k_tan.c 4.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154
  1. #pragma ident "@(#)k_tan.c 1.5 04/04/22 SMI"
  2. /*
  3. * ====================================================
  4. * Copyright 2004 Sun Microsystems, Inc. All Rights Reserved.
  5. *
  6. * Permission to use, copy, modify, and distribute this
  7. * software is freely granted, provided that this notice
  8. * is preserved.
  9. * ====================================================
  10. */
  11. /* INDENT OFF */
  12. /* __kernel_tan( x, y, k )
  13. * kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854
  14. * Input x is assumed to be bounded by ~pi/4 in magnitude.
  15. * Input y is the tail of x.
  16. * Input k indicates whether tan (if k = 1) or -1/tan (if k = -1) is returned.
  17. *
  18. * Algorithm
  19. * 1. Since tan(-x) = -tan(x), we need only to consider positive x.
  20. * 2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0.
  21. * 3. tan(x) is approximated by a odd polynomial of degree 27 on
  22. * [0,0.67434]
  23. * 3 27
  24. * tan(x) ~ x + T1*x + ... + T13*x
  25. * where
  26. *
  27. * |tan(x) 2 4 26 | -59.2
  28. * |----- - (1+T1*x +T2*x +.... +T13*x )| <= 2
  29. * | x |
  30. *
  31. * Note: tan(x+y) = tan(x) + tan'(x)*y
  32. * ~ tan(x) + (1+x*x)*y
  33. * Therefore, for better accuracy in computing tan(x+y), let
  34. * 3 2 2 2 2
  35. * r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
  36. * then
  37. * 3 2
  38. * tan(x+y) = x + (T1*x + (x *(r+y)+y))
  39. *
  40. * 4. For x in [0.67434,pi/4], let y = pi/4 - x, then
  41. * tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
  42. * = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
  43. */
  44. #include "fdlibm.h"
  45. #ifndef _DOUBLE_IS_32BITS
  46. static const double xxx[] = {
  47. 3.33333333333334091986e-01, /* 3FD55555, 55555563 */
  48. 1.33333333333201242699e-01, /* 3FC11111, 1110FE7A */
  49. 5.39682539762260521377e-02, /* 3FABA1BA, 1BB341FE */
  50. 2.18694882948595424599e-02, /* 3F9664F4, 8406D637 */
  51. 8.86323982359930005737e-03, /* 3F8226E3, E96E8493 */
  52. 3.59207910759131235356e-03, /* 3F6D6D22, C9560328 */
  53. 1.45620945432529025516e-03, /* 3F57DBC8, FEE08315 */
  54. 5.88041240820264096874e-04, /* 3F4344D8, F2F26501 */
  55. 2.46463134818469906812e-04, /* 3F3026F7, 1A8D1068 */
  56. 7.81794442939557092300e-05, /* 3F147E88, A03792A6 */
  57. 7.14072491382608190305e-05, /* 3F12B80F, 32F0A7E9 */
  58. -1.85586374855275456654e-05, /* BEF375CB, DB605373 */
  59. 2.59073051863633712884e-05, /* 3EFB2A70, 74BF7AD4 */
  60. /* one */ 1.00000000000000000000e+00, /* 3FF00000, 00000000 */
  61. /* pio4 */ 7.85398163397448278999e-01, /* 3FE921FB, 54442D18 */
  62. /* pio4lo */ 3.06161699786838301793e-17 /* 3C81A626, 33145C07 */
  63. };
  64. #define one xxx[13]
  65. #define pio4 xxx[14]
  66. #define pio4lo xxx[15]
  67. #define T xxx
  68. /* INDENT ON */
  69. double
  70. __kernel_tan(double x, double y, int iy) {
  71. double z, r, v, w, s;
  72. int32_t ix, hx;
  73. GET_HIGH_WORD(hx,x); /* high word of x */
  74. ix = hx & 0x7fffffff; /* high word of |x| */
  75. if (ix < 0x3e300000) { /* x < 2**-28 */
  76. if ((int) x == 0) { /* generate inexact */
  77. uint32_t low;
  78. GET_LOW_WORD(low,x);
  79. if (((ix | low) | (iy + 1)) == 0)
  80. return one / fabs(x);
  81. else {
  82. if (iy == 1)
  83. return x;
  84. else { /* compute -1 / (x+y) carefully */
  85. double a, t;
  86. z = w = x + y;
  87. SET_LOW_WORD(z,0);
  88. v = y - (z - x);
  89. t = a = -one / w;
  90. SET_LOW_WORD(t,0);
  91. s = one + t * z;
  92. return t + a * (s + t * v);
  93. }
  94. }
  95. }
  96. }
  97. if (ix >= 0x3FE59428) { /* |x| >= 0.6744 */
  98. if (hx < 0) {
  99. x = -x;
  100. y = -y;
  101. }
  102. z = pio4 - x;
  103. w = pio4lo - y;
  104. x = z + w;
  105. y = 0.0;
  106. }
  107. z = x * x;
  108. w = z * z;
  109. /*
  110. * Break x^5*(T[1]+x^2*T[2]+...) into
  111. * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
  112. * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
  113. */
  114. r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] +
  115. w * T[11]))));
  116. v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] +
  117. w * T[12])))));
  118. s = z * x;
  119. r = y + z * (s * (r + v) + y);
  120. r += T[0] * s;
  121. w = x + r;
  122. if (ix >= 0x3FE59428) {
  123. v = (double) iy;
  124. return (double) (1 - ((hx >> 30) & 2)) *
  125. (v - 2.0 * (x - (w * w / (w + v) - r)));
  126. }
  127. if (iy == 1)
  128. return w;
  129. else {
  130. /*
  131. * if allow error up to 2 ulp, simply return
  132. * -1.0 / (x+r) here
  133. */
  134. /* compute -1.0 / (x+r) accurately */
  135. double a, t;
  136. z = w;
  137. SET_LOW_WORD(z,0);
  138. v = r - (z - x); /* z+v = r+x */
  139. t = a = -1.0 / w; /* a = -1.0/w */
  140. SET_LOW_WORD(t,0);
  141. s = 1.0 + t * z;
  142. return t + a * (s + t * v);
  143. }
  144. }
  145. #endif /* defined(_DOUBLE_IS_32BITS) */