diff options
author | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2024-10-29 10:02:20 -0300 |
---|---|---|
committer | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2024-11-22 10:52:27 -0300 |
commit | baa495f2311f182d6a829cba1abc10363fd1e2cb (patch) | |
tree | e8ccfa4e73d0e95585f819f0d63a2131991ea827 /sysdeps/ieee754 | |
parent | 994fec2397770e0655d240f7c9f9e5c29dbb0926 (diff) | |
download | glibc-baa495f2311f182d6a829cba1abc10363fd1e2cb.zip glibc-baa495f2311f182d6a829cba1abc10363fd1e2cb.tar.gz glibc-baa495f2311f182d6a829cba1abc10363fd1e2cb.tar.bz2 |
math: Use erfcf from CORE-MATH
The CORE-MATH implementation is correctly rounded (for any rounding mode)
and shows better performance to the generic erfcf.
The code was adapted to glibc style and to use the definition of
math_config.h.
Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (M1,
gcc 13.2.1), and powerpc (POWER10, gcc 13.2.1):
latency master patched improvement
x86_64 98.8796 66.2142 33.04%
x86_64v2 98.9617 67.4221 31.87%
x86_64v3 87.4161 53.1754 39.17%
aarch64 33.8336 22.0781 34.75%
power10 21.1750 13.5864 35.84%
powerpc 21.4694 13.8149 35.65%
reciprocal-throughput master patched improvement
x86_64 48.5620 27.6731 43.01%
x86_64v2 47.9497 28.3804 40.81%
x86_64v3 42.0255 18.1355 56.85%
aarch64 24.3938 13.4041 45.05%
power10 10.4919 6.1881 41.02%
powerpc 11.763 6.76468 42.49%
Signed-off-by: Alexei Sibidanov <sibid@uvic.ca>
Signed-off-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Reviewed-by: DJ Delorie <dj@redhat.com>
Diffstat (limited to 'sysdeps/ieee754')
-rw-r--r-- | sysdeps/ieee754/flt-32/s_erfcf.c | 334 |
1 files changed, 174 insertions, 160 deletions
diff --git a/sysdeps/ieee754/flt-32/s_erfcf.c b/sysdeps/ieee754/flt-32/s_erfcf.c index fdba278..3dae2a0 100644 --- a/sysdeps/ieee754/flt-32/s_erfcf.c +++ b/sysdeps/ieee754/flt-32/s_erfcf.c @@ -1,173 +1,187 @@ -/* s_erfcf.c -- float version of s_erfc.c. - */ +/* Correctly-rounded complementary error function for the binary32 format -/* - * ==================================================== - * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. - * - * Developed at SunPro, a Sun Microsystems, Inc. business. - * Permission to use, copy, modify, and distribute this - * software is freely granted, provided that this notice - * is preserved. - * ==================================================== - */ +Copyright (c) 2023, 2024 Alexei Sibidanov. -#if defined(LIBM_SCCS) && !defined(lint) -static char rcsid[] = "$NetBSD: s_erff.c,v 1.4 1995/05/10 20:47:07 jtc Exp $"; -#endif +This file is part of the CORE-MATH project +project (file src/binary32/erfc/erfcf.c revision bc385c2). + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +*/ #include <errno.h> -#include <fix-int-fp-convert-zero.h> -#include <libm-alias-float.h> -#include <math-narrow-eval.h> #include <math.h> -#include <math_private.h> - +#include <stdint.h> +#include <libm-alias-float.h> +#include "math_config.h" -static const float -tiny = 1e-30, -half= 5.0000000000e-01, /* 0x3F000000 */ -one = 1.0000000000e+00, /* 0x3F800000 */ -two = 2.0000000000e+00, /* 0x40000000 */ - /* c = (subfloat)0.84506291151 */ -erx = 8.4506291151e-01, /* 0x3f58560b */ -/* - * Coefficients for approximation to erf on [0,0.84375] - */ -pp0 = 1.2837916613e-01, /* 0x3e0375d4 */ -pp1 = -3.2504209876e-01, /* 0xbea66beb */ -pp2 = -2.8481749818e-02, /* 0xbce9528f */ -pp3 = -5.7702702470e-03, /* 0xbbbd1489 */ -pp4 = -2.3763017452e-05, /* 0xb7c756b1 */ -qq1 = 3.9791721106e-01, /* 0x3ecbbbce */ -qq2 = 6.5022252500e-02, /* 0x3d852a63 */ -qq3 = 5.0813062117e-03, /* 0x3ba68116 */ -qq4 = 1.3249473704e-04, /* 0x390aee49 */ -qq5 = -3.9602282413e-06, /* 0xb684e21a */ -/* - * Coefficients for approximation to erf in [0.84375,1.25] - */ -pa0 = -2.3621185683e-03, /* 0xbb1acdc6 */ -pa1 = 4.1485610604e-01, /* 0x3ed46805 */ -pa2 = -3.7220788002e-01, /* 0xbebe9208 */ -pa3 = 3.1834661961e-01, /* 0x3ea2fe54 */ -pa4 = -1.1089469492e-01, /* 0xbde31cc2 */ -pa5 = 3.5478305072e-02, /* 0x3d1151b3 */ -pa6 = -2.1663755178e-03, /* 0xbb0df9c0 */ -qa1 = 1.0642088205e-01, /* 0x3dd9f331 */ -qa2 = 5.4039794207e-01, /* 0x3f0a5785 */ -qa3 = 7.1828655899e-02, /* 0x3d931ae7 */ -qa4 = 1.2617121637e-01, /* 0x3e013307 */ -qa5 = 1.3637083583e-02, /* 0x3c5f6e13 */ -qa6 = 1.1984500103e-02, /* 0x3c445aa3 */ -/* - * Coefficients for approximation to erfc in [1.25,1/0.35] - */ -ra0 = -9.8649440333e-03, /* 0xbc21a093 */ -ra1 = -6.9385856390e-01, /* 0xbf31a0b7 */ -ra2 = -1.0558626175e+01, /* 0xc128f022 */ -ra3 = -6.2375331879e+01, /* 0xc2798057 */ -ra4 = -1.6239666748e+02, /* 0xc322658c */ -ra5 = -1.8460508728e+02, /* 0xc3389ae7 */ -ra6 = -8.1287437439e+01, /* 0xc2a2932b */ -ra7 = -9.8143291473e+00, /* 0xc11d077e */ -sa1 = 1.9651271820e+01, /* 0x419d35ce */ -sa2 = 1.3765776062e+02, /* 0x4309a863 */ -sa3 = 4.3456588745e+02, /* 0x43d9486f */ -sa4 = 6.4538726807e+02, /* 0x442158c9 */ -sa5 = 4.2900814819e+02, /* 0x43d6810b */ -sa6 = 1.0863500214e+02, /* 0x42d9451f */ -sa7 = 6.5702495575e+00, /* 0x40d23f7c */ -sa8 = -6.0424413532e-02, /* 0xbd777f97 */ -/* - * Coefficients for approximation to erfc in [1/.35,28] - */ -rb0 = -9.8649431020e-03, /* 0xbc21a092 */ -rb1 = -7.9928326607e-01, /* 0xbf4c9dd4 */ -rb2 = -1.7757955551e+01, /* 0xc18e104b */ -rb3 = -1.6063638306e+02, /* 0xc320a2ea */ -rb4 = -6.3756646729e+02, /* 0xc41f6441 */ -rb5 = -1.0250950928e+03, /* 0xc480230b */ -rb6 = -4.8351919556e+02, /* 0xc3f1c275 */ -sb1 = 3.0338060379e+01, /* 0x41f2b459 */ -sb2 = 3.2579251099e+02, /* 0x43a2e571 */ -sb3 = 1.5367296143e+03, /* 0x44c01759 */ -sb4 = 3.1998581543e+03, /* 0x4547fdbb */ -sb5 = 2.5530502930e+03, /* 0x451f90ce */ -sb6 = 4.7452853394e+02, /* 0x43ed43a7 */ -sb7 = -2.2440952301e+01; /* 0xc1b38712 */ +static const double E[] = + { + 0x1p+0, 0x1.0163da9fb3335p+0, 0x1.02c9a3e778061p+0, + 0x1.04315e86e7f85p+0, 0x1.059b0d3158574p+0, 0x1.0706b29ddf6dep+0, + 0x1.0874518759bc8p+0, 0x1.09e3ecac6f383p+0, 0x1.0b5586cf9890fp+0, + 0x1.0cc922b7247f7p+0, 0x1.0e3ec32d3d1a2p+0, 0x1.0fb66affed31bp+0, + 0x1.11301d0125b51p+0, 0x1.12abdc06c31ccp+0, 0x1.1429aaea92dep+0, + 0x1.15a98c8a58e51p+0, 0x1.172b83c7d517bp+0, 0x1.18af9388c8deap+0, + 0x1.1a35beb6fcb75p+0, 0x1.1bbe084045cd4p+0, 0x1.1d4873168b9aap+0, + 0x1.1ed5022fcd91dp+0, 0x1.2063b88628cd6p+0, 0x1.21f49917ddc96p+0, + 0x1.2387a6e756238p+0, 0x1.251ce4fb2a63fp+0, 0x1.26b4565e27cddp+0, + 0x1.284dfe1f56381p+0, 0x1.29e9df51fdee1p+0, 0x1.2b87fd0dad99p+0, + 0x1.2d285a6e4030bp+0, 0x1.2ecafa93e2f56p+0, 0x1.306fe0a31b715p+0, + 0x1.32170fc4cd831p+0, 0x1.33c08b26416ffp+0, 0x1.356c55f929ff1p+0, + 0x1.371a7373aa9cbp+0, 0x1.38cae6d05d866p+0, 0x1.3a7db34e59ff7p+0, + 0x1.3c32dc313a8e5p+0, 0x1.3dea64c123422p+0, 0x1.3fa4504ac801cp+0, + 0x1.4160a21f72e2ap+0, 0x1.431f5d950a897p+0, 0x1.44e086061892dp+0, + 0x1.46a41ed1d0057p+0, 0x1.486a2b5c13cdp+0, 0x1.4a32af0d7d3dep+0, + 0x1.4bfdad5362a27p+0, 0x1.4dcb299fddd0dp+0, 0x1.4f9b2769d2ca7p+0, + 0x1.516daa2cf6642p+0, 0x1.5342b569d4f82p+0, 0x1.551a4ca5d920fp+0, + 0x1.56f4736b527dap+0, 0x1.58d12d497c7fdp+0, 0x1.5ab07dd485429p+0, + 0x1.5c9268a5946b7p+0, 0x1.5e76f15ad2148p+0, 0x1.605e1b976dc09p+0, + 0x1.6247eb03a5585p+0, 0x1.6434634ccc32p+0, 0x1.6623882552225p+0, + 0x1.68155d44ca973p+0, 0x1.6a09e667f3bcdp+0, 0x1.6c012750bdabfp+0, + 0x1.6dfb23c651a2fp+0, 0x1.6ff7df9519484p+0, 0x1.71f75e8ec5f74p+0, + 0x1.73f9a48a58174p+0, 0x1.75feb564267c9p+0, 0x1.780694fde5d3fp+0, + 0x1.7a11473eb0187p+0, 0x1.7c1ed0130c132p+0, 0x1.7e2f336cf4e62p+0, + 0x1.80427543e1a12p+0, 0x1.82589994cce13p+0, 0x1.8471a4623c7adp+0, + 0x1.868d99b4492edp+0, 0x1.88ac7d98a6699p+0, 0x1.8ace5422aa0dbp+0, + 0x1.8cf3216b5448cp+0, 0x1.8f1ae99157736p+0, 0x1.9145b0b91ffc6p+0, + 0x1.93737b0cdc5e5p+0, 0x1.95a44cbc8520fp+0, 0x1.97d829fde4e5p+0, + 0x1.9a0f170ca07bap+0, 0x1.9c49182a3f09p+0, 0x1.9e86319e32323p+0, + 0x1.a0c667b5de565p+0, 0x1.a309bec4a2d33p+0, 0x1.a5503b23e255dp+0, + 0x1.a799e1330b358p+0, 0x1.a9e6b5579fdbfp+0, 0x1.ac36bbfd3f37ap+0, + 0x1.ae89f995ad3adp+0, 0x1.b0e07298db666p+0, 0x1.b33a2b84f15fbp+0, + 0x1.b59728de5593ap+0, 0x1.b7f76f2fb5e47p+0, 0x1.ba5b030a1064ap+0, + 0x1.bcc1e904bc1d2p+0, 0x1.bf2c25bd71e09p+0, 0x1.c199bdd85529cp+0, + 0x1.c40ab5fffd07ap+0, 0x1.c67f12e57d14bp+0, 0x1.c8f6d9406e7b5p+0, + 0x1.cb720dcef9069p+0, 0x1.cdf0b555dc3fap+0, 0x1.d072d4a07897cp+0, + 0x1.d2f87080d89f2p+0, 0x1.d5818dcfba487p+0, 0x1.d80e316c98398p+0, + 0x1.da9e603db3285p+0, 0x1.dd321f301b46p+0, 0x1.dfc97337b9b5fp+0, + 0x1.e264614f5a129p+0, 0x1.e502ee78b3ff6p+0, 0x1.e7a51fbc74c83p+0, + 0x1.ea4afa2a490dap+0, 0x1.ecf482d8e67f1p+0, 0x1.efa1bee615a27p+0, + 0x1.f252b376bba97p+0, 0x1.f50765b6e454p+0, 0x1.f7bfdad9cbe14p+0, + 0x1.fa7c1819e90d8p+0, 0x1.fd3c22b8f71f1p+0 + }; -float __erfcf(float x) +float +__erfcf (float xf) { - int32_t hx,ix; - float R,S,P,Q,s,y,z,r; - GET_FLOAT_WORD(hx,x); - ix = hx&0x7fffffff; - if(ix>=0x7f800000) { /* erfc(nan)=nan */ - /* erfc(+-inf)=0,2 */ - float ret = (float)(((uint32_t)hx>>31)<<1)+one/x; - if (FIX_INT_FP_CONVERT_ZERO && ret == 0.0f) - return 0.0f; - return ret; - } - - if(ix < 0x3f580000) { /* |x|<0.84375 */ - if(ix < 0x32800000) /* |x|<2**-26 */ - return one-x; - z = x*x; - r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); - s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); - y = r/s; - if(hx < 0x3e800000) { /* x<1/4 */ - return one-(x+x*y); - } else { - r = x*y; - r += (x-half); - return half - r ; - } + float axf = fabsf (xf); + double axd = axf; + double x2 = axd * axd; + uint32_t t = asuint (xf); + unsigned int at = t & (~0u >> 1); + unsigned int sgn = t >> 31; + int64_t i = at > 0x40051000; + /* for x < -0x1.ea8f94p+1, erfc(x) rounds to 2 (to nearest) */ + if (__glibc_unlikely (t > 0xc07547ca)) + { /* xf < -0x1.ea8f94p+1 */ + if (__glibc_unlikely (t >= 0xff800000)) + { /* -Inf or NaN */ + if (t == 0xff800000) + return 2.0f; /* -Inf */ + return xf + xf; /* NaN */ } - if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ - s = fabsf(x)-one; - P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); - Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); - if(hx>=0) { - z = one-erx; return z - P/Q; - } else { - z = erx+P/Q; return one+z; - } + return 2.0f - 0x1p-25f; /* rounds to 2 or nextbelow(2) */ + } + /* at is the absolute value of xf + for x >= 0x1.41bbf8p+3, erfc(x) < 2^-150, thus rounds to 0 or to 2^-149 + depending on the rounding mode */ + if (__glibc_unlikely (at >= 0x4120ddfc)) + { /* |xf| >= 0x1.41bbf8p+3 */ + if (__glibc_unlikely (at >= 0x7f800000)) + { /* +Inf or NaN */ + if (at == 0x7f800000) + return 0.0f; /* +Inf */ + return xf + xf; /* NaN */ } - if (ix < 0x41e00000) { /* |x|<28 */ - x = fabsf(x); - s = one/(x*x); - if(ix< 0x4036DB6D) { /* |x| < 1/.35 ~ 2.857143*/ - R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( - ra5+s*(ra6+s*ra7)))))); - S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( - sa5+s*(sa6+s*(sa7+s*sa8))))))); - } else { /* |x| >= 1/.35 ~ 2.857143 */ - if(hx<0&&ix>=0x40c00000) return two-tiny;/* x < -6 */ - R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*( - rb5+s*rb6))))); - S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*( - sb5+s*(sb6+s*sb7)))))); - } - GET_FLOAT_WORD(ix,x); - SET_FLOAT_WORD(z,ix&0xffffe000); - r = __ieee754_expf(-z*z-(float)0.5625)* - __ieee754_expf((z-x)*(z+x)+R/S); - if(hx>0) { - float ret = math_narrow_eval (r/x); - if (ret == 0) - __set_errno (ERANGE); - return ret; - } else - return two-r/x; - } else { - if(hx>0) { - __set_errno (ERANGE); - return tiny*tiny; - } else - return two-tiny; + __set_errno (ERANGE); + /* 0x1p-149f * 0.25f rounds to 0 or 2^-149 depending on rounding */ + return 0x1p-149f * 0.25f; + } + if (__glibc_unlikely (at <= 0x3db80000)) + { /* |x| <= 0x1.7p-4 */ + if (__glibc_unlikely (t == 0xb76c9f62)) + return 0x1.00010ap+0f + 0x1p-25f; /* exceptional case */ + /* for |x| <= 0x1.c5bf88p-26. erfc(x) rounds to 1 (to nearest) */ + if (__glibc_unlikely (at <= 0x32e2dfc4)) + { /* |x| <= 0x1.c5bf88p-26 */ + if (__glibc_unlikely (at == 0)) + return 1.0f; + static const float d[] = { -0x1p-26, 0x1p-25 }; + return 1.0f + d[sgn]; } + /* around 0, erfc(x) behaves as 1 - (odd polynomial) */ + static const double c[] = + { + 0x1.20dd750429b6dp+0, -0x1.812746b03610bp-2, 0x1.ce2f218831d2fp-4, + -0x1.b82c609607dcbp-6, 0x1.553af09b8008ep-8 + }; + double f0 = xf + * (c[0] + x2 * (c[1] + x2 * (c[2] + x2 * (c[3] + x2 * (c[4]))))); + return 1.0 - f0; + } + + /* now -0x1.ea8f94p+1 <= x <= 0x1.41bbf8p+3, with |x| > 0x1.7p-4 */ + const double iln2 = 0x1.71547652b82fep+0; + const double ln2h = 0x1.62e42fefap-8; + const double ln2l = 0x1.cf79abd6f5dc8p-47; + uint64_t jt = asuint64 (fma (x2, iln2, -(1024 + 0x1p-8))); + int64_t j = (int64_t) (jt << 12) >> 48; + double S = asdouble (((j >> 7) + (0x3ff | sgn << 11)) << 52); + static const double ch[] = + { + -0x1.ffffffffff333p-2, 0x1.5555555556a14p-3, -0x1.55556666659b4p-5, + 0x1.1111074cc7b22p-7 + }; + double d = (x2 + ln2h * j) + ln2l * j; + double d2 = d * d; + double e0 = E[j & 127]; + double f = d + d2 * ((ch[0] + d * ch[1]) + d2 * (ch[2] + d * ch[3])); + static const double ct[][16] = + { + { + 0x1.c162355429b28p-1, 0x1.d99999999999ap+1, 0x1.da951cece2b85p-2, + -0x1.70ef6cff4bcc4p+0, 0x1.3d7f7b3d617dep+1, -0x1.9d0aa47537c51p+1, + 0x1.9754ea9a3fcb1p+1, -0x1.27a5453fcc015p+1, 0x1.1ef2e0531aebap+0, + -0x1.eca090f5a1c06p-3, -0x1.7a3cd173a063cp-4, 0x1.30fa68a68fdddp-4, + 0x1.55ad9a326993ap-10, -0x1.07e7b0bb39fbfp-6, 0x1.2328706c0e95p-10, + 0x1.d6aa0b7b19cfep-9 + }, + { + 0x1.137c8983f8516p+2, 0x1.799999999999ap+1, 0x1.05b53aa241333p-3, + -0x1.a3f53872bf87p-3, 0x1.de4c30742c9d5p-4, -0x1.cb24bfa591986p-5, + 0x1.666aec059ca5fp-6, -0x1.a61250eb26b0bp-8, 0x1.2b28b7924b34dp-10, + 0x1.41b13a9d45013p-15, -0x1.6dd5e8a273613p-14, 0x1.09ce8ea5e8da5p-16, + 0x1.33923b4102981p-18, -0x1.1dfd161e3f984p-19, -0x1.c87618fcae3b3p-23, + 0x1.e8a6ffa0ba2c7p-23 + } + }; + double z = (axd - ct[i][0]) / (axd + ct[i][1]); + double z2 = z * z, z4 = z2 * z2; + double z8 = z4 * z4; + const double *c = ct[i] + 3; + double s = (((c[0] + z * c[1]) + z2 * (c[2] + z * c[3])) + + z4 * ((c[4] + z * c[5]) + z2 * (c[6] + z * c[7]))) + + z8 * (((c[8] + z * c[9]) + z2 * (c[10] + z * c[11])) + z4 * (c[12])); + s = ct[i][2] + z * s; + static const double off[] = { 0, 2 }; + double r = (S * (e0 - f * e0)) * s; + double y = off[sgn] + r; + return y; } libm_alias_float (__erfc, erfc) |