date and time created 88/08/31 23:01:12 by bostic
[unix-history] / usr / src / lib / libm / common / atan2.c
CommitLineData
9b525f39 1/*
5f1375d9 2 * Copyright (c) 1985 Regents of the University of California.
9b525f39
KB
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms are permitted
a399f6c8
KB
6 * provided that the above copyright notice and this paragraph are
7 * duplicated in all such forms and that any documentation,
8 * advertising materials, and other materials related to such
9 * distribution and use acknowledge that the software was developed
10 * by the University of California, Berkeley. The name of the
11 * University may not be used to endorse or promote products derived
12 * from this software without specific prior written permission.
13 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
14 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
15 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
9b525f39
KB
16 *
17 * All recipients should regard themselves as participants in an ongoing
18 * research project and hence should feel obligated to report their
19 * experiences (good or bad) with these elementary function codes, using
20 * the sendbug(8) program, to the authors.
5f1375d9
ZAL
21 */
22
23#ifndef lint
a399f6c8 24static char sccsid[] = "@(#)atan2.c 5.3 (Berkeley) %G%";
9b525f39 25#endif /* not lint */
5f1375d9
ZAL
26
27/* ATAN2(Y,X)
28 * RETURN ARG (X+iY)
29 * DOUBLE PRECISION (VAX D format 56 bits, IEEE DOUBLE 53 BITS)
30 * CODED IN C BY K.C. NG, 1/8/85;
31 * REVISED BY K.C. NG on 2/7/85, 2/13/85, 3/7/85, 3/30/85, 6/29/85.
32 *
33 * Required system supported functions :
34 * copysign(x,y)
35 * scalb(x,y)
36 * logb(x)
37 *
38 * Method :
39 * 1. Reduce y to positive by atan2(y,x)=-atan2(-y,x).
40 * 2. Reduce x to positive by (if x and y are unexceptional):
41 * ARG (x+iy) = arctan(y/x) ... if x > 0,
42 * ARG (x+iy) = pi - arctan[y/(-x)] ... if x < 0,
43 * 3. According to the integer k=4t+0.25 truncated , t=y/x, the argument
44 * is further reduced to one of the following intervals and the
45 * arctangent of y/x is evaluated by the corresponding formula:
46 *
47 * [0,7/16] atan(y/x) = t - t^3*(a1+t^2*(a2+...(a10+t^2*a11)...)
48 * [7/16,11/16] atan(y/x) = atan(1/2) + atan( (y-x/2)/(x+y/2) )
49 * [11/16.19/16] atan(y/x) = atan( 1 ) + atan( (y-x)/(x+y) )
50 * [19/16,39/16] atan(y/x) = atan(3/2) + atan( (y-1.5x)/(x+1.5y) )
51 * [39/16,INF] atan(y/x) = atan(INF) + atan( -x/y )
52 *
53 * Special cases:
54 * Notations: atan2(y,x) == ARG (x+iy) == ARG(x,y).
55 *
56 * ARG( NAN , (anything) ) is NaN;
57 * ARG( (anything), NaN ) is NaN;
58 * ARG(+(anything but NaN), +-0) is +-0 ;
59 * ARG(-(anything but NaN), +-0) is +-PI ;
60 * ARG( 0, +-(anything but 0 and NaN) ) is +-PI/2;
61 * ARG( +INF,+-(anything but INF and NaN) ) is +-0 ;
62 * ARG( -INF,+-(anything but INF and NaN) ) is +-PI;
63 * ARG( +INF,+-INF ) is +-PI/4 ;
64 * ARG( -INF,+-INF ) is +-3PI/4;
65 * ARG( (anything but,0,NaN, and INF),+-INF ) is +-PI/2;
66 *
67 * Accuracy:
68 * atan2(y,x) returns (PI/pi) * the exact ARG (x+iy) nearly rounded,
69 * where
70 *
71 * in decimal:
72 * pi = 3.141592653589793 23846264338327 .....
73 * 53 bits PI = 3.141592653589793 115997963 ..... ,
74 * 56 bits PI = 3.141592653589793 227020265 ..... ,
75 *
76 * in hexadecimal:
77 * pi = 3.243F6A8885A308D313198A2E....
78 * 53 bits PI = 3.243F6A8885A30 = 2 * 1.921FB54442D18 error=.276ulps
79 * 56 bits PI = 3.243F6A8885A308 = 4 * .C90FDAA22168C2 error=.206ulps
80 *
81 * In a test run with 356,000 random argument on [-1,1] * [-1,1] on a
82 * VAX, the maximum observed error was 1.41 ulps (units of the last place)
83 * compared with (PI/pi)*(the exact ARG(x+iy)).
84 *
85 * Note:
86 * We use machine PI (the true pi rounded) in place of the actual
87 * value of pi for all the trig and inverse trig functions. In general,
88 * if trig is one of sin, cos, tan, then computed trig(y) returns the
89 * exact trig(y*pi/PI) nearly rounded; correspondingly, computed arctrig
90 * returns the exact arctrig(y)*PI/pi nearly rounded. These guarantee the
91 * trig functions have period PI, and trig(arctrig(x)) returns x for
92 * all critical values x.
93 *
94 * Constants:
95 * The hexadecimal values are the intended ones for the following constants.
96 * The decimal values may be used, provided that the compiler will convert
97 * from decimal to binary accurately enough to produce the hexadecimal values
98 * shown.
99 */
100
859dc438
ZAL
101#if defined(vax)||defined(tahoe) /* VAX D format */
102#ifdef vax
5a9dac58 103#define _0x(A,B) 0x/**/A/**/B
859dc438 104#else /* vax */
5a9dac58 105#define _0x(A,B) 0x/**/B/**/A
859dc438 106#endif /* vax */
5a9dac58
ZAL
107/*static double */
108/*athfhi = 4.6364760900080611433E-1 , /*Hex 2^ -1 * .ED63382B0DDA7B */
109/*athflo = 1.9338828231967579916E-19 , /*Hex 2^-62 * .E450059CFE92C0 */
110/*PIo4 = 7.8539816339744830676E-1 , /*Hex 2^ 0 * .C90FDAA22168C2 */
111/*at1fhi = 9.8279372324732906796E-1 , /*Hex 2^ 0 * .FB985E940FB4D9 */
112/*at1flo = -3.5540295636764633916E-18 , /*Hex 2^-57 * -.831EDC34D6EAEA */
113/*PIo2 = 1.5707963267948966135E0 , /*Hex 2^ 1 * .C90FDAA22168C2 */
114/*PI = 3.1415926535897932270E0 , /*Hex 2^ 2 * .C90FDAA22168C2 */
115/*a1 = 3.3333333333333473730E-1 , /*Hex 2^ -1 * .AAAAAAAAAAAB75 */
116/*a2 = -2.0000000000017730678E-1 , /*Hex 2^ -2 * -.CCCCCCCCCD946E */
117/*a3 = 1.4285714286694640301E-1 , /*Hex 2^ -2 * .92492492744262 */
118/*a4 = -1.1111111135032672795E-1 , /*Hex 2^ -3 * -.E38E38EBC66292 */
119/*a5 = 9.0909091380563043783E-2 , /*Hex 2^ -3 * .BA2E8BB31BD70C */
120/*a6 = -7.6922954286089459397E-2 , /*Hex 2^ -3 * -.9D89C827C37F18 */
121/*a7 = 6.6663180891693915586E-2 , /*Hex 2^ -3 * .8886B4AE379E58 */
122/*a8 = -5.8772703698290408927E-2 , /*Hex 2^ -4 * -.F0BBA58481A942 */
123/*a9 = 5.2170707402812969804E-2 , /*Hex 2^ -4 * .D5B0F3A1AB13AB */
124/*a10 = -4.4895863157820361210E-2 , /*Hex 2^ -4 * -.B7E4B97FD1048F */
125/*a11 = 3.3006147437343875094E-2 , /*Hex 2^ -4 * .8731743CF72D87 */
126/*a12 = -1.4614844866464185439E-2 ; /*Hex 2^ -6 * -.EF731A2F3476D9 */
127static long athfhix[] = { _0x(6338,3fed), _0x(da7b,2b0d)};
128#define athfhi (*(double *)athfhix)
129static long athflox[] = { _0x(5005,2164), _0x(92c0,9cfe)};
130#define athflo (*(double *)athflox)
131static long PIo4x[] = { _0x(0fda,4049), _0x(68c2,a221)};
132#define PIo4 (*(double *)PIo4x)
133static long at1fhix[] = { _0x(985e,407b), _0x(b4d9,940f)};
134#define at1fhi (*(double *)at1fhix)
135static long at1flox[] = { _0x(1edc,a383), _0x(eaea,34d6)};
136#define at1flo (*(double *)at1flox)
137static long PIo2x[] = { _0x(0fda,40c9), _0x(68c2,a221)};
138#define PIo2 (*(double *)PIo2x)
139static long PIx[] = { _0x(0fda,4149), _0x(68c2,a221)};
140#define PI (*(double *)PIx)
141static long a1x[] = { _0x(aaaa,3faa), _0x(ab75,aaaa)};
142#define a1 (*(double *)a1x)
143static long a2x[] = { _0x(cccc,bf4c), _0x(946e,cccd)};
144#define a2 (*(double *)a2x)
145static long a3x[] = { _0x(4924,3f12), _0x(4262,9274)};
146#define a3 (*(double *)a3x)
147static long a4x[] = { _0x(8e38,bee3), _0x(6292,ebc6)};
148#define a4 (*(double *)a4x)
149static long a5x[] = { _0x(2e8b,3eba), _0x(d70c,b31b)};
150#define a5 (*(double *)a5x)
151static long a6x[] = { _0x(89c8,be9d), _0x(7f18,27c3)};
152#define a6 (*(double *)a6x)
153static long a7x[] = { _0x(86b4,3e88), _0x(9e58,ae37)};
154#define a7 (*(double *)a7x)
155static long a8x[] = { _0x(bba5,be70), _0x(a942,8481)};
156#define a8 (*(double *)a8x)
157static long a9x[] = { _0x(b0f3,3e55), _0x(13ab,a1ab)};
158#define a9 (*(double *)a9x)
159static long a10x[] = { _0x(e4b9,be37), _0x(048f,7fd1)};
160#define a10 (*(double *)a10x)
161static long a11x[] = { _0x(3174,3e07), _0x(2d87,3cf7)};
162#define a11 (*(double *)a11x)
163static long a12x[] = { _0x(731a,bd6f), _0x(76d9,2f34)};
164#define a12 (*(double *)a12x)
859dc438 165#else /* defined(vax)||defined(tahoe) */
5a9dac58 166static double
5f1375d9
ZAL
167athfhi = 4.6364760900080609352E-1 , /*Hex 2^ -2 * 1.DAC670561BB4F */
168athflo = 4.6249969567426939759E-18 , /*Hex 2^-58 * 1.5543B8F253271 */
169PIo4 = 7.8539816339744827900E-1 , /*Hex 2^ -1 * 1.921FB54442D18 */
170at1fhi = 9.8279372324732905408E-1 , /*Hex 2^ -1 * 1.F730BD281F69B */
171at1flo = -2.4407677060164810007E-17 , /*Hex 2^-56 * -1.C23DFEFEAE6B5 */
172PIo2 = 1.5707963267948965580E0 , /*Hex 2^ 0 * 1.921FB54442D18 */
173PI = 3.1415926535897931160E0 , /*Hex 2^ 1 * 1.921FB54442D18 */
174a1 = 3.3333333333333942106E-1 , /*Hex 2^ -2 * 1.55555555555C3 */
175a2 = -1.9999999999979536924E-1 , /*Hex 2^ -3 * -1.9999999997CCD */
176a3 = 1.4285714278004377209E-1 , /*Hex 2^ -3 * 1.24924921EC1D7 */
177a4 = -1.1111110579344973814E-1 , /*Hex 2^ -4 * -1.C71C7059AF280 */
178a5 = 9.0908906105474668324E-2 , /*Hex 2^ -4 * 1.745CE5AA35DB2 */
179a6 = -7.6919217767468239799E-2 , /*Hex 2^ -4 * -1.3B0FA54BEC400 */
180a7 = 6.6614695906082474486E-2 , /*Hex 2^ -4 * 1.10DA924597FFF */
181a8 = -5.8358371008508623523E-2 , /*Hex 2^ -5 * -1.DE125FDDBD793 */
182a9 = 4.9850617156082015213E-2 , /*Hex 2^ -5 * 1.9860524BDD807 */
183a10 = -3.6700606902093604877E-2 , /*Hex 2^ -5 * -1.2CA6C04C6937A */
184a11 = 1.6438029044759730479E-2 ; /*Hex 2^ -6 * 1.0D52174A1BB54 */
859dc438 185#endif /* defined(vax)||defined(tahoe) */
5f1375d9
ZAL
186
187double atan2(y,x)
188double y,x;
189{
190 static double zero=0, one=1, small=1.0E-9, big=1.0E18;
191 double copysign(),logb(),scalb(),t,z,signy,signx,hi,lo;
192 int finite(), k,m;
193
859dc438 194#if !defined(vax)&&!defined(tahoe)
5f1375d9
ZAL
195 /* if x or y is NAN */
196 if(x!=x) return(x); if(y!=y) return(y);
859dc438 197#endif /* !defined(vax)&&!defined(tahoe) */
5f1375d9
ZAL
198
199 /* copy down the sign of y and x */
200 signy = copysign(one,y) ;
201 signx = copysign(one,x) ;
202
203 /* if x is 1.0, goto begin */
204 if(x==1) { y=copysign(y,one); t=y; if(finite(t)) goto begin;}
205
206 /* when y = 0 */
207 if(y==zero) return((signx==one)?y:copysign(PI,signy));
208
209 /* when x = 0 */
210 if(x==zero) return(copysign(PIo2,signy));
211
212 /* when x is INF */
213 if(!finite(x))
214 if(!finite(y))
215 return(copysign((signx==one)?PIo4:3*PIo4,signy));
216 else
217 return(copysign((signx==one)?zero:PI,signy));
218
219 /* when y is INF */
220 if(!finite(y)) return(copysign(PIo2,signy));
5f1375d9
ZAL
221
222 /* compute y/x */
223 x=copysign(x,one);
224 y=copysign(y,one);
225 if((m=(k=logb(y))-logb(x)) > 60) t=big+big;
226 else if(m < -80 ) t=y/x;
227 else { t = y/x ; y = scalb(y,-k); x=scalb(x,-k); }
228
229 /* begin argument reduction */
230begin:
231 if (t < 2.4375) {
232
233 /* truncate 4(t+1/16) to integer for branching */
234 k = 4 * (t+0.0625);
235 switch (k) {
236
237 /* t is in [0,7/16] */
238 case 0:
239 case 1:
240 if (t < small)
241 { big + small ; /* raise inexact flag */
242 return (copysign((signx>zero)?t:PI-t,signy)); }
243
244 hi = zero; lo = zero; break;
245
246 /* t is in [7/16,11/16] */
247 case 2:
248 hi = athfhi; lo = athflo;
249 z = x+x;
250 t = ( (y+y) - x ) / ( z + y ); break;
251
252 /* t is in [11/16,19/16] */
253 case 3:
254 case 4:
255 hi = PIo4; lo = zero;
256 t = ( y - x ) / ( x + y ); break;
257
258 /* t is in [19/16,39/16] */
259 default:
260 hi = at1fhi; lo = at1flo;
261 z = y-x; y=y+y+y; t = x+x;
262 t = ( (z+z)-x ) / ( t + y ); break;
263 }
264 }
265 /* end of if (t < 2.4375) */
266
267 else
268 {
269 hi = PIo2; lo = zero;
270
271 /* t is in [2.4375, big] */
272 if (t <= big) t = - x / y;
273
274 /* t is in [big, INF] */
275 else
276 { big+small; /* raise inexact flag */
277 t = zero; }
278 }
279 /* end of argument reduction */
280
281 /* compute atan(t) for t in [-.4375, .4375] */
282 z = t*t;
859dc438 283#if defined(vax)||defined(tahoe)
5f1375d9
ZAL
284 z = t*(z*(a1+z*(a2+z*(a3+z*(a4+z*(a5+z*(a6+z*(a7+z*(a8+
285 z*(a9+z*(a10+z*(a11+z*a12))))))))))));
859dc438 286#else /* defined(vax)||defined(tahoe) */
5f1375d9
ZAL
287 z = t*(z*(a1+z*(a2+z*(a3+z*(a4+z*(a5+z*(a6+z*(a7+z*(a8+
288 z*(a9+z*(a10+z*a11)))))))))));
859dc438 289#endif /* defined(vax)||defined(tahoe) */
5f1375d9
ZAL
290 z = lo - z; z += t; z += hi;
291
292 return(copysign((signx>zero)?z:PI-z,signy));
293}