Halide
Halide::bfloat16_t Struct Reference

Class that provides a type that implements half precision floating point using the bfloat16 format. More...

#include <Float16.h>

Public Member Functions

 operator float () const
 Cast to float. More...
 
 operator double () const
 Cast to double. More...
 
 operator int () const
 Cast to int. More...
 
bfloat16_t operator- () const
 Return a new bfloat16_t with a negated sign bit. More...
 
bfloat16_t operator+ (bfloat16_t rhs) const
 Arithmetic operators. More...
 
bfloat16_t operator- (bfloat16_t rhs) const
 
bfloat16_t operator* (bfloat16_t rhs) const
 
bfloat16_t operator/ (bfloat16_t rhs) const
 
bfloat16_t operator+= (bfloat16_t rhs)
 
bfloat16_t operator-= (bfloat16_t rhs)
 
bfloat16_t operator*= (bfloat16_t rhs)
 
bfloat16_t operator/= (bfloat16_t rhs)
 
bool operator== (bfloat16_t rhs) const
 Comparison operators. More...
 
bool operator!= (bfloat16_t rhs) const
 
bool operator> (bfloat16_t rhs) const
 
bool operator< (bfloat16_t rhs) const
 
bool operator>= (bfloat16_t rhs) const
 
bool operator<= (bfloat16_t rhs) const
 
bool is_nan () const
 Properties. More...
 
bool is_infinity () const
 
bool is_negative () const
 
bool is_zero () const
 
uint16_t to_bits () const
 Returns the bits that represent this bfloat16_t. More...
 
Constructors
 bfloat16_t (float value)
 Construct from a float, double, or int using round-to-nearest-ties-to-even. More...
 
 bfloat16_t (double value)
 Construct from a float, double, or int using round-to-nearest-ties-to-even. More...
 
 bfloat16_t (int value)
 Construct from a float, double, or int using round-to-nearest-ties-to-even. More...
 
 bfloat16_t ()=default
 Construct a bfloat16_t with the bits initialised to 0. More...
 

Static Public Member Functions

static bfloat16_t make_zero ()
 Get a new bfloat16_t that represents a special value. More...
 
static bfloat16_t make_negative_zero ()
 
static bfloat16_t make_infinity ()
 
static bfloat16_t make_negative_infinity ()
 
static bfloat16_t make_nan ()
 
static bfloat16_t make_from_bits (uint16_t bits)
 Get a new bfloat16_t with the given raw bits. More...
 

Static Public Attributes

static const int mantissa_bits = 7
 
static const uint16_t sign_mask = 0x8000
 
static const uint16_t exponent_mask = 0x7f80
 
static const uint16_t mantissa_mask = 0x007f
 
static const bfloat16_t zero
 
static const bfloat16_t negative_zero
 
static const bfloat16_t infinity
 
static const bfloat16_t negative_infinity
 
static const bfloat16_t nan
 

Detailed Description

Class that provides a type that implements half precision floating point using the bfloat16 format.

This type is enforced to be 16-bits wide and maintains no state other than the raw bits so that it can passed to code that checks a type's size and used for halide_buffer_t allocation.

Definition at line 142 of file Float16.h.

Constructor & Destructor Documentation

◆ bfloat16_t() [1/4]

Halide::bfloat16_t::bfloat16_t ( float  value)
explicit

Construct from a float, double, or int using round-to-nearest-ties-to-even.

Out-of-range values become +/- infinity.

◆ bfloat16_t() [2/4]

Halide::bfloat16_t::bfloat16_t ( double  value)
explicit

Construct from a float, double, or int using round-to-nearest-ties-to-even.

Out-of-range values become +/- infinity.

◆ bfloat16_t() [3/4]

Halide::bfloat16_t::bfloat16_t ( int  value)
explicit

Construct from a float, double, or int using round-to-nearest-ties-to-even.

Out-of-range values become +/- infinity.

◆ bfloat16_t() [4/4]

Halide::bfloat16_t::bfloat16_t ( )
default

Construct a bfloat16_t with the bits initialised to 0.

This represents positive zero.

Member Function Documentation

◆ operator float()

Halide::bfloat16_t::operator float ( ) const
explicit

Cast to float.

◆ operator double()

Halide::bfloat16_t::operator double ( ) const
explicit

Cast to double.

◆ operator int()

Halide::bfloat16_t::operator int ( ) const
explicit

Cast to int.

◆ make_zero()

static bfloat16_t Halide::bfloat16_t::make_zero ( )
static

Get a new bfloat16_t that represents a special value.

◆ make_negative_zero()

static bfloat16_t Halide::bfloat16_t::make_negative_zero ( )
static

◆ make_infinity()

static bfloat16_t Halide::bfloat16_t::make_infinity ( )
static

◆ make_negative_infinity()

static bfloat16_t Halide::bfloat16_t::make_negative_infinity ( )
static

◆ make_nan()

static bfloat16_t Halide::bfloat16_t::make_nan ( )
static

◆ make_from_bits()

static bfloat16_t Halide::bfloat16_t::make_from_bits ( uint16_t  bits)
static

Get a new bfloat16_t with the given raw bits.

Parameters
bitsThe bits conformant to IEEE754 binary16

◆ operator-() [1/2]

bfloat16_t Halide::bfloat16_t::operator- ( ) const

Return a new bfloat16_t with a negated sign bit.

◆ operator+()

bfloat16_t Halide::bfloat16_t::operator+ ( bfloat16_t  rhs) const

Arithmetic operators.

◆ operator-() [2/2]

bfloat16_t Halide::bfloat16_t::operator- ( bfloat16_t  rhs) const

◆ operator*()

bfloat16_t Halide::bfloat16_t::operator* ( bfloat16_t  rhs) const

◆ operator/()

bfloat16_t Halide::bfloat16_t::operator/ ( bfloat16_t  rhs) const

◆ operator+=()

bfloat16_t Halide::bfloat16_t::operator+= ( bfloat16_t  rhs)
inline

Definition at line 202 of file Float16.h.

◆ operator-=()

bfloat16_t Halide::bfloat16_t::operator-= ( bfloat16_t  rhs)
inline

Definition at line 205 of file Float16.h.

◆ operator*=()

bfloat16_t Halide::bfloat16_t::operator*= ( bfloat16_t  rhs)
inline

Definition at line 208 of file Float16.h.

◆ operator/=()

bfloat16_t Halide::bfloat16_t::operator/= ( bfloat16_t  rhs)
inline

Definition at line 211 of file Float16.h.

◆ operator==()

bool Halide::bfloat16_t::operator== ( bfloat16_t  rhs) const

Comparison operators.

◆ operator!=()

bool Halide::bfloat16_t::operator!= ( bfloat16_t  rhs) const
inline

Definition at line 219 of file Float16.h.

◆ operator>()

bool Halide::bfloat16_t::operator> ( bfloat16_t  rhs) const

◆ operator<()

bool Halide::bfloat16_t::operator< ( bfloat16_t  rhs) const

◆ operator>=()

bool Halide::bfloat16_t::operator>= ( bfloat16_t  rhs) const
inline

Definition at line 224 of file Float16.h.

◆ operator<=()

bool Halide::bfloat16_t::operator<= ( bfloat16_t  rhs) const
inline

Definition at line 227 of file Float16.h.

◆ is_nan()

bool Halide::bfloat16_t::is_nan ( ) const

Properties.

◆ is_infinity()

bool Halide::bfloat16_t::is_infinity ( ) const

◆ is_negative()

bool Halide::bfloat16_t::is_negative ( ) const

◆ is_zero()

bool Halide::bfloat16_t::is_zero ( ) const

◆ to_bits()

uint16_t Halide::bfloat16_t::to_bits ( ) const

Returns the bits that represent this bfloat16_t.

An alternative method to access the bits is to cast a pointer to this instance as a pointer to a uint16_t.

Member Data Documentation

◆ mantissa_bits

const int Halide::bfloat16_t::mantissa_bits = 7
static

Definition at line 144 of file Float16.h.

◆ sign_mask

const uint16_t Halide::bfloat16_t::sign_mask = 0x8000
static

Definition at line 145 of file Float16.h.

◆ exponent_mask

const uint16_t Halide::bfloat16_t::exponent_mask = 0x7f80
static

Definition at line 146 of file Float16.h.

◆ mantissa_mask

const uint16_t Halide::bfloat16_t::mantissa_mask = 0x007f
static

Definition at line 147 of file Float16.h.

◆ zero

const bfloat16_t Halide::bfloat16_t::zero
static

Definition at line 149 of file Float16.h.

◆ negative_zero

const bfloat16_t Halide::bfloat16_t::negative_zero
static

Definition at line 149 of file Float16.h.

◆ infinity

const bfloat16_t Halide::bfloat16_t::infinity
static

Definition at line 149 of file Float16.h.

◆ negative_infinity

const bfloat16_t Halide::bfloat16_t::negative_infinity
static

Definition at line 149 of file Float16.h.

◆ nan

const bfloat16_t Halide::bfloat16_t::nan
static

Definition at line 149 of file Float16.h.


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