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 llrintl = __llrintl
  32 #if defined(__sparcv9) || defined(__amd64)
  33 #pragma weak lrintl = __llrintl
  34 #pragma weak __lrintl = __llrintl
  35 #endif
  36 #endif
  37 
  38 #include "libm.h"
  39 
  40 #if defined(__sparc)
  41 
  42 #include "fma.h"
  43 #include "fenv_inlines.h"
  44 
  45 long long
  46 llrintl(long double x) {
  47         union {
  48                 unsigned i[4];
  49                 long double q;
  50         } xx;
  51         union {
  52                 unsigned i[2];
  53                 long long l;
  54         } zz;
  55         union {
  56                 unsigned i;
  57                 float f;
  58         } tt;
  59         unsigned int hx, sx, frac;
  60         unsigned int fsr;
  61         int rm, j;
  62         volatile float dummy;
  63 
  64         xx.q = x;
  65         sx = xx.i[0] & 0x80000000;
  66         hx = xx.i[0] & ~0x80000000;
  67 
  68         /* handle trivial cases */
  69         if (hx > 0x403e0000) { /* |x| > 2^63 + ... or x is nan */
  70                 /* convert an out-of-range float */
  71                 tt.i = sx | 0x7f000000;
  72                 return ((long long) tt.f);
  73         } else if ((hx | xx.i[1] | xx.i[2] | xx.i[3]) == 0) /* x is zero */
  74                 return (0LL);
  75 
  76         /* get the rounding mode */
  77         __fenv_getfsr32(&fsr);
  78         rm = fsr >> 30;
  79 
  80         /* flip the sense of directed roundings if x is negative */
  81         if (sx)
  82                 rm ^= rm >> 1;
  83 
  84         /* handle |x| < 1 */
  85         if (hx < 0x3fff0000) {
  86                 dummy = 1.0e30f; /* x is nonzero, so raise inexact */
  87                 dummy += 1.0e-30f;
  88                 if (rm == FSR_RP || (rm == FSR_RN && (hx >= 0x3ffe0000 &&
  89                         ((hx & 0xffff) | xx.i[1] | xx.i[2] | xx.i[3]))))
  90                         return (sx ? -1LL : 1LL);
  91                 return (0LL);
  92         }
  93 
  94         /* extract the integer and fractional parts of x */
  95         j = 0x406f - (hx >> 16);
  96         xx.i[0] = 0x10000 | (xx.i[0] & 0xffff);
  97         if (j >= 96) {
  98                 zz.i[0] = 0;
  99                 zz.i[1] = xx.i[0] >> (j - 96);
 100                 frac = ((xx.i[0] << 1) << (127 - j)) | (xx.i[1] >> (j - 96));
 101                 if (((xx.i[1] << 1) << (127 - j)) | xx.i[2] | xx.i[3])
 102                         frac |= 1;
 103         } else if (j >= 64) {
 104                 zz.i[0] = xx.i[0] >> (j - 64);
 105                 zz.i[1] = ((xx.i[0] << 1) << (95 - j)) | (xx.i[1] >> (j - 64));
 106                 frac = ((xx.i[1] << 1) << (95 - j)) | (xx.i[2] >> (j - 64));
 107                 if (((xx.i[2] << 1) << (95 - j)) | xx.i[3])
 108                         frac |= 1;
 109         } else {
 110                 zz.i[0] = ((xx.i[0] << 1) << (63 - j)) | (xx.i[1] >> (j - 32));
 111                 zz.i[1] = ((xx.i[1] << 1) << (63 - j)) | (xx.i[2] >> (j - 32));
 112                 frac = ((xx.i[2] << 1) << (63 - j)) | (xx.i[3] >> (j - 32));
 113                 if ((xx.i[3] << 1) << (63 - j))
 114                         frac |= 1;
 115         }
 116 
 117         /* round */
 118         if (frac && (rm == FSR_RP || (rm == FSR_RN && (frac > 0x80000000u ||
 119                 (frac == 0x80000000 && (zz.i[1] & 1)))))) {
 120                 if (++zz.i[1] == 0)
 121                         zz.i[0]++;
 122         }
 123 
 124         /* check for result out of range (note that z is |x| at this point) */
 125         if (zz.i[0] > 0x80000000u || (zz.i[0] == 0x80000000 && (zz.i[1] ||
 126                 !sx))) {
 127                 tt.i = sx | 0x7f000000;
 128                 return ((long long) tt.f);
 129         }
 130 
 131         /* raise inexact if need be */
 132         if (frac) {
 133                 dummy = 1.0e30F;
 134                 dummy += 1.0e-30F;
 135         }
 136 
 137         /* negate result if need be */
 138         if (sx) {
 139                 zz.i[0] = ~zz.i[0];
 140                 zz.i[1] = -zz.i[1];
 141                 if (zz.i[1] == 0)
 142                         zz.i[0]++;
 143         }
 144         return (zz.l);
 145 }
 146 #elif defined(__x86)
 147 long long
 148 llrintl(long double x) {
 149         /*
 150          * Note: The following code works on x86 (in the default rounding
 151          * precision mode), but one ought to just use the fistpll instruction
 152          * instead.
 153          */
 154         union {
 155                 unsigned i[3];
 156                 long double e;
 157         } xx, yy;
 158         int ex;
 159 
 160         xx.e = x;
 161         ex = xx.i[2] & 0x7fff;
 162 
 163         if (ex < 0x403e) { /* |x| < 2^63 */
 164                 /* add and subtract a power of two to round x to an integer */
 165                 yy.i[2] = (xx.i[2] & 0x8000) | 0x403e;
 166                 yy.i[1] = 0x80000000;
 167                 yy.i[0] = 0;
 168                 x = (x + yy.e) - yy.e;
 169         }
 170 
 171         /* now x is nan, inf, or integral */
 172         return ((long long) x);
 173 }
 174 #else
 175 #error Unknown architecture
 176 #endif