usr/src/lib/libm/common/m9x/llrintl.c

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