xref: /freebsd/lib/msun/src/e_sqrt.c (revision 0b87f79976047c8f4332bbf7dc03146f6b0de79f)
1 /* @(#)e_sqrt.c 5.1 93/09/24 */
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 #ifndef lint
14 static char rcsid[] = "$FreeBSD$";
15 #endif
16 
17 /* __ieee754_sqrt(x)
18  * Return correctly rounded sqrt.
19  *           ------------------------------------------
20  *	     |  Use the hardware sqrt if you have one |
21  *           ------------------------------------------
22  * Method:
23  *   Bit by bit method using integer arithmetic. (Slow, but portable)
24  *   1. Normalization
25  *	Scale x to y in [1,4) with even powers of 2:
26  *	find an integer k such that  1 <= (y=x*2^(2k)) < 4, then
27  *		sqrt(x) = 2^k * sqrt(y)
28  *   2. Bit by bit computation
29  *	Let q  = sqrt(y) truncated to i bit after binary point (q = 1),
30  *	     i							 0
31  *                                     i+1         2
32  *	    s  = 2*q , and	y  =  2   * ( y - q  ).		(1)
33  *	     i      i            i                 i
34  *
35  *	To compute q    from q , one checks whether
36  *		    i+1       i
37  *
38  *			      -(i+1) 2
39  *			(q + 2      ) <= y.			(2)
40  *     			  i
41  *							      -(i+1)
42  *	If (2) is false, then q   = q ; otherwise q   = q  + 2      .
43  *		 	       i+1   i             i+1   i
44  *
45  *	With some algebric manipulation, it is not difficult to see
46  *	that (2) is equivalent to
47  *                             -(i+1)
48  *			s  +  2       <= y			(3)
49  *			 i                i
50  *
51  *	The advantage of (3) is that s  and y  can be computed by
52  *				      i      i
53  *	the following recurrence formula:
54  *	    if (3) is false
55  *
56  *	    s     =  s  ,	y    = y   ;			(4)
57  *	     i+1      i		 i+1    i
58  *
59  *	    otherwise,
60  *                         -i                     -(i+1)
61  *	    s	  =  s  + 2  ,  y    = y  -  s  - 2  		(5)
62  *           i+1      i          i+1    i     i
63  *
64  *	One may easily use induction to prove (4) and (5).
65  *	Note. Since the left hand side of (3) contain only i+2 bits,
66  *	      it does not necessary to do a full (53-bit) comparison
67  *	      in (3).
68  *   3. Final rounding
69  *	After generating the 53 bits result, we compute one more bit.
70  *	Together with the remainder, we can decide whether the
71  *	result is exact, bigger than 1/2ulp, or less than 1/2ulp
72  *	(it will never equal to 1/2ulp).
73  *	The rounding mode can be detected by checking whether
74  *	huge + tiny is equal to huge, and whether huge - tiny is
75  *	equal to huge for some floating point number "huge" and "tiny".
76  *
77  * Special cases:
78  *	sqrt(+-0) = +-0 	... exact
79  *	sqrt(inf) = inf
80  *	sqrt(-ve) = NaN		... with invalid signal
81  *	sqrt(NaN) = NaN		... with invalid signal for signaling NaN
82  *
83  * Other methods : see the appended file at the end of the program below.
84  *---------------
85  */
86 
87 #include "math.h"
88 #include "math_private.h"
89 
90 static	const double	one	= 1.0, tiny=1.0e-300;
91 
92 double
93 __generic___ieee754_sqrt(double x)
94 {
95 	double z;
96 	int32_t sign = (int)0x80000000;
97 	int32_t ix0,s0,q,m,t,i;
98 	u_int32_t r,t1,s1,ix1,q1;
99 
100 	EXTRACT_WORDS(ix0,ix1,x);
101 
102     /* take care of Inf and NaN */
103 	if((ix0&0x7ff00000)==0x7ff00000) {
104 	    return x*x+x;		/* sqrt(NaN)=NaN, sqrt(+inf)=+inf
105 					   sqrt(-inf)=sNaN */
106 	}
107     /* take care of zero */
108 	if(ix0<=0) {
109 	    if(((ix0&(~sign))|ix1)==0) return x;/* sqrt(+-0) = +-0 */
110 	    else if(ix0<0)
111 		return (x-x)/(x-x);		/* sqrt(-ve) = sNaN */
112 	}
113     /* normalize x */
114 	m = (ix0>>20);
115 	if(m==0) {				/* subnormal x */
116 	    while(ix0==0) {
117 		m -= 21;
118 		ix0 |= (ix1>>11); ix1 <<= 21;
119 	    }
120 	    for(i=0;(ix0&0x00100000)==0;i++) ix0<<=1;
121 	    m -= i-1;
122 	    ix0 |= (ix1>>(32-i));
123 	    ix1 <<= i;
124 	}
125 	m -= 1023;	/* unbias exponent */
126 	ix0 = (ix0&0x000fffff)|0x00100000;
127 	if(m&1){	/* odd m, double x to make it even */
128 	    ix0 += ix0 + ((ix1&sign)>>31);
129 	    ix1 += ix1;
130 	}
131 	m >>= 1;	/* m = [m/2] */
132 
133     /* generate sqrt(x) bit by bit */
134 	ix0 += ix0 + ((ix1&sign)>>31);
135 	ix1 += ix1;
136 	q = q1 = s0 = s1 = 0;	/* [q,q1] = sqrt(x) */
137 	r = 0x00200000;		/* r = moving bit from right to left */
138 
139 	while(r!=0) {
140 	    t = s0+r;
141 	    if(t<=ix0) {
142 		s0   = t+r;
143 		ix0 -= t;
144 		q   += r;
145 	    }
146 	    ix0 += ix0 + ((ix1&sign)>>31);
147 	    ix1 += ix1;
148 	    r>>=1;
149 	}
150 
151 	r = sign;
152 	while(r!=0) {
153 	    t1 = s1+r;
154 	    t  = s0;
155 	    if((t<ix0)||((t==ix0)&&(t1<=ix1))) {
156 		s1  = t1+r;
157 		if(((t1&sign)==sign)&&(s1&sign)==0) s0 += 1;
158 		ix0 -= t;
159 		if (ix1 < t1) ix0 -= 1;
160 		ix1 -= t1;
161 		q1  += r;
162 	    }
163 	    ix0 += ix0 + ((ix1&sign)>>31);
164 	    ix1 += ix1;
165 	    r>>=1;
166 	}
167 
168     /* use floating add to find out rounding direction */
169 	if((ix0|ix1)!=0) {
170 	    z = one-tiny; /* trigger inexact flag */
171 	    if (z>=one) {
172 	        z = one+tiny;
173 	        if (q1==(u_int32_t)0xffffffff) { q1=0; q += 1;}
174 		else if (z>one) {
175 		    if (q1==(u_int32_t)0xfffffffe) q+=1;
176 		    q1+=2;
177 		} else
178 	            q1 += (q1&1);
179 	    }
180 	}
181 	ix0 = (q>>1)+0x3fe00000;
182 	ix1 =  q1>>1;
183 	if ((q&1)==1) ix1 |= sign;
184 	ix0 += (m <<20);
185 	INSERT_WORDS(z,ix0,ix1);
186 	return z;
187 }
188 
189 /*
190 Other methods  (use floating-point arithmetic)
191 -------------
192 (This is a copy of a drafted paper by Prof W. Kahan
193 and K.C. Ng, written in May, 1986)
194 
195 	Two algorithms are given here to implement sqrt(x)
196 	(IEEE double precision arithmetic) in software.
197 	Both supply sqrt(x) correctly rounded. The first algorithm (in
198 	Section A) uses newton iterations and involves four divisions.
199 	The second one uses reciproot iterations to avoid division, but
200 	requires more multiplications. Both algorithms need the ability
201 	to chop results of arithmetic operations instead of round them,
202 	and the INEXACT flag to indicate when an arithmetic operation
203 	is executed exactly with no roundoff error, all part of the
204 	standard (IEEE 754-1985). The ability to perform shift, add,
205 	subtract and logical AND operations upon 32-bit words is needed
206 	too, though not part of the standard.
207 
208 A.  sqrt(x) by Newton Iteration
209 
210    (1)	Initial approximation
211 
212 	Let x0 and x1 be the leading and the trailing 32-bit words of
213 	a floating point number x (in IEEE double format) respectively
214 
215 	    1    11		     52				  ...widths
216 	   ------------------------------------------------------
217 	x: |s|	  e     |	      f				|
218 	   ------------------------------------------------------
219 	      msb    lsb  msb				      lsb ...order
220 
221 
222 	     ------------------------  	     ------------------------
223 	x0:  |s|   e    |    f1     |	 x1: |          f2           |
224 	     ------------------------  	     ------------------------
225 
226 	By performing shifts and subtracts on x0 and x1 (both regarded
227 	as integers), we obtain an 8-bit approximation of sqrt(x) as
228 	follows.
229 
230 		k  := (x0>>1) + 0x1ff80000;
231 		y0 := k - T1[31&(k>>15)].	... y ~ sqrt(x) to 8 bits
232 	Here k is a 32-bit integer and T1[] is an integer array containing
233 	correction terms. Now magically the floating value of y (y's
234 	leading 32-bit word is y0, the value of its trailing word is 0)
235 	approximates sqrt(x) to almost 8-bit.
236 
237 	Value of T1:
238 	static int T1[32]= {
239 	0,	1024,	3062,	5746,	9193,	13348,	18162,	23592,
240 	29598,	36145,	43202,	50740,	58733,	67158,	75992,	85215,
241 	83599,	71378,	60428,	50647,	41945,	34246,	27478,	21581,
242 	16499,	12183,	8588,	5674,	3403,	1742,	661,	130,};
243 
244     (2)	Iterative refinement
245 
246 	Apply Heron's rule three times to y, we have y approximates
247 	sqrt(x) to within 1 ulp (Unit in the Last Place):
248 
249 		y := (y+x/y)/2		... almost 17 sig. bits
250 		y := (y+x/y)/2		... almost 35 sig. bits
251 		y := y-(y-x/y)/2	... within 1 ulp
252 
253 
254 	Remark 1.
255 	    Another way to improve y to within 1 ulp is:
256 
257 		y := (y+x/y)		... almost 17 sig. bits to 2*sqrt(x)
258 		y := y - 0x00100006	... almost 18 sig. bits to sqrt(x)
259 
260 				2
261 			    (x-y )*y
262 		y := y + 2* ----------	...within 1 ulp
263 			       2
264 			     3y  + x
265 
266 
267 	This formula has one division fewer than the one above; however,
268 	it requires more multiplications and additions. Also x must be
269 	scaled in advance to avoid spurious overflow in evaluating the
270 	expression 3y*y+x. Hence it is not recommended uless division
271 	is slow. If division is very slow, then one should use the
272 	reciproot algorithm given in section B.
273 
274     (3) Final adjustment
275 
276 	By twiddling y's last bit it is possible to force y to be
277 	correctly rounded according to the prevailing rounding mode
278 	as follows. Let r and i be copies of the rounding mode and
279 	inexact flag before entering the square root program. Also we
280 	use the expression y+-ulp for the next representable floating
281 	numbers (up and down) of y. Note that y+-ulp = either fixed
282 	point y+-1, or multiply y by nextafter(1,+-inf) in chopped
283 	mode.
284 
285 		I := FALSE;	... reset INEXACT flag I
286 		R := RZ;	... set rounding mode to round-toward-zero
287 		z := x/y;	... chopped quotient, possibly inexact
288 		If(not I) then {	... if the quotient is exact
289 		    if(z=y) {
290 		        I := i;	 ... restore inexact flag
291 		        R := r;  ... restore rounded mode
292 		        return sqrt(x):=y.
293 		    } else {
294 			z := z - ulp;	... special rounding
295 		    }
296 		}
297 		i := TRUE;		... sqrt(x) is inexact
298 		If (r=RN) then z=z+ulp	... rounded-to-nearest
299 		If (r=RP) then {	... round-toward-+inf
300 		    y = y+ulp; z=z+ulp;
301 		}
302 		y := y+z;		... chopped sum
303 		y0:=y0-0x00100000;	... y := y/2 is correctly rounded.
304 	        I := i;	 		... restore inexact flag
305 	        R := r;  		... restore rounded mode
306 	        return sqrt(x):=y.
307 
308     (4)	Special cases
309 
310 	Square root of +inf, +-0, or NaN is itself;
311 	Square root of a negative number is NaN with invalid signal.
312 
313 
314 B.  sqrt(x) by Reciproot Iteration
315 
316    (1)	Initial approximation
317 
318 	Let x0 and x1 be the leading and the trailing 32-bit words of
319 	a floating point number x (in IEEE double format) respectively
320 	(see section A). By performing shifs and subtracts on x0 and y0,
321 	we obtain a 7.8-bit approximation of 1/sqrt(x) as follows.
322 
323 	    k := 0x5fe80000 - (x0>>1);
324 	    y0:= k - T2[63&(k>>14)].	... y ~ 1/sqrt(x) to 7.8 bits
325 
326 	Here k is a 32-bit integer and T2[] is an integer array
327 	containing correction terms. Now magically the floating
328 	value of y (y's leading 32-bit word is y0, the value of
329 	its trailing word y1 is set to zero) approximates 1/sqrt(x)
330 	to almost 7.8-bit.
331 
332 	Value of T2:
333 	static int T2[64]= {
334 	0x1500,	0x2ef8,	0x4d67,	0x6b02,	0x87be,	0xa395,	0xbe7a,	0xd866,
335 	0xf14a,	0x1091b,0x11fcd,0x13552,0x14999,0x15c98,0x16e34,0x17e5f,
336 	0x18d03,0x19a01,0x1a545,0x1ae8a,0x1b5c4,0x1bb01,0x1bfde,0x1c28d,
337 	0x1c2de,0x1c0db,0x1ba73,0x1b11c,0x1a4b5,0x1953d,0x18266,0x16be0,
338 	0x1683e,0x179d8,0x18a4d,0x19992,0x1a789,0x1b445,0x1bf61,0x1c989,
339 	0x1d16d,0x1d77b,0x1dddf,0x1e2ad,0x1e5bf,0x1e6e8,0x1e654,0x1e3cd,
340 	0x1df2a,0x1d635,0x1cb16,0x1be2c,0x1ae4e,0x19bde,0x1868e,0x16e2e,
341 	0x1527f,0x1334a,0x11051,0xe951,	0xbe01,	0x8e0d,	0x5924,	0x1edd,};
342 
343     (2)	Iterative refinement
344 
345 	Apply Reciproot iteration three times to y and multiply the
346 	result by x to get an approximation z that matches sqrt(x)
347 	to about 1 ulp. To be exact, we will have
348 		-1ulp < sqrt(x)-z<1.0625ulp.
349 
350 	... set rounding mode to Round-to-nearest
351 	   y := y*(1.5-0.5*x*y*y)	... almost 15 sig. bits to 1/sqrt(x)
352 	   y := y*((1.5-2^-30)+0.5*x*y*y)... about 29 sig. bits to 1/sqrt(x)
353 	... special arrangement for better accuracy
354 	   z := x*y			... 29 bits to sqrt(x), with z*y<1
355 	   z := z + 0.5*z*(1-z*y)	... about 1 ulp to sqrt(x)
356 
357 	Remark 2. The constant 1.5-2^-30 is chosen to bias the error so that
358 	(a) the term z*y in the final iteration is always less than 1;
359 	(b) the error in the final result is biased upward so that
360 		-1 ulp < sqrt(x) - z < 1.0625 ulp
361 	    instead of |sqrt(x)-z|<1.03125ulp.
362 
363     (3)	Final adjustment
364 
365 	By twiddling y's last bit it is possible to force y to be
366 	correctly rounded according to the prevailing rounding mode
367 	as follows. Let r and i be copies of the rounding mode and
368 	inexact flag before entering the square root program. Also we
369 	use the expression y+-ulp for the next representable floating
370 	numbers (up and down) of y. Note that y+-ulp = either fixed
371 	point y+-1, or multiply y by nextafter(1,+-inf) in chopped
372 	mode.
373 
374 	R := RZ;		... set rounding mode to round-toward-zero
375 	switch(r) {
376 	    case RN:		... round-to-nearest
377 	       if(x<= z*(z-ulp)...chopped) z = z - ulp; else
378 	       if(x<= z*(z+ulp)...chopped) z = z; else z = z+ulp;
379 	       break;
380 	    case RZ:case RM:	... round-to-zero or round-to--inf
381 	       R:=RP;		... reset rounding mod to round-to-+inf
382 	       if(x<z*z ... rounded up) z = z - ulp; else
383 	       if(x>=(z+ulp)*(z+ulp) ...rounded up) z = z+ulp;
384 	       break;
385 	    case RP:		... round-to-+inf
386 	       if(x>(z+ulp)*(z+ulp)...chopped) z = z+2*ulp; else
387 	       if(x>z*z ...chopped) z = z+ulp;
388 	       break;
389 	}
390 
391 	Remark 3. The above comparisons can be done in fixed point. For
392 	example, to compare x and w=z*z chopped, it suffices to compare
393 	x1 and w1 (the trailing parts of x and w), regarding them as
394 	two's complement integers.
395 
396 	...Is z an exact square root?
397 	To determine whether z is an exact square root of x, let z1 be the
398 	trailing part of z, and also let x0 and x1 be the leading and
399 	trailing parts of x.
400 
401 	If ((z1&0x03ffffff)!=0)	... not exact if trailing 26 bits of z!=0
402 	    I := 1;		... Raise Inexact flag: z is not exact
403 	else {
404 	    j := 1 - [(x0>>20)&1]	... j = logb(x) mod 2
405 	    k := z1 >> 26;		... get z's 25-th and 26-th
406 					    fraction bits
407 	    I := i or (k&j) or ((k&(j+j+1))!=(x1&3));
408 	}
409 	R:= r		... restore rounded mode
410 	return sqrt(x):=z.
411 
412 	If multiplication is cheaper then the foregoing red tape, the
413 	Inexact flag can be evaluated by
414 
415 	    I := i;
416 	    I := (z*z!=x) or I.
417 
418 	Note that z*z can overwrite I; this value must be sensed if it is
419 	True.
420 
421 	Remark 4. If z*z = x exactly, then bit 25 to bit 0 of z1 must be
422 	zero.
423 
424 		    --------------------
425 		z1: |        f2        |
426 		    --------------------
427 		bit 31		   bit 0
428 
429 	Further more, bit 27 and 26 of z1, bit 0 and 1 of x1, and the odd
430 	or even of logb(x) have the following relations:
431 
432 	-------------------------------------------------
433 	bit 27,26 of z1		bit 1,0 of x1	logb(x)
434 	-------------------------------------------------
435 	00			00		odd and even
436 	01			01		even
437 	10			10		odd
438 	10			00		even
439 	11			01		even
440 	-------------------------------------------------
441 
442     (4)	Special cases (see (4) of Section A).
443 
444  */
445 
446