1*24579Szliu /* 2*24579Szliu * Copyright (c) 1985 Regents of the University of California. 3*24579Szliu * 4*24579Szliu * Use and reproduction of this software are granted in accordance with 5*24579Szliu * the terms and conditions specified in the Berkeley Software License 6*24579Szliu * Agreement (in particular, this entails acknowledgement of the programs' 7*24579Szliu * source, and inclusion of this notice) with the additional understanding 8*24579Szliu * that all recipients should regard themselves as participants in an 9*24579Szliu * ongoing research project and hence should feel obligated to report 10*24579Szliu * their experiences (good or bad) with these elementary function codes, 11*24579Szliu * using "sendbug 4bsd-bugs@BERKELEY", to the authors. 12*24579Szliu */ 13*24579Szliu 14*24579Szliu #ifndef lint 15*24579Szliu static char sccsid[] = "@(#)cabs.c 1.1 (ELEFUNT) 09/06/85"; 16*24579Szliu #endif not lint 17*24579Szliu 18*24579Szliu /* CABS(Z) 19*24579Szliu * RETURN THE ABSOLUTE VALUE OF THE COMPLEX NUMBER Z = X + iY 20*24579Szliu * DOUBLE PRECISION (VAX D format 56 bits, IEEE DOUBLE 53 BITS) 21*24579Szliu * CODED IN C BY K.C. NG, 11/28/84. 22*24579Szliu * REVISED BY K.C. NG, 7/12/85. 23*24579Szliu * 24*24579Szliu * Required kernel function : 25*24579Szliu * hypot(x,y) 26*24579Szliu * 27*24579Szliu * Method : 28*24579Szliu * cabs(z) = hypot(x,y) . 29*24579Szliu */ 30*24579Szliu 31*24579Szliu double cabs(z) 32*24579Szliu struct { double x, y;} z; 33*24579Szliu { 34*24579Szliu double hypot(); 35*24579Szliu return(hypot(z.x,z.y)); 36*24579Szliu } 37*24579Szliu 38*24579Szliu 39*24579Szliu /* HYPOT(X,Y) 40*24579Szliu * RETURN THE SQUARE ROOT OF X^2 + Y^2 WHERE Z=X+iY 41*24579Szliu * DOUBLE PRECISION (VAX D format 56 bits, IEEE DOUBLE 53 BITS) 42*24579Szliu * CODED IN C BY K.C. NG, 11/28/84; 43*24579Szliu * REVISED BY K.C. NG, 7/12/85. 44*24579Szliu * 45*24579Szliu * Required system supported functions : 46*24579Szliu * copysign(x,y) 47*24579Szliu * finite(x) 48*24579Szliu * scalb(x,N) 49*24579Szliu * sqrt(x) 50*24579Szliu * 51*24579Szliu * Method : 52*24579Szliu * 1. replace x by |x| and y by |y|, and swap x and 53*24579Szliu * y if y > x (hence x is never smaller than y). 54*24579Szliu * 2. Hypot(x,y) is computed by: 55*24579Szliu * Case I, x/y > 2 56*24579Szliu * 57*24579Szliu * y 58*24579Szliu * hypot = x + ----------------------------- 59*24579Szliu * 2 60*24579Szliu * sqrt ( 1 + [x/y] ) + x/y 61*24579Szliu * 62*24579Szliu * Case II, x/y <= 2 63*24579Szliu * y 64*24579Szliu * hypot = x + -------------------------------------------------- 65*24579Szliu * 2 66*24579Szliu * [x/y] - 2 67*24579Szliu * (sqrt(2)+1) + (x-y)/y + ----------------------------- 68*24579Szliu * 2 69*24579Szliu * sqrt ( 1 + [x/y] ) + sqrt(2) 70*24579Szliu * 71*24579Szliu * 72*24579Szliu * 73*24579Szliu * Special cases: 74*24579Szliu * hypot(x,y) is INF if x or y is +INF or -INF; else 75*24579Szliu * hypot(x,y) is NAN if x or y is NAN. 76*24579Szliu * 77*24579Szliu * Accuracy: 78*24579Szliu * hypot(x,y) returns the sqrt(x^2+y^2) with error less than 1 ulps (units 79*24579Szliu * in the last place). See Kahan's "Interval Arithmetic Options in the 80*24579Szliu * Proposed IEEE Floating Point Arithmetic Standard", Interval Mathematics 81*24579Szliu * 1980, Edited by Karl L.E. Nickel, pp 99-128. (A faster but less accurate 82*24579Szliu * code follows in comments.) In a test run with 500,000 random arguments 83*24579Szliu * on a VAX, the maximum observed error was .959 ulps. 84*24579Szliu * 85*24579Szliu * Constants: 86*24579Szliu * The hexadecimal values are the intended ones for the following constants. 87*24579Szliu * The decimal values may be used, provided that the compiler will convert 88*24579Szliu * from decimal to binary accurately enough to produce the hexadecimal values 89*24579Szliu * shown. 90*24579Szliu */ 91*24579Szliu 92*24579Szliu #ifdef VAX /* VAX D format */ 93*24579Szliu /* static double */ 94*24579Szliu /* r2p1hi = 2.4142135623730950345E0 , Hex 2^ 2 * .9A827999FCEF32 */ 95*24579Szliu /* r2p1lo = 1.4349369327986523769E-17 , Hex 2^-55 * .84597D89B3754B */ 96*24579Szliu /* sqrt2 = 1.4142135623730950622E0 ; Hex 2^ 1 * .B504F333F9DE65 */ 97*24579Szliu static long r2p1hix[] = { 0x8279411a, 0xef3299fc}; 98*24579Szliu static long r2p1lox[] = { 0x597d2484, 0x754b89b3}; 99*24579Szliu static long sqrt2x[] = { 0x04f340b5, 0xde6533f9}; 100*24579Szliu #define r2p1hi (*(double*)r2p1hix) 101*24579Szliu #define r2p1lo (*(double*)r2p1lox) 102*24579Szliu #define sqrt2 (*(double*)sqrt2x) 103*24579Szliu #else /* IEEE double format */ 104*24579Szliu static double 105*24579Szliu r2p1hi = 2.4142135623730949234E0 , /*Hex 2^1 * 1.3504F333F9DE6 */ 106*24579Szliu r2p1lo = 1.2537167179050217666E-16 , /*Hex 2^-53 * 1.21165F626CDD5 */ 107*24579Szliu sqrt2 = 1.4142135623730951455E0 ; /*Hex 2^ 0 * 1.6A09E667F3BCD */ 108*24579Szliu #endif 109*24579Szliu 110*24579Szliu double hypot(x,y) 111*24579Szliu double x, y; 112*24579Szliu { 113*24579Szliu static double zero=0, one=1, 114*24579Szliu small=1.0E-18; /* fl(1+small)==1 */ 115*24579Szliu static ibig=30; /* fl(1+2**(2*ibig))==1 */ 116*24579Szliu double copysign(),scalb(),logb(),sqrt(),t,r; 117*24579Szliu int finite(), exp; 118*24579Szliu 119*24579Szliu if(finite(x)) 120*24579Szliu if(finite(y)) 121*24579Szliu { 122*24579Szliu x=copysign(x,one); 123*24579Szliu y=copysign(y,one); 124*24579Szliu if(y > x) 125*24579Szliu { t=x; x=y; y=t; } 126*24579Szliu if(x == zero) return(zero); 127*24579Szliu if(y == zero) return(x); 128*24579Szliu exp= logb(x); 129*24579Szliu if(exp-(int)logb(y) > ibig ) 130*24579Szliu /* raise inexact flag and return |x| */ 131*24579Szliu { one+small; return(x); } 132*24579Szliu 133*24579Szliu /* start computing sqrt(x^2 + y^2) */ 134*24579Szliu r=x-y; 135*24579Szliu if(r>y) { /* x/y > 2 */ 136*24579Szliu r=x/y; 137*24579Szliu r=r+sqrt(one+r*r); } 138*24579Szliu else { /* 1 <= x/y <= 2 */ 139*24579Szliu r/=y; t=r*(r+2.0); 140*24579Szliu r+=t/(sqrt2+sqrt(2.0+t)); 141*24579Szliu r+=r2p1lo; r+=r2p1hi; } 142*24579Szliu 143*24579Szliu r=y/r; 144*24579Szliu return(x+r); 145*24579Szliu 146*24579Szliu } 147*24579Szliu 148*24579Szliu else if(y==y) /* y is +-INF */ 149*24579Szliu return(copysign(y,one)); 150*24579Szliu else 151*24579Szliu return(y); /* y is NaN and x is finite */ 152*24579Szliu 153*24579Szliu else if(x==x) /* x is +-INF */ 154*24579Szliu return (copysign(x,one)); 155*24579Szliu else if(finite(y)) 156*24579Szliu return(x); /* x is NaN, y is finite */ 157*24579Szliu else if(y!=y) return(y); /* x and y is NaN */ 158*24579Szliu else return(copysign(y,one)); /* y is INF */ 159*24579Szliu } 160*24579Szliu 161*24579Szliu /* A faster but less accurate version of cabs(x,y) */ 162*24579Szliu #if 0 163*24579Szliu double hypot(x,y) 164*24579Szliu double x, y; 165*24579Szliu { 166*24579Szliu static double zero=0, one=1; 167*24579Szliu small=1.0E-18; /* fl(1+small)==1 */ 168*24579Szliu static ibig=30; /* fl(1+2**(2*ibig))==1 */ 169*24579Szliu double copysign(),scalb(),logb(),sqrt(),temp; 170*24579Szliu int finite(), exp; 171*24579Szliu 172*24579Szliu if(finite(x)) 173*24579Szliu if(finite(y)) 174*24579Szliu { 175*24579Szliu x=copysign(x,one); 176*24579Szliu y=copysign(y,one); 177*24579Szliu if(y > x) 178*24579Szliu { temp=x; x=y; y=temp; } 179*24579Szliu if(x == zero) return(zero); 180*24579Szliu if(y == zero) return(x); 181*24579Szliu exp= logb(x); 182*24579Szliu x=scalb(x,-exp); 183*24579Szliu if(exp-(int)logb(y) > ibig ) 184*24579Szliu /* raise inexact flag and return |x| */ 185*24579Szliu { one+small; return(scalb(x,exp)); } 186*24579Szliu else y=scalb(y,-exp); 187*24579Szliu return(scalb(sqrt(x*x+y*y),exp)); 188*24579Szliu } 189*24579Szliu 190*24579Szliu else if(y==y) /* y is +-INF */ 191*24579Szliu return(copysign(y,one)); 192*24579Szliu else 193*24579Szliu return(y); /* y is NaN and x is finite */ 194*24579Szliu 195*24579Szliu else if(x==x) /* x is +-INF */ 196*24579Szliu return (copysign(x,one)); 197*24579Szliu else if(finite(y)) 198*24579Szliu return(x); /* x is NaN, y is finite */ 199*24579Szliu else if(y!=y) return(y); /* x and y is NaN */ 200*24579Szliu else return(copysign(y,one)); /* y is INF */ 201*24579Szliu } 202*24579Szliu #endif 203