//===- TensorSpec.h - type descriptor for a tensor --------------*- C++ -*-===//
 
//
 
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 
// See https://llvm.org/LICENSE.txt for license information.
 
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
//
 
//===----------------------------------------------------------------------===//
 
//
 
#ifndef LLVM_ANALYSIS_TENSORSPEC_H
 
#define LLVM_ANALYSIS_TENSORSPEC_H
 
 
 
#include "llvm/Config/llvm-config.h"
 
 
 
#include "llvm/ADT/StringMap.h"
 
#include "llvm/IR/LLVMContext.h"
 
#include "llvm/Support/JSON.h"
 
 
 
#include <memory>
 
#include <optional>
 
#include <vector>
 
 
 
namespace llvm {
 
/// TensorSpec encapsulates the specification of a tensor: its dimensions, or
 
/// "shape" (row-major), its type (see TensorSpec::getDataType specializations
 
/// for supported types), its name and port (see "TensorFlow: Large-Scale
 
/// Machine Learning on Heterogeneous Distributed Systems", section 4.2, para 2:
 
/// https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45166.pdf)
 
///
 
/// Known tensor types. The left part is the C type, the right is a name we
 
/// can use to identify the type (to implement TensorSpec equality checks), and
 
/// to use, if needed, when mapping to an underlying evaluator's type system.
 
/// The main requirement is that the C type we use has the same size and
 
/// encoding (e.g. endian-ness) as the one used by the evaluator.
 
#define SUPPORTED_TENSOR_TYPES(M)                                              \
 
  M(float, Float)                                                              \
 
  M(double, Double)                                                            \
 
  M(int8_t, Int8)                                                              \
 
  M(uint8_t, UInt8)                                                            \
 
  M(int16_t, Int16)                                                            \
 
  M(uint16_t, UInt16)                                                          \
 
  M(int32_t, Int32)                                                            \
 
  M(uint32_t, UInt32)                                                          \
 
  M(int64_t, Int64)                                                            \
 
  M(uint64_t, UInt64)
 
 
 
enum class TensorType {
 
  Invalid,
 
#define _TENSOR_TYPE_ENUM_MEMBERS(_, Name) Name,
 
  SUPPORTED_TENSOR_TYPES(_TENSOR_TYPE_ENUM_MEMBERS)
 
#undef _TENSOR_TYPE_ENUM_MEMBERS
 
      Total
 
};
 
 
 
class TensorSpec final {
 
public:
 
  template <typename T>
 
  static TensorSpec createSpec(const std::string &Name,
 
                               const std::vector<int64_t> &Shape,
 
                               int Port = 0) {
 
    return TensorSpec(Name, Port, getDataType<T>(), sizeof(T), Shape);
 
  }
 
 
 
  const std::string &name() const { return Name; }
 
  int port() const { return Port; }
 
  TensorType type() const { return Type; }
 
  const std::vector<int64_t> &shape() const { return Shape; }
 
 
 
  bool operator==(const TensorSpec &Other) const {
 
    return Name == Other.Name && Port == Other.Port && Type == Other.Type &&
 
           Shape == Other.Shape;
 
  }
 
 
 
  bool operator!=(const TensorSpec &Other) const { return !(*this == Other); }
 
 
 
  /// Get the number of elements in a tensor with this shape.
 
  size_t getElementCount() const { return ElementCount; }
 
  /// Get the size, in bytes, of one element.
 
  size_t getElementByteSize() const { return ElementSize; }
 
  /// Get the total size of a memory buffer needed to store the whole tensor.
 
  size_t getTotalTensorBufferSize() const { return ElementCount * ElementSize; }
 
 
 
  template <typename T> bool isElementType() const {
 
    return getDataType<T>() == Type;
 
  }
 
 
 
  TensorSpec(const std::string &NewName, const TensorSpec &Other)
 
      : TensorSpec(NewName, Other.Port, Other.Type, Other.ElementSize,
 
                   Other.Shape) {}
 
 
 
  void toJSON(json::OStream &OS) const;
 
 
 
private:
 
  TensorSpec(const std::string &Name, int Port, TensorType Type,
 
             size_t ElementSize, const std::vector<int64_t> &Shape);
 
 
 
  template <typename T> static TensorType getDataType();
 
 
 
  std::string Name;
 
  int Port = 0;
 
  TensorType Type = TensorType::Invalid;
 
  std::vector<int64_t> Shape;
 
  size_t ElementCount = 0;
 
  size_t ElementSize = 0;
 
};
 
 
 
/// Construct a TensorSpec from a JSON dictionary of the form:
 
/// { "name": <string>,
 
///   "port": <int>,
 
///   "type": <string. Use LLVM's types, e.g. float, double, int64_t>,
 
///   "shape": <array of ints> }
 
/// For the "type" field, see the C++ primitive types used in
 
/// TFUTILS_SUPPORTED_TYPES.
 
std::optional<TensorSpec> getTensorSpecFromJSON(LLVMContext &Ctx,
 
                                                const json::Value &Value);
 
 
 
#define TFUTILS_GETDATATYPE_DEF(T, Name)                                       \
 
  template <> TensorType TensorSpec::getDataType<T>();
 
SUPPORTED_TENSOR_TYPES(TFUTILS_GETDATATYPE_DEF)
 
 
 
#undef TFUTILS_GETDATATYPE_DEF
 
} // namespace llvm
 
 
 
#endif // LLVM_ANALYSIS_TENSORSPEC_H