Merge pull request #13858 from Sand3r-/mgallus/conv-bias-pass

[MKLDNN] Fuse Conv + Bias using Pass

paddle/fluid/framework/ir/CMakeLists.txt

@@ -39,6 +39,7 @@ pass_library(seq_concat_fc_fuse_pass inference)
 pass_library(conv_bn_fuse_pass inference)
 if(WITH_MKLDNN)
     pass_library(mkldnn_placement_pass base)
+    pass_library(conv_bias_mkldnn_fuse_pass inference)
     pass_library(conv_relu_mkldnn_fuse_pass inference)
 endif()
 

paddle/fluid/framework/ir/conv_bias_mkldnn_fuse_pass.cc

+// Copyright (c) 2018 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/framework/ir/conv_bias_mkldnn_fuse_pass.h"
+#include <functional>
+#include <string>
+#include <vector>
+#include "paddle/fluid/framework/lod_tensor.h"
+#include "paddle/fluid/platform/enforce.h"
+
+namespace paddle {
+namespace framework {
+namespace ir {
+
+template <typename BinaryOperation>
+LoDTensor tensor_apply_eltwise(const LoDTensor& vec_a, const LoDTensor& vec_b,
+                               BinaryOperation f) {
+  PADDLE_ENFORCE_EQ(vec_a.dims(), vec_b.dims());
+  LoDTensor vec_y;
+  vec_y.Resize(vec_a.dims());
+  const float* a = vec_a.data<float>();
+  const float* b = vec_b.data<float>();
+  float* y = vec_y.mutable_data<float>(platform::CPUPlace());
+  for (int i = 0; i < vec_a.numel(); i++) {
+    y[i] = f(a[i], b[i]);
+  }
+  return vec_y;
+}
+
+std::unique_ptr<ir::Graph> ConvBiasFusePass::ApplyImpl(
+    std::unique_ptr<ir::Graph> graph) const {
+  PADDLE_ENFORCE(graph.get());
+  FusePassBase::Init(name_scope_, graph.get());
+
+  auto* scope = param_scope();
+  PADDLE_ENFORCE(scope);
+
+  GraphPatternDetector gpd;
+  auto* conv_input =
+      gpd.mutable_pattern()
+          ->NewNode(patterns::PDNodeName(name_scope_, "conv_input"))
+          ->AsInput()
+          ->assert_is_op_input("conv2d", "Input");
+  patterns::ConvBias conv_bias_pattern(gpd.mutable_pattern(), name_scope_);
+  conv_bias_pattern(conv_input);
+  int found_conv_bias_count = 0;
+  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
+                     Graph* g) {
+    VLOG(4) << "handle ConvBias fuse";
+    GET_IR_NODE_FROM_SUBGRAPH(conv_weight, conv_weight,
+                              conv_bias_pattern);                      // Filter
+    GET_IR_NODE_FROM_SUBGRAPH(conv_out, conv_out, conv_bias_pattern);  // tmp
+    GET_IR_NODE_FROM_SUBGRAPH(conv, conv, conv_bias_pattern);  // CONV op
+    // bias
+    GET_IR_NODE_FROM_SUBGRAPH(eltwise_bias, eltwise_bias, conv_bias_pattern);
+    // output
+    GET_IR_NODE_FROM_SUBGRAPH(eltwise_out, eltwise_out, conv_bias_pattern);
+    // elementwise_add op
+    GET_IR_NODE_FROM_SUBGRAPH(eltwise, eltwise, conv_bias_pattern);
+
+    PADDLE_ENFORCE(subgraph.count(conv_input));
+
+    // check if fuse can be done and if MKL-DNN should be used
+    FuseOptions fuse_option = FindFuseOption(*conv, *eltwise);
+    if (fuse_option == DO_NOT_FUSE || fuse_option == FUSE_NATIVE) {
+      VLOG(3) << "do not perform conv+bias fuse";
+      return;
+    }
+
+    auto* eltwise_bias_tensor =
+        scope->FindVar(eltwise_bias->Name())->GetMutable<LoDTensor>();
+
+    auto input_names = conv->Op()->InputNames();
+    bool has_bias = std::find(input_names.begin(), input_names.end(), "Bias") !=
+                    input_names.end();
+    if (has_bias && conv->Op()->Input("Bias").size() > 0) {
+      auto conv_bias_names = conv->Op()->Input("Bias");
+      // add eltwise bias to existing conv bias
+      PADDLE_ENFORCE_EQ(conv_bias_names.size(), 1);
+      auto* conv_bias_var = scope->FindVar(conv_bias_names[0]);
+      auto* conv_bias_tensor = conv_bias_var->GetMutable<LoDTensor>();
+      PADDLE_ENFORCE_EQ(conv_bias_tensor->dims(), eltwise_bias_tensor->dims());
+      *conv_bias_tensor = tensor_apply_eltwise(
+          *conv_bias_tensor, *eltwise_bias_tensor, std::plus<float>());
+
+      conv->Op()->SetOutput("Output",
+                            std::vector<std::string>({eltwise_out->Name()}));
+
+      GraphSafeRemoveNodes(graph.get(), {eltwise, conv_out});
+
+      IR_NODE_LINK_TO(conv, eltwise_out);
+    } else {
+      // take eltwise bias as conv bias
+      OpDesc desc;
+
+      desc.SetInput(
+          "Input", std::vector<std::string>({subgraph.at(conv_input)->Name()}));
+      desc.SetInput("Filter", std::vector<std::string>({conv_weight->Name()}));
+      desc.SetInput("Bias", std::vector<std::string>({eltwise_bias->Name()}));
+      desc.SetOutput("Output", std::vector<std::string>({eltwise_out->Name()}));
+      desc.SetType("conv2d");
+
+      for (auto& attr : conv->Op()->GetAttrMap()) {
+        desc.SetAttr(attr.first, attr.second);
+      }
+      auto conv_bias_node = g->CreateOpNode(&desc);
+
+      IR_NODE_LINK_TO(subgraph.at(conv_input), conv_bias_node);
+      IR_NODE_LINK_TO(conv_weight, conv_bias_node);
+      IR_NODE_LINK_TO(eltwise_bias, conv_bias_node);
+      IR_NODE_LINK_TO(conv_bias_node, eltwise_out);
+
+      GraphSafeRemoveNodes(graph.get(), {conv, eltwise, conv_out});
+    }
+
+    found_conv_bias_count++;
+  };
+  gpd(graph.get(), handler);
+  AddStatis(found_conv_bias_count);
+  return graph;
+}
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle
+REGISTER_PASS(conv_bias_mkldnn_fuse_pass,
+              paddle::framework::ir::ConvBiasFusePass);

paddle/fluid/framework/ir/conv_bias_mkldnn_fuse_pass.h

+// Copyright (c) 2018 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 "paddle/fluid/framework/ir/fuse_pass_base.h"
+#include "paddle/fluid/framework/ir/graph.h"
+#include "paddle/fluid/framework/ir/graph_pattern_detector.h"
+#include "paddle/fluid/framework/ir/pass.h"
+namespace paddle {
+namespace framework {
+namespace ir {
+/*
+* Fuse the Conv and Elementwise_add to a ConvBiasOp.
+*/
+class ConvBiasFusePass : public FusePassBase {
+ public:
+  virtual ~ConvBiasFusePass() {}
+
+ protected:
+  std::unique_ptr<ir::Graph> ApplyImpl(std::unique_ptr<ir::Graph> graph) const;
+  const std::string name_scope_{"conv_bias_mkldnn_fuse"};
+};
+}  // namespace ir
+}  // namespace framework
+}  // namespace paddle

paddle/fluid/framework/ir/graph_pattern_detector.cc

@@ -966,6 +966,39 @@ PDNode *patterns::ElewiseAddActInplaceGrad::operator()(
   return ele_add_grad;
 }
 
+PDNode *patterns::ConvBias::operator()(
+    paddle::framework::ir::PDNode *conv_input) {
+  // Create Operators
+  conv_input->assert_is_op_input("conv2d", "Input");
+  auto *conv_op = pattern->NewNode(conv_repr())->assert_is_op("conv2d");
+  auto *eltiwse_op =
+      pattern->NewNode(eltwise_repr())->assert_is_op("elementwise_add");
+  // Create variables
+  // Filter
+  auto *conv_weight_var = pattern->NewNode(conv_weight_repr())
+                              ->AsInput()
+                              ->assert_is_persistable_var()
+                              ->assert_is_op_input("conv2d", "Filter");
+  // intermediate variable, will be removed in the IR after fuse.
+  auto *conv_out_var = pattern->NewNode(conv_out_repr())
+                           ->AsIntermediate()
+                           ->assert_is_only_output_of_op("conv2d")
+                           ->assert_is_op_input("elementwise_add");
+  // Bias stored in elementwise_add
+  auto *eltwise_bias_var = pattern->NewNode(eltwise_bias_repr())
+                               ->AsInput()
+                               ->assert_is_persistable_var()
+                               ->assert_is_op_input("elementwise_add", "Y");
+  // output
+  auto *eltwise_out_var = pattern->NewNode(eltwise_out_repr())
+                              ->AsOutput()
+                              ->assert_is_op_output("elementwise_add");
+  conv_op->LinksFrom({conv_input, conv_weight_var}).LinksTo({conv_out_var});
+  eltiwse_op->LinksFrom({conv_out_var, eltwise_bias_var})
+      .LinksTo({eltwise_out_var});
+  return eltwise_out_var;
+}
+
 }  // namespace ir
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/ir/graph_pattern_detector.h

@@ -578,6 +578,27 @@ struct ElewiseAddActInplaceGrad : public PatternBase {
   PATTERN_DECL_NODE(d_ele_y);
   PATTERN_DECL_NODE(ele_y);
 };
+
+// Conv with Elementwise_add as bias
+// op: conv + elementwise_add
+// named nodes:
+// conv_input, conv_weight,
+// conv_out, conv,
+// eltwise_bias, eltwise_out,
+// elementwise_add
+struct ConvBias : public PatternBase {
+  ConvBias(PDPattern* pattern, const std::string& name_scope)
+      : PatternBase(pattern, name_scope, "conv_bias") {}
+  PDNode* operator()(PDNode* conv_input);
+  // declare operator node's name
+  PATTERN_DECL_NODE(conv);
+  PATTERN_DECL_NODE(eltwise);
+  // declare variable node's name
+  PATTERN_DECL_NODE(conv_weight);
+  PATTERN_DECL_NODE(conv_out);
+  PATTERN_DECL_NODE(eltwise_bias);
+  PATTERN_DECL_NODE(eltwise_out);
+};
 }  // namespace patterns
 
 // Link two ir::Nodes from each other.

paddle/fluid/inference/analysis/analyzer.h

@@ -79,6 +79,7 @@ class Analyzer : public OrderedRegistry<PassManager> {
       "conv_bn_fuse_pass",             //
       "conv_eltwiseadd_bn_fuse_pass",  //
 #ifdef PADDLE_WITH_MKLDNN
+      "conv_bias_mkldnn_fuse_pass",  //
       "conv_relu_mkldnn_fuse_pass",  //
 #endif
   }};