1 /* $OpenBSD: s_exp2l.c,v 1.3 2014/07/21 01:51:11 guenther Exp $ */ 2 /*- 3 * Copyright (c) 2005-2008 David Schultz <das@FreeBSD.ORG> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/types.h> 29 #include <machine/ieee.h> 30 #include <float.h> 31 #include <math.h> 32 #include <stdint.h> 33 34 #define TBLBITS 7 35 #define TBLSIZE (1 << TBLBITS) 36 37 #define BIAS (LDBL_MAX_EXP - 1) 38 #define EXPMASK (BIAS + LDBL_MAX_EXP) 39 40 static const long double huge = 0x1p10000L; 41 #if 0 /* XXX Prevent gcc from erroneously constant folding this. */ 42 static const long double twom10000 = 0x1p-10000L; 43 #else 44 static volatile long double twom10000 = 0x1p-10000L; 45 #endif 46 47 static const double 48 redux = 0x1.8p63 / TBLSIZE, 49 P1 = 0x1.62e42fefa39efp-1, 50 P2 = 0x1.ebfbdff82c58fp-3, 51 P3 = 0x1.c6b08d7049fap-5, 52 P4 = 0x1.3b2ab6fba4da5p-7, 53 P5 = 0x1.5d8804780a736p-10, 54 P6 = 0x1.430918835e33dp-13; 55 56 static const double tbl[TBLSIZE * 2] = { 57 0x1.6a09e667f3bcdp-1, -0x1.bdd3413b2648p-55, 58 0x1.6c012750bdabfp-1, -0x1.2895667ff0cp-57, 59 0x1.6dfb23c651a2fp-1, -0x1.bbe3a683c88p-58, 60 0x1.6ff7df9519484p-1, -0x1.83c0f25860fp-56, 61 0x1.71f75e8ec5f74p-1, -0x1.16e4786887bp-56, 62 0x1.73f9a48a58174p-1, -0x1.0a8d96c65d5p-55, 63 0x1.75feb564267c9p-1, -0x1.0245957316ep-55, 64 0x1.780694fde5d3fp-1, 0x1.866b80a0216p-55, 65 0x1.7a11473eb0187p-1, -0x1.41577ee0499p-56, 66 0x1.7c1ed0130c132p-1, 0x1.f124cd1164ep-55, 67 0x1.7e2f336cf4e62p-1, 0x1.05d02ba157ap-57, 68 0x1.80427543e1a12p-1, -0x1.27c86626d97p-55, 69 0x1.82589994cce13p-1, -0x1.d4c1dd41533p-55, 70 0x1.8471a4623c7adp-1, -0x1.8d684a341cep-56, 71 0x1.868d99b4492edp-1, -0x1.fc6f89bd4f68p-55, 72 0x1.88ac7d98a6699p-1, 0x1.994c2f37cb5p-55, 73 0x1.8ace5422aa0dbp-1, 0x1.6e9f156864bp-55, 74 0x1.8cf3216b5448cp-1, -0x1.0d55e32e9e4p-57, 75 0x1.8f1ae99157736p-1, 0x1.5cc13a2e397p-56, 76 0x1.9145b0b91ffc6p-1, -0x1.dd6792e5825p-55, 77 0x1.93737b0cdc5e5p-1, -0x1.75fc781b58p-58, 78 0x1.95a44cbc8520fp-1, -0x1.64b7c96a5fp-57, 79 0x1.97d829fde4e5p-1, -0x1.d185b7c1b86p-55, 80 0x1.9a0f170ca07bap-1, -0x1.173bd91cee6p-55, 81 0x1.9c49182a3f09p-1, 0x1.c7c46b071f2p-57, 82 0x1.9e86319e32323p-1, 0x1.824ca78e64cp-57, 83 0x1.a0c667b5de565p-1, -0x1.359495d1cd5p-55, 84 0x1.a309bec4a2d33p-1, 0x1.6305c7ddc368p-55, 85 0x1.a5503b23e255dp-1, -0x1.d2f6edb8d42p-55, 86 0x1.a799e1330b358p-1, 0x1.bcb7ecac564p-55, 87 0x1.a9e6b5579fdbfp-1, 0x1.0fac90ef7fdp-55, 88 0x1.ac36bbfd3f37ap-1, -0x1.f9234cae76dp-56, 89 0x1.ae89f995ad3adp-1, 0x1.7a1cd345dcc8p-55, 90 0x1.b0e07298db666p-1, -0x1.bdef54c80e4p-55, 91 0x1.b33a2b84f15fbp-1, -0x1.2805e3084d8p-58, 92 0x1.b59728de5593ap-1, -0x1.c71dfbbba6ep-55, 93 0x1.b7f76f2fb5e47p-1, -0x1.5584f7e54acp-57, 94 0x1.ba5b030a1064ap-1, -0x1.efcd30e5429p-55, 95 0x1.bcc1e904bc1d2p-1, 0x1.23dd07a2d9fp-56, 96 0x1.bf2c25bd71e09p-1, -0x1.efdca3f6b9c8p-55, 97 0x1.c199bdd85529cp-1, 0x1.11065895049p-56, 98 0x1.c40ab5fffd07ap-1, 0x1.b4537e083c6p-55, 99 0x1.c67f12e57d14bp-1, 0x1.2884dff483c8p-55, 100 0x1.c8f6d9406e7b5p-1, 0x1.1acbc48805cp-57, 101 0x1.cb720dcef9069p-1, 0x1.503cbd1e94ap-57, 102 0x1.cdf0b555dc3fap-1, -0x1.dd83b53829dp-56, 103 0x1.d072d4a07897cp-1, -0x1.cbc3743797a8p-55, 104 0x1.d2f87080d89f2p-1, -0x1.d487b719d858p-55, 105 0x1.d5818dcfba487p-1, 0x1.2ed02d75b37p-56, 106 0x1.d80e316c98398p-1, -0x1.11ec18bedep-55, 107 0x1.da9e603db3285p-1, 0x1.c2300696db5p-55, 108 0x1.dd321f301b46p-1, 0x1.2da5778f019p-55, 109 0x1.dfc97337b9b5fp-1, -0x1.1a5cd4f184b8p-55, 110 0x1.e264614f5a129p-1, -0x1.7b627817a148p-55, 111 0x1.e502ee78b3ff6p-1, 0x1.39e8980a9cdp-56, 112 0x1.e7a51fbc74c83p-1, 0x1.2d522ca0c8ep-55, 113 0x1.ea4afa2a490dap-1, -0x1.e9c23179c288p-55, 114 0x1.ecf482d8e67f1p-1, -0x1.c93f3b411ad8p-55, 115 0x1.efa1bee615a27p-1, 0x1.dc7f486a4b68p-55, 116 0x1.f252b376bba97p-1, 0x1.3a1a5bf0d8e8p-55, 117 0x1.f50765b6e454p-1, 0x1.9d3e12dd8a18p-55, 118 0x1.f7bfdad9cbe14p-1, -0x1.dbb12d00635p-55, 119 0x1.fa7c1819e90d8p-1, 0x1.74853f3a593p-56, 120 0x1.fd3c22b8f71f1p-1, 0x1.2eb74966578p-58, 121 0x1p+0, 0x0p+0, 122 0x1.0163da9fb3335p+0, 0x1.b61299ab8cd8p-54, 123 0x1.02c9a3e778061p+0, -0x1.19083535b08p-56, 124 0x1.04315e86e7f85p+0, -0x1.0a31c1977c98p-54, 125 0x1.059b0d3158574p+0, 0x1.d73e2a475b4p-55, 126 0x1.0706b29ddf6dep+0, -0x1.c91dfe2b13cp-55, 127 0x1.0874518759bc8p+0, 0x1.186be4bb284p-57, 128 0x1.09e3ecac6f383p+0, 0x1.14878183161p-54, 129 0x1.0b5586cf9890fp+0, 0x1.8a62e4adc61p-54, 130 0x1.0cc922b7247f7p+0, 0x1.01edc16e24f8p-54, 131 0x1.0e3ec32d3d1a2p+0, 0x1.03a1727c58p-59, 132 0x1.0fb66affed31bp+0, -0x1.b9bedc44ebcp-57, 133 0x1.11301d0125b51p+0, -0x1.6c51039449bp-54, 134 0x1.12abdc06c31ccp+0, -0x1.1b514b36ca8p-58, 135 0x1.1429aaea92dep+0, -0x1.32fbf9af1368p-54, 136 0x1.15a98c8a58e51p+0, 0x1.2406ab9eeabp-55, 137 0x1.172b83c7d517bp+0, -0x1.19041b9d78ap-55, 138 0x1.18af9388c8deap+0, -0x1.11023d1970f8p-54, 139 0x1.1a35beb6fcb75p+0, 0x1.e5b4c7b4969p-55, 140 0x1.1bbe084045cd4p+0, -0x1.95386352ef6p-54, 141 0x1.1d4873168b9aap+0, 0x1.e016e00a264p-54, 142 0x1.1ed5022fcd91dp+0, -0x1.1df98027bb78p-54, 143 0x1.2063b88628cd6p+0, 0x1.dc775814a85p-55, 144 0x1.21f49917ddc96p+0, 0x1.2a97e9494a6p-55, 145 0x1.2387a6e756238p+0, 0x1.9b07eb6c7058p-54, 146 0x1.251ce4fb2a63fp+0, 0x1.ac155bef4f5p-55, 147 0x1.26b4565e27cddp+0, 0x1.2bd339940eap-55, 148 0x1.284dfe1f56381p+0, -0x1.a4c3a8c3f0d8p-54, 149 0x1.29e9df51fdee1p+0, 0x1.612e8afad12p-55, 150 0x1.2b87fd0dad99p+0, -0x1.10adcd6382p-59, 151 0x1.2d285a6e4030bp+0, 0x1.0024754db42p-54, 152 0x1.2ecafa93e2f56p+0, 0x1.1ca0f45d524p-56, 153 0x1.306fe0a31b715p+0, 0x1.6f46ad23183p-55, 154 0x1.32170fc4cd831p+0, 0x1.a9ce78e1804p-55, 155 0x1.33c08b26416ffp+0, 0x1.327218436598p-54, 156 0x1.356c55f929ff1p+0, -0x1.b5cee5c4e46p-55, 157 0x1.371a7373aa9cbp+0, -0x1.63aeabf42ebp-54, 158 0x1.38cae6d05d866p+0, -0x1.e958d3c99048p-54, 159 0x1.3a7db34e59ff7p+0, -0x1.5e436d661f6p-56, 160 0x1.3c32dc313a8e5p+0, -0x1.efff8375d2ap-54, 161 0x1.3dea64c123422p+0, 0x1.ada0911f09fp-55, 162 0x1.3fa4504ac801cp+0, -0x1.7d023f956fap-54, 163 0x1.4160a21f72e2ap+0, -0x1.ef3691c309p-58, 164 0x1.431f5d950a897p+0, -0x1.1c7dde35f7ap-55, 165 0x1.44e086061892dp+0, 0x1.89b7a04ef8p-59, 166 0x1.46a41ed1d0057p+0, 0x1.c944bd1648a8p-54, 167 0x1.486a2b5c13cdp+0, 0x1.3c1a3b69062p-56, 168 0x1.4a32af0d7d3dep+0, 0x1.9cb62f3d1be8p-54, 169 0x1.4bfdad5362a27p+0, 0x1.d4397afec42p-56, 170 0x1.4dcb299fddd0dp+0, 0x1.8ecdbbc6a78p-54, 171 0x1.4f9b2769d2ca7p+0, -0x1.4b309d25958p-54, 172 0x1.516daa2cf6642p+0, -0x1.f768569bd94p-55, 173 0x1.5342b569d4f82p+0, -0x1.07abe1db13dp-55, 174 0x1.551a4ca5d920fp+0, -0x1.d689cefede6p-55, 175 0x1.56f4736b527dap+0, 0x1.9bb2c011d938p-54, 176 0x1.58d12d497c7fdp+0, 0x1.295e15b9a1ep-55, 177 0x1.5ab07dd485429p+0, 0x1.6324c0546478p-54, 178 0x1.5c9268a5946b7p+0, 0x1.c4b1b81698p-60, 179 0x1.5e76f15ad2148p+0, 0x1.ba6f93080e68p-54, 180 0x1.605e1b976dc09p+0, -0x1.3e2429b56de8p-54, 181 0x1.6247eb03a5585p+0, -0x1.383c17e40b48p-54, 182 0x1.6434634ccc32p+0, -0x1.c483c759d89p-55, 183 0x1.6623882552225p+0, -0x1.bb60987591cp-54, 184 0x1.68155d44ca973p+0, 0x1.038ae44f74p-57, 185 }; 186 187 /* 188 * exp2l(x): compute the base 2 exponential of x 189 * 190 * Accuracy: Peak error < 0.511 ulp. 191 * 192 * Method: (equally-spaced tables) 193 * 194 * Reduce x: 195 * x = 2**k + y, for integer k and |y| <= 1/2. 196 * Thus we have exp2l(x) = 2**k * exp2(y). 197 * 198 * Reduce y: 199 * y = i/TBLSIZE + z for integer i near y * TBLSIZE. 200 * Thus we have exp2(y) = exp2(i/TBLSIZE) * exp2(z), 201 * with |z| <= 2**-(TBLBITS+1). 202 * 203 * We compute exp2(i/TBLSIZE) via table lookup and exp2(z) via a 204 * degree-6 minimax polynomial with maximum error under 2**-69. 205 * The table entries each have 104 bits of accuracy, encoded as 206 * a pair of double precision values. 207 */ 208 long double 209 exp2l(long double x) 210 { 211 union { 212 long double e; 213 struct ieee_ext bits; 214 } u, v; 215 long double r, twopk, twopkp10000, z, t_hi, t_lo; 216 uint32_t es, hx, ix, i0; 217 int k; 218 219 /* Filter out exceptional cases. */ 220 u.e = x; 221 hx = (u.bits.ext_sign << 15) | u.bits.ext_exp; 222 ix = hx & EXPMASK; 223 if (ix >= BIAS + 14) { /* |x| >= 16384 or x is NaN */ 224 if (ix == BIAS + LDBL_MAX_EXP) { 225 if ((u.bits.ext_frach != 1U << 31 && 226 u.bits.ext_fracl != 0) || (hx & 0x8000) == 0) 227 return (x + x); /* x is +Inf or NaN */ 228 else 229 return (0.0); /* x is -Inf */ 230 } 231 if (x >= 16384) 232 return (huge * huge); /* overflow */ 233 if (x <= -16446) 234 return (twom10000 * twom10000); /* underflow */ 235 } else if (ix <= BIAS - 66) { /* |x| < 0x1p-66 */ 236 return (1.0 + x); 237 } 238 239 /* 240 * Reduce x, computing z, i0, and k. The low bits of x + redux 241 * contain the 16-bit integer part of the exponent (k) followed by 242 * TBLBITS fractional bits (i0). We use bit tricks to extract these 243 * as integers, then set z to the remainder. 244 * 245 * Example: Suppose x is 0xabc.123456p0 and TBLBITS is 8. 246 * Then the low-order word of x + redux is 0x000abc12, 247 * We split this into k = 0xabc and i0 = 0x12 (adjusted to 248 * index into the table), then we compute z = 0x0.003456p0. 249 * 250 * XXX If the exponent is negative, the computation of k depends on 251 * '>>' doing sign extension. 252 */ 253 u.e = x + redux; 254 i0 = u.bits.ext_fracl + TBLSIZE / 2; 255 k = (int)i0 >> TBLBITS; 256 i0 = (i0 & (TBLSIZE - 1)) << 1; 257 u.e -= redux; 258 z = x - u.e; 259 260 v.bits.ext_frach = 1U << 31; 261 v.bits.ext_fracl = 0; 262 263 if (k >= LDBL_MIN_EXP) { 264 es = LDBL_MAX_EXP - 1 + k; 265 v.bits.ext_exp = es & 0x7fffffff; 266 v.bits.ext_sign = u.bits.ext_sign >> 15; 267 twopk = v.e; 268 } else { 269 es = LDBL_MAX_EXP - 1 + k + 10000; 270 v.bits.ext_exp = es & 0x7fffffff; 271 v.bits.ext_sign = u.bits.ext_sign >> 15; 272 twopkp10000 = v.e; 273 } 274 275 /* Compute r = exp2l(y) = exp2lt[i0] * p(z). */ 276 t_hi = tbl[i0]; 277 t_lo = tbl[i0 + 1]; 278 /* XXX This gives > 1 ulp errors outside of FE_TONEAREST mode */ 279 r = t_lo + (t_hi + t_lo) * z * (P1 + z * (P2 + z * (P3 + z * (P4 280 + z * (P5 + z * P6))))) + t_hi; 281 282 /* Scale by 2**k. */ 283 if (k >= LDBL_MIN_EXP) { 284 if (k == LDBL_MAX_EXP) 285 return (r * 2.0 * 0x1p16383L); 286 return (r * twopk); 287 } else { 288 return (r * twopkp10000 * twom10000); 289 } 290 } 291