1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358
|
/**
* @license Apache-2.0
*
* Copyright (c) 2018 The Stdlib Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*
* ## Notice
*
* The following copyright, license, and long comment were part of the original implementation available as part of [FDLIBM]{@link http://www.netlib.org/fdlibm/s_expm1.c}. The implementation follows the original, but has been modified for JavaScript.
*
* ```text
* Copyright (C) 2004 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.
* ```
*/
'use strict';
// MODULES //
var isnan = require( './../../../../base/assert/is-nan' );
var getHighWord = require( '@stdlib/number/float64/base/get-high-word' );
var setHighWord = require( '@stdlib/number/float64/base/set-high-word' );
var PINF = require( '@stdlib/constants/float64/pinf' );
var NINF = require( '@stdlib/constants/float64/ninf' );
var FLOAT64_EXPONENT_BIAS = require( '@stdlib/constants/float64/exponent-bias' );
var HALF_LN2 = require( '@stdlib/constants/float64/half-ln-two' );
var polyval = require( './polyval_q.js' );
// VARIABLES //
var OVERFLOW_THRESHOLD = 7.09782712893383973096e+02; // 0x40862E42 0xFEFA39EF
// High and low words of ln(2):
var LN2_HI = 6.93147180369123816490e-01; // 0x3FE62E42 0xFEE00000
var LN2_LO = 1.90821492927058770002e-10; // 0x3DEA39EF 0x35793C76
// 1 / ln(2):
var LN2_INV = 1.44269504088896338700e+00; // 0x3FF71547 0x652B82FE
// ln(2) * 56:
var LN2x56 = 3.88162421113569373274e+01; // 0x4043687A 0x9F1AF2B1
// ln(2) * 1.5:
var LN2_HALFX3 = 1.03972077083991796413e+00; // 0x3FF0A2B2 0x3F3BAB73
// MAIN //
/**
* Computes `exp(x) - 1`.
*
* ## Method
*
* 1. Given \\(x\\), we use argument reduction to find \\(r\\) and an integer \\(k\\) such that
*
* ```tex
* x = k \cdot \ln(2) + r
* ```
*
* where
*
* ```tex
* |r| \leq \frac{\ln(2)}{2} \approx 0.34658
* ```
*
* <!-- <note> -->
*
* A correction term \\(c\\) will need to be computed to compensate for the error in \\(r\\) when rounded to a floating-point number.
*
* <!-- </note> -->
*
* 2. To approximate \\(\operatorname{expm1}(r)\\), we use a special rational function on the interval \\(\[0,0.34658]\\). Since
*
* ```tex
* r \frac{e^r + 1}{e^r - 1} = 2 + \frac{r^2}{6} - \frac{r^4}{360} + \ldots
* ```
*
* we define \\(\operatorname{R1}(r^2)\\) by
*
* ```tex
* r \frac{e^r + 1}{e^r - 1} = 2 + \frac{r^2}{6} \operatorname{R1}(r^2)
* ```
*
* That is,
*
* ```tex
* \begin{align*}
* \operatorname{R1}(r^2) &= \frac{6}{r} \biggl(\frac{e^r+1}{e^r-1} - \frac{2}{r}\biggr) \\
* &= \frac{6}{r} \biggl( 1 + 2 \biggl(\frac{1}{e^r-1} - \frac{1}{r}\biggr)\biggr) \\
* &= 1 - \frac{r^2}{60} + \frac{r^4}{2520} - \frac{r^6}{100800} + \ldots
* \end{align*}
* ```
*
* We use a special Remes algorithm on \\(\[0,0.347]\\) to generate a polynomial of degree \\(5\\) in \\(r^2\\) to approximate \\(\mathrm{R1}\\). The maximum error of this polynomial approximation is bounded by \\(2^{-61}\\). In other words,
*
* ```tex
* \operatorname{R1}(z) \approx 1 + \mathrm{Q1} \cdot z + \mathrm{Q2} \cdot z^2 + \mathrm{Q3} \cdot z^3 + \mathrm{Q4} \cdot z^4 + \mathrm{Q5} \cdot z^5
* ```
*
* where
*
* ```tex
* \begin{align*}
* \mathrm{Q1} &= -1.6666666666666567384\mbox{e-}2 \\
* \mathrm{Q2} &= 3.9682539681370365873\mbox{e-}4 \\
* \mathrm{Q3} &= -9.9206344733435987357\mbox{e-}6 \\
* \mathrm{Q4} &= 2.5051361420808517002\mbox{e-}7 \\
* \mathrm{Q5} &= -6.2843505682382617102\mbox{e-}9
* \end{align*}
* ```
*
* where \\(z = r^2\\) and the values of \\(\mathrm{Q1}\\) to \\(\mathrm{Q5}\\) are listed in the source. The error is bounded by
*
* ```tex
* \biggl| 1 + \mathrm{Q1} \cdot z + \ldots + \mathrm{Q5} \cdot z - \operatorname{R1}(z) \biggr| \leq 2^{-61}
* ```
*
* \\(\operatorname{expm1}(r) = e^r - 1\\) is then computed by the following specific way which minimizes the accumulated rounding error
*
* ```tex
* \operatorname{expm1}(r) = r + \frac{r^2}{2} + \frac{r^3}{2} \biggl( \frac{3 - (\mathrm{R1} + \mathrm{R1} \cdot \frac{r}{2})}{6 - r ( 3 - \mathrm{R1} \cdot \frac{r}{2})} \biggr)
* ```
*
* To compensate for the error in the argument reduction, we use
*
* ```tex
* \begin{align*}
* \operatorname{expm1}(r+c) &= \operatorname{expm1}(r) + c + \operatorname{expm1}(r) \cdot c \\
* &\approx \operatorname{expm1}(r) + c + rc
* \end{align*}
* ```
*
* Thus, \\(c + rc\\) will be added in as the correction terms for \\(\operatorname{expm1}(r+c)\\). Now, we can rearrange the term to avoid optimization screw up.
*
* ```tex
* \begin{align*}
* \operatorname{expm1}(r+c) &\approx r - \biggl( \biggl( r + \biggl( \frac{r^2}{2} \biggl( \frac{\mathrm{R1} - (3 - \mathrm{R1} \cdot \frac{r}{2})}{6 - r (3 - \mathrm{R1} \cdot \frac{r}{2})} \biggr) - c \biggr) - c \biggr) - \frac{r^2}{2} \biggr) \\
* &= r - \mathrm{E}
* \end{align*}
* ```
*
* 3. To scale back to obtain \\(\operatorname{expm1}(x)\\), we have (from step 1)
*
* ```tex
* \operatorname{expm1}(x) = \begin{cases}
* 2^k (\operatorname{expm1}(r) + 1) - 1 \\
* 2^k (\operatorname{expm1}(r) + (1-2^{-k}))
* \end{cases}
* ```
*
* ## Special Cases
*
* ```tex
* \begin{align*}
* \operatorname{expm1}(\infty) &= \infty \\
* \operatorname{expm1}(-\infty) &= -1 \\
* \operatorname{expm1}(\mathrm{NaN}) &= \mathrm{NaN}
* \end{align*}
* ```
*
*
* ## Notes
*
* - For finite arguments, only \\(\operatorname{expm1}(0) = 0\\) is exact.
*
* - To save one multiplication, we scale the coefficient \\(\mathrm{Qi}\\) to \\(\mathrm{Qi} \cdot {2^i}\\) and replace \\(z\\) by \\(\frac{x^2}{2}\\).
*
* - To achieve maximum accuracy, we compute \\(\operatorname{expm1}(x)\\) by
*
* - if \\(x < -56 \cdot \ln(2)\\), return \\(-1.0\\) (raise inexact if \\(x\\) does not equal \\(\infty\\))
*
* - if \\(k = 0\\), return \\(r-\mathrm{E}\\)
*
* - if \\(k = -1\\), return \\(\frac{(r-\mathrm{E})-1}{2}\\)
*
* - if \\(k = 1\\),
*
* - if \\(r < -0.25\\), return \\(2((r+0.5)- \mathrm{E})\\)
* - else return \\(1+2(r-\mathrm{E})\\)
*
* - if \\(k < -2\\) or \\(k > 56\\), return \\(2^k(1-(\mathrm{E}-r)) - 1\\) (or \\(e^x-1\\))
*
* - if \\(k \leq 20\\), return \\(2^k((1-2^{-k})-(\mathrm{E}-r))\\)
*
* - else return \\(2^k(1-((\mathrm{E}+2^{-k})-r))\\)
*
* - For IEEE 754 double, if \\(x > 7.09782712893383973096\mbox{e+}02\\), then \\(\operatorname{expm1}(x)\\) will overflow.
*
* - The hexadecimal values listed in the source are the intended ones for the implementation constants. Decimal values may be used, provided that the compiler will convert from decimal to binary accurately enough to produce the intended hexadecimal values.
*
* - According to an error analysis, the error is always less than \\(1\\) ulp (unit in the last place).
*
*
* @param {number} x - input value
* @returns {number} function value
*
* @example
* var v = expm1( 0.2 );
* // returns ~0.221
*
* @example
* var v = expm1( -9.0 );
* // returns ~-0.9999
*
* @example
* var v = expm1( 0.0 );
* // returns 0.0
*
* @example
* var v = expm1( NaN );
* // returns NaN
*/
function expm1( x ) {
var halfX;
var sign;
var hi;
var lo;
var hx;
var r1;
var y;
var z;
var c;
var t;
var e;
var k;
if ( x === PINF || isnan( x ) ) {
return x;
}
if ( x === NINF ) {
return -1.0;
}
if ( x === 0.0 ) {
return x; // handles +-0 (IEEE 754-2008)
}
// Set y = |x|:
if ( x < 0.0 ) {
sign = true;
y = -x;
} else {
sign = false;
y = x;
}
// Filter out huge and non-finite arguments...
if ( y >= LN2x56 ) { // if |x| >= 56*ln(2)
if ( sign ) { // if x <= -56*ln(2)
return -1.0;
}
if ( y >= OVERFLOW_THRESHOLD ) { // if |x| >= 709.78...
return PINF;
}
}
// Extract the more significant bits from |x|:
hx = getHighWord( y )|0; // asm type annotation
// Argument reduction...
if ( y > HALF_LN2 ) { // if |x| > 0.5*ln(2)
if ( y < LN2_HALFX3 ) { // if |x| < 1.5*ln(2)
if ( sign ) {
hi = x + LN2_HI;
lo = -LN2_LO;
k = -1;
} else {
hi = x - LN2_HI;
lo = LN2_LO;
k = 1;
}
} else {
if ( sign ) {
k = (LN2_INV*x) - 0.5;
} else {
k = (LN2_INV*x) + 0.5;
}
k |= 0; // use a bitwise OR to cast `k` to an integer (see also asm.js type annotations: http://asmjs.org/spec/latest/#annotations)
t = k;
hi = x - (t*LN2_HI); // t*ln2_hi is exact here
lo = t * LN2_LO;
}
x = hi - lo;
c = (hi-x) - lo;
}
// if |x| < 2**-54 => high word: 0 01111001001 00000000000000000000 => 0x3c900000 = 1016070144 => exponent = 01111001001 = 969 = 1023-54
else if ( hx < 1016070144 ) {
return x;
}
else {
k = 0;
}
// x is now in primary range...
halfX = 0.5 * x;
z = x * halfX;
r1 = 1.0 + ( z * polyval( z ) );
t = 3.0 - (r1*halfX);
e = z * ( (r1-t) / (6.0 - (x*t)) );
if ( k === 0 ) {
return x - ( (x*e) - z ); // c is 0
}
e = ( x * (e-c) ) - c;
e -= z;
if ( k === -1 ) {
return ( 0.5*(x-e) )- 0.5;
}
if ( k === 1 ) {
if ( x < -0.25 ) {
return -2.0 * ( e - (x+0.5) );
}
return 1 + ( 2.0 * (x-e) );
}
if ( k <= -2 || k > 56 ) { // suffice to return exp(x)-1
y = 1.0 - (e-x);
// Add k to y's exponent:
hi = (getHighWord( y ) + (k<<20))|0; // asm type annotation
y = setHighWord( y, hi );
return y - 1.0;
}
t = 1.0;
if ( k < 20 ) {
// 0x3ff00000 - (0x200000>>k) = 1072693248 - (0x200000>>k) => 0x200000 = 0 00000000010 00000000000000000000
hi = (1072693248 - (0x200000>>k))|0; // asm type annotation
t = setHighWord( t, hi ); // t=1-2^-k
y = t - (e-x);
} else {
hi = ( (FLOAT64_EXPONENT_BIAS-k)<<20 )|0; // asm type annotation
t = setHighWord( t, hi ); // t=2^-k
y = x - (e+t);
y += 1.0;
}
// Add k to y's exponent:
hi = (getHighWord( y ) + (k<<20))|0; // asm type annotation
return setHighWord( y, hi );
}
// EXPORTS //
module.exports = expm1;
|