xref: /csrg-svn/lib/libm/common/atan2.c (revision 24718)
124578Szliu /*
224578Szliu  * Copyright (c) 1985 Regents of the University of California.
324578Szliu  *
424578Szliu  * Use and reproduction of this software are granted  in  accordance  with
524578Szliu  * the terms and conditions specified in  the  Berkeley  Software  License
624578Szliu  * Agreement (in particular, this entails acknowledgement of the programs'
724578Szliu  * source, and inclusion of this notice) with the additional understanding
824578Szliu  * that  all  recipients  should regard themselves as participants  in  an
924578Szliu  * ongoing  research  project and hence should  feel  obligated  to report
1024578Szliu  * their  experiences (good or bad) with these elementary function  codes,
1124578Szliu  * using "sendbug 4bsd-bugs@BERKELEY", to the authors.
1224578Szliu  */
1324578Szliu 
1424578Szliu #ifndef lint
15*24718Selefunt static char sccsid[] =
16*24718Selefunt "@(#)atan2.c	1.3 (Berkeley) 8/21/85; 1.2 (ucb.elefunt) 09/12/85";
1724578Szliu #endif not lint
1824578Szliu 
1924578Szliu /* ATAN2(Y,X)
2024578Szliu  * RETURN ARG (X+iY)
2124578Szliu  * DOUBLE PRECISION (VAX D format 56 bits, IEEE DOUBLE 53 BITS)
2224578Szliu  * CODED IN C BY K.C. NG, 1/8/85;
2324578Szliu  * REVISED BY K.C. NG on 2/7/85, 2/13/85, 3/7/85, 3/30/85, 6/29/85.
2424578Szliu  *
2524578Szliu  * Required system supported functions :
2624578Szliu  *	copysign(x,y)
2724578Szliu  *	scalb(x,y)
2824578Szliu  *	logb(x)
2924578Szliu  *
3024578Szliu  * Method :
3124578Szliu  *	1. Reduce y to positive by atan2(y,x)=-atan2(-y,x).
3224578Szliu  *	2. Reduce x to positive by (if x and y are unexceptional):
3324578Szliu  *		ARG (x+iy) = arctan(y/x)   	   ... if x > 0,
3424578Szliu  *		ARG (x+iy) = pi - arctan[y/(-x)]   ... if x < 0,
3524578Szliu  *	3. According to the integer k=4t+0.25 truncated , t=y/x, the argument
3624578Szliu  *	   is further reduced to one of the following intervals and the
3724578Szliu  *	   arctangent of y/x is evaluated by the corresponding formula:
3824578Szliu  *
3924578Szliu  *         [0,7/16]	   atan(y/x) = t - t^3*(a1+t^2*(a2+...(a10+t^2*a11)...)
4024578Szliu  *	   [7/16,11/16]    atan(y/x) = atan(1/2) + atan( (y-x/2)/(x+y/2) )
4124578Szliu  *	   [11/16.19/16]   atan(y/x) = atan( 1 ) + atan( (y-x)/(x+y) )
4224578Szliu  *	   [19/16,39/16]   atan(y/x) = atan(3/2) + atan( (y-1.5x)/(x+1.5y) )
4324578Szliu  *	   [39/16,INF]     atan(y/x) = atan(INF) + atan( -x/y )
4424578Szliu  *
4524578Szliu  * Special cases:
4624578Szliu  * Notations: atan2(y,x) == ARG (x+iy) == ARG(x,y).
4724578Szliu  *
4824578Szliu  *	ARG( NAN , (anything) ) is NaN;
4924578Szliu  *	ARG( (anything), NaN ) is NaN;
5024578Szliu  *	ARG(+(anything but NaN), +-0) is +-0  ;
5124578Szliu  *	ARG(-(anything but NaN), +-0) is +-PI ;
5224578Szliu  *	ARG( 0, +-(anything but 0 and NaN) ) is +-PI/2;
5324578Szliu  *	ARG( +INF,+-(anything but INF and NaN) ) is +-0 ;
5424578Szliu  *	ARG( -INF,+-(anything but INF and NaN) ) is +-PI;
5524578Szliu  *	ARG( +INF,+-INF ) is +-PI/4 ;
5624578Szliu  *	ARG( -INF,+-INF ) is +-3PI/4;
5724578Szliu  *	ARG( (anything but,0,NaN, and INF),+-INF ) is +-PI/2;
5824578Szliu  *
5924578Szliu  * Accuracy:
6024578Szliu  *	atan2(y,x) returns (PI/pi) * the exact ARG (x+iy) nearly rounded,
6124578Szliu  *	where
6224578Szliu  *
6324578Szliu  *	in decimal:
6424578Szliu  *		pi = 3.141592653589793 23846264338327 .....
6524578Szliu  *    53 bits   PI = 3.141592653589793 115997963 ..... ,
6624578Szliu  *    56 bits   PI = 3.141592653589793 227020265 ..... ,
6724578Szliu  *
6824578Szliu  *	in hexadecimal:
6924578Szliu  *		pi = 3.243F6A8885A308D313198A2E....
7024578Szliu  *    53 bits   PI = 3.243F6A8885A30  =  2 * 1.921FB54442D18	error=.276ulps
7124578Szliu  *    56 bits   PI = 3.243F6A8885A308 =  4 * .C90FDAA22168C2    error=.206ulps
7224578Szliu  *
7324578Szliu  *	In a test run with 356,000 random argument on [-1,1] * [-1,1] on a
7424578Szliu  *	VAX, the maximum observed error was 1.41 ulps (units of the last place)
7524578Szliu  *	compared with (PI/pi)*(the exact ARG(x+iy)).
7624578Szliu  *
7724578Szliu  * Note:
7824578Szliu  *	We use machine PI (the true pi rounded) in place of the actual
7924578Szliu  *	value of pi for all the trig and inverse trig functions. In general,
8024578Szliu  *	if trig is one of sin, cos, tan, then computed trig(y) returns the
8124578Szliu  *	exact trig(y*pi/PI) nearly rounded; correspondingly, computed arctrig
8224578Szliu  *	returns the exact arctrig(y)*PI/pi nearly rounded. These guarantee the
8324578Szliu  *	trig functions have period PI, and trig(arctrig(x)) returns x for
8424578Szliu  *	all critical values x.
8524578Szliu  *
8624578Szliu  * Constants:
8724578Szliu  * The hexadecimal values are the intended ones for the following constants.
8824578Szliu  * The decimal values may be used, provided that the compiler will convert
8924578Szliu  * from decimal to binary accurately enough to produce the hexadecimal values
9024578Szliu  * shown.
9124578Szliu  */
9224578Szliu 
9324578Szliu static double
9424578Szliu #ifdef VAX 	/* VAX D format */
9524578Szliu athfhi =  4.6364760900080611433E-1    , /*Hex  2^ -1   *  .ED63382B0DDA7B */
9624578Szliu athflo =  1.9338828231967579916E-19   , /*Hex  2^-62   *  .E450059CFE92C0 */
9724578Szliu PIo4   =  7.8539816339744830676E-1    , /*Hex  2^  0   *  .C90FDAA22168C2 */
9824578Szliu at1fhi =  9.8279372324732906796E-1    , /*Hex  2^  0   *  .FB985E940FB4D9 */
9924578Szliu at1flo = -3.5540295636764633916E-18   , /*Hex  2^-57   * -.831EDC34D6EAEA */
10024578Szliu PIo2   =  1.5707963267948966135E0     , /*Hex  2^  1   *  .C90FDAA22168C2 */
10124578Szliu PI     =  3.1415926535897932270E0     , /*Hex  2^  2   *  .C90FDAA22168C2 */
10224578Szliu a1     =  3.3333333333333473730E-1    , /*Hex  2^ -1   *  .AAAAAAAAAAAB75 */
10324578Szliu a2     = -2.0000000000017730678E-1    , /*Hex  2^ -2   * -.CCCCCCCCCD946E */
10424578Szliu a3     =  1.4285714286694640301E-1    , /*Hex  2^ -2   *  .92492492744262 */
10524578Szliu a4     = -1.1111111135032672795E-1    , /*Hex  2^ -3   * -.E38E38EBC66292 */
10624578Szliu a5     =  9.0909091380563043783E-2    , /*Hex  2^ -3   *  .BA2E8BB31BD70C */
10724578Szliu a6     = -7.6922954286089459397E-2    , /*Hex  2^ -3   * -.9D89C827C37F18 */
10824578Szliu a7     =  6.6663180891693915586E-2    , /*Hex  2^ -3   *  .8886B4AE379E58 */
10924578Szliu a8     = -5.8772703698290408927E-2    , /*Hex  2^ -4   * -.F0BBA58481A942 */
11024578Szliu a9     =  5.2170707402812969804E-2    , /*Hex  2^ -4   *  .D5B0F3A1AB13AB */
11124578Szliu a10    = -4.4895863157820361210E-2    , /*Hex  2^ -4   * -.B7E4B97FD1048F */
11224578Szliu a11    =  3.3006147437343875094E-2    , /*Hex  2^ -4   *  .8731743CF72D87 */
11324578Szliu a12    = -1.4614844866464185439E-2    ; /*Hex  2^ -6   * -.EF731A2F3476D9 */
11424578Szliu #else 	/* IEEE double */
11524578Szliu athfhi =  4.6364760900080609352E-1    , /*Hex  2^ -2   *  1.DAC670561BB4F */
11624578Szliu athflo =  4.6249969567426939759E-18   , /*Hex  2^-58   *  1.5543B8F253271 */
11724578Szliu PIo4   =  7.8539816339744827900E-1    , /*Hex  2^ -1   *  1.921FB54442D18 */
11824578Szliu at1fhi =  9.8279372324732905408E-1    , /*Hex  2^ -1   *  1.F730BD281F69B */
11924578Szliu at1flo = -2.4407677060164810007E-17   , /*Hex  2^-56   * -1.C23DFEFEAE6B5 */
12024578Szliu PIo2   =  1.5707963267948965580E0     , /*Hex  2^  0   *  1.921FB54442D18 */
12124578Szliu PI     =  3.1415926535897931160E0     , /*Hex  2^  1   *  1.921FB54442D18 */
12224578Szliu a1     =  3.3333333333333942106E-1    , /*Hex  2^ -2   *  1.55555555555C3 */
12324578Szliu a2     = -1.9999999999979536924E-1    , /*Hex  2^ -3   * -1.9999999997CCD */
12424578Szliu a3     =  1.4285714278004377209E-1    , /*Hex  2^ -3   *  1.24924921EC1D7 */
12524578Szliu a4     = -1.1111110579344973814E-1    , /*Hex  2^ -4   * -1.C71C7059AF280 */
12624578Szliu a5     =  9.0908906105474668324E-2    , /*Hex  2^ -4   *  1.745CE5AA35DB2 */
12724578Szliu a6     = -7.6919217767468239799E-2    , /*Hex  2^ -4   * -1.3B0FA54BEC400 */
12824578Szliu a7     =  6.6614695906082474486E-2    , /*Hex  2^ -4   *  1.10DA924597FFF */
12924578Szliu a8     = -5.8358371008508623523E-2    , /*Hex  2^ -5   * -1.DE125FDDBD793 */
13024578Szliu a9     =  4.9850617156082015213E-2    , /*Hex  2^ -5   *  1.9860524BDD807 */
13124578Szliu a10    = -3.6700606902093604877E-2    , /*Hex  2^ -5   * -1.2CA6C04C6937A */
13224578Szliu a11    =  1.6438029044759730479E-2    ; /*Hex  2^ -6   *  1.0D52174A1BB54 */
13324578Szliu #endif
13424578Szliu 
13524578Szliu double atan2(y,x)
13624578Szliu double  y,x;
13724578Szliu {
13824578Szliu 	static double zero=0, one=1, small=1.0E-9, big=1.0E18;
13924578Szliu 	double copysign(),logb(),scalb(),t,z,signy,signx,hi,lo;
14024578Szliu 	int finite(), k,m;
14124578Szliu 
14224578Szliu     /* if x or y is NAN */
14324578Szliu 	if(x!=x) return(x); if(y!=y) return(y);
14424578Szliu 
14524578Szliu     /* copy down the sign of y and x */
14624578Szliu 	signy = copysign(one,y) ;
14724578Szliu 	signx = copysign(one,x) ;
14824578Szliu 
14924578Szliu     /* if x is 1.0, goto begin */
15024578Szliu 	if(x==1) { y=copysign(y,one); t=y; if(finite(t)) goto begin;}
15124578Szliu 
15224578Szliu     /* when y = 0 */
15324578Szliu 	if(y==zero) return((signx==one)?y:copysign(PI,signy));
15424578Szliu 
15524578Szliu     /* when x = 0 */
15624578Szliu 	if(x==zero) return(copysign(PIo2,signy));
15724578Szliu 
15824578Szliu     /* when x is INF */
15924578Szliu 	if(!finite(x))
16024578Szliu 	    if(!finite(y))
16124578Szliu 		return(copysign((signx==one)?PIo4:3*PIo4,signy));
16224578Szliu 	    else
16324578Szliu 		return(copysign((signx==one)?zero:PI,signy));
16424578Szliu 
16524578Szliu     /* when y is INF */
16624578Szliu 	if(!finite(y)) return(copysign(PIo2,signy));
16724578Szliu 
16824578Szliu 
16924578Szliu     /* compute y/x */
17024578Szliu 	x=copysign(x,one);
17124578Szliu 	y=copysign(y,one);
17224578Szliu 	if((m=(k=logb(y))-logb(x)) > 60) t=big+big;
17324578Szliu 	    else if(m < -80 ) t=y/x;
17424578Szliu 	    else { t = y/x ; y = scalb(y,-k); x=scalb(x,-k); }
17524578Szliu 
17624578Szliu     /* begin argument reduction */
17724578Szliu begin:
17824578Szliu 	if (t < 2.4375) {
17924578Szliu 
18024578Szliu 	/* truncate 4(t+1/16) to integer for branching */
18124578Szliu 	    k = 4 * (t+0.0625);
18224578Szliu 	    switch (k) {
18324578Szliu 
18424578Szliu 	    /* t is in [0,7/16] */
18524578Szliu 	    case 0:
18624578Szliu 	    case 1:
18724578Szliu 		if (t < small)
18824578Szliu 		    { big + small ;  /* raise inexact flag */
18924578Szliu 		      return (copysign((signx>zero)?t:PI-t,signy)); }
19024578Szliu 
19124578Szliu 		hi = zero;  lo = zero;  break;
19224578Szliu 
19324578Szliu 	    /* t is in [7/16,11/16] */
19424578Szliu 	    case 2:
19524578Szliu 		hi = athfhi; lo = athflo;
19624578Szliu 		z = x+x;
19724578Szliu 		t = ( (y+y) - x ) / ( z +  y ); break;
19824578Szliu 
19924578Szliu 	    /* t is in [11/16,19/16] */
20024578Szliu 	    case 3:
20124578Szliu 	    case 4:
20224578Szliu 		hi = PIo4; lo = zero;
20324578Szliu 		t = ( y - x ) / ( x + y ); break;
20424578Szliu 
20524578Szliu 	    /* t is in [19/16,39/16] */
20624578Szliu 	    default:
20724578Szliu 		hi = at1fhi; lo = at1flo;
20824578Szliu 		z = y-x; y=y+y+y; t = x+x;
20924578Szliu 		t = ( (z+z)-x ) / ( t + y ); break;
21024578Szliu 	    }
21124578Szliu 	}
21224578Szliu 	/* end of if (t < 2.4375) */
21324578Szliu 
21424578Szliu 	else
21524578Szliu 	{
21624578Szliu 	    hi = PIo2; lo = zero;
21724578Szliu 
21824578Szliu 	    /* t is in [2.4375, big] */
21924578Szliu 	    if (t <= big)  t = - x / y;
22024578Szliu 
22124578Szliu 	    /* t is in [big, INF] */
22224578Szliu 	    else
22324578Szliu 	      { big+small;	/* raise inexact flag */
22424578Szliu 		t = zero; }
22524578Szliu 	}
22624578Szliu     /* end of argument reduction */
22724578Szliu 
22824578Szliu     /* compute atan(t) for t in [-.4375, .4375] */
22924578Szliu 	z = t*t;
23024578Szliu #ifdef VAX
23124578Szliu 	z = t*(z*(a1+z*(a2+z*(a3+z*(a4+z*(a5+z*(a6+z*(a7+z*(a8+
23224578Szliu 			z*(a9+z*(a10+z*(a11+z*a12))))))))))));
23324578Szliu #else	/* IEEE double */
23424578Szliu 	z = t*(z*(a1+z*(a2+z*(a3+z*(a4+z*(a5+z*(a6+z*(a7+z*(a8+
23524578Szliu 			z*(a9+z*(a10+z*a11)))))))))));
23624578Szliu #endif
23724578Szliu 	z = lo - z; z += t; z += hi;
23824578Szliu 
23924578Szliu 	return(copysign((signx>zero)?z:PI-z,signy));
24024578Szliu }
241