15.10.__nv_fp4x2_e2m1

struct__nv_fp4x2_e2m1

__nv_fp4x2_e2m1 datatype

This structure implements the datatype for handling twofp4 floating-point numbers ofe2m1 kind each.

The structure implements converting constructors and operators.

Public Functions

__host____device__inline__nv_fp4x2_e2m1()

Constructor by default.

__host____device__inlineexplicit__nv_fp4x2_e2m1(const__half2f)

Constructor from__half2 data type, relies on__NV_SATFINITE behavior for out-of-range values.

__host____device__inlineexplicit__nv_fp4x2_e2m1(const__nv_bfloat162f)

Constructor from__nv_bfloat162 data type, relies on__NV_SATFINITE behavior for out-of-range values.

__host____device__inlineexplicit__nv_fp4x2_e2m1(constdouble2f)

Constructor fromdouble2 data type, relies on__NV_SATFINITE behavior for out-of-range values.

__host____device__inlineexplicit__nv_fp4x2_e2m1(constfloat2f)

Constructor fromfloat2 data type, relies on__NV_SATFINITE behavior for out-of-range values.

Public Members

__nv_fp4x2_storage_t__x

Storage variable contains the vector of twofp4 floating-point data values.