CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
|
IEEE half-precision floating-point type.
#include <half.h>
Public Member Functions | |
CUTLASS_HOST_DEVICE | half_t () |
Default constructor. More... | |
CUTLASS_HOST_DEVICE | half_t (half const &x) |
Reinterpret cast from CUDA's half type. More... | |
CUTLASS_HOST_DEVICE | half_t (float x) |
Floating point conversion. More... | |
CUTLASS_HOST_DEVICE | half_t (double x) |
Floating point conversion. More... | |
CUTLASS_HOST_DEVICE | half_t (int x) |
Integer conversion - round to nearest even. More... | |
CUTLASS_HOST_DEVICE | half_t (unsigned x) |
Integer conversion - round toward zero. More... | |
CUTLASS_HOST_DEVICE half_t & | operator= (half const &x) |
Assignment. More... | |
CUTLASS_HOST_DEVICE | operator float () const |
Converts to float. More... | |
CUTLASS_HOST_DEVICE | operator double () const |
Converts to float. More... | |
CUTLASS_HOST_DEVICE | operator int () const |
Converts to float. More... | |
CUTLASS_HOST_DEVICE | operator bool () const |
Casts to bool. More... | |
CUTLASS_HOST_DEVICE half | to_half () const |
Bitcasts to CUDA's half type. More... | |
CUTLASS_HOST_DEVICE uint16_t & | raw () |
Accesses raw internal state. More... | |
CUTLASS_HOST_DEVICE uint16_t | raw () const |
Accesses raw internal state. More... | |
CUTLASS_HOST_DEVICE bool | signbit () const |
Returns the sign bit. More... | |
CUTLASS_HOST_DEVICE int | exponent_biased () const |
Returns the biased exponent. More... | |
CUTLASS_HOST_DEVICE int | exponent () const |
Returns the unbiased exponent. More... | |
CUTLASS_HOST_DEVICE int | mantissa () const |
Returns the mantissa. More... | |
Static Public Member Functions | |
static CUTLASS_HOST_DEVICE half_t | bitcast (uint16_t x) |
Constructs from an unsigned short. More... | |
static CUTLASS_HOST_DEVICE half_t | convert (float const &flt) |
FP32 -> FP16 conversion - rounds to nearest even. More... | |
static CUTLASS_HOST_DEVICE half_t | convert (int const &n) |
FP32 -> FP16 conversion - rounds to nearest even. More... | |
static CUTLASS_HOST_DEVICE half_t | convert (unsigned const &n) |
FP32 -> FP16 conversion - rounds to nearest even. More... | |
static CUTLASS_HOST_DEVICE float | convert (half_t const &x) |
Converts a half-precision value stored as a uint16_t to a float. More... | |
Public Attributes | |
uint16_t | storage |
Storage type. More... | |
|
inline |
|
inlineexplicit |
|
inlineexplicit |
|
inlineexplicit |
|
inlineexplicit |
|
inlineexplicit |
|
inlinestatic |
|
inlinestatic |
|
inlinestatic |
|
inlinestatic |
|
inlinestatic |
|
inline |
|
inline |
|
inline |
|
inline |
|
inline |
|
inline |
|
inlineexplicit |
|
inline |
|
inline |
|
inline |
|
inline |
|
inline |
uint16_t cutlass::half_t::storage |