git: d47b524eadb5 - stable/13 - * lib/msun/Makefile b/lib/msun/Makefile: . Disconnect imprecise.c from the build. This file can be deleted. . Add b_tgammal.c to the build for ld80 and ld128 targets. The ld128 is a 'git mv' of imprecise.c to ld128/b_tgammal.c.
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Wed, 26 Jul 2023 23:51:24 UTC
The branch stable/13 has been updated by emaste: URL: https://cgit.FreeBSD.org/src/commit/?id=d47b524eadb5a87506fe6ae03febbf72fad2ee06 commit d47b524eadb5a87506fe6ae03febbf72fad2ee06 Author: Mark Murray <markm@FreeBSD.org> AuthorDate: 2021-12-14 09:08:57 +0000 Commit: Ed Maste <emaste@FreeBSD.org> CommitDate: 2023-07-26 23:07:30 +0000 * lib/msun/Makefile b/lib/msun/Makefile: . Disconnect imprecise.c from the build. This file can be deleted. . Add b_tgammal.c to the build for ld80 and ld128 targets. The ld128 is a 'git mv' of imprecise.c to ld128/b_tgammal.c. * lib/msun/ld80/b_expl.c: . New file. Implement __exp__D for ld80 targets. This is based on bsdsrc/b_exp.c. * lib/msun/ld80/b_logl.c: . New file. Implement __log__D for ld80 targets. This is based on bsdsrc/b_log.c. * lib/msun/ld80/b_tgammal.c b/lib/msun/ld80/b_tgammal.c . New file. Implement tgammal(x) for ld80 targets. Submitted by: Steve Kargl Differential Revision: https://reviews.freebsd.org/D33444 Reviewed by: pfg (cherry picked from commit 03a88e3de9c68182d21df94b1c8c7ced930dbd1f) --- lib/msun/Makefile | 3 +- lib/msun/{src/imprecise.c => ld128/b_tgammal.c} | 0 lib/msun/ld80/b_expl.c | 113 +++++++ lib/msun/ld80/b_logl.c | 375 +++++++++++++++++++++ lib/msun/ld80/b_tgammal.c | 419 ++++++++++++++++++++++++ 5 files changed, 908 insertions(+), 2 deletions(-) diff --git a/lib/msun/Makefile b/lib/msun/Makefile index 24d740d761d5..8cb67f3700b9 100644 --- a/lib/msun/Makefile +++ b/lib/msun/Makefile @@ -69,7 +69,6 @@ COMMON_SRCS= b_tgamma.c \ e_pow.c e_powf.c e_rem_pio2.c \ e_rem_pio2f.c e_remainder.c e_remainderf.c e_scalb.c e_scalbf.c \ e_sinh.c e_sinhf.c e_sqrt.c e_sqrtf.c fenv.c \ - imprecise.c \ k_cos.c k_cosf.c k_exp.c k_expf.c k_rem_pio2.c k_sin.c k_sinf.c \ k_tan.c k_tanf.c \ s_asinh.c s_asinhf.c s_atan.c s_atanf.c s_carg.c s_cargf.c s_cargl.c \ @@ -116,7 +115,7 @@ SYMBOL_MAPS= ${SYM_MAPS} COMMON_SRCS+= s_copysignl.c s_fabsl.c s_llrintl.c s_lrintl.c s_modfl.c .if ${LDBL_PREC} != 53 # If long double != double use these; otherwise, we alias the double versions. -COMMON_SRCS+= catrigl.c \ +COMMON_SRCS+= b_tgammal.c catrigl.c \ e_acoshl.c e_acosl.c e_asinl.c e_atan2l.c e_atanhl.c \ e_coshl.c e_fmodl.c e_hypotl.c \ e_lgammal.c e_lgammal_r.c e_powl.c \ diff --git a/lib/msun/src/imprecise.c b/lib/msun/ld128/b_tgammal.c similarity index 100% rename from lib/msun/src/imprecise.c rename to lib/msun/ld128/b_tgammal.c diff --git a/lib/msun/ld80/b_expl.c b/lib/msun/ld80/b_expl.c new file mode 100644 index 000000000000..21bacf4497a3 --- /dev/null +++ b/lib/msun/ld80/b_expl.c @@ -0,0 +1,113 @@ +/*- + * SPDX-License-Identifier: BSD-3-Clause + * + * Copyright (c) 1985, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +/* + * See bsdsrc/b_exp.c for implementation details. + * + * bsdrc/b_exp.c converted to long double by Steven G. Kargl. + */ + +#include "fpmath.h" +#include "math_private.h" + +static const union IEEEl2bits + p0u = LD80C(0xaaaaaaaaaaaaaaab, -3, 1.66666666666666666671e-01L), + p1u = LD80C(0xb60b60b60b60b59a, -9, -2.77777777777777775377e-03L), + p2u = LD80C(0x8ab355e008a3cfce, -14, 6.61375661375629297465e-05L), + p3u = LD80C(0xddebbc994b0c1376, -20, -1.65343915327882529784e-06L), + p4u = LD80C(0xb354784cb4ef4c41, -25, 4.17535101591534118469e-08L), + p5u = LD80C(0x913e8a718382ce75, -30, -1.05679137034774806475e-09L), + p6u = LD80C(0xe8f0042aa134502e, -36, 2.64819349895429516863e-11L); +#define p1 (p0u.e) +#define p2 (p1u.e) +#define p3 (p2u.e) +#define p4 (p3u.e) +#define p5 (p4u.e) +#define p6 (p5u.e) +#define p7 (p6u.e) + +/* + * lnhuge = (LDBL_MAX_EXP + 9) * log(2.) + * lntiny = (LDBL_MIN_EXP - 64 - 10) * log(2.) + * invln2 = 1 / log(2.) + */ +static const union IEEEl2bits +ln2hiu = LD80C(0xb17217f700000000, -1, 6.93147180369123816490e-01L), +ln2lou = LD80C(0xd1cf79abc9e3b398, -33, 1.90821492927058781614e-10L), +lnhugeu = LD80C(0xb18b0c0330a8fad9, 13, 1.13627617309191834574e+04L), +lntinyu = LD80C(0xb236f28a68bc3bd7, 13, -1.14057368561139000667e+04L), +invln2u = LD80C(0xb8aa3b295c17f0bc, 0, 1.44269504088896340739e+00L); +#define ln2hi (ln2hiu.e) +#define ln2lo (ln2lou.e) +#define lnhuge (lnhugeu.e) +#define lntiny (lntinyu.e) +#define invln2 (invln2u.e) + +/* returns exp(r = x + c) for |c| < |x| with no overlap. */ + +static long double +__exp__D(long double x, long double c) +{ + long double hi, lo, z; + int k; + + if (x != x) /* x is NaN. */ + return(x); + + if (x <= lnhuge) { + if (x >= lntiny) { + /* argument reduction: x --> x - k*ln2 */ + z = invln2 * x; + k = z + copysignl(0.5L, x); + + /* + * Express (x + c) - k * ln2 as hi - lo. + * Let x = hi - lo rounded. + */ + hi = x - k * ln2hi; /* Exact. */ + lo = k * ln2lo - c; + x = hi - lo; + + /* Return 2^k*[1+x+x*c/(2+c)] */ + z = x * x; + c = x - z * (p1 + z * (p2 + z * (p3 + z * (p4 + + z * (p5 + z * (p6 + z * p7)))))); + c = (x * c) / (2 - c); + + return (ldexpl(1 + (hi - (lo - c)), k)); + } else { + /* exp(-INF) is 0. exp(-big) underflows to 0. */ + return (isfinite(x) ? ldexpl(1., -5000) : 0); + } + } else + /* exp(INF) is INF, exp(+big#) overflows to INF */ + return (isfinite(x) ? ldexpl(1., 5000) : x); +} diff --git a/lib/msun/ld80/b_logl.c b/lib/msun/ld80/b_logl.c new file mode 100644 index 000000000000..b11eacbe17a1 --- /dev/null +++ b/lib/msun/ld80/b_logl.c @@ -0,0 +1,375 @@ +/*- + * SPDX-License-Identifier: BSD-3-Clause + * + * Copyright (c) 1992, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +/* + * See bsdsrc/b_log.c for implementation details. + * + * bsdrc/b_log.c converted to long double by Steven G. Kargl. + */ + +#define N 128 + +/* + * Coefficients in the polynomial approximation of log(1+f/F). + * Domain of x is [0,1./256] with 2**(-84.48) precision. + */ +static const union IEEEl2bits + a1u = LD80C(0xaaaaaaaaaaaaaaab, -4, 8.33333333333333333356e-02L), + a2u = LD80C(0xcccccccccccccd29, -7, 1.25000000000000000781e-02L), + a3u = LD80C(0x9249249241ed3764, -9, 2.23214285711721994134e-03L), + a4u = LD80C(0xe38e959e1e7e01cf, -12, 4.34030476540000360640e-04L); +#define A1 (a1u.e) +#define A2 (a2u.e) +#define A3 (a3u.e) +#define A4 (a4u.e) + +/* + * Table of log(Fj) = logF_head[j] + logF_tail[j], for Fj = 1+j/128. + * Used for generation of extend precision logarithms. + * The constant 35184372088832 is 2^45, so the divide is exact. + * It ensures correct reading of logF_head, even for inaccurate + * decimal-to-binary conversion routines. (Everybody gets the + * right answer for integers less than 2^53.) + * Values for log(F) were generated using error < 10^-57 absolute + * with the bc -l package. + */ + +static double logF_head[N+1] = { + 0., + .007782140442060381246, + .015504186535963526694, + .023167059281547608406, + .030771658666765233647, + .038318864302141264488, + .045809536031242714670, + .053244514518837604555, + .060624621816486978786, + .067950661908525944454, + .075223421237524235039, + .082443669210988446138, + .089612158689760690322, + .096729626458454731618, + .103796793681567578460, + .110814366340264314203, + .117783035656430001836, + .124703478501032805070, + .131576357788617315236, + .138402322859292326029, + .145182009844575077295, + .151916042025732167530, + .158605030176659056451, + .165249572895390883786, + .171850256926518341060, + .178407657472689606947, + .184922338493834104156, + .191394852999565046047, + .197825743329758552135, + .204215541428766300668, + .210564769107350002741, + .216873938300523150246, + .223143551314024080056, + .229374101064877322642, + .235566071312860003672, + .241719936886966024758, + .247836163904594286577, + .253915209980732470285, + .259957524436686071567, + .265963548496984003577, + .271933715484010463114, + .277868451003087102435, + .283768173130738432519, + .289633292582948342896, + .295464212893421063199, + .301261330578199704177, + .307025035294827830512, + .312755710004239517729, + .318453731118097493890, + .324119468654316733591, + .329753286372579168528, + .335355541920762334484, + .340926586970454081892, + .346466767346100823488, + .351976423156884266063, + .357455888922231679316, + .362905493689140712376, + .368325561158599157352, + .373716409793814818840, + .379078352934811846353, + .384411698910298582632, + .389716751140440464951, + .394993808240542421117, + .400243164127459749579, + .405465108107819105498, + .410659924985338875558, + .415827895143593195825, + .420969294644237379543, + .426084395310681429691, + .431173464818130014464, + .436236766774527495726, + .441274560805140936281, + .446287102628048160113, + .451274644139630254358, + .456237433481874177232, + .461175715122408291790, + .466089729924533457960, + .470979715219073113985, + .475845904869856894947, + .480688529345570714212, + .485507815781602403149, + .490303988045525329653, + .495077266798034543171, + .499827869556611403822, + .504556010751912253908, + .509261901790523552335, + .513945751101346104405, + .518607764208354637958, + .523248143765158602036, + .527867089620485785417, + .532464798869114019908, + .537041465897345915436, + .541597282432121573947, + .546132437597407260909, + .550647117952394182793, + .555141507540611200965, + .559615787935399566777, + .564070138285387656651, + .568504735352689749561, + .572919753562018740922, + .577315365035246941260, + .581691739635061821900, + .586049045003164792433, + .590387446602107957005, + .594707107746216934174, + .599008189645246602594, + .603290851438941899687, + .607555250224322662688, + .611801541106615331955, + .616029877215623855590, + .620240409751204424537, + .624433288012369303032, + .628608659422752680256, + .632766669570628437213, + .636907462236194987781, + .641031179420679109171, + .645137961373620782978, + .649227946625615004450, + .653301272011958644725, + .657358072709030238911, + .661398482245203922502, + .665422632544505177065, + .669430653942981734871, + .673422675212350441142, + .677398823590920073911, + .681359224807238206267, + .685304003098281100392, + .689233281238557538017, + .693147180560117703862 +}; + +static double logF_tail[N+1] = { + 0., + -.00000000000000543229938420049, + .00000000000000172745674997061, + -.00000000000001323017818229233, + -.00000000000001154527628289872, + -.00000000000000466529469958300, + .00000000000005148849572685810, + -.00000000000002532168943117445, + -.00000000000005213620639136504, + -.00000000000001819506003016881, + .00000000000006329065958724544, + .00000000000008614512936087814, + -.00000000000007355770219435028, + .00000000000009638067658552277, + .00000000000007598636597194141, + .00000000000002579999128306990, + -.00000000000004654729747598444, + -.00000000000007556920687451336, + .00000000000010195735223708472, + -.00000000000017319034406422306, + -.00000000000007718001336828098, + .00000000000010980754099855238, + -.00000000000002047235780046195, + -.00000000000008372091099235912, + .00000000000014088127937111135, + .00000000000012869017157588257, + .00000000000017788850778198106, + .00000000000006440856150696891, + .00000000000016132822667240822, + -.00000000000007540916511956188, + -.00000000000000036507188831790, + .00000000000009120937249914984, + .00000000000018567570959796010, + -.00000000000003149265065191483, + -.00000000000009309459495196889, + .00000000000017914338601329117, + -.00000000000001302979717330866, + .00000000000023097385217586939, + .00000000000023999540484211737, + .00000000000015393776174455408, + -.00000000000036870428315837678, + .00000000000036920375082080089, + -.00000000000009383417223663699, + .00000000000009433398189512690, + .00000000000041481318704258568, + -.00000000000003792316480209314, + .00000000000008403156304792424, + -.00000000000034262934348285429, + .00000000000043712191957429145, + -.00000000000010475750058776541, + -.00000000000011118671389559323, + .00000000000037549577257259853, + .00000000000013912841212197565, + .00000000000010775743037572640, + .00000000000029391859187648000, + -.00000000000042790509060060774, + .00000000000022774076114039555, + .00000000000010849569622967912, + -.00000000000023073801945705758, + .00000000000015761203773969435, + .00000000000003345710269544082, + -.00000000000041525158063436123, + .00000000000032655698896907146, + -.00000000000044704265010452446, + .00000000000034527647952039772, + -.00000000000007048962392109746, + .00000000000011776978751369214, + -.00000000000010774341461609578, + .00000000000021863343293215910, + .00000000000024132639491333131, + .00000000000039057462209830700, + -.00000000000026570679203560751, + .00000000000037135141919592021, + -.00000000000017166921336082431, + -.00000000000028658285157914353, + -.00000000000023812542263446809, + .00000000000006576659768580062, + -.00000000000028210143846181267, + .00000000000010701931762114254, + .00000000000018119346366441110, + .00000000000009840465278232627, + -.00000000000033149150282752542, + -.00000000000018302857356041668, + -.00000000000016207400156744949, + .00000000000048303314949553201, + -.00000000000071560553172382115, + .00000000000088821239518571855, + -.00000000000030900580513238244, + -.00000000000061076551972851496, + .00000000000035659969663347830, + .00000000000035782396591276383, + -.00000000000046226087001544578, + .00000000000062279762917225156, + .00000000000072838947272065741, + .00000000000026809646615211673, + -.00000000000010960825046059278, + .00000000000002311949383800537, + -.00000000000058469058005299247, + -.00000000000002103748251144494, + -.00000000000023323182945587408, + -.00000000000042333694288141916, + -.00000000000043933937969737844, + .00000000000041341647073835565, + .00000000000006841763641591466, + .00000000000047585534004430641, + .00000000000083679678674757695, + -.00000000000085763734646658640, + .00000000000021913281229340092, + -.00000000000062242842536431148, + -.00000000000010983594325438430, + .00000000000065310431377633651, + -.00000000000047580199021710769, + -.00000000000037854251265457040, + .00000000000040939233218678664, + .00000000000087424383914858291, + .00000000000025218188456842882, + -.00000000000003608131360422557, + -.00000000000050518555924280902, + .00000000000078699403323355317, + -.00000000000067020876961949060, + .00000000000016108575753932458, + .00000000000058527188436251509, + -.00000000000035246757297904791, + -.00000000000018372084495629058, + .00000000000088606689813494916, + .00000000000066486268071468700, + .00000000000063831615170646519, + .00000000000025144230728376072, + -.00000000000017239444525614834 +}; +/* + * Extra precision variant, returning struct {double a, b;}; + * log(x) = a + b to 63 bits, with 'a' rounded to 24 bits. + */ +static struct Double +__log__D(long double x) +{ + int m, j; + long double F, f, g, q, u, v, u1, u2; + struct Double r; + + /* + * Argument reduction: 1 <= g < 2; x/2^m = g; + * y = F*(1 + f/F) for |f| <= 2^-8 + */ + g = frexpl(x, &m); + g *= 2; + m--; + if (m == DBL_MIN_EXP - 1) { + j = ilogbl(g); + m += j; + g = ldexpl(g, -j); + } + j = N * (g - 1) + 0.5L; + F = (1.L / N) * j + 1; + f = g - F; + + g = 1 / (2 * F + f); + u = 2 * f * g; + v = u * u; + q = u * v * (A1 + v * (A2 + v * (A3 + v * A4))); + if (m | j) { + u1 = u + 513; + u1 -= 513; + } else { + u1 = (float)u; + } + u2 = (2 * (f - F * u1) - u1 * f) * g; + + u1 += m * (long double)logF_head[N] + logF_head[j]; + + u2 += logF_tail[j]; + u2 += q; + u2 += logF_tail[N] * m; + r.a = (float)(u1 + u2); /* Only difference is here. */ + r.b = (u1 - r.a) + u2; + return (r); +} diff --git a/lib/msun/ld80/b_tgammal.c b/lib/msun/ld80/b_tgammal.c new file mode 100644 index 000000000000..121248adb1b2 --- /dev/null +++ b/lib/msun/ld80/b_tgammal.c @@ -0,0 +1,419 @@ +/*- + * SPDX-License-Identifier: BSD-3-Clause + * + * Copyright (c) 1992, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +/* + * The original code, FreeBSD's old svn r93211, contain the following + * attribution: + * + * This code by P. McIlroy, Oct 1992; + * + * The financial support of UUNET Communications Services is greatfully + * acknowledged. + * + * bsdrc/b_tgamma.c converted to long double by Steven G. Kargl. + */ + +/* + * See bsdsrc/t_tgamma.c for implementation details. + */ + +#include <float.h> + +#if LDBL_MAX_EXP != 0x4000 +#error "Unsupported long double format" +#endif + +#ifdef __i386__ +#include <ieeefp.h> +#endif + +#include "fpmath.h" +#include "math.h" +#include "math_private.h" + +/* Used in b_log.c and below. */ +struct Double { + long double a; + long double b; +}; + +#include "b_logl.c" +#include "b_expl.c" + +static const double zero = 0.; +static const volatile double tiny = 1e-300; +/* + * x >= 6 + * + * Use the asymptotic approximation (Stirling's formula) adjusted for + * equal-ripples: + * + * log(G(x)) ~= (x-0.5)*(log(x)-1) + 0.5(log(2*pi)-1) + 1/x*P(1/(x*x)) + * + * Keep extra precision in multiplying (x-.5)(log(x)-1), to avoid + * premature round-off. + * + * Accurate to max(ulp(1/128) absolute, 2^-66 relative) error. + */ + +/* + * The following is a decomposition of 0.5 * (log(2*pi) - 1) into the + * first 12 bits in ln2pi_hi and the trailing 64 bits in ln2pi_lo. The + * variables are clearly misnamed. + */ +static const union IEEEl2bits +ln2pi_hiu = LD80C(0xd680000000000000, -2, 4.18945312500000000000e-01L), +ln2pi_lou = LD80C(0xe379b414b596d687, -18, -6.77929532725821967032e-06L); +#define ln2pi_hi (ln2pi_hiu.e) +#define ln2pi_lo (ln2pi_lou.e) + +static const union IEEEl2bits + Pa0u = LD80C(0xaaaaaaaaaaaaaaaa, -4, 8.33333333333333333288e-02L), + Pa1u = LD80C(0xb60b60b60b5fcd59, -9, -2.77777777777776516326e-03L), + Pa2u = LD80C(0xd00d00cffbb47014, -11, 7.93650793635429639018e-04L), + Pa3u = LD80C(0x9c09c07c0805343e, -11, -5.95238087960599252215e-04L), + Pa4u = LD80C(0xdca8d31f8e6e5e8f, -11, 8.41749082509607342883e-04L), + Pa5u = LD80C(0xfb4d4289632f1638, -10, -1.91728055205541624556e-03L), + Pa6u = LD80C(0xd15a4ba04078d3f8, -8, 6.38893788027752396194e-03L), + Pa7u = LD80C(0xe877283110bcad95, -6, -2.83771309846297590312e-02L), + Pa8u = LD80C(0x8da97eed13717af8, -3, 1.38341887683837576925e-01L), + Pa9u = LD80C(0xf093b1c1584e30ce, -2, -4.69876818515470146031e-01L); +#define Pa0 (Pa0u.e) +#define Pa1 (Pa1u.e) +#define Pa2 (Pa2u.e) +#define Pa3 (Pa3u.e) +#define Pa4 (Pa4u.e) +#define Pa5 (Pa5u.e) +#define Pa6 (Pa6u.e) +#define Pa7 (Pa7u.e) +#define Pa8 (Pa8u.e) +#define Pa9 (Pa9u.e) + +static struct Double +large_gam(long double x) +{ + long double p, z, thi, tlo, xhi, xlo; + long double logx; + struct Double u; + + z = 1 / (x * x); + p = Pa0 + z * (Pa1 + z * (Pa2 + z * (Pa3 + z * (Pa4 + z * (Pa5 + + z * (Pa6 + z * (Pa7 + z * (Pa8 + z * Pa9)))))))); + p = p / x; + + u = __log__D(x); + u.a -= 1; + + /* Split (x - 0.5) in high and low parts. */ + x -= 0.5L; + xhi = (float)x; + xlo = x - xhi; + + /* Compute t = (x-.5)*(log(x)-1) in extra precision. */ + thi = xhi * u.a; + tlo = xlo * u.a + x * u.b; + + /* Compute thi + tlo + ln2pi_hi + ln2pi_lo + p. */ + tlo += ln2pi_lo; + tlo += p; + u.a = ln2pi_hi + tlo; + u.a += thi; + u.b = thi - u.a; + u.b += ln2pi_hi; + u.b += tlo; + return (u); +} +/* + * Rational approximation, A0 + x * x * P(x) / Q(x), on the interval + * [1.066.., 2.066..] accurate to 4.25e-19. + * + * Returns r.a + r.b = a0 + (z + c)^2 * p / q, with r.a truncated. + */ +static const union IEEEl2bits + a0_hiu = LD80C(0xe2b6e4153a57746c, -1, 8.85603194410888700265e-01L), + a0_lou = LD80C(0x851566d40f32c76d, -66, 1.40907742727049706207e-20L); +#define a0_hi (a0_hiu.e) +#define a0_lo (a0_lou.e) + +static const union IEEEl2bits +P0u = LD80C(0xdb629fb9bbdc1c1d, -2, 4.28486815855585429733e-01L), +P1u = LD80C(0xe6f4f9f5641aa6be, -3, 2.25543885805587730552e-01L), +P2u = LD80C(0xead1bd99fdaf7cc1, -6, 2.86644652514293482381e-02L), +P3u = LD80C(0x9ccc8b25838ab1e0, -8, 4.78512567772456362048e-03L), +P4u = LD80C(0x8f0c4383ef9ce72a, -9, 2.18273781132301146458e-03L), +P5u = LD80C(0xe732ab2c0a2778da, -13, 2.20487522485636008928e-04L), +P6u = LD80C(0xce70b27ca822b297, -16, 2.46095923774929264284e-05L), +P7u = LD80C(0xa309e2e16fb63663, -19, 2.42946473022376182921e-06L), +P8u = LD80C(0xaf9c110efb2c633d, -23, 1.63549217667765869987e-07L), +Q1u = LD80C(0xd4d7422719f48f15, -1, 8.31409582658993993626e-01L), +Q2u = LD80C(0xe13138ea404f1268, -5, -5.49785826915643198508e-02L), +Q3u = LD80C(0xd1c6cc91989352c0, -4, -1.02429960435139887683e-01L), +Q4u = LD80C(0xa7e9435a84445579, -7, 1.02484853505908820524e-02L), +Q5u = LD80C(0x83c7c34db89b7bda, -8, 4.02161632832052872697e-03L), +Q6u = LD80C(0xbed06bf6e1c14e5b, -11, -7.27898206351223022157e-04L), +Q7u = LD80C(0xef05bf841d4504c0, -18, 7.12342421869453515194e-06L), +Q8u = LD80C(0xf348d08a1ff53cb1, -19, 3.62522053809474067060e-06L); +#define P0 (P0u.e) +#define P1 (P1u.e) +#define P2 (P2u.e) +#define P3 (P3u.e) +#define P4 (P4u.e) +#define P5 (P5u.e) +#define P6 (P6u.e) +#define P7 (P7u.e) +#define P8 (P8u.e) +#define Q1 (Q1u.e) +#define Q2 (Q2u.e) +#define Q3 (Q3u.e) +#define Q4 (Q4u.e) +#define Q5 (Q5u.e) +#define Q6 (Q6u.e) +#define Q7 (Q7u.e) +#define Q8 (Q8u.e) + +static struct Double +ratfun_gam(long double z, long double c) +{ + long double p, q, thi, tlo; + struct Double r; + + q = 1 + z * (Q1 + z * (Q2 + z * (Q3 + z * (Q4 + z * (Q5 + + z * (Q6 + z * (Q7 + z * Q8))))))); + p = P0 + z * (P1 + z * (P2 + z * (P3 + z * (P4 + z * (P5 + + z * (P6 + z * (P7 + z * P8))))))); + p = p / q; + + /* Split z into high and low parts. */ + thi = (float)z; + tlo = (z - thi) + c; + tlo *= (thi + z); + + /* Split (z+c)^2 into high and low parts. */ + thi *= thi; + q = thi; + thi = (float)thi; + tlo += (q - thi); + + /* Split p/q into high and low parts. */ + r.a = (float)p; + r.b = p - r.a; + + tlo = tlo * p + thi * r.b + a0_lo; + thi *= r.a; /* t = (z+c)^2*(P/Q) */ + r.a = (float)(thi + a0_hi); + r.b = ((a0_hi - r.a) + thi) + tlo; + return (r); /* r = a0 + t */ +} +/* + * x < 6 + * + * Use argument reduction G(x+1) = xG(x) to reach the range [1.066124, + * 2.066124]. Use a rational approximation centered at the minimum + * (x0+1) to ensure monotonicity. + * + * Good to < 1 ulp. (provably .90 ulp; .87 ulp on 1,000,000 runs.) + * It also has correct monotonicity. + */ +static const union IEEEl2bits + xm1u = LD80C(0xec5b0c6ad7c7edc3, -2, 4.61632144968362341254e-01L); +#define x0 (xm1u.e) + +static const double + left = -0.3955078125; /* left boundary for rat. approx */ + +static long double +small_gam(long double x) +{ + long double t, y, ym1; + struct Double yy, r; + + y = x - 1; + + if (y <= 1 + (left + x0)) { + yy = ratfun_gam(y - x0, 0); + return (yy.a + yy.b); + } + + r.a = (float)y; + yy.a = r.a - 1; + y = y - 1 ; + r.b = yy.b = y - yy.a; + + /* Argument reduction: G(x+1) = x*G(x) */ + for (ym1 = y - 1; ym1 > left + x0; y = ym1--, yy.a--) { + t = r.a * yy.a; + r.b = r.a * yy.b + y * r.b; + r.a = (float)t; + r.b += (t - r.a); + } + + /* Return r*tgamma(y). */ + yy = ratfun_gam(y - x0, 0); + y = r.b * (yy.a + yy.b) + r.a * yy.b; + y += yy.a * r.a; + return (y); +} +/* + * Good on (0, 1+x0+left]. Accurate to 1 ulp. + */ +static long double +smaller_gam(long double x) +{ + long double d, rhi, rlo, t, xhi, xlo; + struct Double r; + + if (x < x0 + left) { + t = (float)x; + d = (t + x) * (x - t); + t *= t; + xhi = (float)(t + x); + xlo = x - xhi; + xlo += t; + xlo += d; + t = 1 - x0; + t += x; + d = 1 - x0; + d -= t; + d += x; + x = xhi + xlo; + } else { + xhi = (float)x; + xlo = x - xhi; + t = x - x0; + d = - x0 - t; + d += x; + } + + r = ratfun_gam(t, d); + d = (float)(r.a / x); + r.a -= d * xhi; + r.a -= d * xlo; + r.a += r.b; + + return (d + r.a / x); +} +/* + * x < 0 + * + * Use reflection formula, G(x) = pi/(sin(pi*x)*x*G(x)). + * At negative integers, return NaN and raise invalid. + */ +static const union IEEEl2bits +piu = LD80C(0xc90fdaa22168c235, 1, 3.14159265358979323851e+00L); +#define pi (piu.e) + +static long double +neg_gam(long double x) +{ + int sgn = 1; + struct Double lg, lsine; + long double y, z; + + y = ceill(x); + if (y == x) /* Negative integer. */ + return ((x - x) / zero); + + z = y - x; + if (z > 0.5) + z = 1 - z; + + y = y / 2; + if (y == ceill(y)) + sgn = -1; + + if (z < 0.25) + z = sinpil(z); + else + z = cospil(0.5 - z); + + /* Special case: G(1-x) = Inf; G(x) may be nonzero. */ + if (x < -1753) { + + if (x < -1760) + return (sgn * tiny * tiny); + y = expl(lgammal(x) / 2); + y *= y; + return (sgn < 0 ? -y : y); + } + + + y = 1 - x; + if (1 - y == x) + y = tgammal(y); + else /* 1-x is inexact */ + y = - x * tgammal(-x); + + if (sgn < 0) y = -y; + return (pi / (y * z)); +} +/* + * xmax comes from lgamma(xmax) - emax * log(2) = 0. + * static const float xmax = 35.040095f + * static const double xmax = 171.624376956302725; + * ld80: LD80C(0xdb718c066b352e20, 10, 1.75554834290446291689e+03L), + * ld128: 1.75554834290446291700388921607020320e+03L, + * + * iota is a sloppy threshold to isolate x = 0. + */ +static const double xmax = 1755.54834290446291689; +static const double iota = 0x1p-116; + +long double +tgammal(long double x) +{ + struct Double u; + + ENTERI(); + + if (x >= 6) { + if (x > xmax) + RETURNI(x / zero); + u = large_gam(x); + RETURNI(__exp__D(u.a, u.b)); + } + + if (x >= 1 + left + x0) + RETURNI(small_gam(x)); + + if (x > iota) + RETURNI(smaller_gam(x)); + + if (x > -iota) { + if (x != 0) + u.a = 1 - tiny; /* raise inexact */ + RETURNI(1 / x); + } + + if (!isfinite(x)) + RETURNI(x - x); /* x is NaN or -Inf */ + *** 2 LINES SKIPPED ***