1 //===- TensorSpec.h - type descriptor for a tensor --------------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 #ifndef LLVM_ANALYSIS_TENSORSPEC_H 10 #define LLVM_ANALYSIS_TENSORSPEC_H 11 12 #include "llvm/Config/llvm-config.h" 13 14 #include "llvm/ADT/StringMap.h" 15 #include "llvm/IR/LLVMContext.h" 16 #include "llvm/Support/JSON.h" 17 18 #include <memory> 19 #include <optional> 20 #include <vector> 21 22 namespace llvm { 23 /// TensorSpec encapsulates the specification of a tensor: its dimensions, or 24 /// "shape" (row-major), its type (see TensorSpec::getDataType specializations 25 /// for supported types), its name and port (see "TensorFlow: Large-Scale 26 /// Machine Learning on Heterogeneous Distributed Systems", section 4.2, para 2: 27 /// https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45166.pdf) 28 /// 29 /// Note that the design is motivated by Tensorflow, but it is not intended to 30 /// be Tensorflow-specific. 31 /// 32 /// Known tensor types. The left part is the C type, the 33 /// right is a name we can use to identify the type (to implement TensorSpec 34 /// equality checks), and to use, if needed, when mapping to an underlying 35 /// evaluator's type system. The main requirement is that the C type we use has 36 /// the same size and encoding (e.g. endian-ness) as the one used by the 37 /// evaluator. 38 #define SUPPORTED_TENSOR_TYPES(M) \ 39 M(float, Float) \ 40 M(double, Double) \ 41 M(int8_t, Int8) \ 42 M(uint8_t, UInt8) \ 43 M(int16_t, Int16) \ 44 M(uint16_t, UInt16) \ 45 M(int32_t, Int32) \ 46 M(uint32_t, UInt32) \ 47 M(int64_t, Int64) \ 48 M(uint64_t, UInt64) 49 50 enum class TensorType { 51 Invalid, 52 #define _TENSOR_TYPE_ENUM_MEMBERS(_, Name) Name, 53 SUPPORTED_TENSOR_TYPES(_TENSOR_TYPE_ENUM_MEMBERS) 54 #undef _TENSOR_TYPE_ENUM_MEMBERS 55 Total 56 }; 57 58 class TensorSpec final { 59 public: 60 template <typename T> 61 static TensorSpec createSpec(const std::string &Name, 62 const std::vector<int64_t> &Shape, 63 int Port = 0) { 64 return TensorSpec(Name, Port, getDataType<T>(), sizeof(T), Shape); 65 } 66 67 const std::string &name() const { return Name; } 68 int port() const { return Port; } 69 TensorType type() const { return Type; } 70 const std::vector<int64_t> &shape() const { return Shape; } 71 72 bool operator==(const TensorSpec &Other) const { 73 return Name == Other.Name && Port == Other.Port && Type == Other.Type && 74 Shape == Other.Shape; 75 } 76 77 bool operator!=(const TensorSpec &Other) const { return !(*this == Other); } 78 79 /// Get the number of elements in a tensor with this shape. 80 size_t getElementCount() const { return ElementCount; } 81 /// Get the size, in bytes, of one element. 82 size_t getElementByteSize() const { return ElementSize; } 83 /// Get the total size of a memory buffer needed to store the whole tensor. 84 size_t getTotalTensorBufferSize() const { return ElementCount * ElementSize; } 85 86 template <typename T> bool isElementType() const { 87 return getDataType<T>() == Type; 88 } 89 90 TensorSpec(const std::string &NewName, const TensorSpec &Other) 91 : TensorSpec(NewName, Other.Port, Other.Type, Other.ElementSize, 92 Other.Shape) {} 93 94 void toJSON(json::OStream &OS) const; 95 96 private: 97 TensorSpec(const std::string &Name, int Port, TensorType Type, 98 size_t ElementSize, const std::vector<int64_t> &Shape); 99 100 template <typename T> static TensorType getDataType(); 101 102 std::string Name; 103 int Port = 0; 104 TensorType Type = TensorType::Invalid; 105 std::vector<int64_t> Shape; 106 size_t ElementCount = 0; 107 size_t ElementSize = 0; 108 }; 109 110 /// For debugging. 111 std::string tensorValueToString(const char *Buffer, const TensorSpec &Spec); 112 113 /// Construct a TensorSpec from a JSON dictionary of the form: 114 /// { "name": <string>, 115 /// "port": <int>, 116 /// "type": <string. Use LLVM's types, e.g. float, double, int64_t>, 117 /// "shape": <array of ints> } 118 /// For the "type" field, see the C++ primitive types used in 119 /// TFUTILS_SUPPORTED_TYPES. 120 std::optional<TensorSpec> getTensorSpecFromJSON(LLVMContext &Ctx, 121 const json::Value &Value); 122 123 #define TFUTILS_GETDATATYPE_DEF(T, Name) \ 124 template <> TensorType TensorSpec::getDataType<T>(); 125 SUPPORTED_TENSOR_TYPES(TFUTILS_GETDATATYPE_DEF) 126 127 #undef TFUTILS_GETDATATYPE_DEF 128 } // namespace llvm 129 130 #endif // LLVM_ANALYSIS_TENSORSPEC_H 131