1
/* k_tanf.c -- float version of k_tan.c
2
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
3
* Optimized by Bruce D. Evans.
7
* ====================================================
8
* Copyright 2004 Sun Microsystems, Inc. All Rights Reserved.
10
* Permission to use, copy, modify, and distribute this
11
* software is freely granted, provided that this notice
13
* ====================================================
16
#ifndef INLINE_KERNEL_TANDF
18
static char rcsid[] = "$FreeBSD: src/lib/msun/src/k_tanf.c,v 1.20 2005/11/28 11:46:20 bde Exp $";
23
#include "math_private.h"
25
/* |tan(x)/x - t(x)| < 2**-25.5 (~[-2e-08, 2e-08]). */
28
0x15554d3418c99f.0p-54, /* 0.333331395030791399758 */
29
0x1112fd38999f72.0p-55, /* 0.133392002712976742718 */
30
0x1b54c91d865afe.0p-57, /* 0.0533812378445670393523 */
31
0x191df3908c33ce.0p-58, /* 0.0245283181166547278873 */
32
0x185dadfcecf44e.0p-61, /* 0.00297435743359967304927 */
33
0x1362b9bf971bcd.0p-59, /* 0.00946564784943673166728 */
36
#ifdef INLINE_KERNEL_TANDF
40
__kernel_tandf(double x, int iy)
46
* Split up the polynomial into small independent terms to give
47
* opportunities for parallel evaluation. The chosen splitting is
48
* micro-optimized for Athlons (XP, X64). It costs 2 multiplications
49
* relative to Horner's method on sequential machines.
51
* We add the small terms from lowest degree up for efficiency on
52
* non-sequential machines (the lowest degree terms tend to be ready
53
* earlier). Apart from this, we don't care about order of
54
* operations, and don't need to to care since we have precision to
55
* spare. However, the chosen splitting is good for accuracy too,
56
* and would give results as accurate as Horner's method if the
57
* small terms were added from highest degree down.
64
r = (x+s*u)+(s*w)*(t+w*r);