summaryrefslogtreecommitdiff
path: root/set_d.c
blob: 1bd0465f21eefee39c8f4607aed70ec8eec56257 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
#include <math.h> /* for isnan */
#include "gmp.h"
#include "gmp-impl.h"
#include "longlong.h"
#include "mpfr.h"

#define NaN sqrt(-1) /* ensures a machine-independent NaN */

/* Included from gmp-2.0.2, patched to support denorms */

#ifdef XDEBUG
#undef _GMP_IEEE_FLOATS
#endif

#ifndef _GMP_IEEE_FLOATS
#define _GMP_IEEE_FLOATS 0
#endif

#define MP_BASE_AS_DOUBLE (2.0 * ((mp_limb_t) 1 << (BITS_PER_MP_LIMB - 1)))
int
#if __STDC__
__mpfr_extract_double (mp_ptr rp, double d, int e)
#else
__mpfr_extract_double (rp, d)
     mp_ptr rp;
     double d;
     int e;
#endif
     /* e=0 iff rp has only one limb */
{
  long exp;
  mp_limb_t manh, manl;

  /* BUGS

     1. Should handle Inf and NaN in IEEE specific code.
     2. Handle Inf and NaN also in default code, to avoid hangs.
     3. Generalize to handle all BITS_PER_MP_LIMB >= 32.
     4. This lits is incomplete and misspelled.
   */

  if (d == 0.0)
    {
      rp[0] = 0;
#if BITS_PER_MP_LIMB == 32
      if (e) rp[1] = 0;
#endif
      return 0;
    }

#if _GMP_IEEE_FLOATS
  {
    union ieee_double_extract x;
    x.d = d;

    exp = x.s.exp;
    if (exp) 
      {
#if BITS_PER_MP_LIMB == 64
	manl = (((mp_limb_t) 1 << 63)
		| ((mp_limb_t) x.s.manh << 43) | ((mp_limb_t) x.s.manl << 11));
#else
	manh = ((mp_limb_t) 1 << 31) | (x.s.manh << 11) | (x.s.manl >> 21);
	manl = x.s.manl << 11;      
#endif
      }
    else
      {
#if BITS_PER_MP_LIMB == 64
	manl = ((mp_limb_t) x.s.manh << 43) | ((mp_limb_t) x.s.manl << 11);
#else
    manh = (x.s.manh << 11) | (x.s.manl >> 21);
	manl = x.s.manl << 11;      
#endif
      }
  }
#else
  {
    /* Unknown (or known to be non-IEEE) double format.  */
    exp = 0;
    if (d >= 1.0)
      {
        if (d * 0.5 == d)
          abort ();

        while (d >= 32768.0)
          {
            d *= (1.0 / 65536.0);
            exp += 16;
          }
        while (d >= 1.0)
          {
            d *= 0.5;
            exp += 1;
          }
      }
    else if (d < 0.5)
      {
        while (d < (1.0 / 65536.0))
          {
            d *=  65536.0;
            exp -= 16;
          }
        while (d < 0.5)
          {
            d *= 2.0;
            exp -= 1;
          }
      }

    d *= MP_BASE_AS_DOUBLE;
#if BITS_PER_MP_LIMB == 64
    manl = d;
#else
    manh = d;
    manl = (d - manh) * MP_BASE_AS_DOUBLE;
#endif

    exp += 1022;
  }
#endif

  if (exp) exp = (unsigned) exp - 1022; else exp = -1021; 

#if BITS_PER_MP_LIMB == 64
      rp[0] = manl;
#else
      if (e) {
	rp[1] = manh;
	rp[0] = manl;
      }
      else {
	rp[0] = manh;
      }
#endif

  return exp;
}

/* End of part included from gmp-2.0.2 */
/* Part included from gmp temporary releases */
double
#if __STDC__
__mpfr_scale2 (double d, int exp)
#else
__mpfr_scale2 (d, exp)
     double d;
     int exp;
#endif
{
#if _GMP_IEEE_FLOATS
  {
    union ieee_double_extract x;
    x.d = d;
    exp += x.s.exp;
    x.s.exp = exp;
    if (exp >= 2047)
      {
        /* Return +-infinity */
        x.s.exp = 2047;
        x.s.manl = x.s.manh = 0;
      }
    else if (exp < 1)
      {
        x.s.exp = 1;            /* smallest exponent (biased) */
        /* Divide result by 2 until we have scaled it to the right IEEE
           denormalized number, but stop if it becomes zero.  */
        while (exp < 1 && x.d != 0)
          {
            x.d *= 0.5;
            exp++;
          }
      }
    return x.d;
  }
#else
  {
    double factor, r;

    factor = 2.0;
    if (exp < 0)
      {
        factor = 0.5;
        exp = -exp;
      }
    r = d;
    if (exp != 0)
      {
        if ((exp & 1) != 0)
          r *= factor;
        exp >>= 1;
        while (exp != 0)
          {
            factor *= factor;
            if ((exp & 1) != 0)
              r *= factor;
            exp >>= 1;
          }
      }
    return r;
  }
#endif
}


/* End of part included from gmp */

void
mpfr_set_d(mpfr_t r, double d, unsigned char rnd_mode)
{
  int negative, sizer; unsigned int cnt;

  if (d == 0)
    {      
      EXP(r) = 0;
      return;
    }
  else if (isnan(d)) { SET_NAN(r); return; }

  negative = d < 0;
  d = ABS (d);

  sizer = MPFR_LIMBS_PER_DOUBLE; if (ABSSIZE(r)<sizer) sizer=ABSSIZE(r);
  /* warning: __mpfr_extract_double requires at least two limbs */
  EXP(r) = __mpfr_extract_double (MANT(r), d, (sizer>=2) );
  
  count_leading_zeros(cnt, MANT(r)[sizer-1]);
  if (cnt) mpn_lshift(MANT(r), MANT(r), sizer, cnt); 
  
  EXP(r) -= cnt; 
  SIZE(r) = sizer; if (negative) CHANGE_SIGN(r);

  mpfr_round(r, rnd_mode, PREC(r)); 
  return; 
}

double
mpfr_get_d2(mpfr_srcptr src, long e)
{
  double res;
  mp_size_t size, i, n_limbs_to_use;
  mp_ptr qp;
  int negative;

  if (FLAG_NAN(src)) { 
#ifdef DEBUG
    printf("recognized NaN\n");
#endif
    return NaN; }
  if (NOTZERO(src)==0) return 0.0;
  size = 1+(PREC(src)-1)/BITS_PER_MP_LIMB;
  qp = MANT(src);
  negative = (SIGN(src)==-1);

  /* Warning: don't compute the abs(res) and set the sign afterwards,
     otherwise the current machine rounding mode will not be taken
     correctly into account. */
  /* res = (negative) ? -(double)qp[size - 1] : qp[size - 1]; */
  res = 0.0;
  /* Warning: an arbitrary number of limbs may be required for an exact 
     rounding. The following code is correct but not optimal since one
     may be able to decide without considering all limbs. */
  /* n_limbs_to_use = MIN (MPFR_LIMBS_PER_DOUBLE, size); */
  n_limbs_to_use = size;
  /* Accumulate the limbs from less significant to most significant
     otherwise due to rounding we may accumulate several ulps,
     especially in rounding towards -/+infinity. */
  for (i = n_limbs_to_use; i>=1; i--)
    res = res / MP_BASE_AS_DOUBLE +
      ((negative) ? -(double)qp[size - i] : qp[size - i]);
  res = __mpfr_scale2 (res, e - BITS_PER_MP_LIMB); 

  return res;
}

double 
mpfr_get_d(mpfr_srcptr src)
{
  mpfr_get_d2(src, EXP(src));
}