Home | History | Annotate | Line # | Download | only in spmath
      1 /*	$NetBSD: dfsqrt.c,v 1.6 2016/02/29 18:22:06 christos Exp $	*/
      2 
      3 /*	$OpenBSD: dfsqrt.c,v 1.5 2001/03/29 03:58:17 mickey Exp $	*/
      4 
      5 /*
      6  * Copyright 1996 1995 by Open Software Foundation, Inc.
      7  *              All Rights Reserved
      8  *
      9  * Permission to use, copy, modify, and distribute this software and
     10  * its documentation for any purpose and without fee is hereby granted,
     11  * provided that the above copyright notice appears in all copies and
     12  * that both the copyright notice and this permission notice appear in
     13  * supporting documentation.
     14  *
     15  * OSF DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE
     16  * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
     17  * FOR A PARTICULAR PURPOSE.
     18  *
     19  * IN NO EVENT SHALL OSF BE LIABLE FOR ANY SPECIAL, INDIRECT, OR
     20  * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
     21  * LOSS OF USE, DATA OR PROFITS, WHETHER IN ACTION OF CONTRACT,
     22  * NEGLIGENCE, OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION
     23  * WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
     24  *
     25  */
     26 /*
     27  * pmk1.1
     28  */
     29 /*
     30  * (c) Copyright 1986 HEWLETT-PACKARD COMPANY
     31  *
     32  * To anyone who acknowledges that this file is provided "AS IS"
     33  * without any express or implied warranty:
     34  *     permission to use, copy, modify, and distribute this file
     35  * for any purpose is hereby granted without fee, provided that
     36  * the above copyright notice and this notice appears in all
     37  * copies, and that the name of Hewlett-Packard Company not be
     38  * used in advertising or publicity pertaining to distribution
     39  * of the software without specific, written prior permission.
     40  * Hewlett-Packard Company makes no representations about the
     41  * suitability of this software for any purpose.
     42  */
     43 
     44 #include <sys/cdefs.h>
     45 __KERNEL_RCSID(0, "$NetBSD: dfsqrt.c,v 1.6 2016/02/29 18:22:06 christos Exp $");
     46 
     47 #include "../spmath/float.h"
     48 #include "../spmath/dbl_float.h"
     49 
     50 /*
     51  *  Double Floating-point Square Root
     52  */
     53 
     54 /*ARGSUSED*/
     55 int
     56 dbl_fsqrt(dbl_floating_point *srcptr, dbl_floating_point *dstptr,
     57     unsigned int *status)
     58 {
     59 	register unsigned int srcp1, srcp2, resultp1, resultp2;
     60 	register unsigned int newbitp1, newbitp2, sump1, sump2;
     61 	register int src_exponent;
     62 	register int guardbit = false, even_exponent;
     63 
     64 	Dbl_copyfromptr(srcptr,srcp1,srcp2);
     65 	/*
     66 	 * check source operand for NaN or infinity
     67 	 */
     68 	if ((src_exponent = Dbl_exponent(srcp1)) == DBL_INFINITY_EXPONENT) {
     69 		/*
     70 		 * is signaling NaN?
     71 		 */
     72 		if (Dbl_isone_signaling(srcp1)) {
     73 			/* trap if INVALIDTRAP enabled */
     74 			if (Is_invalidtrap_enabled()) return(INVALIDEXCEPTION);
     75 			/* make NaN quiet */
     76 			Set_invalidflag();
     77 			Dbl_set_quiet(srcp1);
     78 		}
     79 		/*
     80 		 * Return quiet NaN or positive infinity.
     81 		 *  Fall thru to negative test if negative infinity.
     82 		 */
     83 		if (Dbl_iszero_sign(srcp1) ||
     84 		    Dbl_isnotzero_mantissa(srcp1,srcp2)) {
     85 			Dbl_copytoptr(srcp1,srcp2,dstptr);
     86 			return(NOEXCEPTION);
     87 		}
     88 	}
     89 
     90 	/*
     91 	 * check for zero source operand
     92 	 */
     93 	if (Dbl_iszero_exponentmantissa(srcp1,srcp2)) {
     94 		Dbl_copytoptr(srcp1,srcp2,dstptr);
     95 		return(NOEXCEPTION);
     96 	}
     97 
     98 	/*
     99 	 * check for negative source operand
    100 	 */
    101 	if (Dbl_isone_sign(srcp1)) {
    102 		/* trap if INVALIDTRAP enabled */
    103 		if (Is_invalidtrap_enabled()) return(INVALIDEXCEPTION);
    104 		/* make NaN quiet */
    105 		Set_invalidflag();
    106 		Dbl_makequietnan(srcp1,srcp2);
    107 		Dbl_copytoptr(srcp1,srcp2,dstptr);
    108 		return(NOEXCEPTION);
    109 	}
    110 
    111 	/*
    112 	 * Generate result
    113 	 */
    114 	if (src_exponent > 0) {
    115 		even_exponent = Dbl_hidden(srcp1);
    116 		Dbl_clear_signexponent_set_hidden(srcp1);
    117 	}
    118 	else {
    119 		/* normalize operand */
    120 		Dbl_clear_signexponent(srcp1);
    121 		src_exponent++;
    122 		Dbl_normalize(srcp1,srcp2,src_exponent);
    123 		even_exponent = src_exponent & 1;
    124 	}
    125 	if (even_exponent) {
    126 		/* exponent is even */
    127 		/* Add comment here.  Explain why odd exponent needs correction */
    128 		Dbl_leftshiftby1(srcp1,srcp2);
    129 	}
    130 	/*
    131 	 * Add comment here.  Explain following algorithm.
    132 	 *
    133 	 * Trust me, it works.
    134 	 *
    135 	 */
    136 	Dbl_setzero(resultp1,resultp2);
    137 	Dbl_allp1(newbitp1) = 1 << (DBL_P - 32);
    138 	Dbl_setzero_mantissap2(newbitp2);
    139 	while (Dbl_isnotzero(newbitp1,newbitp2) && Dbl_isnotzero(srcp1,srcp2)) {
    140 		Dbl_addition(resultp1,resultp2,newbitp1,newbitp2,sump1,sump2);
    141 		if(Dbl_isnotgreaterthan(sump1,sump2,srcp1,srcp2)) {
    142 			Dbl_leftshiftby1(newbitp1,newbitp2);
    143 			/* update result */
    144 			Dbl_addition(resultp1,resultp2,newbitp1,newbitp2,
    145 			 resultp1,resultp2);
    146 			Dbl_subtract(srcp1,srcp2,sump1,sump2,srcp1,srcp2);
    147 			Dbl_rightshiftby2(newbitp1,newbitp2);
    148 		}
    149 		else {
    150 			Dbl_rightshiftby1(newbitp1,newbitp2);
    151 		}
    152 		Dbl_leftshiftby1(srcp1,srcp2);
    153 	}
    154 	/* correct exponent for pre-shift */
    155 	if (even_exponent) {
    156 		Dbl_rightshiftby1(resultp1,resultp2);
    157 	}
    158 
    159 	/* check for inexact */
    160 	if (Dbl_isnotzero(srcp1,srcp2)) {
    161 		if (!even_exponent && Dbl_islessthan(resultp1,resultp2,srcp1,srcp2)) {
    162 			Dbl_increment(resultp1,resultp2);
    163 		}
    164 		guardbit = Dbl_lowmantissap2(resultp2);
    165 		Dbl_rightshiftby1(resultp1,resultp2);
    166 
    167 		/*  now round result  */
    168 		switch (Rounding_mode()) {
    169 		case ROUNDPLUS:
    170 		     Dbl_increment(resultp1,resultp2);
    171 		     break;
    172 		case ROUNDNEAREST:
    173 		     /* stickybit is always true, so guardbit
    174 		      * is enough to determine rounding */
    175 		     if (guardbit) {
    176 			    Dbl_increment(resultp1,resultp2);
    177 		     }
    178 		     break;
    179 		}
    180 		/* increment result exponent by 1 if mantissa overflowed */
    181 		if (Dbl_isone_hiddenoverflow(resultp1)) src_exponent+=2;
    182 
    183 		if (Is_inexacttrap_enabled()) {
    184 			Dbl_set_exponent(resultp1,
    185 			 ((src_exponent-DBL_BIAS)>>1)+DBL_BIAS);
    186 			Dbl_copytoptr(resultp1,resultp2,dstptr);
    187 			return(INEXACTEXCEPTION);
    188 		}
    189 		else Set_inexactflag();
    190 	}
    191 	else {
    192 		Dbl_rightshiftby1(resultp1,resultp2);
    193 	}
    194 	Dbl_set_exponent(resultp1,((src_exponent-DBL_BIAS)>>1)+DBL_BIAS);
    195 	Dbl_copytoptr(resultp1,resultp2,dstptr);
    196 	return(NOEXCEPTION);
    197 }
    198