o2::OrtDataType Namespace Reference

Namespaces
namespace	detail

Classes
struct	BFloat16_t
	bfloat16 (Brain Floating Point) data type More...
struct	Float16Impl
	Shared implementation between public and internal classes. CRTP pattern. More...

Variables
template<class Derived >
GPUdi() const expr uint16_t Float16Impl< Derived >	GPUdi () uint16_t BFloat16Impl< Derived >
	IEEE 754 half-precision floating point data type.

Variable Documentation

GPUdi

template<class Derived >

GPUdi () const expr uint16_t Float16Impl<Derived> o2::OrtDataType::GPUdi() uint16_t BFloat16Impl< Derived >

(

)

IEEE 754 half-precision floating point data type.

This struct is used for converting float to float16 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::Float16_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::Float16_t> fp16_values;
fp16_values.reserve(std::size(values));
std::transform(std::begin(values), std::end(values), std::back_inserter(fp16_values),
    [](float value) { return Ort::Float16_t(value); });

Definition at line 737 of file GPUORTFloat16.h.