1 /*- 2 * Copyright (c) 2005 David Schultz <das@FreeBSD.ORG> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <float.h> 31 32 #include "math.h" 33 #include "math_private.h" 34 35 #define TBLBITS 4 36 #define TBLSIZE (1 << TBLBITS) 37 38 static const float 39 redux = 0x1.8p23f / TBLSIZE, 40 P1 = 0x1.62e430p-1f, 41 P2 = 0x1.ebfbe0p-3f, 42 P3 = 0x1.c6b348p-5f, 43 P4 = 0x1.3b2c9cp-7f; 44 45 static volatile float 46 huge = 0x1p100f, 47 twom100 = 0x1p-100f; 48 49 static const double exp2ft[TBLSIZE] = { 50 0x1.6a09e667f3bcdp-1, 51 0x1.7a11473eb0187p-1, 52 0x1.8ace5422aa0dbp-1, 53 0x1.9c49182a3f090p-1, 54 0x1.ae89f995ad3adp-1, 55 0x1.c199bdd85529cp-1, 56 0x1.d5818dcfba487p-1, 57 0x1.ea4afa2a490dap-1, 58 0x1.0000000000000p+0, 59 0x1.0b5586cf9890fp+0, 60 0x1.172b83c7d517bp+0, 61 0x1.2387a6e756238p+0, 62 0x1.306fe0a31b715p+0, 63 0x1.3dea64c123422p+0, 64 0x1.4bfdad5362a27p+0, 65 0x1.5ab07dd485429p+0, 66 }; 67 68 /* 69 * exp2f(x): compute the base 2 exponential of x 70 * 71 * Accuracy: Peak error < 0.501 ulp; location of peak: -0.030110927. 72 * 73 * Method: (equally-spaced tables) 74 * 75 * Reduce x: 76 * x = 2**k + y, for integer k and |y| <= 1/2. 77 * Thus we have exp2f(x) = 2**k * exp2(y). 78 * 79 * Reduce y: 80 * y = i/TBLSIZE + z for integer i near y * TBLSIZE. 81 * Thus we have exp2(y) = exp2(i/TBLSIZE) * exp2(z), 82 * with |z| <= 2**-(TBLSIZE+1). 83 * 84 * We compute exp2(i/TBLSIZE) via table lookup and exp2(z) via a 85 * degree-4 minimax polynomial with maximum error under 1.4 * 2**-33. 86 * Using double precision for everything except the reduction makes 87 * roundoff error insignificant and simplifies the scaling step. 88 * 89 * This method is due to Tang, but I do not use his suggested parameters: 90 * 91 * Tang, P. Table-driven Implementation of the Exponential Function 92 * in IEEE Floating-Point Arithmetic. TOMS 15(2), 144-157 (1989). 93 */ 94 float 95 exp2f(float x) 96 { 97 double tv, twopk, u, z; 98 float t; 99 uint32_t hx, ix, i0; 100 int32_t k; 101 102 /* Filter out exceptional cases. */ 103 GET_FLOAT_WORD(hx, x); 104 ix = hx & 0x7fffffff; /* high word of |x| */ 105 if(ix >= 0x43000000) { /* |x| >= 128 */ 106 if(ix >= 0x7f800000) { 107 if ((ix & 0x7fffff) != 0 || (hx & 0x80000000) == 0) 108 return (x + x); /* x is NaN or +Inf */ 109 else 110 return (0.0); /* x is -Inf */ 111 } 112 if(x >= 0x1.0p7f) 113 return (huge * huge); /* overflow */ 114 if(x <= -0x1.2cp7f) 115 return (twom100 * twom100); /* underflow */ 116 } else if (ix <= 0x33000000) { /* |x| <= 0x1p-25 */ 117 return (1.0f + x); 118 } 119 120 /* Reduce x, computing z, i0, and k. */ 121 STRICT_ASSIGN(float, t, x + redux); 122 GET_FLOAT_WORD(i0, t); 123 i0 += TBLSIZE / 2; 124 k = (i0 >> TBLBITS) << 20; 125 i0 &= TBLSIZE - 1; 126 t -= redux; 127 z = x - t; 128 INSERT_WORDS(twopk, 0x3ff00000 + k, 0); 129 130 /* Compute r = exp2(y) = exp2ft[i0] * p(z). */ 131 tv = exp2ft[i0]; 132 u = tv * z; 133 tv = tv + u * (P1 + z * P2) + u * (z * z) * (P3 + z * P4); 134 135 /* Scale by 2**(k>>20). */ 136 return (tv * twopk); 137 } 138