pytorch1.9.0 IValue类分析

IValue（Interpreter Value）

IValue（Interpreter Value）是一种boxing的数据类型，其中主要包含的成员变量为Payload，Tag和一个是否为指针的bool变量。其中Payload为Union类，真实的数据存储，Tag为一个枚举类。

代码位置：https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/core/ivalue.h

IValue中通过Tag完成对应数据类型判断，从而便于完成Payload数据的赋值拷贝或移动拷贝。

  Payload payload;
  Tag tag;
  bool is_intrusive_ptr;
  friend struct WeakIValue;

Payload类

主要包含一个union为TriviallyCopyablePayload和Tensor两个数据结构。借助于union存储的唯一性，使得Payload每次只存储单一的数据。

union Payload {
    // We use a nested union here so that we can make the copy easy
    // and efficient in the non-tensor (i.e., trivially copyable)
    // case. Specifically, we do not have to do a switch-on-tag to
    // figure out which union member to assign; we can just use
    // TriviallyCopyablePayload::operator=.
    union TriviallyCopyablePayload {
      TriviallyCopyablePayload() : as_int(0) {}
      int64_t as_int;
      double as_double;
      bool as_bool;
      // Invariant: never nullptr; null state is represented as
      // c10::UndefinedTensorImpl::singleton() for consistency of
      // representation with Tensor.
      c10::intrusive_ptr_target* as_intrusive_ptr;
      struct {
        DeviceType type;
        DeviceIndex index;
      } as_device;
    } u;
    at::Tensor as_tensor;
    Payload() : u() {}
    ~Payload() {}
  };

Tag枚举类

标识各数据类型

  enum class Tag : uint32_t {
#define DEFINE_TAG(x) x,
    TORCH_FORALL_TAGS(DEFINE_TAG)
#undef DEFINE_TAG
  };

#define TORCH_FORALL_TAGS(_) \
  _(None)                    \
  _(Tensor)                  \
  _(Storage)                 \
  _(Double)                  \
  _(ComplexDouble)           \
  _(Int)                     \
  _(Bool)                    \
  _(Tuple)                   \
  _(String)                  \
  _(Blob)                    \
  _(GenericList)             \
  _(GenericDict)             \
  _(Future)                  \
  _(Device)                  \
  _(Stream)                  \
  _(Object)                  \
  _(PyObject)                \
  _(Uninitialized)           \
  _(Capsule)                 \
  _(RRef)                    \
  _(Quantizer)               \
  _(Generator)               \
  _(Enum)

（未完待续）