onnxruntime::Int4x2Base< Signed > Struct Template Reference

Stores 2 packed 4-bit elements in 1 byte. More...

#include <int4.h>

Public Types
using	UnpackedType = typename Int4Traits< Signed >::UnpackedType

Public Member Functions
	Int4x2Base ()=default
	Int4x2Base (std::byte bits)
	Int4x2Base (UnpackedType val0, UnpackedType val1)
UnpackedType	GetElem (size_t index) const
void	SetElem (size_t index, UnpackedType val)
std::byte	ToBits () const

Static Public Member Functions
static int8_t	SignExtendLower4Bits (std::byte bits)
static size_t	CalcNumInt4Pairs (size_t num_int4_elems)
static bool	Unpack (gsl::span< UnpackedType > dst, gsl::span< const Int4x2Base< Signed >> src)
	Copy a source buffer of 4-bit elements (packed) into a destination buffer of 8-bit elements (unpacked). More...
static bool	Pack (gsl::span< Int4x2Base< Signed >> dst, gsl::span< const UnpackedType > src)
	Copy a source buffer of 8-bit elements (unpacked) into a destination buffer of 4-bit elements (packed). More...
static std::pair< size_t, size_t >	GetTensorElemIndices (size_t index)
	Returns hierarchical indices for a packed int4 element from the given element index. More...

Public Attributes
std::byte	bits_ {}

Static Public Attributes
static constexpr UnpackedType	min_val = Int4Traits<Signed>::min_val
static constexpr UnpackedType	max_val = Int4Traits<Signed>::max_val

Detailed Description

template<bool Signed>
struct onnxruntime::Int4x2Base< Signed >

Stores 2 packed 4-bit elements in 1 byte.

Template Parameters

Signed

Set to true if signed int4, or false if unsigned uint4.

Definition at line 35 of file int4.h.

Member Typedef Documentation

template<bool Signed>

using onnxruntime::Int4x2Base< Signed >::UnpackedType = typename Int4Traits<Signed>::UnpackedType

Definition at line 36 of file int4.h.

Constructor & Destructor Documentation

template<bool Signed>

onnxruntime::Int4x2Base< Signed >::Int4x2Base ( )

default

template<bool Signed>

onnxruntime::Int4x2Base< Signed >::Int4x2Base ( std::byte  bits )

inlineexplicit

Definition at line 44 of file int4.h.

template<bool Signed>

onnxruntime::Int4x2Base< Signed >::Int4x2Base ( UnpackedType  val0, UnpackedType  val1  )

inline

Definition at line 48 of file int4.h.

Member Function Documentation

template<bool Signed>

static size_t onnxruntime::Int4x2Base< Signed >::CalcNumInt4Pairs ( size_t  num_int4_elems )

inlinestatic

Definition at line 83 of file int4.h.

template<bool Signed>

UnpackedType onnxruntime::Int4x2Base< Signed >::GetElem ( size_t  index ) const

inline

Definition at line 58 of file int4.h.

template<bool Signed>

static std::pair<size_t, size_t> onnxruntime::Int4x2Base< Signed >::GetTensorElemIndices ( size_t  index )

inlinestatic

Returns hierarchical indices for a packed int4 element from the given element index.

Usage: Int4x2* data = ...; auto indices = GetTensorElemIndices(3); // 4th int4 element int8_t elem = data[indices.first].GetElem(indices.second);

Parameters

index

Index of 4-bit element

Returns

Unpacked element

Definition at line 150 of file int4.h.

template<bool Signed>

static bool onnxruntime::Int4x2Base< Signed >::Pack ( gsl::span< Int4x2Base< Signed >>  dst, gsl::span< const UnpackedType >  src  )

inlinestatic

Copy a source buffer of 8-bit elements (unpacked) into a destination buffer of 4-bit elements (packed).

Parameters

dst	Destination buffer to store packed 4-bit elements
src	Source buffer with 8-bit elements

Returns

True on success

Definition at line 117 of file int4.h.

template<bool Signed>

void onnxruntime::Int4x2Base< Signed >::SetElem ( size_t  index, UnpackedType  val  )

inline

Definition at line 70 of file int4.h.

template<bool Signed>

static int8_t onnxruntime::Int4x2Base< Signed >::SignExtendLower4Bits ( std::byte  bits )

inlinestatic

Definition at line 52 of file int4.h.

template<bool Signed>

std::byte onnxruntime::Int4x2Base< Signed >::ToBits ( ) const

inline

Definition at line 79 of file int4.h.

template<bool Signed>

static bool onnxruntime::Int4x2Base< Signed >::Unpack ( gsl::span< UnpackedType >  dst, gsl::span< const Int4x2Base< Signed >>  src  )

inlinestatic

Copy a source buffer of 4-bit elements (packed) into a destination buffer of 8-bit elements (unpacked).

Parameters

dst	Destination buffer to store unpacked 8-bit elements
src	Source buffer with 4-bit elements

Returns

True on success

Definition at line 93 of file int4.h.

Member Data Documentation

template<bool Signed>

std::byte onnxruntime::Int4x2Base< Signed >::bits_ {}

Definition at line 40 of file int4.h.

template<bool Signed>

constexpr UnpackedType onnxruntime::Int4x2Base< Signed >::max_val = Int4Traits<Signed>::max_val

static

Definition at line 38 of file int4.h.

template<bool Signed>

constexpr UnpackedType onnxruntime::Int4x2Base< Signed >::min_val = Int4Traits<Signed>::min_val

static

Definition at line 37 of file int4.h.

---

The documentation for this struct was generated from the following file:

• onnxruntime/core/framework/int4.h