xref: /csrg-svn/lib/libm/common/atan2.c (revision 34127)
1 /*
2  * Copyright (c) 1985 Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are permitted
6  * provided that this notice is preserved and that due credit is given
7  * to the University of California at Berkeley. The name of the University
8  * may not be used to endorse or promote products derived from this
9  * software without specific prior written permission. This software
10  * is provided ``as is'' without express or implied warranty.
11  *
12  * All recipients should regard themselves as participants in an ongoing
13  * research project and hence should feel obligated to report their
14  * experiences (good or bad) with these elementary function codes, using
15  * the sendbug(8) program, to the authors.
16  */
17 
18 #ifndef lint
19 static char sccsid[] = "@(#)atan2.c	5.2 (Berkeley) 04/29/88";
20 #endif /* not lint */
21 
22 /* ATAN2(Y,X)
23  * RETURN ARG (X+iY)
24  * DOUBLE PRECISION (VAX D format 56 bits, IEEE DOUBLE 53 BITS)
25  * CODED IN C BY K.C. NG, 1/8/85;
26  * REVISED BY K.C. NG on 2/7/85, 2/13/85, 3/7/85, 3/30/85, 6/29/85.
27  *
28  * Required system supported functions :
29  *	copysign(x,y)
30  *	scalb(x,y)
31  *	logb(x)
32  *
33  * Method :
34  *	1. Reduce y to positive by atan2(y,x)=-atan2(-y,x).
35  *	2. Reduce x to positive by (if x and y are unexceptional):
36  *		ARG (x+iy) = arctan(y/x)   	   ... if x > 0,
37  *		ARG (x+iy) = pi - arctan[y/(-x)]   ... if x < 0,
38  *	3. According to the integer k=4t+0.25 truncated , t=y/x, the argument
39  *	   is further reduced to one of the following intervals and the
40  *	   arctangent of y/x is evaluated by the corresponding formula:
41  *
42  *         [0,7/16]	   atan(y/x) = t - t^3*(a1+t^2*(a2+...(a10+t^2*a11)...)
43  *	   [7/16,11/16]    atan(y/x) = atan(1/2) + atan( (y-x/2)/(x+y/2) )
44  *	   [11/16.19/16]   atan(y/x) = atan( 1 ) + atan( (y-x)/(x+y) )
45  *	   [19/16,39/16]   atan(y/x) = atan(3/2) + atan( (y-1.5x)/(x+1.5y) )
46  *	   [39/16,INF]     atan(y/x) = atan(INF) + atan( -x/y )
47  *
48  * Special cases:
49  * Notations: atan2(y,x) == ARG (x+iy) == ARG(x,y).
50  *
51  *	ARG( NAN , (anything) ) is NaN;
52  *	ARG( (anything), NaN ) is NaN;
53  *	ARG(+(anything but NaN), +-0) is +-0  ;
54  *	ARG(-(anything but NaN), +-0) is +-PI ;
55  *	ARG( 0, +-(anything but 0 and NaN) ) is +-PI/2;
56  *	ARG( +INF,+-(anything but INF and NaN) ) is +-0 ;
57  *	ARG( -INF,+-(anything but INF and NaN) ) is +-PI;
58  *	ARG( +INF,+-INF ) is +-PI/4 ;
59  *	ARG( -INF,+-INF ) is +-3PI/4;
60  *	ARG( (anything but,0,NaN, and INF),+-INF ) is +-PI/2;
61  *
62  * Accuracy:
63  *	atan2(y,x) returns (PI/pi) * the exact ARG (x+iy) nearly rounded,
64  *	where
65  *
66  *	in decimal:
67  *		pi = 3.141592653589793 23846264338327 .....
68  *    53 bits   PI = 3.141592653589793 115997963 ..... ,
69  *    56 bits   PI = 3.141592653589793 227020265 ..... ,
70  *
71  *	in hexadecimal:
72  *		pi = 3.243F6A8885A308D313198A2E....
73  *    53 bits   PI = 3.243F6A8885A30  =  2 * 1.921FB54442D18	error=.276ulps
74  *    56 bits   PI = 3.243F6A8885A308 =  4 * .C90FDAA22168C2    error=.206ulps
75  *
76  *	In a test run with 356,000 random argument on [-1,1] * [-1,1] on a
77  *	VAX, the maximum observed error was 1.41 ulps (units of the last place)
78  *	compared with (PI/pi)*(the exact ARG(x+iy)).
79  *
80  * Note:
81  *	We use machine PI (the true pi rounded) in place of the actual
82  *	value of pi for all the trig and inverse trig functions. In general,
83  *	if trig is one of sin, cos, tan, then computed trig(y) returns the
84  *	exact trig(y*pi/PI) nearly rounded; correspondingly, computed arctrig
85  *	returns the exact arctrig(y)*PI/pi nearly rounded. These guarantee the
86  *	trig functions have period PI, and trig(arctrig(x)) returns x for
87  *	all critical values x.
88  *
89  * Constants:
90  * The hexadecimal values are the intended ones for the following constants.
91  * The decimal values may be used, provided that the compiler will convert
92  * from decimal to binary accurately enough to produce the hexadecimal values
93  * shown.
94  */
95 
96 #if defined(vax)||defined(tahoe) 	/* VAX D format */
97 #ifdef vax
98 #define _0x(A,B)	0x/**/A/**/B
99 #else	/* vax */
100 #define _0x(A,B)	0x/**/B/**/A
101 #endif	/* vax */
102 /*static double */
103 /*athfhi =  4.6364760900080611433E-1    , /*Hex  2^ -1   *  .ED63382B0DDA7B */
104 /*athflo =  1.9338828231967579916E-19   , /*Hex  2^-62   *  .E450059CFE92C0 */
105 /*PIo4   =  7.8539816339744830676E-1    , /*Hex  2^  0   *  .C90FDAA22168C2 */
106 /*at1fhi =  9.8279372324732906796E-1    , /*Hex  2^  0   *  .FB985E940FB4D9 */
107 /*at1flo = -3.5540295636764633916E-18   , /*Hex  2^-57   * -.831EDC34D6EAEA */
108 /*PIo2   =  1.5707963267948966135E0     , /*Hex  2^  1   *  .C90FDAA22168C2 */
109 /*PI     =  3.1415926535897932270E0     , /*Hex  2^  2   *  .C90FDAA22168C2 */
110 /*a1     =  3.3333333333333473730E-1    , /*Hex  2^ -1   *  .AAAAAAAAAAAB75 */
111 /*a2     = -2.0000000000017730678E-1    , /*Hex  2^ -2   * -.CCCCCCCCCD946E */
112 /*a3     =  1.4285714286694640301E-1    , /*Hex  2^ -2   *  .92492492744262 */
113 /*a4     = -1.1111111135032672795E-1    , /*Hex  2^ -3   * -.E38E38EBC66292 */
114 /*a5     =  9.0909091380563043783E-2    , /*Hex  2^ -3   *  .BA2E8BB31BD70C */
115 /*a6     = -7.6922954286089459397E-2    , /*Hex  2^ -3   * -.9D89C827C37F18 */
116 /*a7     =  6.6663180891693915586E-2    , /*Hex  2^ -3   *  .8886B4AE379E58 */
117 /*a8     = -5.8772703698290408927E-2    , /*Hex  2^ -4   * -.F0BBA58481A942 */
118 /*a9     =  5.2170707402812969804E-2    , /*Hex  2^ -4   *  .D5B0F3A1AB13AB */
119 /*a10    = -4.4895863157820361210E-2    , /*Hex  2^ -4   * -.B7E4B97FD1048F */
120 /*a11    =  3.3006147437343875094E-2    , /*Hex  2^ -4   *  .8731743CF72D87 */
121 /*a12    = -1.4614844866464185439E-2    ; /*Hex  2^ -6   * -.EF731A2F3476D9 */
122 static long athfhix[] = { _0x(6338,3fed), _0x(da7b,2b0d)};
123 #define athfhi	(*(double *)athfhix)
124 static long athflox[] = { _0x(5005,2164), _0x(92c0,9cfe)};
125 #define athflo	(*(double *)athflox)
126 static long   PIo4x[] = { _0x(0fda,4049), _0x(68c2,a221)};
127 #define   PIo4	(*(double *)PIo4x)
128 static long at1fhix[] = { _0x(985e,407b), _0x(b4d9,940f)};
129 #define at1fhi	(*(double *)at1fhix)
130 static long at1flox[] = { _0x(1edc,a383), _0x(eaea,34d6)};
131 #define at1flo	(*(double *)at1flox)
132 static long   PIo2x[] = { _0x(0fda,40c9), _0x(68c2,a221)};
133 #define   PIo2	(*(double *)PIo2x)
134 static long     PIx[] = { _0x(0fda,4149), _0x(68c2,a221)};
135 #define     PI	(*(double *)PIx)
136 static long     a1x[] = { _0x(aaaa,3faa), _0x(ab75,aaaa)};
137 #define     a1	(*(double *)a1x)
138 static long     a2x[] = { _0x(cccc,bf4c), _0x(946e,cccd)};
139 #define     a2	(*(double *)a2x)
140 static long     a3x[] = { _0x(4924,3f12), _0x(4262,9274)};
141 #define     a3	(*(double *)a3x)
142 static long     a4x[] = { _0x(8e38,bee3), _0x(6292,ebc6)};
143 #define     a4	(*(double *)a4x)
144 static long     a5x[] = { _0x(2e8b,3eba), _0x(d70c,b31b)};
145 #define     a5	(*(double *)a5x)
146 static long     a6x[] = { _0x(89c8,be9d), _0x(7f18,27c3)};
147 #define     a6	(*(double *)a6x)
148 static long     a7x[] = { _0x(86b4,3e88), _0x(9e58,ae37)};
149 #define     a7	(*(double *)a7x)
150 static long     a8x[] = { _0x(bba5,be70), _0x(a942,8481)};
151 #define     a8	(*(double *)a8x)
152 static long     a9x[] = { _0x(b0f3,3e55), _0x(13ab,a1ab)};
153 #define     a9	(*(double *)a9x)
154 static long    a10x[] = { _0x(e4b9,be37), _0x(048f,7fd1)};
155 #define    a10	(*(double *)a10x)
156 static long    a11x[] = { _0x(3174,3e07), _0x(2d87,3cf7)};
157 #define    a11	(*(double *)a11x)
158 static long    a12x[] = { _0x(731a,bd6f), _0x(76d9,2f34)};
159 #define    a12	(*(double *)a12x)
160 #else 	/* defined(vax)||defined(tahoe) */
161 static double
162 athfhi =  4.6364760900080609352E-1    , /*Hex  2^ -2   *  1.DAC670561BB4F */
163 athflo =  4.6249969567426939759E-18   , /*Hex  2^-58   *  1.5543B8F253271 */
164 PIo4   =  7.8539816339744827900E-1    , /*Hex  2^ -1   *  1.921FB54442D18 */
165 at1fhi =  9.8279372324732905408E-1    , /*Hex  2^ -1   *  1.F730BD281F69B */
166 at1flo = -2.4407677060164810007E-17   , /*Hex  2^-56   * -1.C23DFEFEAE6B5 */
167 PIo2   =  1.5707963267948965580E0     , /*Hex  2^  0   *  1.921FB54442D18 */
168 PI     =  3.1415926535897931160E0     , /*Hex  2^  1   *  1.921FB54442D18 */
169 a1     =  3.3333333333333942106E-1    , /*Hex  2^ -2   *  1.55555555555C3 */
170 a2     = -1.9999999999979536924E-1    , /*Hex  2^ -3   * -1.9999999997CCD */
171 a3     =  1.4285714278004377209E-1    , /*Hex  2^ -3   *  1.24924921EC1D7 */
172 a4     = -1.1111110579344973814E-1    , /*Hex  2^ -4   * -1.C71C7059AF280 */
173 a5     =  9.0908906105474668324E-2    , /*Hex  2^ -4   *  1.745CE5AA35DB2 */
174 a6     = -7.6919217767468239799E-2    , /*Hex  2^ -4   * -1.3B0FA54BEC400 */
175 a7     =  6.6614695906082474486E-2    , /*Hex  2^ -4   *  1.10DA924597FFF */
176 a8     = -5.8358371008508623523E-2    , /*Hex  2^ -5   * -1.DE125FDDBD793 */
177 a9     =  4.9850617156082015213E-2    , /*Hex  2^ -5   *  1.9860524BDD807 */
178 a10    = -3.6700606902093604877E-2    , /*Hex  2^ -5   * -1.2CA6C04C6937A */
179 a11    =  1.6438029044759730479E-2    ; /*Hex  2^ -6   *  1.0D52174A1BB54 */
180 #endif 	/* defined(vax)||defined(tahoe) */
181 
182 double atan2(y,x)
183 double  y,x;
184 {
185 	static double zero=0, one=1, small=1.0E-9, big=1.0E18;
186 	double copysign(),logb(),scalb(),t,z,signy,signx,hi,lo;
187 	int finite(), k,m;
188 
189 #if !defined(vax)&&!defined(tahoe)
190     /* if x or y is NAN */
191 	if(x!=x) return(x); if(y!=y) return(y);
192 #endif	/* !defined(vax)&&!defined(tahoe) */
193 
194     /* copy down the sign of y and x */
195 	signy = copysign(one,y) ;
196 	signx = copysign(one,x) ;
197 
198     /* if x is 1.0, goto begin */
199 	if(x==1) { y=copysign(y,one); t=y; if(finite(t)) goto begin;}
200 
201     /* when y = 0 */
202 	if(y==zero) return((signx==one)?y:copysign(PI,signy));
203 
204     /* when x = 0 */
205 	if(x==zero) return(copysign(PIo2,signy));
206 
207     /* when x is INF */
208 	if(!finite(x))
209 	    if(!finite(y))
210 		return(copysign((signx==one)?PIo4:3*PIo4,signy));
211 	    else
212 		return(copysign((signx==one)?zero:PI,signy));
213 
214     /* when y is INF */
215 	if(!finite(y)) return(copysign(PIo2,signy));
216 
217     /* compute y/x */
218 	x=copysign(x,one);
219 	y=copysign(y,one);
220 	if((m=(k=logb(y))-logb(x)) > 60) t=big+big;
221 	    else if(m < -80 ) t=y/x;
222 	    else { t = y/x ; y = scalb(y,-k); x=scalb(x,-k); }
223 
224     /* begin argument reduction */
225 begin:
226 	if (t < 2.4375) {
227 
228 	/* truncate 4(t+1/16) to integer for branching */
229 	    k = 4 * (t+0.0625);
230 	    switch (k) {
231 
232 	    /* t is in [0,7/16] */
233 	    case 0:
234 	    case 1:
235 		if (t < small)
236 		    { big + small ;  /* raise inexact flag */
237 		      return (copysign((signx>zero)?t:PI-t,signy)); }
238 
239 		hi = zero;  lo = zero;  break;
240 
241 	    /* t is in [7/16,11/16] */
242 	    case 2:
243 		hi = athfhi; lo = athflo;
244 		z = x+x;
245 		t = ( (y+y) - x ) / ( z +  y ); break;
246 
247 	    /* t is in [11/16,19/16] */
248 	    case 3:
249 	    case 4:
250 		hi = PIo4; lo = zero;
251 		t = ( y - x ) / ( x + y ); break;
252 
253 	    /* t is in [19/16,39/16] */
254 	    default:
255 		hi = at1fhi; lo = at1flo;
256 		z = y-x; y=y+y+y; t = x+x;
257 		t = ( (z+z)-x ) / ( t + y ); break;
258 	    }
259 	}
260 	/* end of if (t < 2.4375) */
261 
262 	else
263 	{
264 	    hi = PIo2; lo = zero;
265 
266 	    /* t is in [2.4375, big] */
267 	    if (t <= big)  t = - x / y;
268 
269 	    /* t is in [big, INF] */
270 	    else
271 	      { big+small;	/* raise inexact flag */
272 		t = zero; }
273 	}
274     /* end of argument reduction */
275 
276     /* compute atan(t) for t in [-.4375, .4375] */
277 	z = t*t;
278 #if defined(vax)||defined(tahoe)
279 	z = t*(z*(a1+z*(a2+z*(a3+z*(a4+z*(a5+z*(a6+z*(a7+z*(a8+
280 			z*(a9+z*(a10+z*(a11+z*a12))))))))))));
281 #else	/* defined(vax)||defined(tahoe) */
282 	z = t*(z*(a1+z*(a2+z*(a3+z*(a4+z*(a5+z*(a6+z*(a7+z*(a8+
283 			z*(a9+z*(a10+z*a11)))))))))));
284 #endif	/* defined(vax)||defined(tahoe) */
285 	z = lo - z; z += t; z += hi;
286 
287 	return(copysign((signx>zero)?z:PI-z,signy));
288 }
289