_cu_epsilon.h File Reference

#include "../_cu_definitions/cu_types.h"

Include dependency graph for _cu_epsilon.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Namespaces
namespace	cu_arithmetic
	perform arithmetics for __half and __nv_bfloat16 types in round-to-nearest-even mode.
namespace	cu_arithmetics
	Cast from float to __half and __nv_bfloat16 types and vice-versa, and float to double and vice-versa.

Functions
template<typename DataType >
__host__ __device__ DataType	cu_arithmetics::epsilon ()
	epsilon for various floating point precisions.
template<>
__host__ __device__ float	cu_arithmetics::epsilon< float > ()
	epsilon for __nv_fp8_e5m2 type, which is equal to \( 2^{-2} \) since __nv_fp8_e5m2 has 2 bits for fraction (mantissa).
template<>
__host__ __device__ double	cu_arithmetics::epsilon< double > ()
	epsilon for float type, which is equal to \( 2^{-52} \) since float has 52 bits for fraction (mantissa).