incubate/lite: batch norm pattern detect (#18021)

* bn pattern

* 1. add conv bn fusion pass(conv2d, depthwise_conv2d)
2. modify the graph to ssagraph
test=develop

paddle/fluid/lite/core/mir/CMakeLists.txt

@@ -3,10 +3,12 @@ cc_library(mir_ssa_graph SRCS ssa_graph.cc DEPS mir_node)
 cc_library(mir_pass SRCS pass.cc DEPS mir_ssa_graph)
 cc_library(mir_pass_manager SRCS pass_manager.cc DEPS mir_pass mir_ssa_graph mir_passes)
 cc_library(mir_pass_registry SRCS pass_registry.cc DEPS mir_pass_manager)
+
 add_subdirectory(fusion)
 cc_library(mir_passes
         SRCS fc_fuse_pass.cc
         conv_elementwise_add_relu_fuse_pass.cc
+        conv_bn_fuse_pass.cc 
         static_kernel_pick_pass.cc
         variable_place_inference_pass.cc
         type_target_transform_pass.cc
@@ -55,6 +57,7 @@ lite_cc_test(test_pattern_matcher_lite SRCS pattern_matcher_test.cc DEPS pattern
 
 lite_cc_library(pattern_matcher_high_api SRCS pattern_matcher_high_api.cc DEPS pattern_matcher_lite)
 
+
 # TODO(wz) replace framework/proto to lite proto.
 if (NOT LITE_WITH_LIGHT_WEIGHT_FRAMEWORK)
     # it depends on the fluid/framework/proto, that is too heavy for mobile execution.

paddle/fluid/lite/core/mir/conv_bn_fuse_pass.cc

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#include "paddle/fluid/lite/core/mir/conv_bn_fuse_pass.h"
+#include <memory>
+#include <vector>
+#include "paddle/fluid/lite/core/mir/fusion/conv_bn_fuser.h"
+#include "paddle/fluid/lite/core/mir/pass_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace mir {
+
+void ConvBNFusePass::Apply(const std::unique_ptr<SSAGraph>& graph) {
+  fusion::ConvBNFuser fuser("conv2d");
+  fuser(graph.get());
+
+  fusion::ConvBNFuser fuser2("depthwise_conv2d");
+  fuser2(graph.get());
+}
+
+}  // namespace mir
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_MIR_PASS(lite_conv_bn_fuse_pass, paddle::lite::mir::ConvBNFusePass);

paddle/fluid/lite/core/mir/conv_bn_fuse_pass.h

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#pragma once
+
+#include <memory>
+#include <string>
+#include "paddle/fluid/lite/core/mir/pass.h"
+
+namespace paddle {
+namespace lite {
+namespace mir {
+
+class ConvBNFusePass : public ProgramPass {
+ public:
+  void Apply(const std::unique_ptr<SSAGraph>& graph) override;
+};
+
+}  // namespace mir
+}  // namespace lite
+}  // namespace paddle

paddle/fluid/lite/core/mir/conv_elementwise_add_relu_fuse_pass_test.cc

@@ -40,8 +40,8 @@ std::unique_ptr<SSAGraph> BuildGraph(framework::ProgramDesc* program_desc,
   auto* conv2d_1 = main_block->AppendOp();
   auto* conv2d_2 = main_block->AppendOp();
   auto* add_1 = main_block->AppendOp();
-  auto* add_2 = main_block->AppendOp();
   auto* relu_1 = main_block->AppendOp();
+  auto* add_2 = main_block->AppendOp();
   auto* relu_2 = main_block->AppendOp();
 
   main_block->Var("input_1");
@@ -123,8 +123,8 @@ TEST(conv_elementwise_add_relu_fuse_pass, graph_test) {
   auto scope = std::make_shared<Scope>();
   auto graph = BuildGraph(&program_desc, scope, places);
 
-  ASSERT_EQ(graph->nodes().size(),
-            11UL /*vars*/ + 6UL /*ops*/ + 2UL /*feed op + fetch op*/);
+  Visualize(graph.get());
+  ASSERT_EQ(graph->nodes().size(), 11UL /*vars*/ + 6UL /*ops*/);
   Visualize(graph.get());
 }
 

paddle/fluid/lite/core/mir/fusion/CMakeLists.txt

 cc_library(fuse_fc
         SRCS fc_fuser.cc
         DEPS pattern_matcher_high_api)
-cc_library(conv_elementwise_add_relu
+cc_library(fuse_conv_elementwise_add_relu
         SRCS conv_elementwise_add_relu_fuser.cc
         DEPS pattern_matcher_high_api)
+cc_library(fuse_conv_bn
+        SRCS conv_bn_fuser.cc
+        DEPS pattern_matcher_high_api)
 
 set(mir_fusers 
     fuse_fc 
-    conv_elementwise_add_relu
+    fuse_conv_elementwise_add_relu
+    fuse_conv_bn
     CACHE INTERNAL "fusers")
+
+lite_cc_test(test_lite_conv_bn_fuse SRCS conv_bn_fuse_pass_test.cc
+   DEPS elementwise_ops_lite batch_norm_op_lite conv_op_lite proto_desc compatible_pb_lite program_lite mir_pass mir_pass_manager pattern_matcher_high_api)

paddle/fluid/lite/core/mir/fusion/conv_bn_fuse_pass_test.cc

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#include "paddle/fluid/lite/core/mir/conv_bn_fuse_pass.h"
+#include <gflags/gflags.h>
+#include <gtest/gtest.h>
+#include <vector>
+#include "paddle/fluid/framework/program_desc.h"
+#include "paddle/fluid/lite/core/compatible_tensor.h"
+#include "paddle/fluid/lite/core/mir/graph_visualize_pass.h"
+#include "paddle/fluid/lite/core/program.h"
+
+namespace paddle {
+namespace lite {
+namespace mir {
+namespace fusion {
+
+std::unique_ptr<SSAGraph> BuildGraph(framework::ProgramDesc* program_desc,
+                                     const std::shared_ptr<Scope>& scope,
+                                     const std::vector<Place>& valid_places) {
+  auto* main_block = program_desc->MutableBlock(0);
+  auto* conv_op = main_block->AppendOp();
+  auto* bn_op = main_block->AppendOp();
+  main_block->Var("conv_i");
+  main_block->Var("conv_param");
+  main_block->Var("conv_out");
+
+  main_block->Var("bn_scale");
+  main_block->Var("bn_bias");
+  main_block->Var("bn_mean");
+  main_block->Var("bn_var");
+  main_block->Var("bn_out");
+  main_block->Var("bn_mean_out");
+  main_block->Var("bn_var_out");
+  main_block->Var("bn_saved_mean");
+  main_block->Var("bn_saved_var");
+
+  scope->Var("conv_i")->GetMutable<lite::Tensor>();
+  auto conv_param_t = scope->Var("conv_param")->GetMutable<lite::Tensor>();
+  std::vector<int64_t> conv_param_shape = {3, 1, 2, 2};
+  conv_param_t->Resize(lite::DDim(conv_param_shape));
+  conv_param_t->mutable_data<float>();
+  scope->Var("conv_out")->GetMutable<lite::Tensor>();
+  auto bn_scale_t = scope->Var("bn_scale")->GetMutable<lite::Tensor>();
+  std::vector<int64_t> bn_scale_shape = {3};
+  bn_scale_t->Resize(lite::DDim(bn_scale_shape));
+  bn_scale_t->mutable_data<float>();
+
+  auto bn_bias_t = scope->Var("bn_bias")->GetMutable<lite::Tensor>();
+  std::vector<int64_t> bn_bias_shape = {3};
+  bn_bias_t->Resize(lite::DDim(bn_bias_shape));
+  bn_bias_t->mutable_data<float>();
+
+  auto bn_mean_t = scope->Var("bn_mean")->GetMutable<lite::Tensor>();
+  bn_mean_t->Resize(lite::DDim(bn_bias_shape));
+  bn_mean_t->mutable_data<float>();
+
+  auto bn_var_t = scope->Var("bn_var")->GetMutable<lite::Tensor>();
+  bn_var_t->Resize(lite::DDim(bn_bias_shape));
+  bn_var_t->mutable_data<float>();
+
+  scope->Var("bn_out")->GetMutable<lite::Tensor>();
+  scope->Var("bn_mean_out")->GetMutable<lite::Tensor>();
+  scope->Var("bn_var_out")->GetMutable<lite::Tensor>();
+  scope->Var("bn_saved_mean")->GetMutable<lite::Tensor>();
+  scope->Var("bn_saved_var")->GetMutable<lite::Tensor>();
+
+  conv_op->SetType("conv2d");
+  conv_op->SetInput("Input", {"conv_i"});
+  conv_op->SetInput("Filter", {"conv_param"});
+  conv_op->SetOutput("Output", {"conv_out"});
+  const std::vector<int> strides({1, 1});
+  const std::vector<int> paddings({1, 1});
+  const std::vector<int> dilations({1, 1});
+  const int groups = 1;
+  conv_op->SetAttr("strides", strides);
+  conv_op->SetAttr("paddings", paddings);
+  conv_op->SetAttr("dilations", dilations);
+  conv_op->SetAttr("groups", groups);
+
+  bn_op->SetType("batch_norm");
+  bn_op->SetInput("X", {"conv_out"});
+  bn_op->SetInput("Bias", {"bn_bias"});
+  bn_op->SetInput("Mean", {"bn_mean"});
+  bn_op->SetInput("Scale", {"bn_scale"});
+  bn_op->SetInput("Variance", {"bn_var"});
+
+  bn_op->SetOutput("Y", {"bn_out"});
+  bn_op->SetOutput("MeanOut", {"bn_mean_out"});
+  bn_op->SetOutput("VarianceOut", {"bn_var_out"});
+  bn_op->SetOutput("SavedMean", {"bn_saved_mean"});
+  bn_op->SetOutput("SavedVariance", {"bn_saved_var"});
+  float eps = 1e-5;
+  bn_op->SetAttr("epsilon", eps);
+
+  program_desc->Flush();
+
+  lite::Program program(*program_desc->Proto(), scope, valid_places);
+  auto graph = std::unique_ptr<SSAGraph>(new SSAGraph());
+  graph->Build(program, valid_places);
+
+  return graph;
+}
+
+TEST(pattern_matcher2, test) {
+  framework::ProgramDesc program_desc;
+  std::vector<Place> places{{TARGET(kHost), PRECISION(kFloat)}};
+  auto scope = std::make_shared<Scope>();
+  auto graph = BuildGraph(&program_desc, scope, places);
+  const int num_nodes = graph->nodes().size();
+  auto* fuser = new ConvBNFusePass;
+  fuser->Apply(graph);
+  ASSERT_EQ(graph->nodes().size(),
+            num_nodes - 8UL /*nodes removed */ + 1UL /* eltwise_add node*/);
+}
+
+}  // namespace fusion
+}  // namespace mir
+}  // namespace lite
+}  // namespace paddle
+
+USE_LITE_OP(conv2d);
+USE_LITE_OP(batch_norm);
+USE_LITE_OP(elementwise_add);

paddle/fluid/lite/core/mir/fusion/conv_bn_fuser.cc

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#include "paddle/fluid/lite/core/mir/fusion/conv_bn_fuser.h"
+#include <memory>
+#include <vector>
+
+namespace paddle {
+namespace lite {
+namespace mir {
+namespace fusion {
+
+void ConvBNFuser::BuildPattern() {
+  auto* conv_input =
+      VarNode("conv_input")->assert_is_op_input(conv_type_, "Input")->AsInput();
+  auto* conv_weight = VarNode("conv_weight")
+                          ->assert_is_op_input(conv_type_, "Filter")
+                          ->AsInput();
+  auto* conv = OpNode("conv2d", conv_type_)->assert_is_op(conv_type_);
+  auto* conv_out = VarNode("conv_out")
+                       ->assert_is_op_output(conv_type_, "Output")
+                       ->assert_is_op_input("batch_norm", "X");
+
+  auto* bn_scale = VarNode("bn_scale")
+                       ->assert_is_op_input("batch_norm", "Scale")
+                       ->AsIntermediate();
+  auto* bn_bias =
+      VarNode("bn_bias")->assert_is_op_input("batch_norm", "Bias")->AsInput();
+  auto* bn_mean = VarNode("bn_mean")
+                      ->assert_is_op_input("batch_norm", "Mean")
+                      ->AsIntermediate();
+  auto* bn_var = VarNode("bn_variance")
+                     ->assert_is_op_input("batch_norm", "Variance")
+                     ->AsIntermediate();
+  auto* bn =
+      OpNode("bn", "batch_norm")->assert_is_op("batch_norm")->AsIntermediate();
+
+  auto* bn_out =
+      VarNode("bn_out")->assert_is_op_output("batch_norm", "Y")->AsOutput();
+  auto* bn_mean_out = VarNode("bn_mean_out")
+                          ->assert_is_op_output("batch_norm", "MeanOut")
+                          ->AsIntermediate();
+  auto* bn_var_out = VarNode("bn_var_out")
+                         ->assert_is_op_output("batch_norm", "VarianceOut")
+                         ->AsIntermediate();
+  auto* bn_saved_mean = VarNode("bn_saved_mean")
+                            ->assert_is_op_output("batch_norm", "SavedMean")
+                            ->AsIntermediate();
+  auto* bn_saved_var = VarNode("bn_saved_var")
+                           ->assert_is_op_output("batch_norm", "SavedVariance")
+                           ->AsIntermediate();
+
+  conv->LinksFrom({conv_input, conv_weight}).LinksTo({conv_out});
+
+  bn->LinksFrom({conv_out, bn_scale, bn_bias, bn_mean, bn_var})
+      .LinksTo({bn_out, bn_mean_out, bn_saved_mean, bn_saved_var, bn_var_out});
+}
+
+void ConvBNFuser::InsertNewNode(SSAGraph* graph, const key2nodes_t& matched) {
+  auto op_desc = GenOpDesc(matched);
+  auto eltwise_op = LiteOpRegistry::Global().Create("elementwise_add");
+  auto conv = matched.at("conv2d")->stmt()->op;
+  auto* scope = conv->scope();
+  auto& valid_places = conv->valid_places();
+
+  auto conv_weight_t = scope->FindVar(matched.at("conv_weight")->arg()->name)
+                           ->GetMutable<lite::Tensor>();
+  auto conv_weight_d = conv_weight_t->mutable_data<float>();
+  auto conv_weight_dims = conv_weight_t->dims();
+  size_t weight_num = conv_weight_t->data_size();
+
+  auto bn_scale_t = scope->FindVar(matched.at("bn_scale")->arg()->name)
+                        ->GetMutable<lite::Tensor>();
+  size_t bias_size = bn_scale_t->data_size();
+  auto bn_scale_d = bn_scale_t->mutable_data<float>();
+  PADDLE_ENFORCE(bias_size == conv_weight_dims[0],
+                 "The BN bias's size should be equal to the size of the first "
+                 "dim size of the conv weights");
+
+  auto bn_mean_t = scope->FindVar(matched.at("bn_mean")->arg()->name)
+                       ->GetMutable<lite::Tensor>();
+  auto bn_mean_d = bn_mean_t->mutable_data<float>();
+
+  auto bn_var_t = scope->FindVar(matched.at("bn_variance")->arg()->name)
+                      ->GetMutable<lite::Tensor>();
+  auto bn_var_d = bn_var_t->mutable_data<float>();
+
+  auto bn_bias_t = scope->FindVar(matched.at("bn_bias")->arg()->name)
+                       ->GetMutable<lite::Tensor>();
+  auto bn_bias_d = bn_bias_t->mutable_data<float>();
+  auto eps = matched.at("bn")->stmt()->op_info()->GetAttr<float>("epsilon");
+
+  ComputeFusedWeight(bn_scale_d, bn_mean_d, bn_var_d, bn_bias_d, conv_weight_d,
+                     eps, bias_size, weight_num / bias_size);
+
+  eltwise_op->Attach(op_desc, scope);
+  auto* new_op_node = graph->GraphCreateInstructNode(eltwise_op, valid_places);
+
+  IR_NODE_LINK_TO(matched.at("conv_out"), new_op_node);
+  IR_NODE_LINK_TO(matched.at("bn_bias"), new_op_node);
+  IR_NODE_LINK_TO(new_op_node, matched.at("bn_out"));
+}
+
+cpp::OpDesc ConvBNFuser::GenOpDesc(const key2nodes_t& matched) {
+  cpp::OpDesc op_desc;
+  op_desc.SetType("elementwise_add");
+  op_desc.SetInput("X", {matched.at("conv_out")->arg()->name});
+  op_desc.SetInput("Y", {matched.at("bn_bias")->arg()->name});
+  op_desc.SetOutput("Out", {matched.at("bn_out")->arg()->name});
+  op_desc.SetAttr("axis", 1);
+  return op_desc;
+}
+
+}  // namespace fusion
+}  // namespace mir
+}  // namespace lite
+}  // namespace paddle

paddle/fluid/lite/core/mir/fusion/conv_bn_fuser.h

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#pragma once
+
+#include <memory>
+#include <string>
+#include "paddle/fluid/lite/core/mir/pattern_matcher_high_api.h"
+
+namespace paddle {
+namespace lite {
+namespace mir {
+namespace fusion {
+
+class ConvBNFuser : public FuseBase {
+ public:
+  explicit ConvBNFuser(const std::string& conv_type) : conv_type_(conv_type) {}
+  void BuildPattern() override;
+  void InsertNewNode(SSAGraph* graph, const key2nodes_t& matched) override;
+
+ private:
+  cpp::OpDesc GenOpDesc(const key2nodes_t& matched) override;
+  void ComputeFusedWeight(float* scale_d, float* mean_d, float* var_d,
+                          float* bias_d, float* conv_weight_d, float eps, int h,
+                          int w) {
+    for (int i = 0; i < h; i++) {
+      var_d[i] = scale_d[i] / std::sqrt(var_d[i] + eps);
+    }
+    for (int i = 0; i < h; i++) {
+      bias_d[i] += (-mean_d[i]) * var_d[i];
+    }
+    for (int i = 0; i < h; i++) {
+      for (int j = 0; j < w; j++) {
+        conv_weight_d[i * w + j] *= var_d[i];
+      }
+    }
+  }
+
+ private:
+  std::string conv_type_{"conv2d"};
+};
+
+}  // namespace fusion
+}  // namespace mir
+}  // namespace lite
+}  // namespace paddle

paddle/fluid/lite/core/mir/node.h

@@ -71,12 +71,20 @@ class Node {
 
   struct Arg {
     std::string name;
+    int id{0};
     const Type* type{};
     // Weight is a special kind of argument, it is marked as weight explicitly
     // so that some weight related optimization can take place.
     bool is_weight{false};
   };
 
+  Arg& AsArg(const std::string& name, int id) {
+    auto& x = AsArg();
+    x.name = name;
+    x.id = id;
+    return x;
+  }
+
   Arg& AsArg(const std::string& name) {
     auto& x = AsArg();
     x.name = name;

paddle/fluid/lite/core/mir/passes.h

@@ -31,3 +31,4 @@ USE_MIR_PASS(generate_program_pass);
 USE_MIR_PASS(io_copy_kernel_pick_pass);
 USE_MIR_PASS(argument_type_display_pass);
 USE_MIR_PASS(runtime_context_assign_pass);
+USE_MIR_PASS(lite_conv_bn_fuse_pass);

paddle/fluid/lite/core/mir/pattern_matcher.cc

@@ -407,37 +407,60 @@ PMNode *PMNode::assert_is_op_output(const std::string &op_type) {
   return this;
 }
 
-PMNode *PMNode::assert_is_op_input(const std::string &op_type) {
+bool IsNthOutput(const Node *var, const Node *op, const std::string &argument,
+                 size_t nth) {
+  PADDLE_ENFORCE(var->IsArg());
+  PADDLE_ENFORCE(op->IsStmt());
+  auto op_info = op->stmt()->op_info();
+  if (op_info->Output(argument).size() <= nth) return false;
+  return var->arg()->name == op_info->Output(argument)[nth];
+}
+
+bool IsNthInput(const Node *var, const Node *op, const std::string &argument,
+                size_t nth) {
+  PADDLE_ENFORCE(var->IsArg());
+  PADDLE_ENFORCE(op->IsStmt());
+  auto op_info = op->stmt()->op_info();
+  if (op_info->Input(argument).size() <= nth) return false;
+  return var->arg()->name == op_info->Input(argument)[nth];
+}
+
+PMNode *PMNode::assert_is_op_input(const std::string &op_type,
+                                   const std::string &argument) {
+  assert_is_var();
+  assert_is_op_nth_input(op_type, argument, 0);
+  return this;
+}
+
+PMNode *PMNode::assert_is_op_nth_input(const std::string &op_type,
+                                       const std::string &argument, int nth) {
   assert_is_var();
+  assert_is_op_input(op_type);
   asserts_.emplace_back([=](const Node *x) {
     for (auto *op : x->outlinks) {
-      if (op && op->IsStmt()) {
-        auto *op_info = op->stmt()->op_info();
-        if (op_info->Type() == op_type) {
+      if (op && op->IsStmt() && op->stmt()->op_info()->Type() == op_type &&
+          IsNthInput(x, op, argument, nth))
         return true;
     }
-      }
-    }
     return false;
   });
   return this;
 }
 
-PMNode *PMNode::assert_is_op_input(const std::string &op_type,
+PMNode *PMNode::assert_is_op_output(const std::string &op_type,
                                     const std::string &argument) {
   assert_is_var();
-  assert_is_op_nth_input(op_type, argument, 0);
+  assert_is_op_nth_output(op_type, argument, 0);
   return this;
 }
 
-PMNode *PMNode::assert_is_op_nth_input(const std::string &op_type,
+PMNode *PMNode::assert_is_op_nth_output(const std::string &op_type,
                                         const std::string &argument, int nth) {
   assert_is_var();
-  assert_is_op_input(op_type);
   asserts_.emplace_back([=](const Node *x) {
-    for (auto *op : x->outlinks) {
-      if (op->IsStmt() && op->stmt()->op_info()->Type() == op_type &&
-          IsNthInput(*x, *op, argument, nth))
+    for (auto *op : x->inlinks) {
+      if (op && op->IsStmt() && op->stmt()->op_info()->Type() == op_type &&
+          IsNthOutput(x, op, argument, nth))
         return true;
     }
     return false;
@@ -445,14 +468,20 @@ PMNode *PMNode::assert_is_op_nth_input(const std::string &op_type,
   return this;
 }
 
-bool IsNthInput(const Node &var, const Node &op, const std::string &argument,
-                int nth) {
-  CHECK(var.IsArg());
-  CHECK(op.IsStmt());
-  if (!HasInput(op, argument) ||
-      static_cast<int>(op.stmt()->op_info()->Input(argument).size()) <= nth)
+PMNode *PMNode::assert_is_op_input(const std::string &op_type) {
+  assert_is_var();
+  asserts_.emplace_back([=](const Node *x) {
+    for (auto *op : x->outlinks) {
+      if (op && op->IsStmt()) {
+        auto *op_info = op->stmt()->op_info();
+        if (op_info->Type() == op_type) {
+          return true;
+        }
+      }
+    }
     return false;
-  return var.arg()->name == op.stmt()->op_info()->Input(argument)[nth];
+  });
+  return this;
 }
 
 bool HasInput(const Node &op, const std::string &argument) {

paddle/fluid/lite/core/mir/pattern_matcher.h

@@ -129,8 +129,13 @@ struct PMNode {
   PMNode* assert_is_op_input(const std::string& op_type);
   PMNode* assert_is_op_input(const std::string& op_type,
                              const std::string& argument);
+  PMNode* assert_is_op_output(const std::string& op_type,
+                              const std::string& argument);
+
   PMNode* assert_is_op_nth_input(const std::string& op_type,
                                  const std::string& argument, int nth);
+  PMNode* assert_is_op_nth_output(const std::string& op_type,
+                                  const std::string& argument, int nth);
 
   template <typename T>
   PMNode* assert_op_attr(const std::string& attr_name, const T& attr) {

paddle/fluid/lite/core/mir/pattern_matcher_high_api_test.cc

@@ -124,8 +124,7 @@ TEST(pattern_matcher_high_api, graph_test) {
   auto scope = std::make_shared<Scope>();
   auto graph = BuildGraph(&program_desc, scope, places);
 
-  ASSERT_EQ(graph->nodes().size(),
-            7UL /*real nodes*/ + 2UL /*feed op + fetch op*/);
+  ASSERT_EQ(graph->nodes().size(), 7UL /*real nodes*/);
   Visualize(graph.get());
 }
 

paddle/fluid/lite/core/mir/ssa_graph.cc

@@ -16,6 +16,7 @@
 #include <algorithm>
 #include <memory>
 #include <set>
+#include <unordered_map>
 #include <utility>
 
 namespace paddle {
@@ -25,6 +26,8 @@ namespace mir {
 bool SSAGraph::CheckBidirectionalConnection() {
   LOG(INFO) << "node count " << node_storage_.size();
   for (auto &node : node_storage_) {
+    if (node.IsStmt()) LOG(INFO) << node.AsStmt().op_info()->Type();
+    if (node.IsArg()) LOG(INFO) << node.AsArg().name << " " << node.AsArg().id;
     for (auto *in : node.inlinks) {
       CHECK(in->outlinks.end() !=
             std::find(in->outlinks.begin(), in->outlinks.end(), &node));
@@ -93,31 +96,6 @@ std::vector<mir::Node *> SSAGraph::StmtTopologicalOrder() {
   return res;
 }
 
-void SSAGraph::GraphCreateTmpVarNodes(const Program &program) {
-  for (const auto &name : program.tmp_vars()) {
-    CHECK(!arguments_.count(name)) << "duplicate creating temp variable: "
-                                   << name;
-    VLOG(5) << "create arg node " << name;
-    node_storage_.emplace_back();
-    auto &new_node = node_storage_.back();
-    new_node.AsArg(name);
-    arguments_[name] = &new_node;
-  }
-}
-
-void SSAGraph::GraphCreateWeightVarNodes(const Program &program) {
-  // create weight nodes.
-  for (const auto &name : program.weights()) {
-    CHECK(!arguments_.count(name)) << "duplicate creating weight variable: "
-                                   << name;
-    VLOG(5) << "create arg node " << name;
-    node_storage_.emplace_back();
-    auto &new_node = node_storage_.back();
-    new_node.AsArg(name);
-    arguments_[name] = &new_node;
-  }
-}
-
 Node *SSAGraph::GraphCreateInstructNode(
     const std::shared_ptr<OpLite> &op, const std::vector<Place> &valid_places) {
   node_storage_.emplace_back();
@@ -135,29 +113,50 @@ Node *SSAGraph::GraphCreateInstructNode(
 void SSAGraph::Build(const Program &program,
                      const std::vector<Place> &valid_places) {
   CHECK(node_storage_.empty());
-  GraphCreateTmpVarNodes(program);
-  GraphCreateWeightVarNodes(program);
-  CHECK(CheckNodesRoleSet());
 
+  auto weights_name = program.weights();
+  auto is_weights = [&](const std::string &name) -> bool {
+    auto it = std::find(weights_name.begin(), weights_name.end(), name);
+    if (it == weights_name.end()) return false;
+    return true;
+  };
+
+  std::unordered_map<std::string, mir::Node *> arg_update_node_map_;
   for (auto &op : program.ops()) {
+    LOG(INFO) << op->op_info()->Type();
     auto *op_node = GraphCreateInstructNode(op, valid_places);
+    LOG(INFO) << "input:";
     for (const std::string &name : op->op_info()->input_names()) {
-      auto *arg = Argument(name);
-      CHECK(arg->IsRoleSet());
-      DirectedLink(arg, op_node);
+      LOG(INFO) << name;
+      mir::Node *arg_node = nullptr;
+      if (arg_update_node_map_.count(name)) {
+        arg_node = arg_update_node_map_.at(name);
+      } else {
+        node_storage_.emplace_back();
+        arg_node = &node_storage_.back();
+        arg_node->AsArg(name, node_storage_.size() - 1);
+        arg_update_node_map_[name] = arg_node;
       }
-    for (const std::string &name : op->op_info()->output_names()) {
-      if (!arguments_.count(name)) {
-        NewArgumentNode(name);
+      if (is_weights(name)) arg_node->AsArg().is_weight = true;
+      CHECK(arg_node->IsRoleSet());
+      DirectedLink(arg_node, op_node);
     }
-      auto *arg = arguments_.at(name);
-      CHECK(arg->IsRoleSet());
-      DirectedLink(op_node, arg);
+    LOG(INFO) << "output:";
+    for (const std::string &name : op->op_info()->output_names()) {
+      LOG(INFO) << name;
+      node_storage_.emplace_back();
+      auto *arg_node = &node_storage_.back();
+      arg_node->AsArg(name, node_storage_.size() - 1);
+      arg_update_node_map_[name] = arg_node;
+
+      if (is_weights(name)) arg_node->AsArg().is_weight = true;
+      CHECK(arg_node->IsRoleSet());
+      DirectedLink(op_node, arg_node);
     }
     CHECK(CheckLinksRoleSet());
   }
 
-  MarkArgumentWeights(program);
+  CHECK(CheckNodesRoleSet());
   CheckValid();
 }
 
@@ -227,10 +226,9 @@ bool SSAGraph::CheckLinksRoleSet() {
 
 Node *SSAGraph::NewArgumentNode(const std::string &name) {
   node_storage_.emplace_back();
-  CHECK(!arguments_.count(name)) << "duplicate argument called " << name;
-  arguments_[name] = &node_storage_.back();
-  node_storage_.back().AsArg(name);
-  return &node_storage_.back();
+  auto &arg_node = node_storage_.back();
+  arg_node.AsArg(name, node_storage_.size() - 1);
+  return &arg_node;
 }
 
 Node *SSAGraph::NewInstructNode() {

paddle/fluid/lite/core/mir/ssa_graph.h

@@ -40,8 +40,6 @@ class SSAGraph : GraphBase {
   void Build(const Program &program, const std::vector<Place> &valid_places);
   void RemoveNode(const mir::Node *node);
 
-  mir::Node *Argument(const std::string &name);
-
   std::vector<mir::Node *> StmtTopologicalOrder();
 
   // The inputs of the graph.
@@ -68,9 +66,7 @@ class SSAGraph : GraphBase {
                                 const std::vector<Place> &valid_places);
 
  private:
-  void GraphCreateTmpVarNodes(const Program &program);
-  void GraphCreateWeightVarNodes(const Program &program);
-
+  mir::Node *Argument(const std::string &name);
   // Check the bidirectional connection.
   bool CheckBidirectionalConnection();
   bool CheckNodesRoleSet();

paddle/fluid/lite/core/mir/type_target_transform_pass.cc

@@ -65,20 +65,22 @@ void TypeTargetTransformPass::ComplementInputs(SSAGraph* graph, Node* inst_node,
               << " for kernel " << inst.op->DebugString() << " "
               << *in->AsArg().type << " -> " << *decl_arg_type;
     // Add an IoCopy instruction to make the input compatible with other dist.
-    AddIoCopyInst(*in->AsArg().type, *decl_arg_type, in->AsArg().name, graph,
-                  inst_node, valid_places_);
+    AddIoCopyInst(*in->AsArg().type, *decl_arg_type, in, graph, inst_node,
+                  valid_places_);
   }
 }
 
 void TypeTargetTransformPass::AddIoCopyInst(
-    const Type& from, const Type& to, const std::string& var, SSAGraph* graph,
+    const Type& from, const Type& to, Node* in, SSAGraph* graph,
     Node* inst_node, const std::vector<Place>& valid_places) {
   CHECK(!valid_places.empty()) << "valid_place should be set";
   // var -> new_transform_op -> new_var -> inst
   // So there will be a new Argument node and a new IoCopy Statement Node.
 
+  CHECK(in->IsArg());
   auto node_id = [&] { return graph->nodes().size(); };
-  auto io_copy_output_name = var + "/trans/" + std::to_string(node_id());
+  auto io_copy_output_name =
+      in->AsArg().name + "/trans/" + std::to_string(node_id());
   auto* io_copy_output_arg = graph->NewArgumentNode(io_copy_output_name);
   auto* io_copy_inst = graph->NewInstructNode();
 
@@ -92,7 +94,7 @@ void TypeTargetTransformPass::AddIoCopyInst(
   // Create IoCopy Instruction.
   cpp::OpDesc op_desc;
   op_desc.SetType("io_copy");
-  op_desc.SetInput("Input", {var});
+  op_desc.SetInput("Input", {in->AsArg().name});
   op_desc.SetOutput("Out", {io_copy_output_name});
 
   io_copy_op->Attach(op_desc, inst_node->AsStmt().op->scope());
@@ -100,18 +102,18 @@ void TypeTargetTransformPass::AddIoCopyInst(
   io_copy_inst->AsStmt("io_copy", std::move(kernels), io_copy_op);
 
   // Remove the old link
-  RemoveDirectedLink(graph->Argument(var), inst_node);
+  RemoveDirectedLink(in, inst_node);
 
   // Update the original instruction OpDesc.
   // Update its input to the io_copy_output_name
 
   // Add new link, var -> new_inst, new_inst->newarg, newarg->inst
-  DirectedLink(graph->Argument(var), io_copy_inst);
+  DirectedLink(in, io_copy_inst);
   DirectedLink(io_copy_inst, io_copy_output_arg);
   DirectedLink(io_copy_output_arg, inst_node);
 
   // reset opdesc and update kernel information
-  UpdateInputTo(inst_node->AsStmt().op->mutable_op_info(), var,
+  UpdateInputTo(inst_node->AsStmt().op->mutable_op_info(), in->AsArg().name,
                 io_copy_output_name);
 
   inst_node->AsStmt().op->Attach(*inst_node->AsStmt().op->op_info(),

paddle/fluid/lite/core/mir/type_target_transform_pass.h

@@ -45,7 +45,7 @@ class TypeTargetTransformPass : public ProgramPass {
 
   void ComplementInputs(SSAGraph* graph, Node* inst_node, Node* in);
 
-  void AddIoCopyInst(const Type& from, const Type& to, const std::string& var,
+  void AddIoCopyInst(const Type& from, const Type& to, Node* in,
                      SSAGraph* graph, Node* inst_node,
                      const std::vector<Place>& valid_places);
 

paddle/fluid/lite/core/mir/variable_place_inference_pass.h

@@ -13,7 +13,10 @@
 // limitations under the License.
 
 #pragma once
+#include <map>
 #include <memory>
+#include <string>
+#include <vector>
 #include "paddle/fluid/lite/core/mir/pass.h"
 #include "paddle/fluid/lite/core/target_wrapper.h"
 
@@ -60,40 +63,44 @@ class VariablePlaceInferencePass : public DebugPass {
       // LOG(INFO) << "- inferencing type " <<
       // deal with inputs
       VLOG(4) << "inferencing op " << inst.op_type;
-      for (auto& arg_name : inst.op_info()->input_argnames()) {
+      // TODO(zhaolong): Add check if the node's name in op's arguments.
+
+      auto get_argname = [&](
+          const std::string& node_name,
+          const std::map<std::string, std::vector<std::string>>& argname_map)
+          -> std::string {
+            for (auto& ele : argname_map) {
+              auto it =
+                  std::find(ele.second.begin(), ele.second.end(), node_name);
+              if (it != ele.second.end()) return ele.first;
+            }
+            return "";
+          };
+
+      for (auto* x_in : x->inlinks) {
+        std::string node_name = x_in->AsArg().name;
+        std::string arg_name = get_argname(node_name, inst.op_info()->inputs());
+        CHECK(arg_name.size() > 0) << "can not found op arguments for node "
+                                   << node_name;
         VLOG(3) << "-- input arg_name " << arg_name;
-        // check if inputs's place is set, if not set, update them with the
-        // kernel's declaration.
         auto type = inst.picked_kernel().GetInputDeclType(arg_name);
-        auto arg_names = inst.op_info()->inputs().at(arg_name);
-
-        for (auto& arg_name : arg_names) {
-          VLOG(3) << "--- var " << arg_name;
-          auto* node = graph->RetrieveArgument(arg_name);
-          CHECK(node) << "argument " << arg_name << " not exists in the graph";
-          auto& arg_node = node->AsArg();
-          if (!arg_node.type) {
-            VLOG(4) << "set type " << *type << " " << node;
-            arg_node.type = type;
-          }
+        if (!x_in->AsArg().type) {
+          VLOG(4) << "set type " << *type << " " << x_in;
+          x_in->AsArg().type = type;
         }
       }
 
-      for (auto& arg_name : inst.op_info()->output_argnames()) {
+      for (auto* x_out : x->outlinks) {
+        std::string node_name = x_out->AsArg().name;
+        std::string arg_name =
+            get_argname(node_name, inst.op_info()->outputs());
+        CHECK(arg_name.size() > 0) << "can not found op arguments for node "
+                                   << node_name;
         VLOG(3) << "-- output arg_name " << arg_name;
         auto type = inst.picked_kernel().GetOutputDeclType(arg_name);
-        auto arg_names = inst.op_info()->outputs().at(arg_name);
-        // check if outputs's place is set, if not set, update them with the
-        // kernel's declaration.
-        for (auto& arg_name : arg_names) {
-          VLOG(3) << "--- var " << arg_name;
-          auto* node = graph->RetrieveArgument(arg_name);
-          CHECK(node) << "argument " << arg_name << " not exists in the graph";
-          auto& arg_node = node->AsArg();
-          if (!arg_node.type) {
-            node->AsArg().type = type;
-            VLOG(3) << "set type " << *type;
-          }
+        if (!x_out->AsArg().type) {
+          VLOG(4) << "set type " << *type << " " << x_out;
+          x_out->AsArg().type = type;
         }
       }
     }

paddle/fluid/lite/core/optimizer.h

@@ -49,6 +49,8 @@ class Optimizer {
 #ifndef LITE_WITH_LIGHT_WEIGHT_FRAMEWORK
     if (passes.empty()) {
       RunPasses(std::vector<std::string>{{
+          "lite_conv_bn_fuse_pass",                   //
+          "lite_conv_elementwise_add_act_fuse_pass",  //
           "lite_fc_fuse_pass",                        //
           "static_kernel_pick_pass",                  //
           "variable_place_inference_pass",            //

paddle/fluid/lite/kernels/x86/conv_compute.cc

@@ -74,6 +74,7 @@ class Conv2dCompute : public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
     lite::Tensor col_matrix;
     if (is_expand) {
       col.Resize(col_shape);
+      col.mutable_data<T>();
       col_matrix.ShareDataWith(col);
       col_matrix.Resize(col_matrix_shape);
     }
@@ -155,7 +156,6 @@ REGISTER_LITE_KERNEL(conv2d, kX86, kFloat, kNCHW,
     .BindInput("Input", {LiteType::GetTensorTy(TARGET(kX86))})
     .BindInput("Filter", {LiteType::GetTensorTy(TARGET(kX86))})
     .BindInput("Bias", {LiteType::GetTensorTy(TARGET(kX86))})
-    .BindInput("Input", {LiteType::GetTensorTy(TARGET(kX86))})
     .BindOutput("Output", {LiteType::GetTensorTy(TARGET(kX86))})
     .Finalize();
 
@@ -164,6 +164,5 @@ REGISTER_LITE_KERNEL(depthwise_conv2d, kX86, kFloat, kNCHW,
     .BindInput("Input", {LiteType::GetTensorTy(TARGET(kX86))})
     .BindInput("Filter", {LiteType::GetTensorTy(TARGET(kX86))})
     .BindInput("Bias", {LiteType::GetTensorTy(TARGET(kX86))})
-    .BindInput("Input", {LiteType::GetTensorTy(TARGET(kX86))})
     .BindOutput("Output", {LiteType::GetTensorTy(TARGET(kX86))})
     .Finalize();

paddle/fluid/lite/operators/CMakeLists.txt

@@ -19,6 +19,7 @@ cc_library(dropout_op_lite SRCS dropout_op.cc DEPS ${op_DEPS})
 cc_library(concat_op_lite SRCS concat_op.cc DEPS ${op_DEPS})
 cc_library(conv_op_lite SRCS conv_op.cc DEPS ${op_DEPS})
 cc_library(pool_op_lite SRCS pool_op.cc DEPS ${op_DEPS})
+cc_library(batch_norm_op_lite SRCS batch_norm.cc DEPS ${op_DEPS})
 
 set(ops_lite
     fc_op_lite
@@ -38,6 +39,7 @@ set(ops_lite
     concat_op_lite
     conv_op_lite
     pool_op_lite
+    batch_norm_op_lite
     CACHE INTERNAL "ops lite")
 
 lite_cc_test(test_fc_op_lite SRCS fc_op_test.cc 

paddle/fluid/lite/operators/batch_norm.cc

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#include "paddle/fluid/lite/operators/batch_norm.h"
+#include <vector>
+#include "paddle/fluid/lite/core/op_registry.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+bool BatchNormOpLite::CheckShape() const { return true; }
+
+bool BatchNormOpLite::InferShape() const { return true; }
+
+}  // namespace operators
+}  // namespace lite
+}  // namespace paddle
+
+REGISTER_LITE_OP(batch_norm, paddle::lite::operators::BatchNormOpLite);

paddle/fluid/lite/operators/batch_norm.h

+// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#pragma once
+#include <string>
+#include <vector>
+#include "paddle/fluid/lite/core/kernel.h"
+#include "paddle/fluid/lite/core/op_lite.h"
+#include "paddle/fluid/lite/core/scope.h"
+#include "paddle/fluid/lite/operators/op_params.h"
+#include "paddle/fluid/lite/utils/all.h"
+
+namespace paddle {
+namespace lite {
+namespace operators {
+
+class BatchNormOpLite : public OpLite {
+ public:
+  BatchNormOpLite() {}
+
+  explicit BatchNormOpLite(const std::string &type) : OpLite(type) {}
+
+  bool CheckShape() const override;
+
+  bool InferShape() const override;
+
+  void AttachKernel(KernelBase *kernel) override { kernel->SetParam(param_); }
+  // TODO(Superjomn) replace framework::OpDesc with a lite one.
+  bool AttachImpl(const cpp::OpDesc &op_desc, lite::Scope *scope) override {
+    auto x = op_desc.Input("X").front();
+    auto bias = op_desc.Input("Bias").front();
+    auto mean = op_desc.Input("Mean").front();
+    auto scale = op_desc.Input("Scale").front();
+    auto variance = op_desc.Input("Variance").front();
+
+    auto out = op_desc.Output("Y").front();
+    auto mean_out = op_desc.Output("MeanOut").front();
+    auto var_out = op_desc.Output("VarianceOut").front();
+    auto saved_mean = op_desc.Output("SavedMean").front();
+    auto saved_var = op_desc.Output("SavedVariance").front();
+
+    auto *var = scope->FindVar(x);
+    param_.x = var->GetMutable<Tensor>();
+    var = scope->FindVar(bias);
+    param_.bias = var->GetMutable<Tensor>();
+    var = scope->FindVar(mean);
+    param_.mean = var->GetMutable<Tensor>();
+    var = scope->FindVar(scale);
+    param_.scale = var->GetMutable<Tensor>();
+    var = scope->FindVar(variance);
+    param_.var = var->GetMutable<Tensor>();
+    var = scope->FindVar(out);
+    param_.out = var->GetMutable<Tensor>();
+    var = scope->FindVar(mean_out);
+    param_.mean_out = var->GetMutable<Tensor>();
+    var = scope->FindVar(var_out);
+    param_.var_out = var->GetMutable<Tensor>();
+    var = scope->FindVar(saved_mean);
+    param_.saved_mean = var->GetMutable<Tensor>();
+    var = scope->FindVar(saved_var);
+    param_.saved_var = var->GetMutable<Tensor>();
+
+    param_.eps = op_desc.GetAttr<float>("epsilon");
+
+    return true;
+  }
+
+  std::string DebugString() const override { return "batch_norm"; }
+
+ private:
+  mutable BatchNormParam param_;
+};
+
+}  // namespace operators
+}  // namespace lite
+}  // namespace paddle

paddle/fluid/lite/operators/op_params.h

@@ -237,6 +237,22 @@ struct SGDParam {
   lite::Tensor* ParamOut{};
 };
 
+//
+struct BatchNormParam {
+  lite::Tensor* x{};
+  lite::Tensor* bias{};
+  lite::Tensor* mean{};
+  lite::Tensor* scale{};
+  lite::Tensor* var{};
+  lite::Tensor* out{};
+  lite::Tensor* mean_out{};
+  lite::Tensor* var_out{};
+  lite::Tensor* saved_mean{};
+  lite::Tensor* saved_var{};
+
+  float eps{1e-5};
+};
+
 }  // namespace operators
 }  // namespace lite
 }  // namespace paddle