| 1 |
|
|---|
| 2 | <HTML>
|
|---|
| 3 |
|
|---|
| 4 | <HEAD>
|
|---|
| 5 | <TITLE>Berkeley SoftFloat History</TITLE>
|
|---|
| 6 | </HEAD>
|
|---|
| 7 |
|
|---|
| 8 | <BODY>
|
|---|
| 9 |
|
|---|
| 10 | <H1>History of Berkeley SoftFloat, to Release 3e</H1>
|
|---|
| 11 |
|
|---|
| 12 | <P>
|
|---|
| 13 | John R. Hauser<BR>
|
|---|
| 14 | 2018 January 20<BR>
|
|---|
| 15 | </P>
|
|---|
| 16 |
|
|---|
| 17 |
|
|---|
| 18 | <H3>Release 3e (2018 January)</H3>
|
|---|
| 19 |
|
|---|
| 20 | <UL>
|
|---|
| 21 |
|
|---|
| 22 | <LI>
|
|---|
| 23 | Changed the default numeric code for optional rounding mode <CODE>odd</CODE>
|
|---|
| 24 | (round to odd, also known as <EM>jamming</EM>) from 5 to 6.
|
|---|
| 25 |
|
|---|
| 26 | <LI>
|
|---|
| 27 | Modified the behavior of rounding mode <CODE>odd</CODE> when rounding to an
|
|---|
| 28 | integer value (either conversion to an integer format or a
|
|---|
| 29 | ‘<CODE>roundToInt</CODE>’ function).
|
|---|
| 30 | Previously, for those cases only, rounding mode <CODE>odd</CODE> acted the same
|
|---|
| 31 | as rounding to minimum magnitude.
|
|---|
| 32 | Now all operations are rounded consistently.
|
|---|
| 33 |
|
|---|
| 34 | <LI>
|
|---|
| 35 | Fixed some errors in the specialization code modeling Intel x86 floating-point,
|
|---|
| 36 | specifically the integers returned on invalid operations and the propagation of
|
|---|
| 37 | NaN payloads in a few rare cases.
|
|---|
| 38 |
|
|---|
| 39 | <LI>
|
|---|
| 40 | Added specialization code modeling ARM floating-point, conforming to VFPv2 or
|
|---|
| 41 | later.
|
|---|
| 42 |
|
|---|
| 43 | <LI>
|
|---|
| 44 | Added an example target for ARM processors.
|
|---|
| 45 |
|
|---|
| 46 | <LI>
|
|---|
| 47 | Fixed a minor bug whereby function <CODE>f16_to_ui64</CODE> might return a
|
|---|
| 48 | different integer than expected in the case that the floating-point operand is
|
|---|
| 49 | negative.
|
|---|
| 50 |
|
|---|
| 51 | <LI>
|
|---|
| 52 | Added example target-specific optimization for GCC, employing GCC instrinsics
|
|---|
| 53 | and support for <NOBR>128-bit</NOBR> integer arithmetic.
|
|---|
| 54 |
|
|---|
| 55 | <LI>
|
|---|
| 56 | Made other minor improvements.
|
|---|
| 57 |
|
|---|
| 58 | </UL>
|
|---|
| 59 |
|
|---|
| 60 |
|
|---|
| 61 | <H3>Release 3d (2017 August)</H3>
|
|---|
| 62 |
|
|---|
| 63 | <UL>
|
|---|
| 64 |
|
|---|
| 65 | <LI>
|
|---|
| 66 | Fixed bugs in the square root functions for <NOBR>64-bit</NOBR>
|
|---|
| 67 | double-precision, <NOBR>80-bit</NOBR> double-extended-precision, and
|
|---|
| 68 | <NOBR>128-bit</NOBR> quadruple-precision.
|
|---|
| 69 | For <NOBR>64-bit</NOBR> double-precision (<CODE>f64_sqrt</CODE>), the result
|
|---|
| 70 | could sometimes be off by <NOBR>1 unit</NOBR> in the last place
|
|---|
| 71 | (<NOBR>1 ulp</NOBR>) from what it should be.
|
|---|
| 72 | For the larger formats, the square root could be wrong in a large portion of
|
|---|
| 73 | the less-significant bits.
|
|---|
| 74 | (A bug in <CODE>f128_sqrt</CODE> was first reported by Alexei Sibidanov.)
|
|---|
| 75 |
|
|---|
| 76 | </UL>
|
|---|
| 77 |
|
|---|
| 78 |
|
|---|
| 79 | <H3>Release 3c (2017 February)</H3>
|
|---|
| 80 |
|
|---|
| 81 | <UL>
|
|---|
| 82 |
|
|---|
| 83 | <LI>
|
|---|
| 84 | Added optional rounding mode <CODE>odd</CODE> (round to odd, also known as
|
|---|
| 85 | <EM>jamming</EM>).
|
|---|
| 86 |
|
|---|
| 87 | <LI>
|
|---|
| 88 | Corrected the documentation concerning non-canonical representations in
|
|---|
| 89 | <NOBR>80-bit</NOBR> double-extended-precision.
|
|---|
| 90 |
|
|---|
| 91 | </UL>
|
|---|
| 92 |
|
|---|
| 93 |
|
|---|
| 94 | <H3>Release 3b (2016 July)</H3>
|
|---|
| 95 |
|
|---|
| 96 | <UL>
|
|---|
| 97 |
|
|---|
| 98 | <LI>
|
|---|
| 99 | Implemented the common <NOBR>16-bit</NOBR> “half-precision”
|
|---|
| 100 | floating-point format (<CODE>float16_t</CODE>).
|
|---|
| 101 |
|
|---|
| 102 | <LI>
|
|---|
| 103 | Made the integer values returned on invalid conversions to integer formats
|
|---|
| 104 | be determined by the port-specific specialization instead of being the same for
|
|---|
| 105 | all ports.
|
|---|
| 106 |
|
|---|
| 107 | <LI>
|
|---|
| 108 | Added preprocessor macro <CODE>THREAD_LOCAL</CODE> to allow the floating-point
|
|---|
| 109 | state (modes and exception flags) to be made per-thread.
|
|---|
| 110 |
|
|---|
| 111 | <LI>
|
|---|
| 112 | Modified the provided Makefiles to allow some options to be overridden from the
|
|---|
| 113 | <CODE>make</CODE> command.
|
|---|
| 114 |
|
|---|
| 115 | <LI>
|
|---|
| 116 | Made other minor improvements.
|
|---|
| 117 |
|
|---|
| 118 | </UL>
|
|---|
| 119 |
|
|---|
| 120 |
|
|---|
| 121 | <H3>Release 3a (2015 October)</H3>
|
|---|
| 122 |
|
|---|
| 123 | <UL>
|
|---|
| 124 |
|
|---|
| 125 | <LI>
|
|---|
| 126 | Replaced the license text supplied by the University of California, Berkeley.
|
|---|
| 127 |
|
|---|
| 128 | </UL>
|
|---|
| 129 |
|
|---|
| 130 |
|
|---|
| 131 | <H3>Release 3 (2015 February)</H3>
|
|---|
| 132 |
|
|---|
| 133 | <UL>
|
|---|
| 134 |
|
|---|
| 135 | <LI>
|
|---|
| 136 | Complete rewrite, funded by the University of California, Berkeley, and
|
|---|
| 137 | consequently having a different use license than earlier releases.
|
|---|
| 138 | Major changes included renaming most types and functions, upgrading some
|
|---|
| 139 | algorithms, restructuring the source files, and making SoftFloat into a true
|
|---|
| 140 | library.
|
|---|
| 141 |
|
|---|
| 142 | <LI>
|
|---|
| 143 | Added functions to convert between floating-point and unsigned integers, both
|
|---|
| 144 | <NOBR>32-bit</NOBR> and <NOBR>64-bit</NOBR> (<CODE>uint32_t</CODE> and
|
|---|
| 145 | <CODE>uint64_t</CODE>).
|
|---|
| 146 |
|
|---|
| 147 | <LI>
|
|---|
| 148 | Added functions for fused multiply-add, for all supported floating-point
|
|---|
| 149 | formats except <NOBR>80-bit</NOBR> double-extended-precision.
|
|---|
| 150 |
|
|---|
| 151 | <LI>
|
|---|
| 152 | Added support for a fifth rounding mode, <CODE>near_maxMag</CODE> (round to
|
|---|
| 153 | nearest, with ties to maximum magnitude, away from zero).
|
|---|
| 154 |
|
|---|
| 155 | <LI>
|
|---|
| 156 | Dropped the <CODE>timesoftfloat</CODE> program (now part of the Berkeley
|
|---|
| 157 | TestFloat package).
|
|---|
| 158 |
|
|---|
| 159 | </UL>
|
|---|
| 160 |
|
|---|
| 161 |
|
|---|
| 162 | <H3>Release 2c (2015 January)</H3>
|
|---|
| 163 |
|
|---|
| 164 | <UL>
|
|---|
| 165 |
|
|---|
| 166 | <LI>
|
|---|
| 167 | Fixed mistakes affecting some <NOBR>64-bit</NOBR> processors.
|
|---|
| 168 |
|
|---|
| 169 | <LI>
|
|---|
| 170 | Further improved the documentation and the wording for the legal restrictions
|
|---|
| 171 | on using SoftFloat releases <NOBR>through 2c</NOBR> (not applicable to
|
|---|
| 172 | <NOBR>Release 3</NOBR> or later).
|
|---|
| 173 |
|
|---|
| 174 | </UL>
|
|---|
| 175 |
|
|---|
| 176 |
|
|---|
| 177 | <H3>Release 2b (2002 May)</H3>
|
|---|
| 178 |
|
|---|
| 179 | <UL>
|
|---|
| 180 |
|
|---|
| 181 | <LI>
|
|---|
| 182 | Made minor updates to the documentation, including improved wording for the
|
|---|
| 183 | legal restrictions on using SoftFloat.
|
|---|
| 184 |
|
|---|
| 185 | </UL>
|
|---|
| 186 |
|
|---|
| 187 |
|
|---|
| 188 | <H3>Release 2a (1998 December)</H3>
|
|---|
| 189 |
|
|---|
| 190 | <UL>
|
|---|
| 191 |
|
|---|
| 192 | <LI>
|
|---|
| 193 | Added functions to convert between <NOBR>64-bit</NOBR> integers
|
|---|
| 194 | (<CODE>int64</CODE>) and all supported floating-point formats.
|
|---|
| 195 |
|
|---|
| 196 | <LI>
|
|---|
| 197 | Fixed a bug in all <NOBR>64-bit</NOBR>-version square root functions except
|
|---|
| 198 | <CODE>float32_sqrt</CODE> that caused the result sometimes to be off by
|
|---|
| 199 | <NOBR>1 unit</NOBR> in the last place (<NOBR>1 ulp</NOBR>) from what it should
|
|---|
| 200 | be.
|
|---|
| 201 | (Bug discovered by Paul Donahue.)
|
|---|
| 202 |
|
|---|
| 203 | <LI>
|
|---|
| 204 | Improved the Makefiles.
|
|---|
| 205 | </UL>
|
|---|
| 206 |
|
|---|
| 207 |
|
|---|
| 208 | <H3>Release 2 (1997 June)</H3>
|
|---|
| 209 |
|
|---|
| 210 | <UL>
|
|---|
| 211 |
|
|---|
| 212 | <LI>
|
|---|
| 213 | Created the <NOBR>64-bit</NOBR> (<CODE>bits64</CODE>) version, adding the
|
|---|
| 214 | <CODE>floatx80</CODE> and <CODE>float128</CODE> formats.
|
|---|
| 215 |
|
|---|
| 216 | <LI>
|
|---|
| 217 | Changed the source directory structure, splitting the sources into a
|
|---|
| 218 | <CODE>bits32</CODE> and a <CODE>bits64</CODE> version.
|
|---|
| 219 | Renamed <CODE>environment.h</CODE> to <CODE>milieu.h</CODE> to avoid confusion
|
|---|
| 220 | with environment variables.
|
|---|
| 221 |
|
|---|
| 222 | <LI>
|
|---|
| 223 | Fixed a small error that caused <CODE>float64_round_to_int</CODE> often to
|
|---|
| 224 | round the wrong way in nearest/even mode when the operand was between
|
|---|
| 225 | 2<SUP>20</SUP> and 2<SUP>21</SUP> and halfway between two integers.
|
|---|
| 226 |
|
|---|
| 227 | </UL>
|
|---|
| 228 |
|
|---|
| 229 |
|
|---|
| 230 | <H3>Release 1a (1996 July)</H3>
|
|---|
| 231 |
|
|---|
| 232 | <UL>
|
|---|
| 233 |
|
|---|
| 234 | <LI>
|
|---|
| 235 | Corrected a mistake that caused borderline underflow cases not to raise the
|
|---|
| 236 | underflow flag when they should have.
|
|---|
| 237 | (Problem reported by Doug Priest.)
|
|---|
| 238 |
|
|---|
| 239 | <LI>
|
|---|
| 240 | Added the <CODE>float_detect_tininess</CODE> variable to control whether
|
|---|
| 241 | tininess is detected before or after rounding.
|
|---|
| 242 |
|
|---|
| 243 | </UL>
|
|---|
| 244 |
|
|---|
| 245 |
|
|---|
| 246 | <H3>Release 1 (1996 July)</H3>
|
|---|
| 247 |
|
|---|
| 248 | <UL>
|
|---|
| 249 |
|
|---|
| 250 | <LI>
|
|---|
| 251 | Original release, based on work done for the International Computer Science
|
|---|
| 252 | Institute (ICSI) in Berkeley, California.
|
|---|
| 253 |
|
|---|
| 254 | </UL>
|
|---|
| 255 |
|
|---|
| 256 |
|
|---|
| 257 | </BODY>
|
|---|
| 258 |
|
|---|
