1 /* 2 * ==================================================== 3 * Copyright 2004 Sun Microsystems, Inc. All Rights Reserved. 4 * Copyright (c) 2008 Steven G. Kargl, David Schultz, Bruce D. Evans. 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 12 /* 13 * ld80 version of k_tan.c. See ../src/k_tan.c for most comments. 14 */ 15 16 #include "math.h" 17 #include "math_private.h" 18 19 /* 20 * Domain [-0.67434, 0.67434], range ~[-2.25e-22, 1.921e-22] 21 * |tan(x)/x - t(x)| < 2**-71.9 22 * 23 * See k_cosl.c for more details about the polynomial. 24 */ 25 #if defined(__amd64__) || defined(__i386__) 26 /* Long double constants are slow on these arches, and broken on i386. */ 27 static const volatile double 28 T3hi = 0.33333333333333331, /* 0x15555555555555.0p-54 */ 29 T3lo = 1.8350121769317163e-17, /* 0x15280000000000.0p-108 */ 30 T5hi = 0.13333333333333336, /* 0x11111111111112.0p-55 */ 31 T5lo = 1.3051083651294260e-17, /* 0x1e180000000000.0p-109 */ 32 T7hi = 0.053968253968250494, /* 0x1ba1ba1ba1b827.0p-57 */ 33 T7lo = 3.1509625637859973e-18, /* 0x1d100000000000.0p-111 */ 34 pio4_hi = 0.78539816339744828, /* 0x1921fb54442d18.0p-53 */ 35 pio4_lo = 3.0628711372715500e-17, /* 0x11a80000000000.0p-107 */ 36 pio4lo_hi = -1.2541394031670831e-20, /* -0x1d9cceba3f91f2.0p-119 */ 37 pio4lo_lo = 6.1493048227390915e-37; /* 0x1a280000000000.0p-173 */ 38 #define T3 ((long double)T3hi + T3lo) 39 #define T5 ((long double)T5hi + T5lo) 40 #define T7 ((long double)T7hi + T7lo) 41 #define pio4 ((long double)pio4_hi + pio4_lo) 42 #define pio4lo ((long double)pio4lo_hi + pio4lo_lo) 43 #else 44 static const long double 45 T3 = 0.333333333333333333180L, /* 0xaaaaaaaaaaaaaaa5.0p-65 */ 46 T5 = 0.133333333333333372290L, /* 0x88888888888893c3.0p-66 */ 47 T7 = 0.0539682539682504975744L, /* 0xdd0dd0dd0dc13ba2.0p-68 */ 48 pio4 = 0.785398163397448309628L, /* 0xc90fdaa22168c235.0p-64 */ 49 pio4lo = -1.25413940316708300586e-20L; /* -0xece675d1fc8f8cbb.0p-130 */ 50 #endif 51 52 static const double 53 T9 = 0.021869488536312216, /* 0x1664f4882cc1c2.0p-58 */ 54 T11 = 0.0088632355256619590, /* 0x1226e355c17612.0p-59 */ 55 T13 = 0.0035921281113786528, /* 0x1d6d3d185d7ff8.0p-61 */ 56 T15 = 0.0014558334756312418, /* 0x17da354aa3f96b.0p-62 */ 57 T17 = 0.00059003538700862256, /* 0x13559358685b83.0p-63 */ 58 T19 = 0.00023907843576635544, /* 0x1f56242026b5be.0p-65 */ 59 T21 = 0.000097154625656538905, /* 0x1977efc26806f4.0p-66 */ 60 T23 = 0.000038440165747303162, /* 0x14275a09b3ceac.0p-67 */ 61 T25 = 0.000018082171885432524, /* 0x12f5e563e5487e.0p-68 */ 62 T27 = 0.0000024196006108814377, /* 0x144c0d80cc6896.0p-71 */ 63 T29 = 0.0000078293456938132840, /* 0x106b59141a6cb3.0p-69 */ 64 T31 = -0.0000032609076735050182, /* -0x1b5abef3ba4b59.0p-71 */ 65 T33 = 0.0000023261313142559411; /* 0x13835436c0c87f.0p-71 */ 66 67 long double 68 __kernel_tanl(long double x, long double y, int iy) { 69 long double z, r, v, w, s; 70 long double osign; 71 int i; 72 73 iy = (iy == 1 ? -1 : 1); /* XXX recover original interface */ 74 osign = (x >= 0 ? 1.0 : -1.0); /* XXX slow, probably wrong for -0 */ 75 if (fabsl(x) >= 0.67434) { 76 if (x < 0) { 77 x = -x; 78 y = -y; 79 } 80 z = pio4 - x; 81 w = pio4lo - y; 82 x = z + w; 83 y = 0.0; 84 i = 1; 85 } else 86 i = 0; 87 z = x * x; 88 w = z * z; 89 r = T5 + w * (T9 + w * (T13 + w * (T17 + w * (T21 + 90 w * (T25 + w * (T29 + w * T33)))))); 91 v = z * (T7 + w * (T11 + w * (T15 + w * (T19 + w * (T23 + 92 w * (T27 + w * T31)))))); 93 s = z * x; 94 r = y + z * (s * (r + v) + y); 95 r += T3 * s; 96 w = x + r; 97 if (i == 1) { 98 v = (long double) iy; 99 return osign * 100 (v - 2.0 * (x - (w * w / (w + v) - r))); 101 } 102 if (iy == 1) 103 return w; 104 else { 105 /* 106 * if allow error up to 2 ulp, simply return 107 * -1.0 / (x+r) here 108 */ 109 /* compute -1.0 / (x+r) accurately */ 110 long double a, t; 111 z = w; 112 z = z + 0x1p32 - 0x1p32; 113 v = r - (z - x); /* z+v = r+x */ 114 t = a = -1.0 / w; /* a = -1.0/w */ 115 t = t + 0x1p32 - 0x1p32; 116 s = 1.0 + t * z; 117 return t + a * (s + t * v); 118 } 119 } 120