| 1 | /* |
| 2 | * IBM Accurate Mathematical Library |
| 3 | * written by International Business Machines Corp. |
| 4 | * Copyright (C) 2001-2017 Free Software Foundation, Inc. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU Lesser General Public License as published by |
| 8 | * the Free Software Foundation; either version 2.1 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public License |
| 17 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
| 18 | */ |
| 19 | /****************************************************************************/ |
| 20 | /* */ |
| 21 | /* MODULE_NAME:usncs.c */ |
| 22 | /* */ |
| 23 | /* FUNCTIONS: usin */ |
| 24 | /* ucos */ |
| 25 | /* slow */ |
| 26 | /* slow1 */ |
| 27 | /* slow2 */ |
| 28 | /* sloww */ |
| 29 | /* sloww1 */ |
| 30 | /* sloww2 */ |
| 31 | /* bsloww */ |
| 32 | /* bsloww1 */ |
| 33 | /* bsloww2 */ |
| 34 | /* cslow2 */ |
| 35 | /* FILES NEEDED: dla.h endian.h mpa.h mydefs.h usncs.h */ |
| 36 | /* branred.c sincos32.c dosincos.c mpa.c */ |
| 37 | /* sincos.tbl */ |
| 38 | /* */ |
| 39 | /* An ultimate sin and routine. Given an IEEE double machine number x */ |
| 40 | /* it computes the correctly rounded (to nearest) value of sin(x) or cos(x) */ |
| 41 | /* Assumption: Machine arithmetic operations are performed in */ |
| 42 | /* round to nearest mode of IEEE 754 standard. */ |
| 43 | /* */ |
| 44 | /****************************************************************************/ |
| 45 | |
| 46 | |
| 47 | #include <errno.h> |
| 48 | #include <float.h> |
| 49 | #include "endian.h" |
| 50 | #include "mydefs.h" |
| 51 | #include "usncs.h" |
| 52 | #include "MathLib.h" |
| 53 | #include <math.h> |
| 54 | #include <math_private.h> |
| 55 | #include <fenv.h> |
| 56 | |
| 57 | /* Helper macros to compute sin of the input values. */ |
| 58 | #define POLYNOMIAL2(xx) ((((s5 * (xx) + s4) * (xx) + s3) * (xx) + s2) * (xx)) |
| 59 | |
| 60 | #define POLYNOMIAL(xx) (POLYNOMIAL2 (xx) + s1) |
| 61 | |
| 62 | /* The computed polynomial is a variation of the Taylor series expansion for |
| 63 | sin(a): |
| 64 | |
| 65 | a - a^3/3! + a^5/5! - a^7/7! + a^9/9! + (1 - a^2) * da / 2 |
| 66 | |
| 67 | The constants s1, s2, s3, etc. are pre-computed values of 1/3!, 1/5! and so |
| 68 | on. The result is returned to LHS and correction in COR. */ |
| 69 | #define TAYLOR_SIN(xx, a, da, cor) \ |
| 70 | ({ \ |
| 71 | double t = ((POLYNOMIAL (xx) * (a) - 0.5 * (da)) * (xx) + (da)); \ |
| 72 | double res = (a) + t; \ |
| 73 | (cor) = ((a) - res) + t; \ |
| 74 | res; \ |
| 75 | }) |
| 76 | |
| 77 | /* This is again a variation of the Taylor series expansion with the term |
| 78 | x^3/3! expanded into the following for better accuracy: |
| 79 | |
| 80 | bb * x ^ 3 + 3 * aa * x * x1 * x2 + aa * x1 ^ 3 + aa * x2 ^ 3 |
| 81 | |
| 82 | The correction term is dx and bb + aa = -1/3! |
| 83 | */ |
| 84 | #define TAYLOR_SLOW(x0, dx, cor) \ |
| 85 | ({ \ |
| 86 | static const double th2_36 = 206158430208.0; /* 1.5*2**37 */ \ |
| 87 | double xx = (x0) * (x0); \ |
| 88 | double x1 = ((x0) + th2_36) - th2_36; \ |
| 89 | double y = aa * x1 * x1 * x1; \ |
| 90 | double r = (x0) + y; \ |
| 91 | double x2 = ((x0) - x1) + (dx); \ |
| 92 | double t = (((POLYNOMIAL2 (xx) + bb) * xx + 3.0 * aa * x1 * x2) \ |
| 93 | * (x0) + aa * x2 * x2 * x2 + (dx)); \ |
| 94 | t = (((x0) - r) + y) + t; \ |
| 95 | double res = r + t; \ |
| 96 | (cor) = (r - res) + t; \ |
| 97 | res; \ |
| 98 | }) |
| 99 | |
| 100 | #define SINCOS_TABLE_LOOKUP(u, sn, ssn, cs, ccs) \ |
| 101 | ({ \ |
| 102 | int4 k = u.i[LOW_HALF] << 2; \ |
| 103 | sn = __sincostab.x[k]; \ |
| 104 | ssn = __sincostab.x[k + 1]; \ |
| 105 | cs = __sincostab.x[k + 2]; \ |
| 106 | ccs = __sincostab.x[k + 3]; \ |
| 107 | }) |
| 108 | |
| 109 | #ifndef SECTION |
| 110 | # define SECTION |
| 111 | #endif |
| 112 | |
| 113 | extern const union |
| 114 | { |
| 115 | int4 i[880]; |
| 116 | double x[440]; |
| 117 | } __sincostab attribute_hidden; |
| 118 | |
| 119 | static const double |
| 120 | sn3 = -1.66666666666664880952546298448555E-01, |
| 121 | sn5 = 8.33333214285722277379541354343671E-03, |
| 122 | cs2 = 4.99999999999999999999950396842453E-01, |
| 123 | cs4 = -4.16666666666664434524222570944589E-02, |
| 124 | cs6 = 1.38888874007937613028114285595617E-03; |
| 125 | |
| 126 | static const double t22 = 0x1.8p22; |
| 127 | |
| 128 | void __dubsin (double x, double dx, double w[]); |
| 129 | void __docos (double x, double dx, double w[]); |
| 130 | double __mpsin (double x, double dx, bool reduce_range); |
| 131 | double __mpcos (double x, double dx, bool reduce_range); |
| 132 | static double slow (double x); |
| 133 | static double slow1 (double x); |
| 134 | static double slow2 (double x); |
| 135 | static double sloww (double x, double dx, double orig, bool shift_quadrant); |
| 136 | static double sloww1 (double x, double dx, double orig, bool shift_quadrant); |
| 137 | static double sloww2 (double x, double dx, double orig, int n); |
| 138 | static double bsloww (double x, double dx, double orig, int n); |
| 139 | static double bsloww1 (double x, double dx, double orig, int n); |
| 140 | static double bsloww2 (double x, double dx, double orig, int n); |
| 141 | int __branred (double x, double *a, double *aa); |
| 142 | static double cslow2 (double x); |
| 143 | |
| 144 | /* Given a number partitioned into X and DX, this function computes the cosine |
| 145 | of the number by combining the sin and cos of X (as computed by a variation |
| 146 | of the Taylor series) with the values looked up from the sin/cos table to |
| 147 | get the result in RES and a correction value in COR. */ |
| 148 | static inline double |
| 149 | __always_inline |
| 150 | do_cos (double x, double dx, double *corp) |
| 151 | { |
| 152 | mynumber u; |
| 153 | |
| 154 | if (x < 0) |
| 155 | dx = -dx; |
| 156 | |
| 157 | u.x = big + fabs (x); |
| 158 | x = fabs (x) - (u.x - big) + dx; |
| 159 | |
| 160 | double xx, s, sn, ssn, c, cs, ccs, res, cor; |
| 161 | xx = x * x; |
| 162 | s = x + x * xx * (sn3 + xx * sn5); |
| 163 | c = xx * (cs2 + xx * (cs4 + xx * cs6)); |
| 164 | SINCOS_TABLE_LOOKUP (u, sn, ssn, cs, ccs); |
| 165 | cor = (ccs - s * ssn - cs * c) - sn * s; |
| 166 | res = cs + cor; |
| 167 | cor = (cs - res) + cor; |
| 168 | *corp = cor; |
| 169 | return res; |
| 170 | } |
| 171 | |
| 172 | /* A more precise variant of DO_COS. EPS is the adjustment to the correction |
| 173 | COR. */ |
| 174 | static inline double |
| 175 | __always_inline |
| 176 | do_cos_slow (double x, double dx, double eps, double *corp) |
| 177 | { |
| 178 | mynumber u; |
| 179 | |
| 180 | if (x <= 0) |
| 181 | dx = -dx; |
| 182 | |
| 183 | u.x = big + fabs (x); |
| 184 | x = fabs (x) - (u.x - big); |
| 185 | |
| 186 | double xx, y, x1, x2, e1, e2, res, cor; |
| 187 | double s, sn, ssn, c, cs, ccs; |
| 188 | xx = x * x; |
| 189 | s = x * xx * (sn3 + xx * sn5); |
| 190 | c = x * dx + xx * (cs2 + xx * (cs4 + xx * cs6)); |
| 191 | SINCOS_TABLE_LOOKUP (u, sn, ssn, cs, ccs); |
| 192 | x1 = (x + t22) - t22; |
| 193 | x2 = (x - x1) + dx; |
| 194 | e1 = (sn + t22) - t22; |
| 195 | e2 = (sn - e1) + ssn; |
| 196 | cor = (ccs - cs * c - e1 * x2 - e2 * x) - sn * s; |
| 197 | y = cs - e1 * x1; |
| 198 | cor = cor + ((cs - y) - e1 * x1); |
| 199 | res = y + cor; |
| 200 | cor = (y - res) + cor; |
| 201 | cor = 1.0005 * cor + __copysign (eps, cor); |
| 202 | *corp = cor; |
| 203 | return res; |
| 204 | } |
| 205 | |
| 206 | /* Given a number partitioned into X and DX, this function computes the sine of |
| 207 | the number by combining the sin and cos of X (as computed by a variation of |
| 208 | the Taylor series) with the values looked up from the sin/cos table to get |
| 209 | the result in RES and a correction value in COR. */ |
| 210 | static inline double |
| 211 | __always_inline |
| 212 | do_sin (double x, double dx, double *corp) |
| 213 | { |
| 214 | mynumber u; |
| 215 | |
| 216 | if (x <= 0) |
| 217 | dx = -dx; |
| 218 | u.x = big + fabs (x); |
| 219 | x = fabs (x) - (u.x - big); |
| 220 | |
| 221 | double xx, s, sn, ssn, c, cs, ccs, cor, res; |
| 222 | xx = x * x; |
| 223 | s = x + (dx + x * xx * (sn3 + xx * sn5)); |
| 224 | c = x * dx + xx * (cs2 + xx * (cs4 + xx * cs6)); |
| 225 | SINCOS_TABLE_LOOKUP (u, sn, ssn, cs, ccs); |
| 226 | cor = (ssn + s * ccs - sn * c) + cs * s; |
| 227 | res = sn + cor; |
| 228 | cor = (sn - res) + cor; |
| 229 | *corp = cor; |
| 230 | return res; |
| 231 | } |
| 232 | |
| 233 | /* A more precise variant of DO_SIN. EPS is the adjustment to the correction |
| 234 | COR. */ |
| 235 | static inline double |
| 236 | __always_inline |
| 237 | do_sin_slow (double x, double dx, double eps, double *corp) |
| 238 | { |
| 239 | mynumber u; |
| 240 | |
| 241 | if (x <= 0) |
| 242 | dx = -dx; |
| 243 | u.x = big + fabs (x); |
| 244 | x = fabs (x) - (u.x - big); |
| 245 | |
| 246 | double xx, y, x1, x2, c1, c2, res, cor; |
| 247 | double s, sn, ssn, c, cs, ccs; |
| 248 | xx = x * x; |
| 249 | s = x * xx * (sn3 + xx * sn5); |
| 250 | c = xx * (cs2 + xx * (cs4 + xx * cs6)); |
| 251 | SINCOS_TABLE_LOOKUP (u, sn, ssn, cs, ccs); |
| 252 | x1 = (x + t22) - t22; |
| 253 | x2 = (x - x1) + dx; |
| 254 | c1 = (cs + t22) - t22; |
| 255 | c2 = (cs - c1) + ccs; |
| 256 | cor = (ssn + s * ccs + cs * s + c2 * x + c1 * x2 - sn * x * dx) - sn * c; |
| 257 | y = sn + c1 * x1; |
| 258 | cor = cor + ((sn - y) + c1 * x1); |
| 259 | res = y + cor; |
| 260 | cor = (y - res) + cor; |
| 261 | cor = 1.0005 * cor + __copysign (eps, cor); |
| 262 | *corp = cor; |
| 263 | return res; |
| 264 | } |
| 265 | |
| 266 | /* Reduce range of X and compute sin of a + da. When SHIFT_QUADRANT is true, |
| 267 | the routine returns the cosine of a + da by rotating the quadrant once and |
| 268 | computing the sine of the result. */ |
| 269 | static inline double |
| 270 | __always_inline |
| 271 | reduce_and_compute (double x, bool shift_quadrant) |
| 272 | { |
| 273 | double retval = 0, a, da; |
| 274 | unsigned int n = __branred (x, &a, &da); |
| 275 | int4 k = (n + shift_quadrant) % 4; |
| 276 | switch (k) |
| 277 | { |
| 278 | case 2: |
| 279 | a = -a; |
| 280 | da = -da; |
| 281 | /* Fall through. */ |
| 282 | case 0: |
| 283 | if (a * a < 0.01588) |
| 284 | retval = bsloww (a, da, x, n); |
| 285 | else |
| 286 | retval = bsloww1 (a, da, x, n); |
| 287 | break; |
| 288 | |
| 289 | case 1: |
| 290 | case 3: |
| 291 | retval = bsloww2 (a, da, x, n); |
| 292 | break; |
| 293 | } |
| 294 | return retval; |
| 295 | } |
| 296 | |
| 297 | static inline int4 |
| 298 | __always_inline |
| 299 | reduce_sincos_1 (double x, double *a, double *da) |
| 300 | { |
| 301 | mynumber v; |
| 302 | |
| 303 | double t = (x * hpinv + toint); |
| 304 | double xn = t - toint; |
| 305 | v.x = t; |
| 306 | double y = (x - xn * mp1) - xn * mp2; |
| 307 | int4 n = v.i[LOW_HALF] & 3; |
| 308 | double db = xn * mp3; |
| 309 | double b = y - db; |
| 310 | db = (y - b) - db; |
| 311 | |
| 312 | *a = b; |
| 313 | *da = db; |
| 314 | |
| 315 | return n; |
| 316 | } |
| 317 | |
| 318 | /* Compute sin (A + DA). cos can be computed by passing SHIFT_QUADRANT as |
| 319 | true, which results in shifting the quadrant N clockwise. */ |
| 320 | static double |
| 321 | __always_inline |
| 322 | do_sincos_1 (double a, double da, double x, int4 n, bool shift_quadrant) |
| 323 | { |
| 324 | double xx, retval, res, cor; |
| 325 | double eps = fabs (x) * 1.2e-30; |
| 326 | |
| 327 | int k1 = (n + shift_quadrant) & 3; |
| 328 | switch (k1) |
| 329 | { /* quarter of unit circle */ |
| 330 | case 2: |
| 331 | a = -a; |
| 332 | da = -da; |
| 333 | /* Fall through. */ |
| 334 | case 0: |
| 335 | xx = a * a; |
| 336 | if (xx < 0.01588) |
| 337 | { |
| 338 | /* Taylor series. */ |
| 339 | res = TAYLOR_SIN (xx, a, da, cor); |
| 340 | cor = 1.02 * cor + __copysign (eps, cor); |
| 341 | retval = (res == res + cor) ? res : sloww (a, da, x, shift_quadrant); |
| 342 | } |
| 343 | else |
| 344 | { |
| 345 | res = do_sin (a, da, &cor); |
| 346 | cor = 1.035 * cor + __copysign (eps, cor); |
| 347 | retval = ((res == res + cor) ? __copysign (res, a) |
| 348 | : sloww1 (a, da, x, shift_quadrant)); |
| 349 | } |
| 350 | break; |
| 351 | |
| 352 | case 1: |
| 353 | case 3: |
| 354 | res = do_cos (a, da, &cor); |
| 355 | cor = 1.025 * cor + __copysign (eps, cor); |
| 356 | retval = ((res == res + cor) ? ((n & 2) ? -res : res) |
| 357 | : sloww2 (a, da, x, n)); |
| 358 | break; |
| 359 | } |
| 360 | |
| 361 | return retval; |
| 362 | } |
| 363 | |
| 364 | static inline int4 |
| 365 | __always_inline |
| 366 | reduce_sincos_2 (double x, double *a, double *da) |
| 367 | { |
| 368 | mynumber v; |
| 369 | |
| 370 | double t = (x * hpinv + toint); |
| 371 | double xn = t - toint; |
| 372 | v.x = t; |
| 373 | double xn1 = (xn + 8.0e22) - 8.0e22; |
| 374 | double xn2 = xn - xn1; |
| 375 | double y = ((((x - xn1 * mp1) - xn1 * mp2) - xn2 * mp1) - xn2 * mp2); |
| 376 | int4 n = v.i[LOW_HALF] & 3; |
| 377 | double db = xn1 * pp3; |
| 378 | t = y - db; |
| 379 | db = (y - t) - db; |
| 380 | db = (db - xn2 * pp3) - xn * pp4; |
| 381 | double b = t + db; |
| 382 | db = (t - b) + db; |
| 383 | |
| 384 | *a = b; |
| 385 | *da = db; |
| 386 | |
| 387 | return n; |
| 388 | } |
| 389 | |
| 390 | /* Compute sin (A + DA). cos can be computed by passing SHIFT_QUADRANT as |
| 391 | true, which results in shifting the quadrant N clockwise. */ |
| 392 | static double |
| 393 | __always_inline |
| 394 | do_sincos_2 (double a, double da, double x, int4 n, bool shift_quadrant) |
| 395 | { |
| 396 | double res, retval, cor, xx; |
| 397 | |
| 398 | double eps = 1.0e-24; |
| 399 | |
| 400 | int4 k = (n + shift_quadrant) & 3; |
| 401 | |
| 402 | switch (k) |
| 403 | { |
| 404 | case 2: |
| 405 | a = -a; |
| 406 | da = -da; |
| 407 | /* Fall through. */ |
| 408 | case 0: |
| 409 | xx = a * a; |
| 410 | if (xx < 0.01588) |
| 411 | { |
| 412 | /* Taylor series. */ |
| 413 | res = TAYLOR_SIN (xx, a, da, cor); |
| 414 | cor = 1.02 * cor + __copysign (eps, cor); |
| 415 | retval = (res == res + cor) ? res : bsloww (a, da, x, n); |
| 416 | } |
| 417 | else |
| 418 | { |
| 419 | res = do_sin (a, da, &cor); |
| 420 | cor = 1.035 * cor + __copysign (eps, cor); |
| 421 | retval = ((res == res + cor) ? __copysign (res, a) |
| 422 | : bsloww1 (a, da, x, n)); |
| 423 | } |
| 424 | break; |
| 425 | |
| 426 | case 1: |
| 427 | case 3: |
| 428 | res = do_cos (a, da, &cor); |
| 429 | cor = 1.025 * cor + __copysign (eps, cor); |
| 430 | retval = ((res == res + cor) ? ((n & 2) ? -res : res) |
| 431 | : bsloww2 (a, da, x, n)); |
| 432 | break; |
| 433 | } |
| 434 | |
| 435 | return retval; |
| 436 | } |
| 437 | |
| 438 | /*******************************************************************/ |
| 439 | /* An ultimate sin routine. Given an IEEE double machine number x */ |
| 440 | /* it computes the correctly rounded (to nearest) value of sin(x) */ |
| 441 | /*******************************************************************/ |
| 442 | #ifdef IN_SINCOS |
| 443 | static double |
| 444 | #else |
| 445 | double |
| 446 | SECTION |
| 447 | #endif |
| 448 | __sin (double x) |
| 449 | { |
| 450 | double xx, res, t, cor; |
| 451 | mynumber u; |
| 452 | int4 k, m; |
| 453 | double retval = 0; |
| 454 | |
| 455 | #ifndef IN_SINCOS |
| 456 | SET_RESTORE_ROUND_53BIT (FE_TONEAREST); |
| 457 | #endif |
| 458 | |
| 459 | u.x = x; |
| 460 | m = u.i[HIGH_HALF]; |
| 461 | k = 0x7fffffff & m; /* no sign */ |
| 462 | if (k < 0x3e500000) /* if x->0 =>sin(x)=x */ |
| 463 | { |
| 464 | math_check_force_underflow (x); |
| 465 | retval = x; |
| 466 | } |
| 467 | /*---------------------------- 2^-26 < |x|< 0.25 ----------------------*/ |
| 468 | else if (k < 0x3fd00000) |
| 469 | { |
| 470 | xx = x * x; |
| 471 | /* Taylor series. */ |
| 472 | t = POLYNOMIAL (xx) * (xx * x); |
| 473 | res = x + t; |
| 474 | cor = (x - res) + t; |
| 475 | retval = (res == res + 1.07 * cor) ? res : slow (x); |
| 476 | } /* else if (k < 0x3fd00000) */ |
| 477 | /*---------------------------- 0.25<|x|< 0.855469---------------------- */ |
| 478 | else if (k < 0x3feb6000) |
| 479 | { |
| 480 | res = do_sin (x, 0, &cor); |
| 481 | retval = (res == res + 1.096 * cor) ? res : slow1 (x); |
| 482 | retval = __copysign (retval, x); |
| 483 | } /* else if (k < 0x3feb6000) */ |
| 484 | |
| 485 | /*----------------------- 0.855469 <|x|<2.426265 ----------------------*/ |
| 486 | else if (k < 0x400368fd) |
| 487 | { |
| 488 | |
| 489 | t = hp0 - fabs (x); |
| 490 | res = do_cos (t, hp1, &cor); |
| 491 | retval = (res == res + 1.020 * cor) ? res : slow2 (x); |
| 492 | retval = __copysign (retval, x); |
| 493 | } /* else if (k < 0x400368fd) */ |
| 494 | |
| 495 | #ifndef IN_SINCOS |
| 496 | /*-------------------------- 2.426265<|x|< 105414350 ----------------------*/ |
| 497 | else if (k < 0x419921FB) |
| 498 | { |
| 499 | double a, da; |
| 500 | int4 n = reduce_sincos_1 (x, &a, &da); |
| 501 | retval = do_sincos_1 (a, da, x, n, false); |
| 502 | } /* else if (k < 0x419921FB ) */ |
| 503 | |
| 504 | /*---------------------105414350 <|x|< 281474976710656 --------------------*/ |
| 505 | else if (k < 0x42F00000) |
| 506 | { |
| 507 | double a, da; |
| 508 | |
| 509 | int4 n = reduce_sincos_2 (x, &a, &da); |
| 510 | retval = do_sincos_2 (a, da, x, n, false); |
| 511 | } /* else if (k < 0x42F00000 ) */ |
| 512 | |
| 513 | /* -----------------281474976710656 <|x| <2^1024----------------------------*/ |
| 514 | else if (k < 0x7ff00000) |
| 515 | retval = reduce_and_compute (x, false); |
| 516 | |
| 517 | /*--------------------- |x| > 2^1024 ----------------------------------*/ |
| 518 | else |
| 519 | { |
| 520 | if (k == 0x7ff00000 && u.i[LOW_HALF] == 0) |
| 521 | __set_errno (EDOM); |
| 522 | retval = x / x; |
| 523 | } |
| 524 | #endif |
| 525 | |
| 526 | return retval; |
| 527 | } |
| 528 | |
| 529 | |
| 530 | /*******************************************************************/ |
| 531 | /* An ultimate cos routine. Given an IEEE double machine number x */ |
| 532 | /* it computes the correctly rounded (to nearest) value of cos(x) */ |
| 533 | /*******************************************************************/ |
| 534 | |
| 535 | #ifdef IN_SINCOS |
| 536 | static double |
| 537 | #else |
| 538 | double |
| 539 | SECTION |
| 540 | #endif |
| 541 | __cos (double x) |
| 542 | { |
| 543 | double y, xx, res, cor, a, da; |
| 544 | mynumber u; |
| 545 | int4 k, m; |
| 546 | |
| 547 | double retval = 0; |
| 548 | |
| 549 | #ifndef IN_SINCOS |
| 550 | SET_RESTORE_ROUND_53BIT (FE_TONEAREST); |
| 551 | #endif |
| 552 | |
| 553 | u.x = x; |
| 554 | m = u.i[HIGH_HALF]; |
| 555 | k = 0x7fffffff & m; |
| 556 | |
| 557 | /* |x|<2^-27 => cos(x)=1 */ |
| 558 | if (k < 0x3e400000) |
| 559 | retval = 1.0; |
| 560 | |
| 561 | else if (k < 0x3feb6000) |
| 562 | { /* 2^-27 < |x| < 0.855469 */ |
| 563 | res = do_cos (x, 0, &cor); |
| 564 | retval = (res == res + 1.020 * cor) ? res : cslow2 (x); |
| 565 | } /* else if (k < 0x3feb6000) */ |
| 566 | |
| 567 | else if (k < 0x400368fd) |
| 568 | { /* 0.855469 <|x|<2.426265 */ ; |
| 569 | y = hp0 - fabs (x); |
| 570 | a = y + hp1; |
| 571 | da = (y - a) + hp1; |
| 572 | xx = a * a; |
| 573 | if (xx < 0.01588) |
| 574 | { |
| 575 | res = TAYLOR_SIN (xx, a, da, cor); |
| 576 | cor = 1.02 * cor + __copysign (1.0e-31, cor); |
| 577 | retval = (res == res + cor) ? res : sloww (a, da, x, true); |
| 578 | } |
| 579 | else |
| 580 | { |
| 581 | res = do_sin (a, da, &cor); |
| 582 | cor = 1.035 * cor + __copysign (1.0e-31, cor); |
| 583 | retval = ((res == res + cor) ? __copysign (res, a) |
| 584 | : sloww1 (a, da, x, true)); |
| 585 | } |
| 586 | |
| 587 | } /* else if (k < 0x400368fd) */ |
| 588 | |
| 589 | |
| 590 | #ifndef IN_SINCOS |
| 591 | else if (k < 0x419921FB) |
| 592 | { /* 2.426265<|x|< 105414350 */ |
| 593 | double a, da; |
| 594 | int4 n = reduce_sincos_1 (x, &a, &da); |
| 595 | retval = do_sincos_1 (a, da, x, n, true); |
| 596 | } /* else if (k < 0x419921FB ) */ |
| 597 | |
| 598 | else if (k < 0x42F00000) |
| 599 | { |
| 600 | double a, da; |
| 601 | |
| 602 | int4 n = reduce_sincos_2 (x, &a, &da); |
| 603 | retval = do_sincos_2 (a, da, x, n, true); |
| 604 | } /* else if (k < 0x42F00000 ) */ |
| 605 | |
| 606 | /* 281474976710656 <|x| <2^1024 */ |
| 607 | else if (k < 0x7ff00000) |
| 608 | retval = reduce_and_compute (x, true); |
| 609 | |
| 610 | else |
| 611 | { |
| 612 | if (k == 0x7ff00000 && u.i[LOW_HALF] == 0) |
| 613 | __set_errno (EDOM); |
| 614 | retval = x / x; /* |x| > 2^1024 */ |
| 615 | } |
| 616 | #endif |
| 617 | |
| 618 | return retval; |
| 619 | } |
| 620 | |
| 621 | /************************************************************************/ |
| 622 | /* Routine compute sin(x) for 2^-26 < |x|< 0.25 by Taylor with more */ |
| 623 | /* precision and if still doesn't accurate enough by mpsin or dubsin */ |
| 624 | /************************************************************************/ |
| 625 | |
| 626 | static inline double |
| 627 | __always_inline |
| 628 | slow (double x) |
| 629 | { |
| 630 | double res, cor, w[2]; |
| 631 | res = TAYLOR_SLOW (x, 0, cor); |
| 632 | if (res == res + 1.0007 * cor) |
| 633 | return res; |
| 634 | |
| 635 | __dubsin (fabs (x), 0, w); |
| 636 | if (w[0] == w[0] + 1.000000001 * w[1]) |
| 637 | return __copysign (w[0], x); |
| 638 | |
| 639 | return __copysign (__mpsin (fabs (x), 0, false), x); |
| 640 | } |
| 641 | |
| 642 | /*******************************************************************************/ |
| 643 | /* Routine compute sin(x) for 0.25<|x|< 0.855469 by __sincostab.tbl and Taylor */ |
| 644 | /* and if result still doesn't accurate enough by mpsin or dubsin */ |
| 645 | /*******************************************************************************/ |
| 646 | |
| 647 | static inline double |
| 648 | __always_inline |
| 649 | slow1 (double x) |
| 650 | { |
| 651 | double w[2], cor, res; |
| 652 | |
| 653 | res = do_sin_slow (x, 0, 0, &cor); |
| 654 | if (res == res + cor) |
| 655 | return res; |
| 656 | |
| 657 | __dubsin (fabs (x), 0, w); |
| 658 | if (w[0] == w[0] + 1.000000005 * w[1]) |
| 659 | return w[0]; |
| 660 | |
| 661 | return __mpsin (fabs (x), 0, false); |
| 662 | } |
| 663 | |
| 664 | /**************************************************************************/ |
| 665 | /* Routine compute sin(x) for 0.855469 <|x|<2.426265 by __sincostab.tbl */ |
| 666 | /* and if result still doesn't accurate enough by mpsin or dubsin */ |
| 667 | /**************************************************************************/ |
| 668 | static inline double |
| 669 | __always_inline |
| 670 | slow2 (double x) |
| 671 | { |
| 672 | double w[2], y, y1, y2, cor, res; |
| 673 | |
| 674 | double t = hp0 - fabs (x); |
| 675 | res = do_cos_slow (t, hp1, 0, &cor); |
| 676 | if (res == res + cor) |
| 677 | return res; |
| 678 | |
| 679 | y = fabs (x) - hp0; |
| 680 | y1 = y - hp1; |
| 681 | y2 = (y - y1) - hp1; |
| 682 | __docos (y1, y2, w); |
| 683 | if (w[0] == w[0] + 1.000000005 * w[1]) |
| 684 | return w[0]; |
| 685 | |
| 686 | return __mpsin (fabs (x), 0, false); |
| 687 | } |
| 688 | |
| 689 | /* Compute sin(x + dx) where X is small enough to use Taylor series around zero |
| 690 | and (x + dx) in the first or third quarter of the unit circle. ORIG is the |
| 691 | original value of X for computing error of the result. If the result is not |
| 692 | accurate enough, the routine calls mpsin or dubsin. SHIFT_QUADRANT rotates |
| 693 | the unit circle by 1 to compute the cosine instead of sine. */ |
| 694 | static inline double |
| 695 | __always_inline |
| 696 | sloww (double x, double dx, double orig, bool shift_quadrant) |
| 697 | { |
| 698 | double y, t, res, cor, w[2], a, da, xn; |
| 699 | mynumber v; |
| 700 | int4 n; |
| 701 | res = TAYLOR_SLOW (x, dx, cor); |
| 702 | |
| 703 | double eps = fabs (orig) * 3.1e-30; |
| 704 | |
| 705 | cor = 1.0005 * cor + __copysign (eps, cor); |
| 706 | |
| 707 | if (res == res + cor) |
| 708 | return res; |
| 709 | |
| 710 | a = fabs (x); |
| 711 | da = (x > 0) ? dx : -dx; |
| 712 | __dubsin (a, da, w); |
| 713 | eps = fabs (orig) * 1.1e-30; |
| 714 | cor = 1.000000001 * w[1] + __copysign (eps, w[1]); |
| 715 | |
| 716 | if (w[0] == w[0] + cor) |
| 717 | return __copysign (w[0], x); |
| 718 | |
| 719 | t = (orig * hpinv + toint); |
| 720 | xn = t - toint; |
| 721 | v.x = t; |
| 722 | y = (orig - xn * mp1) - xn * mp2; |
| 723 | n = (v.i[LOW_HALF] + shift_quadrant) & 3; |
| 724 | da = xn * pp3; |
| 725 | t = y - da; |
| 726 | da = (y - t) - da; |
| 727 | y = xn * pp4; |
| 728 | a = t - y; |
| 729 | da = ((t - a) - y) + da; |
| 730 | |
| 731 | if (n & 2) |
| 732 | { |
| 733 | a = -a; |
| 734 | da = -da; |
| 735 | } |
| 736 | x = fabs (a); |
| 737 | dx = (a > 0) ? da : -da; |
| 738 | __dubsin (x, dx, w); |
| 739 | eps = fabs (orig) * 1.1e-40; |
| 740 | cor = 1.000000001 * w[1] + __copysign (eps, w[1]); |
| 741 | |
| 742 | if (w[0] == w[0] + cor) |
| 743 | return __copysign (w[0], a); |
| 744 | |
| 745 | return shift_quadrant ? __mpcos (orig, 0, true) : __mpsin (orig, 0, true); |
| 746 | } |
| 747 | |
| 748 | /* Compute sin(x + dx) where X is in the first or third quarter of the unit |
| 749 | circle. ORIG is the original value of X for computing error of the result. |
| 750 | If the result is not accurate enough, the routine calls mpsin or dubsin. |
| 751 | SHIFT_QUADRANT rotates the unit circle by 1 to compute the cosine instead of |
| 752 | sine. */ |
| 753 | static inline double |
| 754 | __always_inline |
| 755 | sloww1 (double x, double dx, double orig, bool shift_quadrant) |
| 756 | { |
| 757 | double w[2], cor, res; |
| 758 | |
| 759 | res = do_sin_slow (x, dx, 3.1e-30 * fabs (orig), &cor); |
| 760 | |
| 761 | if (res == res + cor) |
| 762 | return __copysign (res, x); |
| 763 | |
| 764 | dx = (x > 0 ? dx : -dx); |
| 765 | __dubsin (fabs (x), dx, w); |
| 766 | |
| 767 | double eps = 1.1e-30 * fabs (orig); |
| 768 | cor = 1.000000005 * w[1] + __copysign (eps, w[1]); |
| 769 | |
| 770 | if (w[0] == w[0] + cor) |
| 771 | return __copysign (w[0], x); |
| 772 | |
| 773 | return shift_quadrant ? __mpcos (orig, 0, true) : __mpsin (orig, 0, true); |
| 774 | } |
| 775 | |
| 776 | /***************************************************************************/ |
| 777 | /* Routine compute sin(x+dx) (Double-Length number) where x in second or */ |
| 778 | /* fourth quarter of unit circle.Routine receive also the original value */ |
| 779 | /* and quarter(n= 1or 3)of x for computing error of result.And if result not*/ |
| 780 | /* accurate enough routine calls mpsin1 or dubsin */ |
| 781 | /***************************************************************************/ |
| 782 | |
| 783 | static inline double |
| 784 | __always_inline |
| 785 | sloww2 (double x, double dx, double orig, int n) |
| 786 | { |
| 787 | double w[2], cor, res; |
| 788 | |
| 789 | res = do_cos_slow (x, dx, 3.1e-30 * fabs (orig), &cor); |
| 790 | |
| 791 | if (res == res + cor) |
| 792 | return (n & 2) ? -res : res; |
| 793 | |
| 794 | dx = x > 0 ? dx : -dx; |
| 795 | __docos (fabs (x), dx, w); |
| 796 | |
| 797 | double eps = 1.1e-30 * fabs (orig); |
| 798 | cor = 1.000000005 * w[1] + __copysign (eps, w[1]); |
| 799 | |
| 800 | if (w[0] == w[0] + cor) |
| 801 | return (n & 2) ? -w[0] : w[0]; |
| 802 | |
| 803 | return (n & 1) ? __mpsin (orig, 0, true) : __mpcos (orig, 0, true); |
| 804 | } |
| 805 | |
| 806 | /***************************************************************************/ |
| 807 | /* Routine compute sin(x+dx) or cos(x+dx) (Double-Length number) where x */ |
| 808 | /* is small enough to use Taylor series around zero and (x+dx) */ |
| 809 | /* in first or third quarter of unit circle.Routine receive also */ |
| 810 | /* (right argument) the original value of x for computing error of */ |
| 811 | /* result.And if result not accurate enough routine calls other routines */ |
| 812 | /***************************************************************************/ |
| 813 | |
| 814 | static inline double |
| 815 | __always_inline |
| 816 | bsloww (double x, double dx, double orig, int n) |
| 817 | { |
| 818 | double res, cor, w[2], a, da; |
| 819 | |
| 820 | res = TAYLOR_SLOW (x, dx, cor); |
| 821 | cor = 1.0005 * cor + __copysign (1.1e-24, cor); |
| 822 | if (res == res + cor) |
| 823 | return res; |
| 824 | |
| 825 | a = fabs (x); |
| 826 | da = (x > 0) ? dx : -dx; |
| 827 | __dubsin (a, da, w); |
| 828 | cor = 1.000000001 * w[1] + __copysign (1.1e-24, w[1]); |
| 829 | |
| 830 | if (w[0] == w[0] + cor) |
| 831 | return __copysign (w[0], x); |
| 832 | |
| 833 | return (n & 1) ? __mpcos (orig, 0, true) : __mpsin (orig, 0, true); |
| 834 | } |
| 835 | |
| 836 | /***************************************************************************/ |
| 837 | /* Routine compute sin(x+dx) or cos(x+dx) (Double-Length number) where x */ |
| 838 | /* in first or third quarter of unit circle.Routine receive also */ |
| 839 | /* (right argument) the original value of x for computing error of result.*/ |
| 840 | /* And if result not accurate enough routine calls other routines */ |
| 841 | /***************************************************************************/ |
| 842 | |
| 843 | static inline double |
| 844 | __always_inline |
| 845 | bsloww1 (double x, double dx, double orig, int n) |
| 846 | { |
| 847 | double w[2], cor, res; |
| 848 | |
| 849 | res = do_sin_slow (x, dx, 1.1e-24, &cor); |
| 850 | if (res == res + cor) |
| 851 | return (x > 0) ? res : -res; |
| 852 | |
| 853 | dx = (x > 0) ? dx : -dx; |
| 854 | __dubsin (fabs (x), dx, w); |
| 855 | |
| 856 | cor = 1.000000005 * w[1] + __copysign (1.1e-24, w[1]); |
| 857 | |
| 858 | if (w[0] == w[0] + cor) |
| 859 | return __copysign (w[0], x); |
| 860 | |
| 861 | return (n & 1) ? __mpcos (orig, 0, true) : __mpsin (orig, 0, true); |
| 862 | } |
| 863 | |
| 864 | /***************************************************************************/ |
| 865 | /* Routine compute sin(x+dx) or cos(x+dx) (Double-Length number) where x */ |
| 866 | /* in second or fourth quarter of unit circle.Routine receive also the */ |
| 867 | /* original value and quarter(n= 1or 3)of x for computing error of result. */ |
| 868 | /* And if result not accurate enough routine calls other routines */ |
| 869 | /***************************************************************************/ |
| 870 | |
| 871 | static inline double |
| 872 | __always_inline |
| 873 | bsloww2 (double x, double dx, double orig, int n) |
| 874 | { |
| 875 | double w[2], cor, res; |
| 876 | |
| 877 | res = do_cos_slow (x, dx, 1.1e-24, &cor); |
| 878 | if (res == res + cor) |
| 879 | return (n & 2) ? -res : res; |
| 880 | |
| 881 | dx = (x > 0) ? dx : -dx; |
| 882 | __docos (fabs (x), dx, w); |
| 883 | |
| 884 | cor = 1.000000005 * w[1] + __copysign (1.1e-24, w[1]); |
| 885 | |
| 886 | if (w[0] == w[0] + cor) |
| 887 | return (n & 2) ? -w[0] : w[0]; |
| 888 | |
| 889 | return (n & 1) ? __mpsin (orig, 0, true) : __mpcos (orig, 0, true); |
| 890 | } |
| 891 | |
| 892 | /************************************************************************/ |
| 893 | /* Routine compute cos(x) for 2^-27 < |x|< 0.25 by Taylor with more */ |
| 894 | /* precision and if still doesn't accurate enough by mpcos or docos */ |
| 895 | /************************************************************************/ |
| 896 | |
| 897 | static inline double |
| 898 | __always_inline |
| 899 | cslow2 (double x) |
| 900 | { |
| 901 | double w[2], cor, res; |
| 902 | |
| 903 | res = do_cos_slow (x, 0, 0, &cor); |
| 904 | if (res == res + cor) |
| 905 | return res; |
| 906 | |
| 907 | __docos (fabs (x), 0, w); |
| 908 | if (w[0] == w[0] + 1.000000005 * w[1]) |
| 909 | return w[0]; |
| 910 | |
| 911 | return __mpcos (x, 0, false); |
| 912 | } |
| 913 | |
| 914 | #ifndef __cos |
| 915 | weak_alias (__cos, cos) |
| 916 | # ifdef NO_LONG_DOUBLE |
| 917 | strong_alias (__cos, __cosl) |
| 918 | weak_alias (__cos, cosl) |
| 919 | # endif |
| 920 | #endif |
| 921 | #ifndef __sin |
| 922 | weak_alias (__sin, sin) |
| 923 | # ifdef NO_LONG_DOUBLE |
| 924 | strong_alias (__sin, __sinl) |
| 925 | weak_alias (__sin, sinl) |
| 926 | # endif |
| 927 | #endif |
| 928 | |