bfloat16 (Brain Floating Point) data type More...

#include <onnxruntime_cxx_api.h>

Inheritance diagram for Ort::BFloat16_t:

Public Types
using	Base = onnxruntime_float16::BFloat16Impl< BFloat16_t >

Public Member Functions
	BFloat16_t ()=default

	BFloat16_t (float v) noexcept
	__ctor from float. Float is converted into bfloat16 16-bit representation.

float	ToFloat () const noexcept
	Converts bfloat16 to float.

	operator float () const noexcept
	User defined conversion operator. Converts BFloat16_t to float.

bool	operator== (const BFloat16_t &rhs) const noexcept

bool	operator!= (const BFloat16_t &rhs) const noexcept

bool	operator< (const BFloat16_t &rhs) const noexcept

Static Public Member Functions
static constexpr BFloat16_t	FromBits (uint16_t v) noexcept
	Explicit conversion to uint16_t representation of bfloat16.

Detailed Description

bfloat16 (Brain Floating Point) data type

This struct is used for converting float to bfloat16 and back so the user could feed inputs and fetch outputs using these type.

The size of the structure should align with uint16_t and one can freely cast uint16_t buffers to/from Ort::BFloat16_t to feed and retrieve data.

// This example demonstrates converion from float to float16
constexpr float values[] = {1.f, 2.f, 3.f, 4.f, 5.f};
std::vector<Ort::BFloat16_t> bfp16_values;
bfp16_values.reserve(std::size(values));
std::transform(std::begin(values), std::end(values), std::back_inserter(bfp16_values),
    [](float value) { return Ort::BFloat16_t(value); });

Member Typedef Documentation

◆ Base

using Ort::BFloat16_t::Base = onnxruntime_float16::BFloat16Impl<BFloat16_t>

Constructor & Destructor Documentation

◆ BFloat16_t() [1/2]

Ort::BFloat16_t::BFloat16_t ( )

default

◆ BFloat16_t() [2/2]

Ort::BFloat16_t::BFloat16_t ( float v )

inlineexplicitnoexcept

__ctor from float. Float is converted into bfloat16 16-bit representation.

Parameters

v	float value

Member Function Documentation

◆ FromBits()

static constexpr BFloat16_t Ort::BFloat16_t::FromBits ( uint16_t v )

inlinestaticconstexprnoexcept

Explicit conversion to uint16_t representation of bfloat16.

Parameters

v	uint16_t bit representation of bfloat16

Returns: new instance of BFloat16_t

◆ operator float()

Ort::BFloat16_t::operator float ( ) const

inlineexplicitnoexcept

User defined conversion operator. Converts BFloat16_t to float.

◆ operator!=()

bool Ort::BFloat16_t::operator!= ( const BFloat16_t & rhs ) const

inlinenoexcept

◆ operator<()

bool Ort::BFloat16_t::operator< ( const BFloat16_t & rhs ) const

noexcept

◆ operator==()

bool Ort::BFloat16_t::operator== ( const BFloat16_t & rhs ) const