1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  24  */
  25 /*
  26  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  27  * Use is subject to license terms.
  28  */
  29 
  30 #if defined(ELFOBJ)
  31 #pragma weak nearbyint = __nearbyint
  32 #endif
  33 
  34 /*
  35  * nearbyint(x) returns the nearest fp integer to x in the direction
  36  * corresponding to the current rounding direction without raising
  37  * the inexact exception.
  38  *
  39  * nearbyint(x) is x unchanged if x is +/-0 or +/-inf.  If x is NaN,
  40  * nearbyint(x) is also NaN.
  41  */
  42 
  43 #include "libm.h"
  44 #include "fenv_synonyms.h"
  45 #include <fenv.h>
  46 
  47 double
  48 __nearbyint(double x) {
  49         union {
  50                 unsigned i[2];
  51                 double d;
  52         } xx;
  53         unsigned hx, sx, i, frac;
  54         int rm, j;
  55 
  56         xx.d = x;
  57         sx = xx.i[HIWORD] & 0x80000000;
  58         hx = xx.i[HIWORD] & ~0x80000000;
  59 
  60         /* handle trivial cases */
  61         if (hx >= 0x43300000) {      /* x is nan, inf, or already integral */
  62                 if (hx >= 0x7ff00000)        /* x is inf or nan */
  63 #if defined(FPADD_TRAPS_INCOMPLETE_ON_NAN)
  64                         return (hx >= 0x7ff80000 ? x : x + x);
  65                         /* assumes sparc-like QNaN */
  66 #else
  67                         return (x + x);
  68 #endif
  69                 return (x);
  70         } else if ((hx | xx.i[LOWORD]) == 0)    /* x is zero */
  71                 return (x);
  72 
  73         /* get the rounding mode */
  74         rm = fegetround();
  75 
  76         /* flip the sense of directed roundings if x is negative */
  77         if (sx && (rm == FE_UPWARD || rm == FE_DOWNWARD))
  78                 rm = (FE_UPWARD + FE_DOWNWARD) - rm;
  79 
  80         /* handle |x| < 1 */
  81         if (hx < 0x3ff00000) {
  82                 if (rm == FE_UPWARD || (rm == FE_TONEAREST &&
  83                         (hx >= 0x3fe00000 && ((hx & 0xfffff) | xx.i[LOWORD]))))
  84                         xx.i[HIWORD] = sx | 0x3ff00000;
  85                 else
  86                         xx.i[HIWORD] = sx;
  87                 xx.i[LOWORD] = 0;
  88                 return (xx.d);
  89         }
  90 
  91         /* round x at the integer bit */
  92         j = 0x433 - (hx >> 20);
  93         if (j >= 32) {
  94                 i = 1 << (j - 32);
  95                 frac = ((xx.i[HIWORD] << 1) << (63 - j)) |
  96                         (xx.i[LOWORD] >> (j - 32));
  97                 if (xx.i[LOWORD] & (i - 1))
  98                         frac |= 1;
  99                 if (!frac)
 100                         return (x);
 101                 xx.i[LOWORD] = 0;
 102                 xx.i[HIWORD] &= ~(i - 1);
 103                 if (rm == FE_UPWARD || (rm == FE_TONEAREST &&
 104                         (frac > 0x80000000u || (frac == 0x80000000) &&
 105                         (xx.i[HIWORD] & i))))
 106                         xx.i[HIWORD] += i;
 107         } else {
 108                 i = 1 << j;
 109                 frac = (xx.i[LOWORD] << 1) << (31 - j);
 110                 if (!frac)
 111                         return (x);
 112                 xx.i[LOWORD] &= ~(i - 1);
 113                 if (rm == FE_UPWARD || (rm == FE_TONEAREST &&
 114                         (frac > 0x80000000u || (frac == 0x80000000) &&
 115                         (xx.i[LOWORD] & i)))) {
 116                         xx.i[LOWORD] += i;
 117                         if (xx.i[LOWORD] == 0)
 118                                 xx.i[HIWORD]++;
 119                 }
 120         }
 121         return (xx.d);
 122 }
 123 
 124 #if 0
 125 
 126 /*
 127 *  Alternate implementations for SPARC, x86, using fp ops.  These may
 128 *  be faster depending on how expensive saving and restoring the fp
 129 *  modes and status flags is.
 130 */
 131 
 132 #include "libm.h"
 133 #include "fma.h"
 134 
 135 #if defined(__sparc)
 136 
 137 double
 138 __nearbyint(double x) {
 139         union {
 140                 unsigned i[2];
 141                 double d;
 142         } xx, yy;
 143         double z;
 144         unsigned hx, sx, fsr, oldfsr;
 145         int rm;
 146 
 147         xx.d = x;
 148         sx = xx.i[0] & 0x80000000;
 149         hx = xx.i[0] & ~0x80000000;
 150 
 151         /* handle trivial cases */
 152         if (hx >= 0x43300000)        /* x is nan, inf, or already integral */
 153                 return (x + 0.0);
 154         else if ((hx | xx.i[1]) == 0)   /* x is zero */
 155                 return (x);
 156 
 157         /* save the fsr */
 158         __fenv_getfsr(&oldfsr);
 159 
 160         /* handle |x| < 1 */
 161         if (hx < 0x3ff00000) {
 162                 /* flip the sense of directed roundings if x is negative */
 163                 rm = oldfsr >> 30;
 164                 if (sx)
 165                         rm ^= rm >> 1;
 166                 if (rm == FSR_RP || (rm == FSR_RN && (hx >= 0x3fe00000 &&
 167                         ((hx & 0xfffff) | xx.i[1]))))
 168                         xx.i[0] = sx | 0x3ff00000;
 169                 else
 170                         xx.i[0] = sx;
 171                 xx.i[1] = 0;
 172                 return (xx.d);
 173         }
 174 
 175         /* clear the inexact trap */
 176         fsr = oldfsr & ~FSR_NXM;
 177         __fenv_setfsr(&fsr);
 178 
 179         /* round x at the integer bit */
 180         yy.i[0] = sx | 0x43300000;
 181         yy.i[1] = 0;
 182         z = (x + yy.d) - yy.d;
 183 
 184         /* restore the old fsr */
 185         __fenv_setfsr(&oldfsr);
 186 
 187         return (z);
 188 }
 189 
 190 #elif defined(__x86)
 191 
 192 /* inline template */
 193 extern long double frndint(long double);
 194 
 195 double
 196 __nearbyint(double x) {
 197         long double z;
 198         unsigned oldcwsw, cwsw;
 199 
 200         /* save the control and status words, mask the inexact exception */
 201         __fenv_getcwsw(&oldcwsw);
 202         cwsw = oldcwsw | 0x00200000;
 203         __fenv_setcwsw(&cwsw);
 204 
 205         z = frndint((long double) x);
 206 
 207         /*
 208          * restore the control and status words, preserving all but the
 209          * inexact flag
 210          */
 211         __fenv_getcwsw(&cwsw);
 212         oldcwsw |= (cwsw & 0x1f);
 213         __fenv_setcwsw(&oldcwsw);
 214 
 215         /* note: the value of z is representable in double precision */
 216         return (z);
 217 }
 218 
 219 #else
 220 #error Unknown architecture
 221 #endif
 222 
 223 #endif