operator.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 <atomic>
#include <memory>
#include <mutex>  // NOLINT
#include <string>
#include <tuple>
#include <unordered_map>
#include <utility>
#include <vector>

#include "glog/logging.h"  // For VLOG
#include "paddle/fluid/framework/attribute.h"
#include "paddle/fluid/framework/block_desc.h"
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/op_info.h"
#include "paddle/fluid/framework/op_kernel_type.h"
#include "paddle/fluid/framework/pten_utils.h"
#include "paddle/fluid/framework/scope.h"
#include "paddle/fluid/framework/selected_rows.h"
#include "paddle/fluid/framework/tensor.h"
#include "paddle/fluid/framework/unused_var_check.h"
#include "paddle/fluid/memory/malloc.h"
#include "paddle/fluid/platform/device_context.h"
#include "paddle/fluid/platform/variant.h"
#include "paddle/utils/flat_hash_map.h"

#include "paddle/pten/core/arg_map_context.h"
#include "paddle/pten/include/core.h"

namespace paddle {
namespace framework {
class InferShapeContext;
class OpInfo;
class Scope;
class Variable;
}  // namespace framework
}  // namespace paddle

DECLARE_int32(inner_op_parallelism);

namespace paddle {
namespace framework {

/// If a variable is a empty variable, that name will be used.
constexpr char kEmptyVarName[] = "@EMPTY@";

/// If a variable is a temporary variable, that name will be set in Python,
/// but it will be convert to a unique name in scope after OpCreator.
constexpr char kTempVarName[] = "@TEMP@";

/// If a variable's name has a certain suffix, it means that the
/// variable is the gradient of another variable.
/// e.g. Variable "x@GRAD" is the gradient of variable "x".
constexpr char kGradVarSuffix[] = "@GRAD";

constexpr size_t kGradVarSuffixSize = 5U;

/// Variables with this suffix are supposed to be filled up with zeros.
constexpr char kZeroVarSuffix[] = "@ZERO";

/// Variables with this suffix are the new Gradient.
constexpr char kNewGradSuffix[] = "@NEWGRAD@";

/// RuntimeContext is used to relate input/output names of Operator with
/// the corresponding variables in name scope.
/// If an Op has attribute kEnableCacheRuntimeContext, it means that in a same
/// name scope, since the input/output names of this Op do not change in the
/// execution, RuntimeContext could be created only at the first iteration of
/// this Op's execution to save the elapsed time.
constexpr char kEnableCacheRuntimeContext[] = "@ENABLE_CACHE_RUNTIME_CONTEXT@";

/// If an Op has this attribute, all its kernels should calculate output
/// variable's shape in the corresponding Compute() function. And
/// OperatorWithKernel::RunImpl() would skip call this Op's InferShape()
/// function in its runtime for speedup.
/// TODO(luotao): Note that this temporal attribute would be deleted after all
/// ops contain it.
constexpr char kAllKernelsMustComputeRuntimeShape[] =
    "@ALL_KERNELS_MUST_COMPUTE_RUNTIME_SHAPE@";

// define some kernel priority
/* Define multiple kernel type fallback order*/
extern std::vector<std::tuple<platform::Place, LibraryType>> kKernelPriority;

inline std::string GradVarName(const std::string& var_name) {
  std::string result;
  result.reserve(var_name.size() + kGradVarSuffixSize);
  result += var_name;
  result += kGradVarSuffix;
  return result;
}

inline std::string GradOriginalVarName(const std::string& grad_var_name) {
  std::size_t pos = grad_var_name.rfind(kGradVarSuffix);
  if (pos == std::string::npos) {
    return grad_var_name;
  } else {
    return grad_var_name.substr(0, pos);
  }
}

inline bool VarIsTensor(const Variable& var) {
  return var.IsType<LoDTensor>() || var.IsType<SelectedRows>();
}

const Tensor* GetLoDTensorOrSelectedRowsValueFromVar(const Variable& var);
Tensor* GetMutableLoDTensorOrSelectedRowsValueFromVar(Variable* var);

class ExecutionContext;
class OperatorBase;

class RuntimeContext {
 public:
  RuntimeContext(const VariableNameMap& innames,
                 const VariableNameMap& outnames, const Scope& scope);

  RuntimeContext(const VariableValueMap& invars,
                 const VariableValueMap& outvars)
      : inputs(invars), outputs(outvars) {}

  VariableValueMap inputs;
  VariableValueMap outputs;
};

/**
 * OperatorBase has the basic elements that Net will call to do computation.
 * Only CreateOperator from OpRegistry will new Operator directly. User
 * should always construct a proto message OpDesc and call
 * OpRegistry::CreateOp(op_desc) to get an Operator instance.
 */
class OperatorBase {
 public:
  OperatorBase(const std::string& type, const VariableNameMap& inputs,
               const VariableNameMap& outputs, const AttributeMap& attrs);

  virtual ~OperatorBase() {}

  /// Executor will call this interface function to Run an op.
  //  The implementation should be written at RunImpl
  void Run(const Scope& scope, const platform::Place& place);

  // FIXME(typhoonzero): this is only used for recv_op to stop event_loop.
  virtual void Stop() {}

  /// if scope is not null, also show dimensions of arguments
  virtual std::string DebugStringEx(const ScopeBase* scope) const;
  std::string DebugString() const { return DebugStringEx(nullptr); }

  virtual bool SupportGPU() const { return false; }
  virtual bool SupportNPU() const { return false; }
  virtual bool SupportMLU() const { return false; }

  const std::string& Type() const { return type_; }

  bool HasAttr(const std::string& name) const { return attrs_.count(name); }
  template <typename T>
  inline const T& Attr(const std::string& name) const {
    PADDLE_ENFORCE_NE(
        attrs_.find(name), attrs_.end(),
        platform::errors::NotFound("(%s) is not found in AttributeMap.", name));
    return BOOST_GET_CONST(T, attrs_.at(name));
  }
  void SetAttr(const std::string& name, const Attribute& v) {
    PADDLE_ENFORCE_EQ(
        HasAttr(name), true,
        platform::errors::NotFound(
            "The attribute %s is not found in operator %s", name, Type()));

    attrs_[name] = v;
  }
  const AttributeMap& Attrs() const { return attrs_; }

  const VariableNameMap& Inputs() const { return inputs_; }
  const VariableNameMap& Outputs() const { return outputs_; }

  const OpInfo& Info() const {
    PADDLE_ENFORCE_NOT_NULL(
        info_, platform::errors::NotFound(
                   "OpInfo of operator (%s) is not found.", type_));
    return *info_;
  }

  bool HasInputs(const std::string& name) const;
  //! Get a input with argument's name described in `op_proto`
  std::string Input(const std::string& name) const;
  //! Get a input which has multiple variables.
  const std::vector<std::string>& Inputs(const std::string& name) const;
  //! Get all inputs variable names
  std::vector<std::string> InputVars() const;

  bool HasOutputs(const std::string& name) const;
  //! Get a output with argument's name described in `op_proto`
  std::string Output(const std::string& name) const;
  //! Get an output which has multiple variables.
  //! TODO add a vector_view to prevent memory copy.
  const std::vector<std::string>& Outputs(const std::string& name) const;
  //! Get all outputs variable names
  virtual std::vector<std::string> OutputVars(bool has_intermediate) const;

  void SetIsCalledByExecutor(bool x) { run_by_executor_ = x; }

  virtual void RuntimeInferShape(const Scope& scope,
                                 const platform::Place& place,
                                 const RuntimeContext& ctx) const {}

  virtual platform::Place GetExecutionPlace(
      const platform::Place& place) const {
    return place;
  }

 protected:
  std::string type_;
  // NOTE: in case of OpGrad, inputs_ contains:
  // I (Inputs)
  // O (Outputs)
  // OG (Output Gradients)
  VariableNameMap inputs_;

  // NOTE: in case of OpGrad, outputs_ contains
  // IG (Inputs Gradients)
  VariableNameMap outputs_;
  AttributeMap attrs_;

  // OpInfo
  const OpInfo* info_;

  // Whether this operator executes in an Executor.
  bool run_by_executor_{true};

 private:
  void GenerateTemporaryNames();
  void CheckAllInputOutputSet() const;
  virtual void RunImpl(const Scope& scope,
                       const platform::Place& place) const = 0;
};

class ExecutionContext {
 public:
  ExecutionContext(const OperatorBase& op, const Scope& scope,
                   const platform::DeviceContext& device_context,
                   const RuntimeContext& ctx)
      : op_(op), scope_(scope), device_context_(device_context), ctx_(ctx) {}
  virtual ~ExecutionContext() {}

  virtual std::string InputName(const std::string& name) const {
    return op_.Input(name);
  }
  virtual std::vector<std::string> InputNames(const std::string& name) const {
    return op_.Inputs(name);
  }
  virtual std::string OutputName(const std::string& name) const {
    return op_.Output(name);
  }

  virtual std::vector<std::string> OutputNames(const std::string& name) const {
    return op_.Outputs(name);
  }

  virtual bool HasAttr(const std::string& name) const {
    return op_.HasAttr(name);
  }
  virtual const AttributeMap& Attrs() const { return op_.Attrs(); }

  const std::string& Type() const { return op_.Type(); }

  const Scope& scope() const { return scope_; }

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

  virtual const Attribute& GetAttr(const std::string& name) const {
    return op_.Attrs().at(name);
  }

  virtual bool HasInput(const std::string& name) const;

  virtual bool HasOutput(const std::string& name) const;

  virtual size_t InputSize(const std::string& name) const {
    return op_.Inputs(name).size();
  }

  virtual size_t OutputSize(const std::string& name) const {
    return op_.Outputs(name).size();
  }

  virtual const Variable* InputVar(const std::string& name) const;

  virtual Variable* OutputVar(const std::string& name) const;

  virtual const std::vector<Variable*> MultiInputVar(
      const std::string& name) const {
    LogVarUsageIfUnusedVarCheckEnabled(name);

    auto it = ctx_.inputs.find(name);
    if (it == ctx_.inputs.end()) {
      return {};
    }
    return {it->second.begin(), it->second.end()};
  }

  virtual std::vector<Variable*> MultiOutputVar(const std::string& name) const {
    auto it = ctx_.outputs.find(name);
    if (it == ctx_.outputs.end()) {
      return {};
    }
    return it->second;
  }

  virtual std::vector<std::string> InNameList() const {
    std::vector<std::string> vec_temp;
    vec_temp.reserve(ctx_.inputs.size());

    for (auto& input : ctx_.inputs) {
      vec_temp.push_back(input.first);
    }

    return vec_temp;
  }

  template <typename T>
  const T* Input(const std::string& name) const {
    auto* var = InputVar(name);
    return var == nullptr ? nullptr : &var->Get<T>();
  }

  template <typename T>
  T* Output(const std::string& name) const {
    auto var = OutputVar(name);
    return var == nullptr ? nullptr : var->GetMutable<T>();
  }

  template <typename T>
  const std::vector<const T*> MultiInput(const std::string& name) const {
    LogVarUsageIfUnusedVarCheckEnabled(name);

    auto vars = MultiInputVar(name);
    if (vars.size() == 0) {
      return {};
    }
    std::vector<const T*> res;
    res.reserve(vars.size());
    std::transform(vars.begin(), vars.end(), std::back_inserter(res),
                   [&](const Variable* var) -> const T* {
                     return var == nullptr ? nullptr : &var->Get<T>();
                   });
    return res;
  }

  template <typename T>
  std::vector<T*> MultiOutput(const std::string& name) const {
    auto vars = MultiOutputVar(name);

    if (vars.size() == 0) {
      return {};
    }

    std::vector<T*> res;
    res.reserve(vars.size());
    std::transform(vars.begin(), vars.end(), std::back_inserter(res),
                   [&](Variable* var) -> T* {
                     return var == nullptr ? nullptr : var->GetMutable<T>();
                   });

    return res;
  }

  platform::Place GetPlace() const { return device_context_.GetPlace(); }

  template <typename DeviceContextType>
  const DeviceContextType& device_context() const {
    return *reinterpret_cast<const DeviceContextType*>(&device_context_);
  }

  const platform::DeviceContext& device_context() const {
    return device_context_;
  }

#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  const inline platform::CUDADeviceContext& cuda_device_context() const {
    PADDLE_ENFORCE_EQ(platform::is_gpu_place(device_context_.GetPlace()), true,
                      platform::errors::PreconditionNotMet(
                          "Current device context place is not GPUPlace."));
    return *reinterpret_cast<const platform::CUDADeviceContext*>(
        &device_context_);
  }
#endif

  template <typename T, typename DevContext>
  Tensor AllocateTmpTensor(const framework::DDim& dim,
                           const DevContext& dev_ctx) const {
    auto tmp_allocation_ptr = memory::Alloc(dev_ctx, product(dim) * sizeof(T));
    auto& deleter = tmp_allocation_ptr.get_deleter();
    auto* allocation_ptr = tmp_allocation_ptr.release();
    auto shared_allocation =
        std::shared_ptr<pten::Allocation>(allocation_ptr, deleter);

    PADDLE_ENFORCE_GE(
        allocation_ptr->size(), framework::product(dim) * sizeof(T),
        platform::errors::PreconditionNotMet(
            "The data memory size(%d) is less than the tensor needed memory "
            "size(%d).",
            allocation_ptr->size(), framework::product(dim) * sizeof(T)));

    paddle::framework::Tensor temp_tensor(
        framework::ToDataType(std::type_index(typeid(T))));
    temp_tensor.Resize(dim);
    temp_tensor.ResetHolder(std::move(shared_allocation));
    return temp_tensor;
  }

  const RuntimeContext Context() const { return ctx_; }

  std::string DebugString() const { return op_.DebugString(); }
  const OperatorBase& GetOp() const { return op_; }

 private:
  const OperatorBase& op_;
  const Scope& scope_;
  const platform::DeviceContext& device_context_;
  const RuntimeContext& ctx_;
};

// TODO(chenweihang): split impl based OpProto or Dygraph if needed
class ExecutionArgumentMappingContext : public pten::ArgumentMappingContext {
 public:
  explicit ExecutionArgumentMappingContext(const ExecutionContext& ctx)
      : ctx_(ctx) {}

  bool HasInput(const std::string& name) const override {
    return ctx_.HasInput(name);
  }

  bool HasOutput(const std::string& name) const override {
    return ctx_.HasOutput(name);
  }

  bool HasAttr(const std::string& name) const override {
    return ctx_.HasAttr(name);
  }

  size_t InputSize(const std::string& name) const override {
    return ctx_.InputSize(name);
  }

  size_t OutputSize(const std::string& name) const override {
    return ctx_.OutputSize(name);
  }

  bool IsDenseTensorInput(const std::string& name) const override {
    return ctx_.InputVar(name)->IsType<framework::Tensor>() ||
           ctx_.InputVar(name)->IsType<framework::LoDTensor>();
  }

  bool IsSelectedRowsInput(const std::string& name) const override {
    return ctx_.InputVar(name)->IsType<framework::SelectedRows>();
  }

 private:
  const ExecutionContext& ctx_;
};

template <>
const Tensor* ExecutionContext::Input<Tensor>(const std::string& name) const;

template <>
const std::vector<const Tensor*> ExecutionContext::MultiInput<Tensor>(
    const std::string& name) const;

template <>
Tensor* ExecutionContext::Output<Tensor>(const std::string& name) const;

template <>
std::vector<Tensor*> ExecutionContext::MultiOutput<Tensor>(
    const std::string& name) const;

class OpKernelBase {
 public:
  /**
   * ExecutionContext is the only parameter of Kernel Run function.
   * Run will get input/output variables, state such as momentum and
   * device resource such as CUDA stream, cublas handle, etc. from
   * ExecutionContext. User should construct it before run the Operator.
   */

  virtual void Compute(const ExecutionContext& context) const = 0;

  virtual ~OpKernelBase() = default;
};

template <typename T>
class OpKernel : public OpKernelBase {
 public:
  using ELEMENT_TYPE = T;
};

class OperatorWithKernel : public OperatorBase {
 public:
  using OpKernelFunc = std::function<void(const ExecutionContext&)>;
  using OpKernelMap =
      std::unordered_map<OpKernelType, OpKernelFunc, OpKernelType::Hash>;

  OperatorWithKernel(const std::string& type, const VariableNameMap& inputs,
                     const VariableNameMap& outputs, const AttributeMap& attrs)
      : OperatorBase(type, inputs, outputs, attrs) {}

  static paddle::flat_hash_map<std::string /* op_type */, OpKernelMap>&
  AllOpKernels() {
    static paddle::flat_hash_map<std::string, OpKernelMap> g_all_op_kernels;
    return g_all_op_kernels;
  }

  bool SupportGPU() const override {
    auto& op_kernels = OperatorWithKernel::AllOpKernels().at(type_);
    return std::any_of(op_kernels.begin(), op_kernels.end(),
                       [](OpKernelMap::const_reference kern_pair) {
                         return platform::is_gpu_place(kern_pair.first.place_);
                       });
  }
  bool SupportNPU() const override {
    auto& op_kernels = OperatorWithKernel::AllOpKernels().at(type_);
    return std::any_of(op_kernels.begin(), op_kernels.end(),
                       [](OpKernelMap::const_reference kern_pair) {
                         return platform::is_npu_place(kern_pair.first.place_);
                       });
  }
  bool SupportMLU() const override {
    auto& op_kernels = OperatorWithKernel::AllOpKernels().at(type_);
    return std::any_of(op_kernels.begin(), op_kernels.end(),
                       [](OpKernelMap::const_reference kern_pair) {
                         return platform::is_mlu_place(kern_pair.first.place_);
                       });
  }
  bool SupportsMKLDNN(proto::VarType::Type data_type) const;

  bool CanMKLDNNBeUsed(const framework::ExecutionContext& ctx,
                       proto::VarType::Type data_type) const;

  virtual void InferShape(InferShapeContext* ctx) const = 0;

  void RuntimeInferShape(const Scope& scope, const platform::Place& place,
                         const RuntimeContext& ctx) const override;

  proto::VarType::Type IndicateVarDataType(const ExecutionContext& ctx,
                                           const std::string& name) const;

  proto::VarType::Type IndicateOrPromoteVarDataTypes(
      const ExecutionContext& ctx, const std::string& name1,
      const std::string& name2) const;

  virtual OpKernelType GetExpectedKernelType(const ExecutionContext& ctx) const;

  // change this to public so that in dygraph mode we can call it to check if we
  // need transform data
  virtual OpKernelType GetKernelTypeForVar(
      const std::string& var_name, const Tensor& tensor,
      const OpKernelType& expected_kernel_type) const;

  platform::Place GetExecutionPlace(
      const platform::Place& platform) const override {
    return kernel_type_->place_;
  }

  /* member functions for adapting to pten lib */
  /** In the Tensor calculation library, the new Kernel adopts a clearer and
    * more streamlined design. The arguments of the Kernel and the input and
    * output arguments registered in the original OpMaker do not match in some
    * cases, so we use map to record the arguments required by the kernel.
    * When selecting Kernel during Op execution, select the arguments of the
    * original Op according to the GetExpectedPtenKernelArgs returned arguments.
    */
  virtual KernelSignature GetExpectedPtenKernelArgs(
      const ExecutionContext& ctx) const;

  /* member functions for adapting to pten lib */
  void ChoosePtenKernel(const ExecutionContext& ctx) const;

  void BuildPtenKernelContext(const RuntimeContext& ctx,
                              platform::DeviceContext* dev_ctx) const;

  void WriteBackToOutputs(RuntimeContext* ctx) const;

  pten::Kernel* PtenKernel() const { return pt_kernel_.get(); }

  pten::KernelContext* PtenKernelContext() const {
    return pt_kernel_context_.get();
  }

  const OpKernelType* kernel_type() const { return kernel_type_.get(); }

 private:
  void RunImpl(const Scope& scope, const platform::Place& place) const final;
  void RunImpl(const Scope& scope, const platform::Place& place,
               RuntimeContext* runtime_ctx) const;

  /**
   * Transfer data from scope to a transferred scope. If there is no data need
   * to
   * be tranfered, it returns nullptr.
   *
   * * transfered_inplace_vars is a output vector.
   */
  Scope* PrepareData(const Scope& scope,
                     const OpKernelType& expected_kernel_key,
                     std::vector<std::string>* transfered_inplace_vars,
                     RuntimeContext* ctx) const;

  void TransferInplaceVarsBack(const Scope& scope,
                               const std::vector<std::string>& inplace_vars,
                               const Scope& exec_scope) const;

  OpKernelType InnerGetExpectedKernelType(const ExecutionContext& ctx) const;

  void ChooseKernel(const ExecutionContext& ctx) const;

  void HandleComplexGradToRealGrad(const Scope& scope,
                                   RuntimeContext* ctx) const;

  /* Inner assist methods */
  // indicate kernel DataType by input data.
  // By default all input data must be same.
  proto::VarType::Type IndicateDataType(const ExecutionContext& ctx) const;
  // used for IndicateDataType
  void ParseInputDataType(const std::vector<Variable*>& vars,
                          const std::string& name,
                          proto::VarType::Type* data_type) const;
  // used for IndicateOrPromoteVarDataTypes
  Tensor* GetTensorFormInputSafely(const ExecutionContext& ctx,
                                   const std::string& name) const;

 protected:
  mutable std::unique_ptr<OpKernelType> kernel_type_;
  mutable std::unique_ptr<OpKernelFunc> kernel_func_;
  mutable std::unique_ptr<RuntimeContext> runtime_ctx_;
  mutable const Scope* pre_scope_ = nullptr;
  mutable bool need_prepare_data_ = true;
  mutable bool enable_cache_runtime_context_ = false;
  mutable bool all_kernels_must_compute_runtime_shape_ = false;
  mutable std::mutex cache_update_mutex_;
  mutable bool enable_cache_transfer_scope_ = false;
  // NOTE(chenweihang): Similar op members are used to adapt to
  // new pten kernel, if there is a better design in the future,
  // we may polish the implementation here
  mutable bool run_pten_kernel_ = false;
  mutable std::unique_ptr<KernelSignature> pt_kernel_signature_;
  mutable std::unique_ptr<pten::Kernel> pt_kernel_;
  // In order to reduce the compatibility phase
  // performance overhead, temporarily cache KernelContext
  mutable std::unique_ptr<pten::KernelContext> pt_kernel_context_;
};

extern bool OpSupportGPU(const std::string& op_type);

}  // namespace framework
}  // namespace paddle