k_tanf.c (152870) | k_tanf.c (152881) |
---|---|
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. 4 */ 5 6/* 7 * ==================================================== 8 * Copyright 2004 Sun Microsystems, Inc. All Rights Reserved. 9 * 10 * Permission to use, copy, modify, and distribute this 11 * software is freely granted, provided that this notice 12 * is preserved. 13 * ==================================================== 14 */ 15 16#ifndef INLINE_KERNEL_TANDF 17#ifndef lint | 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. 4 */ 5 6/* 7 * ==================================================== 8 * Copyright 2004 Sun Microsystems, Inc. All Rights Reserved. 9 * 10 * Permission to use, copy, modify, and distribute this 11 * software is freely granted, provided that this notice 12 * is preserved. 13 * ==================================================== 14 */ 15 16#ifndef INLINE_KERNEL_TANDF 17#ifndef lint |
18static char rcsid[] = "$FreeBSD: head/lib/msun/src/k_tanf.c 152870 2005-11-28 05:35:32Z bde $"; | 18static char rcsid[] = "$FreeBSD: head/lib/msun/src/k_tanf.c 152881 2005-11-28 11:46:20Z bde $"; |
19#endif 20#endif 21 22#include "math.h" 23#include "math_private.h" 24 25/* |tan(x)/x - t(x)| < 2**-25.5 (~[-2e-08, 2e-08]). */ 26static const double --- 7 unchanged lines hidden (view full) --- 34}; 35 36#ifdef INLINE_KERNEL_TANDF 37extern inline 38#endif 39float 40__kernel_tandf(double x, int iy) 41{ | 19#endif 20#endif 21 22#include "math.h" 23#include "math_private.h" 24 25/* |tan(x)/x - t(x)| < 2**-25.5 (~[-2e-08, 2e-08]). */ 26static const double --- 7 unchanged lines hidden (view full) --- 34}; 35 36#ifdef INLINE_KERNEL_TANDF 37extern inline 38#endif 39float 40__kernel_tandf(double x, int iy) 41{ |
42 double z,r,w,s; | 42 double z,r,w,s,t,u; |
43 44 z = x*x; | 43 44 z = x*x; |
45 w = z*z; 46 /* Break x^5*(T[1]+x^2*T[2]+...) into 47 * x^5*(T[1]+x^4*T[3]+x^8*T[5]) + 48 * x^5*(x^2*(T[2]+x^4*T[4])) 49 */ 50 r = (T[1]+w*(T[3]+w*T[5])) + z*(T[2]+w*T[4]); | 45 /* 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. 50 * 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. 58 */ 59 r = T[4]+z*T[5]; 60 t = T[2]+z*T[3]; 61 w = z*z; |
51 s = z*x; | 62 s = z*x; |
52 r = (x+s*T[0])+(s*z)*r; | 63 u = T[0]+z*T[1]; 64 r = (x+s*u)+(s*w)*(t+w*r); |
53 if(iy==1) return r; 54 else return -1.0/r; 55} | 65 if(iy==1) return r; 66 else return -1.0/r; 67} |