summaryrefslogtreecommitdiff
path: root/sysdeps/powerpc/power4/fpu/mpa.c
blob: d741de63dd623e59ceeedabf3720b21ee592f506 (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

/*
 * IBM Accurate Mathematical Library
 * written by International Business Machines Corp.
 * Copyright (C) 2001-2014 Free Software Foundation, Inc.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published by
 * the Free Software Foundation; either version 2.1 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public License
 * along with this program; if not, see <http://www.gnu.org/licenses/>.
 */

/* Define __mul and __sqr and use the rest from generic code.  */
#define NO__MUL
#define NO__SQR

#include <sysdeps/ieee754/dbl-64/mpa.c>

/* Multiply *X and *Y and store result in *Z.  X and Y may overlap but not X
   and Z or Y and Z.  For P in [1, 2, 3], the exact result is truncated to P
   digits.  In case P > 3 the error is bounded by 1.001 ULP.  */
void
__mul (const mp_no *x, const mp_no *y, mp_no *z, int p)
{
  long i, i1, i2, j, k, k2;
  long p2 = p;
  double u, zk, zk2;

  /* Is z=0?  */
  if (__glibc_unlikely (X[0] * Y[0] == 0))
    {
      Z[0] = 0;
      return;
    }

  /* Multiply, add and carry */
  k2 = (p2 < 3) ? p2 + p2 : p2 + 3;
  zk = Z[k2] = 0;
  for (k = k2; k > 1;)
    {
      if (k > p2)
	{
	  i1 = k - p2;
	  i2 = p2 + 1;
	}
      else
	{
	  i1 = 1;
	  i2 = k;
	}
#if 1
      /* Rearrange this inner loop to allow the fmadd instructions to be
         independent and execute in parallel on processors that have
         dual symmetrical FP pipelines.  */
      if (i1 < (i2 - 1))
	{
	  /* Make sure we have at least 2 iterations.  */
	  if (((i2 - i1) & 1L) == 1L)
	    {
	      /* Handle the odd iterations case.  */
	      zk2 = x->d[i2 - 1] * y->d[i1];
	    }
	  else
	    zk2 = 0.0;
	  /* Do two multiply/adds per loop iteration, using independent
	     accumulators; zk and zk2.  */
	  for (i = i1, j = i2 - 1; i < i2 - 1; i += 2, j -= 2)
	    {
	      zk += x->d[i] * y->d[j];
	      zk2 += x->d[i + 1] * y->d[j - 1];
	    }
	  zk += zk2;		/* Final sum.  */
	}
      else
	{
	  /* Special case when iterations is 1.  */
	  zk += x->d[i1] * y->d[i1];
	}
#else
      /* The original code.  */
      for (i = i1, j = i2 - 1; i < i2; i++, j--)
	zk += X[i] * Y[j];
#endif

      u = (zk + CUTTER) - CUTTER;
      if (u > zk)
	u -= RADIX;
      Z[k] = zk - u;
      zk = u * RADIXI;
      --k;
    }
  Z[k] = zk;

  int e = EX + EY;
  /* Is there a carry beyond the most significant digit?  */
  if (Z[1] == 0)
    {
      for (i = 1; i <= p2; i++)
	Z[i] = Z[i + 1];
      e--;
    }

  EZ = e;
  Z[0] = X[0] * Y[0];
}

/* Square *X and store result in *Y.  X and Y may not overlap.  For P in
   [1, 2, 3], the exact result is truncated to P digits.  In case P > 3 the
   error is bounded by 1.001 ULP.  This is a faster special case of
   multiplication.  */
void
__sqr (const mp_no *x, mp_no *y, int p)
{
  long i, j, k, ip;
  double u, yk;

  /* Is z=0?  */
  if (__glibc_unlikely (X[0] == 0))
    {
      Y[0] = 0;
      return;
    }

  /* We need not iterate through all X's since it's pointless to
     multiply zeroes.  */
  for (ip = p; ip > 0; ip--)
    if (X[ip] != 0)
      break;

  k = (__glibc_unlikely (p < 3)) ? p + p : p + 3;

  while (k > 2 * ip + 1)
    Y[k--] = 0;

  yk = 0;

  while (k > p)
    {
      double yk2 = 0.0;
      long lim = k / 2;

      if (k % 2 == 0)
        {
	  yk += X[lim] * X[lim];
	  lim--;
	}

      /* In __mul, this loop (and the one within the next while loop) run
         between a range to calculate the mantissa as follows:

         Z[k] = X[k] * Y[n] + X[k+1] * Y[n-1] ... + X[n-1] * Y[k+1]
		+ X[n] * Y[k]

         For X == Y, we can get away with summing halfway and doubling the
	 result.  For cases where the range size is even, the mid-point needs
	 to be added separately (above).  */
      for (i = k - p, j = p; i <= lim; i++, j--)
	yk2 += X[i] * X[j];

      yk += 2.0 * yk2;

      u = (yk + CUTTER) - CUTTER;
      if (u > yk)
	u -= RADIX;
      Y[k--] = yk - u;
      yk = u * RADIXI;
    }

  while (k > 1)
    {
      double yk2 = 0.0;
      long lim = k / 2;

      if (k % 2 == 0)
        {
	  yk += X[lim] * X[lim];
	  lim--;
	}

      /* Likewise for this loop.  */
      for (i = 1, j = k - 1; i <= lim; i++, j--)
	yk2 += X[i] * X[j];

      yk += 2.0 * yk2;

      u = (yk + CUTTER) - CUTTER;
      if (u > yk)
	u -= RADIX;
      Y[k--] = yk - u;
      yk = u * RADIXI;
    }
  Y[k] = yk;

  /* Squares are always positive.  */
  Y[0] = 1.0;

  int e = EX * 2;
  /* Is there a carry beyond the most significant digit?  */
  if (__glibc_unlikely (Y[1] == 0))
    {
      for (i = 1; i <= p; i++)
	Y[i] = Y[i + 1];
      e--;
    }
  EY = e;
}