 
     @fns
     def after(input0, input1, input2, input3, input4):
-        return make_tuple(FusedMulApplyMomentum(input0, input1, input2, input3, input4, constant))
+        return make_tuple(tuple_getitem(FusedMulApplyMomentum(input0, input1, input2, input3, input4, constant), 0))
 
     return fns[tag]