Inplace results (#1162)

* inplace results

* fix parameter propagation

* fix python tests

src/ngraph/runtime/cpu/cpu_external_function.cpp

@@ -659,7 +659,10 @@ using namespace ngraph::runtime;
             {
                 shared_ptr<descriptor::TensorView> itv =
                     res->get_inputs().at(0).get_output().get_tensor_view();
+
+                auto output_name = ss.str();
                 m_variable_name_map[itv->get_tensor().get_name()] = ss.str();
+                propagate_in_place_output(&(res->get_inputs().at(0).get_output()), output_name);
             }
         }
 
@@ -954,6 +957,43 @@ using namespace ngraph::runtime;
     }
 }
 
+void runtime::cpu::CPU_ExternalFunction::propagate_in_place_output(
+    ngraph::descriptor::Output* res_src_output, std::string output_name)
+{
+    //we start with a particular output
+    //which is an argument to a given op::Result
+    size_t offset = res_src_output->get_tensor().get_pool_offset();
+    auto it = res_src_output;
+
+    bool propagate_further = false;
+    do
+    {
+        propagate_further = false;
+        auto arg = std::dynamic_pointer_cast<ngraph::op::Op>(it->get_node());
+        if (!arg)
+        {
+            break;
+        }
+        if (auto op_annotations = arg->get_op_annotations())
+        {
+            auto oi_pairs = op_annotations->get_in_place_oi_pairs();
+            if (oi_pairs.count(it->get_index()) != 0)
+            {
+                size_t input_index = oi_pairs.at(it->get_index());
+                auto& input_tensor = arg->get_inputs().at(input_index).get_tensor();
+                if (input_tensor.get_pool_offset() == offset &&
+                    !arg->get_inputs().at(input_index).get_output().get_node()->is_parameter())
+                {
+                    NGRAPH_DEBUG << "Reusing " << output_name << " for " << input_tensor.get_name();
+                    m_variable_name_map[input_tensor.get_name()] = output_name;
+                    it = &arg->get_inputs().at(input_index).get_output();
+                    propagate_further = true;
+                }
+            }
+        }
+    } while (propagate_further);
+}
+
 void runtime::cpu::CPU_ExternalFunction::build()
 {
     if (m_is_built)

src/ngraph/runtime/cpu/cpu_external_function.hpp

@@ -111,6 +111,8 @@ namespace ngraph
                 void compile();
 
             private:
+                void propagate_in_place_output(ngraph::descriptor::Output* res_src_output,
+                                               std::string output_name);
                 void emit_debug_function_entry(codegen::CodeWriter& writer,
                                                Node* node,
                                                const std::vector<TensorViewWrapper>& in,