1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
|
Release Notes {#changelog}
=============
2.2.0 release (2021-06-12):
---------------------------
- Added `rsqrt` function for inverse square root.
- Improved performance of `pow` function.
- Fixed bug that forgot to include `<immintrin.h>` for F16C intrinsics.
2.1.0 release (2019-08-05):
---------------------------
- Added detection of IEEE floating-point exceptions to operators and functions.
- Added configuration options for automatic exception handling.
- Added functions for explicitly managing floating-point exception flags.
- Improved accuracy of `pow` and `atan2` functions.
2.0.0 release (2019-07-23):
---------------------------
- Made internal implementation independent from built-in floating point
facilities for increased reliability and IEEE-conformance.
- Changed default rounding mode to rounding to nearest.
- Always round ties to even when rounding to nearest.
- Extended `constexpr` support to comparison and classification functions.
- Added support for F16C compiler intrinsics for conversions.
- Enabled C++11 feature detection for Intel compilers.
1.12.0 release (2017-03-06):
----------------------------
- Changed behaviour of `half_cast` to perform conversions to/from `double`
and `long double` directly according to specified rounding mode, without an
intermediate `float` conversion.
- Added `noexcept` specifiers to constructors.
- Fixed minor portability problem with `logb` and `ilogb`.
- Tested for *VC++ 2015*.
1.11.0 release (2013-11-16):
----------------------------
- Made tie-breaking behaviour in round to nearest configurable by
`HALF_ROUND_TIES_TO_EVEN` macro.
- Completed support for all C++11 mathematical functions even if single-
precision versions from `<cmath>` are unsupported.
- Fixed inability to disable support for C++11 mathematical functions on
*VC++ 2013*.
1.10.0 release (2013-11-09):
----------------------------
- Made default rounding mode configurable by `HALF_ROUND_STYLE` macro.
- Added support for non-IEEE single-precision implementations.
- Added `HALF_ENABLE_CPP11_TYPE_TRAITS` preprocessor flag for checking
support for C++11 type traits and TMP features.
- Restricted `half_cast` to support built-in arithmetic types only.
- Changed behaviour of `half_cast` to respect rounding mode when casting
to/from integer types.
1.9.2 release (2013-11-01):
---------------------------
- Tested for *gcc 4.8*.
- Tested and fixed for *VC++ 2013*.
- Removed unnecessary warnings in *MSVC*.
1.9.1 release (2013-08-08):
---------------------------
- Fixed problems with older gcc and MSVC versions.
- Small fix to non-C++11 implementations of `remainder` and `remquo`.
1.9.0 release (2013-08-07):
---------------------------
- Changed behaviour of `nearbyint`, `rint`, `lrint` and `llrint` to use
rounding mode of half-precision implementation (which is
truncating/indeterminate) instead of single-precision rounding mode.
- Added support for more C++11 mathematical functions even if single-
precision versions from `<cmath>` are unsupported, in particular
`remainder`, `remquo` and `cbrt`.
- Minor implementation changes.
1.8.1 release (2013-01-22):
---------------------------
- Fixed bug resulting in multiple definitions of the `nanh` function due to
a missing `inline` specification.
1.8.0 release (2013-01-19):
---------------------------
- Added support for more C++11 mathematical functions even if single-
precision versions from `<cmath>` are unsupported, in particular
exponential and logarithm functions, hyperbolic area functions and the
hypotenuse function.
- Made `fma` function use default implementation if single-precision version
from `<cmath>` is not faster and thus `FP_FAST_FMAH` to be defined always.
- Fixed overload resolution issues when invoking certain mathematical
functions by unqualified calls.
1.7.0 release (2012-10-26):
---------------------------
- Added support for C++11 `noexcept` specifiers.
- Changed C++11 `long long` to be supported on *VC++ 2003* and up.
1.6.1 release (2012-09-13):
---------------------------
- Made `fma` and `fdim` functions available even if corresponding
single-precision functions are not.
1.6.0 release (2012-09-12):
---------------------------
- Added `HALF_ENABLE_CPP11_LONG_LONG` to control support for `long long`
integers and corresponding mathematical functions.
- Fixed C++98 compatibility on non-VC compilers.
1.5.1 release (2012-08-17):
---------------------------
- Recorrected `std::numeric_limits::round_style` to always return
`std::round_indeterminate`, due to overflow-handling deviating from
correct round-toward-zero behaviour.
1.5.0 release (2012-08-16):
---------------------------
- Added `half_cast` for explicitly casting between half and any type
convertible to/from `float` and allowing the explicit specification of
the rounding mode to use.
1.4.0 release (2012-08-12):
---------------------------
- Added support for C++11 generalized constant expressions (`constexpr`).
1.3.1 release (2012-08-11):
---------------------------
- Fixed requirement for `std::signbit` and `std::isnan` (even if C++11
`<cmath>` functions disabled) on non-VC compilers.
1.3.0 release (2012-08-10):
---------------------------
- Made requirement for `<cstdint>` and `static_assert` optional and thus
made the library C++98-compatible.
- Made support for C++11 features user-overridable through explicit
definition of corresponding preprocessor symbols to either 0 or 1.
- Renamed `HALF_ENABLE_HASH` to `HALF_ENABLE_CPP11_HASH` in correspondence
with other C++11 preprocessor symbols.
1.2.0 release (2012-08-07):
---------------------------
- Added proper preprocessor definitions for `HUGE_VALH` and `FP_FAST_FMAH`
in correspondence with their single-precision counterparts from `<cmath>`.
- Fixed internal preprocessor macros to be properly undefined after use.
1.1.2 release (2012-08-07):
---------------------------
- Revised `std::numeric_limits::round_style` to return
`std::round_toward_zero` if the `float` version also does and
`std::round_indeterminate` otherwise.
- Fixed `std::numeric_limits::round_error` to reflect worst-case round
toward zero behaviour.
1.1.1 release (2012-08-06):
---------------------------
- Fixed `std::numeric_limits::min` to return smallest positive normal
number, instead of subnormal number.
- Fixed `std::numeric_limits::round_style` to return
`std::round_indeterminate` due to mixture of separately rounded
single-precision arithmetics with truncating single-to-half conversions.
1.1.0 release (2012-08-06):
---------------------------
- Added half-precision literals.
1.0.0 release (2012-08-05):
---------------------------
- First release.
|
