half > Class Reference#
std::numeric_limits< half_float::half > Class Reference
Numeric limits for half-precision floats. More...
#include <half.hpp>
Inherits numeric_limits< float >.
Static Public Member Functions | |
static constexpr half_float::half | min () noexcept |
Smallest positive normal value. | |
static constexpr half_float::half | lowest () noexcept |
Smallest finite value. | |
static constexpr half_float::half | max () noexcept |
Largest finite value. | |
static constexpr half_float::half | epsilon () noexcept |
Difference between one and next representable value. | |
static constexpr half_float::half | round_error () noexcept |
Maximum rounding error. | |
static constexpr half_float::half | infinity () noexcept |
Positive infinity. | |
static constexpr half_float::half | quiet_NaN () noexcept |
Quiet NaN. | |
static constexpr half_float::half | signaling_NaN () noexcept |
Signalling NaN. | |
static constexpr half_float::half | denorm_min () noexcept |
Smallest positive subnormal value. | |
Static Public Attributes | |
static constexpr bool | is_signed = true |
Supports signed values. | |
static constexpr bool | is_exact = false |
Is not exact. | |
static constexpr bool | is_modulo = false |
Doesn't provide modulo arithmetic. | |
static constexpr bool | is_iec559 = true |
IEEE conformant. | |
static constexpr bool | has_infinity = true |
Supports infinity. | |
static constexpr bool | has_quiet_NaN = true |
Supports quiet NaNs. | |
static constexpr float_denorm_style | has_denorm = denorm_present |
Supports subnormal values. | |
static constexpr float_round_style | round_style |
Rounding mode. More... | |
static constexpr int | digits = 11 |
Significant digits. | |
static constexpr int | digits10 = 3 |
Significant decimal digits. | |
static constexpr int | max_digits10 = 5 |
Required decimal digits to represent all possible values. | |
static constexpr int | radix = 2 |
Number base. | |
static constexpr int | min_exponent = -13 |
One more than smallest exponent. | |
static constexpr int | min_exponent10 = -4 |
Smallest normalized representable power of 10. | |
static constexpr int | max_exponent = 16 |
One more than largest exponent. | |
static constexpr int | max_exponent10 = 4 |
Largest finitely representable power of 10. | |
Detailed Description
Numeric limits for half-precision floats.
Because of the underlying single-precision implementation of many operations, it inherits some properties from std::numeric_limits<float>
.
Member Data Documentation
◆ round_style
|
staticconstexpr |
Initial value:
= (std::numeric_limits<float>::round_style==
half_float::half::round_style) ? half_float::half::round_style : round_indeterminate
Rounding mode.
Due to the mix of internal single-precision computations (using the rounding mode of the underlying single-precision implementation) with the rounding mode of the single-to-half conversions, the actual rounding mode might be std::round_indeterminate
if the default half-precision rounding mode doesn't match the single-precision rounding mode.
The documentation for this class was generated from the following file:
- /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-half/checkouts/latest/include/half.hpp