paddlepaddle--paddle/paddle/phi/common/place.h

/* Copyright (c) 2022 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 <cstdint>
#include <ostream>
#include <string>
#include <unordered_map>

#include "paddle/common/enforce.h"
#include "paddle/common/macros.h"
#include "paddle/utils/test_macros.h"

namespace paddle {
enum class PlaceType;
}

namespace phi {

enum class AllocationType : int8_t {
  UNDEFINED = 0,
  CPU = 1,
  GPU = 2,
  CUDA = GPU,
  GPUPINNED = 3,
  XPU = 4,
  XPUPINNED = 5,
  IPU = 7,
  CUSTOM = 9,
};

class CustomRegisteredDeviceMap {
 public:
  PADDLE_API static CustomRegisteredDeviceMap& Instance();

  PADDLE_API size_t
  GetOrRegisterGlobalDeviceTypeId(const std::string& device_type);

  PADDLE_API std::string GetGlobalDeviceType(size_t device_type_id_);

 private:
  CustomRegisteredDeviceMap() = default;
  std::unordered_map<std::string, size_t> registered_device_type_id_;
  std::unordered_map<size_t, std::string> registered_device_type_;
};

PADDLE_API const char* AllocationTypeStr(AllocationType type);
PADDLE_API std::ostream& operator<<(std::ostream& os, AllocationType type);

/// \brief The place is used to specify where the data is stored.
class PADDLE_API Place {
 public:
  Place()
      : device(0), alloc_type_(AllocationType::UNDEFINED), device_type_id_(0) {}

  explicit Place(AllocationType type,
                 int8_t id,
                 const std::string& dev_type = "")
      : device(id),
        alloc_type_(type),
        device_type_id_(phi::CustomRegisteredDeviceMap::Instance()
                            .GetOrRegisterGlobalDeviceTypeId(dev_type)) {}

  explicit Place(AllocationType type, const std::string& dev_type = "");

  // See NOTE [ Why need to temporarily adapt to PlaceType? ]
  Place(paddle::PlaceType type);  // NOLINT

  void Reset(AllocationType type,
             int8_t device_id = 0,
             const std::string& dev_type = "") noexcept {
    alloc_type_ = type;
    device = device_id;
    if (!dev_type.empty()) {
      device_type_id_ = phi::CustomRegisteredDeviceMap::Instance()
                            .GetOrRegisterGlobalDeviceTypeId(dev_type);
    }
  }

  AllocationType GetType() const { return alloc_type_; }

  int8_t GetDeviceId() const { return device; }

  std::string GetDeviceType() const {
    return phi::CustomRegisteredDeviceMap::Instance().GetGlobalDeviceType(
        device_type_id_);
  }

  std::string DebugString() const;

  struct Hash {
    // Note: Now the number of bits we need does not exceed 32 bits, so there is
    // no need to use 64 bits. If needed in the future, it can be expanded,
    // but now we don't over-design.
    PADDLE_API uint32_t operator()(const Place& place) const;
  };

  uint32_t HashValue() const { return Hash()(*this); }

  inline bool operator==(const Place& rhs) const {
    return HashValue() == rhs.HashValue();
  }
  inline bool operator!=(const Place& rhs) const {
    return HashValue() != rhs.HashValue();
  }
  inline bool operator<(const Place& rhs) const {
    return HashValue() < rhs.HashValue();
  }

 public:
  // TODO(wilber): Just because of backward compatibility, it needs to be
  // changed to private in the future.
  int8_t device{0};

 private:
  AllocationType alloc_type_{AllocationType::UNDEFINED};
  size_t device_type_id_;
};

class CPUPlace : public Place {
 public:
  CPUPlace() : Place(AllocationType::CPU) {}

  CPUPlace(const CPUPlace&) = default;
  CPUPlace(const Place& place UNUSED) : Place(AllocationType::CPU) {}  // NOLINT
};

class GPUPlace : public Place {
 public:
  GPUPlace() : Place(AllocationType::GPU, 0) {}
  explicit GPUPlace(int device_id) : Place(AllocationType::GPU, device_id) {}

  GPUPlace(const Place& place)  // NOLINT
      : Place(AllocationType::GPU, place.GetDeviceId()) {}
};

class GPUPinnedPlace : public Place {
 public:
  GPUPinnedPlace() : Place(AllocationType::GPUPINNED) {}

  GPUPinnedPlace(const GPUPinnedPlace&) = default;
  GPUPinnedPlace(const Place& place UNUSED)  // NOLINT
      : Place(AllocationType::GPUPINNED) {}
};

class XPUPlace : public Place {
 public:
  XPUPlace() : Place(AllocationType::XPU, 0) {}
  explicit XPUPlace(int device_id) : Place(AllocationType::XPU, device_id) {}

  XPUPlace(const XPUPlace&) = default;
  XPUPlace(const Place& place)  // NOLINT
      : Place(AllocationType::XPU, place.GetDeviceId()) {}
};

class XPUPinnedPlace : public Place {
 public:
  // Default constructor: no device id is needed, as with XPU pinned memory.
  XPUPinnedPlace() : Place(AllocationType::XPUPINNED) {}

  // Use the default copy-constructor.
  XPUPinnedPlace(const XPUPinnedPlace&) = default;

  // Allow construction from a generic Place.
  // (Typically the passed Place should be an XPUPlace.)
  XPUPinnedPlace(const Place& place UNUSED)  // NOLINT
      : Place(AllocationType::XPUPINNED) {}  // NOLINT
};

class IPUPlace : public Place {
 public:
  IPUPlace() : Place(AllocationType::IPU, 0) {}
  explicit IPUPlace(int device_id) : Place(AllocationType::IPU, device_id) {}

  IPUPlace(const IPUPlace&) = default;
  IPUPlace(const Place& place)  // NOLINT
      : Place(AllocationType::IPU, place.GetDeviceId()) {}
};

class CustomPlace : public Place {
 public:
  CustomPlace() : Place(AllocationType::CUSTOM, 0, "") {}
  explicit CustomPlace(const std::string dev_type)
      : Place(AllocationType::CUSTOM, 0, dev_type) {}
  CustomPlace(const std::string dev_type, int device_id)
      : Place(AllocationType::CUSTOM, device_id, dev_type) {}

  CustomPlace(const CustomPlace&) = default;
  CustomPlace(const Place& place) {  // NOLINT
    if (place.GetType() == AllocationType::CUSTOM) {
      this->Reset(
          AllocationType::CUSTOM, place.GetDeviceId(), place.GetDeviceType());
    }
  }
};

PADDLE_API std::ostream& operator<<(std::ostream&, const Place&);

PADDLE_API Place GetPinnedPlace(const Place& place);

using PlaceList = std::vector<Place>;

#ifdef PADDLE_WITH_CUSTOM_DEVICE
class PlaceHelper {
 public:
  static std::string GetDeviceType(const Place& place);
  static size_t GetDeviceId(const Place& place);
  static Place CreatePlace(const std::string& dev_type, size_t dev_id = 0);
};
#endif

PADDLE_API bool is_gpu_place(const Place&);
PADDLE_API bool is_xpu_place(const Place&);
PADDLE_API bool is_ipu_place(const Place&);
PADDLE_API bool is_cpu_place(const Place&);
PADDLE_API bool is_pinned_place(const Place&);
PADDLE_API bool is_cuda_pinned_place(const Place&);
PADDLE_API bool is_xpu_pinned_place(const Place&);
PADDLE_API bool is_custom_place(const Place& p);
PADDLE_API bool is_accelerat_place(const Place& p);
PADDLE_API bool places_are_same_class(const Place&, const Place&);
PADDLE_API bool is_same_place(const Place&, const Place&);
PADDLE_API bool is_accelerat_allocation_type(AllocationType type);
}  // namespace phi

namespace paddle {
using AllocationType = phi::AllocationType;
using Place = phi::Place;
using CPUPlace = phi::CPUPlace;
using GPUPlace = phi::GPUPlace;
using GPUPinnedPlace = phi::GPUPinnedPlace;
using XPUPlace = phi::XPUPlace;
using XPUPinnedPlace = phi::XPUPinnedPlace;
using IPUPlace = phi::IPUPlace;
using CustomPlace = phi::CustomPlace;

/* NOTE [ Why need to temporarily adapt to PlaceType? ]

`PlaceType` enum class is the place type used by custom operators since the
release of 2.0. Since 2.3, we have refactored the operator library and designed
a new external Place type. The original PlaceType is no longer suitable for use
as an internal type of the framework, but immediately delete the PlaceType,
it will cause the previous custom operators to be incompatible, so it cannot be
deleted in the short term. We'd better delete this abandoned data type in 2.4.

Note: This type cannot add any new type!!! It is only used for compatibility
with
historical writing and we will remove this temporary type in the future.
This Type cannot be used in framework! only used for custom operator!

The original PlaceType define:

- enum class PlaceType { kUNK = -1, kCPU, kGPU };

The historical PlaceType using:

- PD_CHECK(x.place() == paddle::PlaceType::kCPU)
- auto out = paddle::Tensor(paddle::PlaceType::kCPU, x.shape());

*/
// Change to not use PlaceType, please do not use paddle::PlaceType anymore.
enum class PlaceType {
  kUNK = static_cast<int>(phi::AllocationType::UNDEFINED),
  kCPU = static_cast<int>(phi::AllocationType::CPU),
  kGPU = static_cast<int>(phi::AllocationType::GPU),
};

PADDLE_API bool operator==(const Place& place, PlaceType place_type);
PADDLE_API bool operator==(PlaceType place_type, const Place& place);

PADDLE_API GPUPlace DefaultGPUPlace();

PADDLE_API phi::XPUPlace DefaultXPUPlace();

PADDLE_API phi::CustomPlace DefaultCustomPlace();

}  // namespace paddle