[pass] add elementwise_add + matmul/matmul_v2 fusion in fc

lite/core/optimizer/mir/fusion/fc_fuse_pass.cc

@@ -23,21 +23,24 @@ namespace lite {
 namespace mir {
 
 void FcFusePass::Apply(const std::unique_ptr<SSAGraph>& graph) {
+  std::vector<std::string> mul_types{"mul", "matmul", "matmul_v2"};
+  for (auto op_type : mul_types) {
 #if defined(LITE_WITH_X86) || defined(LITE_WITH_CUDA)
 #ifdef LITE_WITH_MLU
-  fusion::FcFuser fuser(false);
-  fuser(graph.get());
+    fusion::FcFuser fuser(op_type, false);
+    fuser(graph.get());
 #else
-  fusion::FcFuser fuser(true);
-  fuser(graph.get());
+    fusion::FcFuser fuser(op_type, true);
+    fuser(graph.get());
 #endif
 #endif
-  fusion::FcFuser fuser2(false);
-  fuser2(graph.get());
+    fusion::FcFuser fuser2(op_type, false);
+    fuser2(graph.get());
 #ifdef LITE_WITH_FPGA
-  fusion::FcFuser fpga_fuser(true);
-  fpga_fuser(graph.get());
+    fusion::FcFuser fpga_fuser(op_type, true);
+    fpga_fuser(graph.get());
 #endif
+  }
 }
 
 }  // namespace mir

lite/core/optimizer/mir/fusion/fc_fuser.cc

@@ -41,16 +41,25 @@ void FcFuser::BuildPattern() {
     size_t b_rank = b_shape.size();
     return b_rank == 2 || b_rank == 1;
   };
+  auto input_attr_teller = [](const Node* node) -> bool {
+    auto op_desc = *const_cast<Node*>(node)->stmt()->op_info();
+    bool trans_x = op_desc.GetAttr<bool>("trans_x");
+    bool trans_y = op_desc.GetAttr<bool>("trans_y") return trans_x == false &&
+                   trans_y == false;
+  };
 
   // create nodes.
-  auto* x = VarNode("x")->assert_is_op_input("mul", "X");
-  auto* W = VarNode("W")->assert_is_op_input("mul", "Y");
+  auto* x = VarNode("x")->assert_is_op_input(op_type_, "X");
+  auto* W = VarNode("W")->assert_is_op_input(op_type_, "Y");
   auto* b = VarNode("b")->assert_is_persistable_var();
-  auto* mul = OpNode("mul", "mul")->assert_node_satisfied(inputs_teller0);
+  auto* mul = OpNode("mul", op_type_)->assert_node_satisfied(inputs_teller0);
   auto* mul_out = VarNode("mul_out");
   auto* add =
       OpNode("add", "elementwise_add")->assert_node_satisfied(inputs_teller1);
   auto* Out = VarNode("Out");
+  if (op_type_ == "matmul" || op_type_ == "matmul_v2") {
+    mul = OpNode("mul", op_type_)->assert_node_satisfied(input_attr_teller);
+  }
 
   // create topology.
   std::vector<PMNode*> mul_inputs{W, x};
@@ -129,9 +138,14 @@ cpp::OpDesc FcFuser::GenOpDesc(const key2nodes_t& matched) {
   op_desc.SetInput("W", {matched.at("W")->arg()->name});
   op_desc.SetInput("Bias", {matched.at("b")->arg()->name});
   op_desc.SetOutput("Out", {matched.at("Out")->arg()->name});
-  op_desc.SetAttr(
-      "in_num_col_dims",
-      matched.at("mul")->stmt()->op_info()->GetAttr<int>("x_num_col_dims"));
+  if (op_type_ == "mul") {
+    op_desc.SetAttr(
+        "in_num_col_dims",
+        matched.at("mul")->stmt()->op_info()->GetAttr<int>("x_num_col_dims"));
+  } else {
+    op_desc.SetAttr("in_num_col_dims", 1);
+  }
+
   if (with_relu_) {
     op_desc.SetAttr("activation_type", std::string{"relu"});
   }

lite/core/optimizer/mir/fusion/fc_fuser.h

@@ -25,12 +25,14 @@ namespace fusion {
 
 class FcFuser : public FuseBase {
  public:
-  explicit FcFuser(bool with_relu) : with_relu_(with_relu) {}
+  explicit FcFuser(std::string op_type, bool with_relu)
+      : op_type_(op_type), with_relu_(with_relu) {}
   void BuildPattern() override;
   void InsertNewNode(SSAGraph* graph, const key2nodes_t& matched) override;
 
  private:
   cpp::OpDesc GenOpDesc(const key2nodes_t& matched) override;
+  std::string op_type_;
   bool with_relu_;
 };
 

lite/tests/unittest_py/pass/test_fc_fuse_pass.py

@@ -64,6 +64,7 @@ def is_program_valid(self,
 
     def sample_program_configs(self, draw):
         has_relu = draw(st.sampled_from([True, False]))
+        op_type = draw(st.sampled_from(["mul", "matmul", "matmul_v2"]))
         mul_x_in_shape = draw(
             st.lists(
                 st.integers(
@@ -105,6 +106,24 @@ def sample_program_configs(self, draw):
                 "x_num_col_dims": x_num_col_dims_data,
                 "y_num_col_dims": 1
             })
+        inputs_data = {
+            "mul_x_data": TensorConfig(shape=mul_x_in_shape),
+            "mul_y_data": TensorConfig(shape=[x1, y1])
+        }
+        if op_type == "matmul" or op_type == "matmul_v2":
+            mul_op = OpConfig(
+                type=op_type,
+                inputs={"X": ["mul_x_data"],
+                        "Y": ["mul_y_data"]},
+                outputs={"Out": ["mul_output_data"]},
+                attrs={"trans_x": False,
+                       "trans_y": False})
+            inputs_data = {
+                "mul_x_data": TensorConfig(
+                    shape=[draw(st.integers(
+                        min_value=2, max_value=100)), x1]),
+                "mul_y_data": TensorConfig(shape=[x1, y1])
+            }
 
         elementwise_add_op = OpConfig(
             type="elementwise_add",