| 1 | /* origin: OpenBSD /usr/src/lib/libm/src/ld80/e_tgammal.c */ |
| 2 | /* |
| 3 | * Copyright (c) 2008 Stephen L. Moshier <steve@moshier.net> |
| 4 | * |
| 5 | * Permission to use, copy, modify, and distribute this software for any |
| 6 | * purpose with or without fee is hereby granted, provided that the above |
| 7 | * copyright notice and this permission notice appear in all copies. |
| 8 | * |
| 9 | * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES |
| 10 | * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF |
| 11 | * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR |
| 12 | * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES |
| 13 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN |
| 14 | * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF |
| 15 | * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. |
| 16 | */ |
| 17 | /* |
| 18 | * Gamma function |
| 19 | * |
| 20 | * |
| 21 | * SYNOPSIS: |
| 22 | * |
| 23 | * long double x, y, tgammal(); |
| 24 | * |
| 25 | * y = tgammal( x ); |
| 26 | * |
| 27 | * |
| 28 | * DESCRIPTION: |
| 29 | * |
| 30 | * Returns gamma function of the argument. The result is |
| 31 | * correctly signed. |
| 32 | * |
| 33 | * Arguments |x| <= 13 are reduced by recurrence and the function |
| 34 | * approximated by a rational function of degree 7/8 in the |
| 35 | * interval (2,3). Large arguments are handled by Stirling's |
| 36 | * formula. Large negative arguments are made positive using |
| 37 | * a reflection formula. |
| 38 | * |
| 39 | * |
| 40 | * ACCURACY: |
| 41 | * |
| 42 | * Relative error: |
| 43 | * arithmetic domain # trials peak rms |
| 44 | * IEEE -40,+40 10000 3.6e-19 7.9e-20 |
| 45 | * IEEE -1755,+1755 10000 4.8e-18 6.5e-19 |
| 46 | * |
| 47 | * Accuracy for large arguments is dominated by error in powl(). |
| 48 | * |
| 49 | */ |
| 50 | |
| 51 | #include "libm.h" |
| 52 | |
| 53 | #if LDBL_MANT_DIG == 53 && LDBL_MAX_EXP == 1024 |
| 54 | long double tgammal(long double x) |
| 55 | { |
| 56 | 	return tgamma(x); |
| 57 | } |
| 58 | #elif LDBL_MANT_DIG == 64 && LDBL_MAX_EXP == 16384 |
| 59 | /* |
| 60 | tgamma(x+2) = tgamma(x+2) P(x)/Q(x) |
| 61 | 0 <= x <= 1 |
| 62 | Relative error |
| 63 | n=7, d=8 |
| 64 | Peak error = 1.83e-20 |
| 65 | Relative error spread = 8.4e-23 |
| 66 | */ |
| 67 | static const long double P[8] = { |
| 68 | 4.212760487471622013093E-5L, |
| 69 | 4.542931960608009155600E-4L, |
| 70 | 4.092666828394035500949E-3L, |
| 71 | 2.385363243461108252554E-2L, |
| 72 | 1.113062816019361559013E-1L, |
| 73 | 3.629515436640239168939E-1L, |
| 74 | 8.378004301573126728826E-1L, |
| 75 | 1.000000000000000000009E0L, |
| 76 | }; |
| 77 | static const long double Q[9] = { |
| 78 | -1.397148517476170440917E-5L, |
| 79 | 2.346584059160635244282E-4L, |
| 80 | -1.237799246653152231188E-3L, |
| 81 | -7.955933682494738320586E-4L, |
| 82 | 2.773706565840072979165E-2L, |
| 83 | -4.633887671244534213831E-2L, |
| 84 | -2.243510905670329164562E-1L, |
| 85 | 4.150160950588455434583E-1L, |
| 86 | 9.999999999999999999908E-1L, |
| 87 | }; |
| 88 | |
| 89 | /* |
| 90 | static const long double P[] = { |
| 91 | -3.01525602666895735709e0L, |
| 92 | -3.25157411956062339893e1L, |
| 93 | -2.92929976820724030353e2L, |
| 94 | -1.70730828800510297666e3L, |
| 95 | -7.96667499622741999770e3L, |
| 96 | -2.59780216007146401957e4L, |
| 97 | -5.99650230220855581642e4L, |
| 98 | -7.15743521530849602425e4L |
| 99 | }; |
| 100 | static const long double Q[] = { |
| 101 | 1.00000000000000000000e0L, |
| 102 | -1.67955233807178858919e1L, |
| 103 | 8.85946791747759881659e1L, |
| 104 | 5.69440799097468430177e1L, |
| 105 | -1.98526250512761318471e3L, |
| 106 | 3.31667508019495079814e3L, |
| 107 | 1.60577839621734713377e4L, |
| 108 | -2.97045081369399940529e4L, |
| 109 | -7.15743521530849602412e4L |
| 110 | }; |
| 111 | */ |
| 112 | #define MAXGAML 1755.455L |
| 113 | /*static const long double LOGPI = 1.14472988584940017414L;*/ |
| 114 | |
| 115 | /* Stirling's formula for the gamma function |
| 116 | tgamma(x) = sqrt(2 pi) x^(x-.5) exp(-x) (1 + 1/x P(1/x)) |
| 117 | z(x) = x |
| 118 | 13 <= x <= 1024 |
| 119 | Relative error |
| 120 | n=8, d=0 |
| 121 | Peak error = 9.44e-21 |
| 122 | Relative error spread = 8.8e-4 |
| 123 | */ |
| 124 | static const long double STIR[9] = { |
| 125 | 7.147391378143610789273E-4L, |
| 126 | -2.363848809501759061727E-5L, |
| 127 | -5.950237554056330156018E-4L, |
| 128 | 6.989332260623193171870E-5L, |
| 129 | 7.840334842744753003862E-4L, |
| 130 | -2.294719747873185405699E-4L, |
| 131 | -2.681327161876304418288E-3L, |
| 132 | 3.472222222230075327854E-3L, |
| 133 | 8.333333333333331800504E-2L, |
| 134 | }; |
| 135 | |
| 136 | #define MAXSTIR 1024.0L |
| 137 | static const long double SQTPI = 2.50662827463100050242E0L; |
| 138 | |
| 139 | /* 1/tgamma(x) = z P(z) |
| 140 | * z(x) = 1/x |
| 141 | * 0 < x < 0.03125 |
| 142 | * Peak relative error 4.2e-23 |
| 143 | */ |
| 144 | static const long double S[9] = { |
| 145 | -1.193945051381510095614E-3L, |
| 146 | 7.220599478036909672331E-3L, |
| 147 | -9.622023360406271645744E-3L, |
| 148 | -4.219773360705915470089E-2L, |
| 149 | 1.665386113720805206758E-1L, |
| 150 | -4.200263503403344054473E-2L, |
| 151 | -6.558780715202540684668E-1L, |
| 152 | 5.772156649015328608253E-1L, |
| 153 | 1.000000000000000000000E0L, |
| 154 | }; |
| 155 | |
| 156 | /* 1/tgamma(-x) = z P(z) |
| 157 | * z(x) = 1/x |
| 158 | * 0 < x < 0.03125 |
| 159 | * Peak relative error 5.16e-23 |
| 160 | * Relative error spread = 2.5e-24 |
| 161 | */ |
| 162 | static const long double SN[9] = { |
| 163 | 1.133374167243894382010E-3L, |
| 164 | 7.220837261893170325704E-3L, |
| 165 | 9.621911155035976733706E-3L, |
| 166 | -4.219773343731191721664E-2L, |
| 167 | -1.665386113944413519335E-1L, |
| 168 | -4.200263503402112910504E-2L, |
| 169 | 6.558780715202536547116E-1L, |
| 170 | 5.772156649015328608727E-1L, |
| 171 | -1.000000000000000000000E0L, |
| 172 | }; |
| 173 | |
| 174 | static const long double PIL = 3.1415926535897932384626L; |
| 175 | |
| 176 | /* Gamma function computed by Stirling's formula. |
| 177 | */ |
| 178 | static long double stirf(long double x) |
| 179 | { |
| 180 | 	long double y, w, v; |
| 181 | |
| 182 | 	w = 1.0/x; |
| 183 | 	/* For large x, use rational coefficients from the analytical expansion. */ |
| 184 | 	if (x > 1024.0) |
| 185 | 		w = (((((6.97281375836585777429E-5L * w |
| 186 | 		 + 7.84039221720066627474E-4L) * w |
| 187 | 		 - 2.29472093621399176955E-4L) * w |
| 188 | 		 - 2.68132716049382716049E-3L) * w |
| 189 | 		 + 3.47222222222222222222E-3L) * w |
| 190 | 		 + 8.33333333333333333333E-2L) * w |
| 191 | 		 + 1.0; |
| 192 | 	else |
| 193 | 		w = 1.0 + w * __polevll(w, STIR, 8); |
| 194 | 	y = expl(x); |
| 195 | 	if (x > MAXSTIR) { /* Avoid overflow in pow() */ |
| 196 | 		v = powl(x, 0.5L * x - 0.25L); |
| 197 | 		y = v * (v / y); |
| 198 | 	} else { |
| 199 | 		y = powl(x, x - 0.5L) / y; |
| 200 | 	} |
| 201 | 	y = SQTPI * y * w; |
| 202 | 	return y; |
| 203 | } |
| 204 | |
| 205 | long double tgammal(long double x) |
| 206 | { |
| 207 | 	long double p, q, z; |
| 208 | |
| 209 | 	if (!isfinite(x)) |
| 210 | 		return x + INFINITY; |
| 211 | |
| 212 | 	q = fabsl(x); |
| 213 | 	if (q > 13.0) { |
| 214 | 		if (x < 0.0) { |
| 215 | 			p = floorl(q); |
| 216 | 			z = q - p; |
| 217 | 			if (z == 0) |
| 218 | 				return 0 / z; |
| 219 | 			if (q > MAXGAML) { |
| 220 | 				z = 0; |
| 221 | 			} else { |
| 222 | 				if (z > 0.5) { |
| 223 | 					p += 1.0; |
| 224 | 					z = q - p; |
| 225 | 				} |
| 226 | 				z = q * sinl(PIL * z); |
| 227 | 				z = fabsl(z) * stirf(q); |
| 228 | 				z = PIL/z; |
| 229 | 			} |
| 230 | 			if (0.5 * p == floorl(q * 0.5)) |
| 231 | 				z = -z; |
| 232 | 		} else if (x > MAXGAML) { |
| 233 | 			z = x * 0x1p16383L; |
| 234 | 		} else { |
| 235 | 			z = stirf(x); |
| 236 | 		} |
| 237 | 		return z; |
| 238 | 	} |
| 239 | |
| 240 | 	z = 1.0; |
| 241 | 	while (x >= 3.0) { |
| 242 | 		x -= 1.0; |
| 243 | 		z *= x; |
| 244 | 	} |
| 245 | 	while (x < -0.03125L) { |
| 246 | 		z /= x; |
| 247 | 		x += 1.0; |
| 248 | 	} |
| 249 | 	if (x <= 0.03125L) |
| 250 | 		goto small; |
| 251 | 	while (x < 2.0) { |
| 252 | 		z /= x; |
| 253 | 		x += 1.0; |
| 254 | 	} |
| 255 | 	if (x == 2.0) |
| 256 | 		return z; |
| 257 | |
| 258 | 	x -= 2.0; |
| 259 | 	p = __polevll(x, P, 7); |
| 260 | 	q = __polevll(x, Q, 8); |
| 261 | 	z = z * p / q; |
| 262 | 	return z; |
| 263 | |
| 264 | small: |
| 265 | 	/* z==1 if x was originally +-0 */ |
| 266 | 	if (x == 0 && z != 1) |
| 267 | 		return x / x; |
| 268 | 	if (x < 0.0) { |
| 269 | 		x = -x; |
| 270 | 		q = z / (x * __polevll(x, SN, 8)); |
| 271 | 	} else |
| 272 | 		q = z / (x * __polevll(x, S, 8)); |
| 273 | 	return q; |
| 274 | } |
| 275 | #elif LDBL_MANT_DIG == 113 && LDBL_MAX_EXP == 16384 |
| 276 | // TODO: broken implementation to make things compile |
| 277 | long double tgammal(long double x) |
| 278 | { |
| 279 | 	return tgamma(x); |
| 280 | } |
| 281 | #endif |