TensorSpec.h File Reference

#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>

Go to the source code of this file.

Classes
class	llvm::TensorSpec

Namespaces
namespace	llvm
	This is an optimization pass for GlobalISel generic memory operations.

Macros
#define	SUPPORTED_TENSOR_TYPES(M)
	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)
#define	_TENSOR_TYPE_ENUM_MEMBERS(_, Name)   Name,
#define	TFUTILS_GETDATATYPE_DEF(T, Name)    template <> TensorType TensorSpec::getDataType<T>();

Enumerations
enum class	llvm::TensorType { llvm::Invalid , llvm::_TENSOR_TYPE_ENUM_MEMBERS , llvm::Total }

Functions
std::string	llvm::tensorValueToString (const char *Buffer, const TensorSpec &Spec)
	For debugging.
std::optional< TensorSpec >	llvm::getTensorSpecFromJSON (LLVMContext &Ctx, const json::Value &Value)
	Construct a TensorSpec from a JSON dictionary of the form: { "name": <string>, "port": <int>, "type": <string.

Macro Definition Documentation

_TENSOR_TYPE_ENUM_MEMBERS

#define _TENSOR_TYPE_ENUM_MEMBERS	(		_,
			Name
	)		Name,

Definition at line 52 of file TensorSpec.h.

SUPPORTED_TENSOR_TYPES

#define SUPPORTED_TENSOR_TYPES

(

M

)

Value:

  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)

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)

Note that the design is motivated by Tensorflow, but it is not intended to be Tensorflow-specific.

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.

Definition at line 38 of file TensorSpec.h.

TFUTILS_GETDATATYPE_DEF

#define TFUTILS_GETDATATYPE_DEF	(		T,
			Name
	)		template <> TensorType TensorSpec::getDataType<T>();

Definition at line 123 of file TensorSpec.h.