| 1 | /* expm1l.c |
| 2 | * |
| 3 | * Exponential function, minus 1 |
| 4 | * 128-bit long double precision |
| 5 | * |
| 6 | * |
| 7 | * |
| 8 | * SYNOPSIS: |
| 9 | * |
| 10 | * long double x, y, expm1l(); |
| 11 | * |
| 12 | * y = expm1l( x ); |
| 13 | * |
| 14 | * |
| 15 | * |
| 16 | * DESCRIPTION: |
| 17 | * |
| 18 | * Returns e (2.71828...) raised to the x power, minus one. |
| 19 | * |
| 20 | * Range reduction is accomplished by separating the argument |
| 21 | * into an integer k and fraction f such that |
| 22 | * |
| 23 | * x k f |
| 24 | * e = 2 e. |
| 25 | * |
| 26 | * An expansion x + .5 x^2 + x^3 R(x) approximates exp(f) - 1 |
| 27 | * in the basic range [-0.5 ln 2, 0.5 ln 2]. |
| 28 | * |
| 29 | * |
| 30 | * ACCURACY: |
| 31 | * |
| 32 | * Relative error: |
| 33 | * arithmetic domain # trials peak rms |
| 34 | * IEEE -79,+MAXLOG 100,000 1.7e-34 4.5e-35 |
| 35 | * |
| 36 | */ |
| 37 | |
| 38 | /* Copyright 2001 by Stephen L. Moshier |
| 39 | |
| 40 | This library is free software; you can redistribute it and/or |
| 41 | modify it under the terms of the GNU Lesser General Public |
| 42 | License as published by the Free Software Foundation; either |
| 43 | version 2.1 of the License, or (at your option) any later version. |
| 44 | |
| 45 | This library is distributed in the hope that it will be useful, |
| 46 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 47 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 48 | Lesser General Public License for more details. |
| 49 | |
| 50 | You should have received a copy of the GNU Lesser General Public |
| 51 | License along with this library; if not, see |
| 52 | <https://www.gnu.org/licenses/>. */ |
| 53 | |
| 54 | |
| 55 | |
| 56 | #include <errno.h> |
| 57 | #include <float.h> |
| 58 | #include <math.h> |
| 59 | #include <math_private.h> |
| 60 | #include <math-underflow.h> |
| 61 | #include <libm-alias-ldouble.h> |
| 62 | |
| 63 | /* exp(x) - 1 = x + 0.5 x^2 + x^3 P(x)/Q(x) |
| 64 | -.5 ln 2 < x < .5 ln 2 |
| 65 | Theoretical peak relative error = 8.1e-36 */ |
| 66 | |
| 67 | static const _Float128 |
| 68 | P0 = L(2.943520915569954073888921213330863757240E8), |
| 69 | P1 = L(-5.722847283900608941516165725053359168840E7), |
| 70 | P2 = L(8.944630806357575461578107295909719817253E6), |
| 71 | P3 = L(-7.212432713558031519943281748462837065308E5), |
| 72 | P4 = L(4.578962475841642634225390068461943438441E4), |
| 73 | P5 = L(-1.716772506388927649032068540558788106762E3), |
| 74 | P6 = L(4.401308817383362136048032038528753151144E1), |
| 75 | P7 = L(-4.888737542888633647784737721812546636240E-1), |
| 76 | Q0 = L(1.766112549341972444333352727998584753865E9), |
| 77 | Q1 = L(-7.848989743695296475743081255027098295771E8), |
| 78 | Q2 = L(1.615869009634292424463780387327037251069E8), |
| 79 | Q3 = L(-2.019684072836541751428967854947019415698E7), |
| 80 | Q4 = L(1.682912729190313538934190635536631941751E6), |
| 81 | Q5 = L(-9.615511549171441430850103489315371768998E4), |
| 82 | Q6 = L(3.697714952261803935521187272204485251835E3), |
| 83 | Q7 = L(-8.802340681794263968892934703309274564037E1), |
| 84 | /* Q8 = 1.000000000000000000000000000000000000000E0 */ |
| 85 | /* C1 + C2 = ln 2 */ |
| 86 | |
| 87 | C1 = L(6.93145751953125E-1), |
| 88 | C2 = L(1.428606820309417232121458176568075500134E-6), |
| 89 | /* ln 2^-114 */ |
| 90 | minarg = L(-7.9018778583833765273564461846232128760607E1), big = L(1e4932); |
| 91 | |
| 92 | |
| 93 | _Float128 |
| 94 | __expm1l (_Float128 x) |
| 95 | { |
| 96 | _Float128 px, qx, xx; |
| 97 | int32_t ix, sign; |
| 98 | ieee854_long_double_shape_type u; |
| 99 | int k; |
| 100 | |
| 101 | /* Detect infinity and NaN. */ |
| 102 | u.value = x; |
| 103 | ix = u.parts32.w0; |
| 104 | sign = ix & 0x80000000; |
| 105 | ix &= 0x7fffffff; |
| 106 | if (!sign && ix >= 0x40060000) |
| 107 | { |
| 108 | /* If num is positive and exp >= 6 use plain exp. */ |
| 109 | return __expl (x); |
| 110 | } |
| 111 | if (ix >= 0x7fff0000) |
| 112 | { |
| 113 | /* Infinity (which must be negative infinity). */ |
| 114 | if (((ix & 0xffff) | u.parts32.w1 | u.parts32.w2 | u.parts32.w3) == 0) |
| 115 | return -1; |
| 116 | /* NaN. Invalid exception if signaling. */ |
| 117 | return x + x; |
| 118 | } |
| 119 | |
| 120 | /* expm1(+- 0) = +- 0. */ |
| 121 | if ((ix == 0) && (u.parts32.w1 | u.parts32.w2 | u.parts32.w3) == 0) |
| 122 | return x; |
| 123 | |
| 124 | /* Minimum value. */ |
| 125 | if (x < minarg) |
| 126 | return (4.0/big - 1); |
| 127 | |
| 128 | /* Avoid internal underflow when result does not underflow, while |
| 129 | ensuring underflow (without returning a zero of the wrong sign) |
| 130 | when the result does underflow. */ |
| 131 | if (fabsl (x) < L(0x1p-113)) |
| 132 | { |
| 133 | math_check_force_underflow (x); |
| 134 | return x; |
| 135 | } |
| 136 | |
| 137 | /* Express x = ln 2 (k + remainder), remainder not exceeding 1/2. */ |
| 138 | xx = C1 + C2; /* ln 2. */ |
| 139 | px = floorl (0.5 + x / xx); |
| 140 | k = px; |
| 141 | /* remainder times ln 2 */ |
| 142 | x -= px * C1; |
| 143 | x -= px * C2; |
| 144 | |
| 145 | /* Approximate exp(remainder ln 2). */ |
| 146 | px = (((((((P7 * x |
| 147 | + P6) * x |
| 148 | + P5) * x + P4) * x + P3) * x + P2) * x + P1) * x + P0) * x; |
| 149 | |
| 150 | qx = (((((((x |
| 151 | + Q7) * x |
| 152 | + Q6) * x + Q5) * x + Q4) * x + Q3) * x + Q2) * x + Q1) * x + Q0; |
| 153 | |
| 154 | xx = x * x; |
| 155 | qx = x + (0.5 * xx + xx * px / qx); |
| 156 | |
| 157 | /* exp(x) = exp(k ln 2) exp(remainder ln 2) = 2^k exp(remainder ln 2). |
| 158 | |
| 159 | We have qx = exp(remainder ln 2) - 1, so |
| 160 | exp(x) - 1 = 2^k (qx + 1) - 1 |
| 161 | = 2^k qx + 2^k - 1. */ |
| 162 | |
| 163 | px = __ldexpl (1, k); |
| 164 | x = px * qx + (px - 1.0); |
| 165 | return x; |
| 166 | } |
| 167 | libm_hidden_def (__expm1l) |
| 168 | libm_alias_ldouble (__expm1, expm1) |
| 169 | |