Rev 1906 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 3362 | Serge | 1 | |
| 2 | /* |
||
| 3 | * ==================================================== |
||
| 4 | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
||
| 5 | * |
||
| 6 | * Developed at SunPro, a Sun Microsystems, Inc. business. |
||
| 7 | * Permission to use, copy, modify, and distribute this |
||
| 8 | * software is freely granted, provided that this notice |
||
| 9 | * is preserved. |
||
| 10 | * ==================================================== |
||
| 11 | */ |
||
| 12 | |||
| 13 | |||
| 14 | * |
||
| 15 | * Method : |
||
| 16 | * If (assume round-to-nearest) z=x*x+y*y |
||
| 17 | * has error less than sqrt(2)/2 ulp, than |
||
| 18 | * sqrt(z) has error less than 1 ulp (exercise). |
||
| 19 | * |
||
| 20 | * So, compute sqrt(x*x+y*y) with some care as |
||
| 21 | * follows to get the error below 1 ulp: |
||
| 22 | * |
||
| 23 | * Assume x>y>0; |
||
| 24 | * (if possible, set rounding to round-to-nearest) |
||
| 25 | * 1. if x > 2y use |
||
| 26 | * x1*x1+(y*y+(x2*(x+x1))) for x*x+y*y |
||
| 27 | * where x1 = x with lower 32 bits cleared, x2 = x-x1; else |
||
| 28 | * 2. if x <= 2y use |
||
| 29 | * t1*y1+((x-y)*(x-y)+(t1*y2+t2*y)) |
||
| 30 | * where t1 = 2x with lower 32 bits cleared, t2 = 2x-t1, |
||
| 31 | * y1= y with lower 32 bits chopped, y2 = y-y1. |
||
| 32 | * |
||
| 33 | * NOTE: scaling may be necessary if some argument is too |
||
| 34 | * large or too tiny |
||
| 35 | * |
||
| 36 | * Special cases: |
||
| 37 | * hypot(x,y) is INF if x or y is +INF or -INF; else |
||
| 38 | * hypot(x,y) is NAN if x or y is NAN. |
||
| 39 | * |
||
| 40 | * Accuracy: |
||
| 41 | * hypot(x,y) returns sqrt(x^2+y^2) with error less |
||
| 42 | * than 1 ulps (units in the last place) |
||
| 43 | */ |
||
| 44 | |||
| 45 | |||
| 46 | |||
| 47 | |||
| 48 | |||
| 49 | |||
| 50 | double __ieee754_hypot(double x, double y) |
||
| 51 | #else |
||
| 52 | double __ieee754_hypot(x,y) |
||
| 53 | double x, y; |
||
| 54 | #endif |
||
| 55 | { |
||
| 56 | double a=x,b=y,t1,t2,y1,y2,w; |
||
| 57 | __int32_t j,k,ha,hb; |
||
| 58 | |||
| 59 | |||
| 60 | ha &= 0x7fffffff; |
||
| 61 | GET_HIGH_WORD(hb,y); |
||
| 62 | hb &= 0x7fffffff; |
||
| 63 | if(hb > ha) {a=y;b=x;j=ha; ha=hb;hb=j;} else {a=x;b=y;} |
||
| 64 | SET_HIGH_WORD(a,ha); /* a <- |a| */ |
||
| 65 | SET_HIGH_WORD(b,hb); /* b <- |b| */ |
||
| 66 | if((ha-hb)>0x3c00000) {return a+b;} /* x/y > 2**60 */ |
||
| 67 | k=0; |
||
| 68 | if(ha > 0x5f300000) { /* a>2**500 */ |
||
| 69 | if(ha >= 0x7ff00000) { /* Inf or NaN */ |
||
| 70 | __uint32_t low; |
||
| 71 | w = a+b; /* for sNaN */ |
||
| 72 | GET_LOW_WORD(low,a); |
||
| 73 | if(((ha&0xfffff)|low)==0) w = a; |
||
| 74 | GET_LOW_WORD(low,b); |
||
| 75 | if(((hb^0x7ff00000)|low)==0) w = b; |
||
| 76 | return w; |
||
| 77 | } |
||
| 78 | /* scale a and b by 2**-600 */ |
||
| 79 | ha -= 0x25800000; hb -= 0x25800000; k += 600; |
||
| 80 | SET_HIGH_WORD(a,ha); |
||
| 81 | SET_HIGH_WORD(b,hb); |
||
| 82 | } |
||
| 83 | if(hb < 0x20b00000) { /* b < 2**-500 */ |
||
| 84 | if(hb <= 0x000fffff) { /* subnormal b or 0 */ |
||
| 85 | __uint32_t low; |
||
| 86 | GET_LOW_WORD(low,b); |
||
| 87 | if((hb|low)==0) return a; |
||
| 88 | t1=0; |
||
| 89 | SET_HIGH_WORD(t1,0x7fd00000); /* t1=2^1022 */ |
||
| 90 | b *= t1; |
||
| 91 | a *= t1; |
||
| 92 | k -= 1022; |
||
| 93 | } else { /* scale a and b by 2^600 */ |
||
| 94 | ha += 0x25800000; /* a *= 2^600 */ |
||
| 95 | hb += 0x25800000; /* b *= 2^600 */ |
||
| 96 | k -= 600; |
||
| 97 | SET_HIGH_WORD(a,ha); |
||
| 98 | SET_HIGH_WORD(b,hb); |
||
| 99 | } |
||
| 100 | } |
||
| 101 | /* medium size a and b */ |
||
| 102 | w = a-b; |
||
| 103 | if (w>b) { |
||
| 104 | t1 = 0; |
||
| 105 | SET_HIGH_WORD(t1,ha); |
||
| 106 | t2 = a-t1; |
||
| 107 | w = __ieee754_sqrt(t1*t1-(b*(-b)-t2*(a+t1))); |
||
| 108 | } else { |
||
| 109 | a = a+a; |
||
| 110 | y1 = 0; |
||
| 111 | SET_HIGH_WORD(y1,hb); |
||
| 112 | y2 = b - y1; |
||
| 113 | t1 = 0; |
||
| 114 | SET_HIGH_WORD(t1,ha+0x00100000); |
||
| 115 | t2 = a - t1; |
||
| 116 | w = __ieee754_sqrt(t1*y1-(w*(-w)-(t1*y2+t2*b))); |
||
| 117 | } |
||
| 118 | if(k!=0) { |
||
| 119 | __uint32_t high; |
||
| 120 | t1 = 1.0; |
||
| 121 | GET_HIGH_WORD(high,t1); |
||
| 122 | SET_HIGH_WORD(t1,high+(k<<20)); |
||
| 123 | return t1*w; |
||
| 124 | } else return w; |
||
| 125 | } |
||
| 126 | |||
| 127 | |||
| 128 | > |