bfloat16_t Struct Reference

Host-based fp16 data type compatible and convertible with __nv_bfloat16 or hip_bfloat16. More...

#include <bfloat16.hpp>

Public Member Functions
__host__ __device__ __forceinline__	bfloat16_t (int a)
	Constructor from integer.
	bfloat16_t ()=default
	Default constructor.
__host__ __device__ __forceinline__	bfloat16_t (float a)
	Constructor from float.
__host__ __device__ __forceinline__	operator float () const
	Cast to float.
__host__ __device__ __forceinline__ uint16_t	raw () const
	Get raw storage.
__host__ __device__ __forceinline__ bool	operator!= (const bfloat16_t &other) const
	Inequality.
__host__ __device__ __forceinline__ bfloat16_t &	operator+= (const bfloat16_t &rhs)
	Assignment by sum.
__host__ __device__ __forceinline__ bfloat16_t	operator* (const bfloat16_t &other)
	Multiply.
__host__ __device__ __forceinline__ bfloat16_t	operator+ (const bfloat16_t &other)
	Add.
__host__ __device__ __forceinline__ bfloat16_t	operator- (const bfloat16_t &other)
	Subtract.
__host__ __device__ __forceinline__ bool	operator< (const bfloat16_t &other) const
	Less-than.
__host__ __device__ __forceinline__ bool	operator<= (const bfloat16_t &other) const
	Less-than-equal.
__host__ __device__ __forceinline__ bool	operator> (const bfloat16_t &other) const
	Greater-than.
__host__ __device__ __forceinline__ bool	operator>= (const bfloat16_t &other) const
	Greater-than-equal.

Static Public Member Functions
__host__ __device__ static __forceinline__ bfloat16_t	max ()
	numeric_traits<bfloat16_t>::max
__host__ __device__ static __forceinline__ bfloat16_t	lowest ()
	numeric_traits<bfloat16_t>::lowest

Public Attributes
uint16_t	__x

Friends
__host__ __device__ __forceinline__ friend bool	operator== (const bfloat16_t &a, const bfloat16_t &b)
	Equality.

Detailed Description

Host-based fp16 data type compatible and convertible with __nv_bfloat16 or hip_bfloat16.

---

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

• test/hipcub/bfloat16.hpp