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 lrintl = __lrintl
  32 #endif
  33 
  34 #include <sys/isa_defs.h> /* _ILP32 */
  35 #include "libm.h"
  36 
  37 #if defined(_ILP32)
  38 #if defined(__sparc)
  39 
  40 #include "fma.h"
  41 #include "fenv_inlines.h"
  42 
  43 long
  44 lrintl(long double x) {
  45         union {
  46                 unsigned int i[4];
  47                 long double q;
  48         } xx;
  49         union {
  50                 unsigned int i;
  51                 float f;
  52         } tt;
  53         unsigned int hx, sx, frac, l, fsr;
  54         int rm, j;
  55         volatile float dummy;
  56 
  57         xx.q = x;
  58         sx = xx.i[0] & 0x80000000;
  59         hx = xx.i[0] & ~0x80000000;
  60 
  61         /* handle trivial cases */
  62         if (hx > 0x401e0000) { /* |x| > 2^31 + ... or x is nan */
  63                 /* convert an out-of-range float */
  64                 tt.i = sx | 0x7f000000;
  65                 return ((long) tt.f);
  66         } else if ((hx | xx.i[1] | xx.i[2] | xx.i[3]) == 0) /* x is zero */
  67                 return (0L);
  68 
  69         /* get the rounding mode */
  70         __fenv_getfsr32(&fsr);
  71         rm = fsr >> 30;
  72 
  73         /* flip the sense of directed roundings if x is negative */
  74         if (sx)
  75                 rm ^= rm >> 1;
  76 
  77         /* handle |x| < 1 */
  78         if (hx < 0x3fff0000) {
  79                 dummy = 1.0e30F; /* x is nonzero, so raise inexact */
  80                 dummy += 1.0e-30F;
  81                 if (rm == FSR_RP || (rm == FSR_RN && (hx >= 0x3ffe0000 &&
  82                         ((hx & 0xffff) | xx.i[1] | xx.i[2] | xx.i[3]))))
  83                         return (sx ? -1L : 1L);
  84                 return (0L);
  85         }
  86 
  87         /* extract the integer and fractional parts of x */
  88         j = 0x406f - (hx >> 16);          /* 91 <= j <= 112 */
  89         xx.i[0] = 0x10000 | (xx.i[0] & 0xffff);
  90         if (j >= 96) {                               /* 96 <= j <= 112 */
  91                 l = xx.i[0] >> (j - 96);
  92                 frac = ((xx.i[0] << 1) << (127 - j)) | (xx.i[1] >> (j - 96));
  93                 if (((xx.i[1] << 1) << (127 - j)) | xx.i[2] | xx.i[3])
  94                         frac |= 1;
  95         } else {                                /* 91 <= j <= 95 */
  96                 l = (xx.i[0] << (96 - j)) | (xx.i[1] >> (j - 64));
  97                 frac = (xx.i[1] << (96 - j)) | (xx.i[2] >> (j - 64));
  98                 if ((xx.i[2] << (96 - j)) | xx.i[3])
  99                         frac |= 1;
 100         }
 101 
 102         /* round */
 103         if (frac && (rm == FSR_RP || (rm == FSR_RN && (frac > 0x80000000U ||
 104                 (frac == 0x80000000 && (l & 1))))))
 105                 l++;
 106 
 107         /* check for result out of range (note that z is |x| at this point) */
 108         if (l > 0x80000000U || (l == 0x80000000U && !sx)) {
 109                 tt.i = sx | 0x7f000000;
 110                 return ((long) tt.f);
 111         }
 112 
 113         /* raise inexact if need be */
 114         if (frac) {
 115                 dummy = 1.0e30F;
 116                 dummy += 1.0e-30F;
 117         }
 118 
 119         /* negate result if need be */
 120         if (sx)
 121                 l = -l;
 122         return ((long) l);
 123 }
 124 #elif defined(__x86)
 125 long
 126 lrintl(long double x) {
 127         /*
 128          * Note: The following code works on x86 (in the default rounding
 129          * precision mode), but one ought to just use the fistpl instruction
 130          * instead.
 131          */
 132         union {
 133                 unsigned i[3];
 134                 long double e;
 135         } xx, yy;
 136         int ex;
 137 
 138         xx.e = x;
 139         ex = xx.i[2] & 0x7fff;
 140         if (ex < 0x403e) {   /* |x| < 2^63 */
 141                 /* add and subtract a power of two to round x to an integer */
 142                 yy.i[2] = (xx.i[2] & 0x8000) | 0x403e;
 143                 yy.i[1] = 0x80000000;
 144                 yy.i[0] = 0;
 145                 x = (x + yy.e) - yy.e;
 146         }
 147 
 148         /* now x is nan, inf, or integral */
 149         return ((long) x);
 150 }
 151 #else
 152 #error Unknown architecture
 153 #endif  /* defined(__sparc) || defined(__x86) */
 154 #else
 155 #error Unsupported architecture
 156 #endif  /* defined(_ILP32) */