grad_op_desc_maker.h

/* Copyright (c) 2016 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 <algorithm>
#include <memory>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <vector>
#include "paddle/fluid/framework/op_desc.h"
#include "paddle/fluid/framework/operator.h"
#include "paddle/fluid/imperative/dygraph_grad_maker.h"
#include "paddle/fluid/imperative/layer.h"
#include "paddle/fluid/imperative/type_defs.h"

namespace paddle {
namespace framework {

namespace details {

template <typename T>
struct GradOpPtrTrait {};

template <>
struct GradOpPtrTrait<OpDesc> {
  using Type = OpDesc*;
};

template <>
struct GradOpPtrTrait<imperative::OpBase> {
  using Type = imperative::TracedGradOp*;
};

}  // namespace details

template <typename T>
using GradOpPtr = typename details::GradOpPtrTrait<T>::Type;

/*
  This functor class is responsible for creating the gradient ops for the given
  operator fwd_op. After it is called (through operator()), the pairs of
  (gradient variable, corresponding input variable of fwd_op) will be added to
  grad_to_var. If an input variable of fwd_op is contained in no_grad_set, its
  gradient varialbe will be ignored or kEmptyVarName depending on the template
  argument DropEmptyIG in the derived classes.
 */
class GradOpDescMakerBase {
 public:
  explicit GradOpDescMakerBase(
      const OpDesc& fwd_op, const std::unordered_set<std::string>& no_grad_set,
      std::unordered_map<std::string, std::string>* grad_to_var,
      const std::vector<BlockDesc*>& grad_block = std::vector<BlockDesc*>())
      : fwd_op_(fwd_op),
        no_grad_set_(no_grad_set),
        grad_to_var_(grad_to_var),
        grad_block_(grad_block) {}

  static std::unique_ptr<OpDesc> CreateOp() {
    return std::unique_ptr<OpDesc>(new OpDesc());
  }

  virtual ~GradOpDescMakerBase() = default;
  virtual std::vector<std::unique_ptr<OpDesc>> operator()() const = 0;

 protected:
  std::vector<std::string> InputGrad(const std::string& name,
                                     bool drop_empty_grad = true) const {
    std::vector<std::string> ret_val;
    auto var_names = this->Input(name);
    ret_val.reserve(var_names.size());
    std::transform(var_names.begin(), var_names.end(),
                   std::back_inserter(ret_val),
                   [this](const std::string& fwd_var_name) -> std::string {
                     auto g_name = GradVarName(fwd_var_name);
                     if (no_grad_set_.empty() || !no_grad_set_.count(g_name)) {
                       (*this->grad_to_var_)[g_name] = fwd_var_name;
                       return g_name;
                     } else {
                       return kEmptyVarName;
                     }
                   });
    if (!drop_empty_grad) {
      return ret_val;
    }
    PADDLE_ENFORCE_LE(var_names.size(), 1UL,
                      "BUG from operator developer:"
                      " for input argument with a list of variables, "
                      " drop_empty_grad is not allowed because it makes"
                      " the correspondence bewteen a variable and its gradient"
                      " ambiguous."
                      " Op type %s",
                      fwd_op_.Type());

    std::vector<std::string> dropped_ret_val;
    dropped_ret_val.reserve(ret_val.size());
    std::copy_if(ret_val.begin(), ret_val.end(),
                 std::back_inserter(dropped_ret_val),
                 [](const std::string& str) { return str != kEmptyVarName; });
    return dropped_ret_val;
  }

  std::vector<std::string> OutputGrad(const std::string& name) const {
    std::vector<std::string> ret_val;
    auto onames = this->Output(name);
    ret_val.reserve(onames.size());
    std::transform(onames.begin(), onames.end(), std::back_inserter(ret_val),
                   [this](const std::string& fwd_var_name) -> std::string {
                     auto g_name = GradVarName(fwd_var_name);
                     (*this->grad_to_var_)[g_name] = fwd_var_name;
                     return g_name;
                   });
    return ret_val;
  }

  static std::vector<std::string> EmptyInput() { return {}; }

  static std::vector<std::string> EmptyOutput() { return {}; }

  static std::vector<std::string> EmptyInputGrad() { return {}; }

  static std::vector<std::string> EmptyOutputGrad() { return {}; }

  std::vector<std::string> InputNames() const {
    return this->fwd_op_.InputNames();
  }

  std::vector<std::string> OutputNames() const {
    return this->fwd_op_.OutputNames();
  }

  std::vector<std::string> Input(const std::string& name) const {
    return fwd_op_.Input(name);
  }

  std::vector<std::string> Output(const std::string& name) const {
    return fwd_op_.Output(name);
  }

  const std::unordered_map<std::string, Attribute>& Attrs() const {
    return fwd_op_.GetAttrMap();
  }

  const Attribute& GetAttr(const std::string& name) const {
    auto& map = fwd_op_.GetAttrMap();
    auto it = map.find(name);
    PADDLE_ENFORCE(it != map.end(), "Cannot find attribute %s", name);
    return it->second;
  }

  template <typename T>
  inline const T& Attr(const std::string& name) const {
    return boost::get<T>(GetAttr(name));
  }

  std::string ForwardOpType() const { return this->fwd_op_.Type(); }

 protected:
  bool HasInput(const std::string& name) const {
    return (fwd_op_.Inputs().count(name) > 0);
  }

  bool HasOutput(const std::string& name) const {
    return (fwd_op_.Outputs().count(name) > 0);
  }

 private:
  const OpDesc& fwd_op_;
  const std::unordered_set<std::string>& no_grad_set_;
  std::unordered_map<std::string, std::string>* grad_to_var_;

 protected:
  std::vector<BlockDesc*> grad_block_;
};

template <typename T>
class SingleGradOpMaker {};

template <>
class SingleGradOpMaker<OpDesc> : public GradOpDescMakerBase {
 public:
  using GradOpDescMakerBase::GradOpDescMakerBase;

  std::vector<std::unique_ptr<OpDesc>> operator()() const final {
    std::vector<std::unique_ptr<OpDesc>> retv;
    retv.emplace_back(new OpDesc());
    this->Apply(retv.front().get());
    return retv;
  }

 protected:
  virtual void Apply(GradOpPtr<OpDesc> op) const = 0;
};

template <>
class SingleGradOpMaker<imperative::OpBase>
    : public imperative::GradOpBaseMakerBase {
 public:
  using GradOpBaseMakerBase::GradOpBaseMakerBase;

  std::shared_ptr<imperative::GradOpNode> operator()() const final {
    auto node = this->NewGradNode();
    {
      imperative::TracedGradOp traced_grad_op(node);
      this->Apply(&traced_grad_op);
    }
    return node->empty() ? nullptr : node;
  }

 protected:
  virtual void Apply(GradOpPtr<imperative::OpBase> op) const = 0;
};

template <typename T, bool DropEmptyIG = true>
class DefaultGradOpMaker final : public SingleGradOpMaker<T> {
 public:
  using SingleGradOpMaker<T>::SingleGradOpMaker;

 protected:
  void Apply(GradOpPtr<T> grad) const final {
    grad->SetType(this->ForwardOpType() + "_grad");

    for (auto& input_param : this->InputNames()) {
      grad->SetInput(input_param, this->Input(input_param));
      grad->SetOutput(GradVarName(input_param),
                      this->InputGrad(input_param, DropEmptyIG));
    }

    for (auto& output_param : this->OutputNames()) {
      grad->SetInput(output_param, this->Output(output_param));
      grad->SetInput(GradVarName(output_param), this->OutputGrad(output_param));
    }

    grad->SetAttrMap(this->Attrs());
  }
};

template <typename T>
class EmptyGradOpMaker {};

template <>
class EmptyGradOpMaker<OpDesc> final : public GradOpDescMakerBase {
 public:
  using GradOpDescMakerBase::GradOpDescMakerBase;
  std::vector<std::unique_ptr<OpDesc>> operator()() const final { return {}; }
};

template <>
class EmptyGradOpMaker<imperative::OpBase> final
    : public imperative::GradOpBaseMakerBase {
 public:
  using GradOpBaseMakerBase::GradOpBaseMakerBase;

  std::shared_ptr<imperative::GradOpNode> operator()() const final {
    return nullptr;
  }
};

}  // namespace framework

namespace operators {

template <typename T>
using GradOpPtr = framework::GradOpPtr<T>;

}  // namespace operators

}  // namespace paddle