xref: /netbsd-src/external/lgpl3/mpc/dist/src/fma.c (revision 367b82799ab709709d3c3b541df56a2a14644d3e)
1 /* mpc_fma -- Fused multiply-add of three complex numbers
2 
3 Copyright (C) 2011, 2012, 2022 INRIA
4 
5 This file is part of GNU MPC.
6 
7 GNU MPC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU Lesser General Public License as published by the
9 Free Software Foundation; either version 3 of the License, or (at your
10 option) any later version.
11 
12 GNU MPC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
14 FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for
15 more details.
16 
17 You should have received a copy of the GNU Lesser General Public License
18 along with this program. If not, see http://www.gnu.org/licenses/ .
19 */
20 
21 #include "mpc-impl.h"
22 
23 int
mpc_fma_naive(mpc_ptr r,mpc_srcptr a,mpc_srcptr b,mpc_srcptr c,mpc_rnd_t rnd)24 mpc_fma_naive (mpc_ptr r, mpc_srcptr a, mpc_srcptr b, mpc_srcptr c, mpc_rnd_t rnd)
25 {
26   mpfr_t rea_reb, rea_imb, ima_reb, ima_imb;
27   mpfr_ptr sum [3];
28   int inex_re, inex_im;
29 
30   mpfr_init2 (rea_reb, mpfr_get_prec (mpc_realref(a)) + mpfr_get_prec (mpc_realref(b)));
31   mpfr_init2 (rea_imb, mpfr_get_prec (mpc_realref(a)) + mpfr_get_prec (mpc_imagref(b)));
32   mpfr_init2 (ima_reb, mpfr_get_prec (mpc_imagref(a)) + mpfr_get_prec (mpc_realref(b)));
33   mpfr_init2 (ima_imb, mpfr_get_prec (mpc_imagref(a)) + mpfr_get_prec (mpc_imagref(b)));
34 
35   mpfr_mul (rea_reb, mpc_realref(a), mpc_realref(b), MPFR_RNDZ); /* exact */
36   mpfr_mul (rea_imb, mpc_realref(a), mpc_imagref(b), MPFR_RNDZ); /* exact */
37   mpfr_mul (ima_reb, mpc_imagref(a), mpc_realref(b), MPFR_RNDZ); /* exact */
38   mpfr_mul (ima_imb, mpc_imagref(a), mpc_imagref(b), MPFR_RNDZ); /* exact */
39 
40   mpfr_neg (ima_imb, ima_imb, MPFR_RNDZ);
41   sum [0] = rea_reb;
42   sum [1] = ima_imb;
43   sum [2] = (mpfr_ptr) mpc_realref (c);
44   inex_re = mpfr_sum (mpc_realref (r), sum, 3, MPC_RND_RE (rnd));
45   sum [0] = rea_imb;
46   sum [1] = ima_reb;
47   sum [2] = (mpfr_ptr) mpc_imagref (c);
48   inex_im = mpfr_sum (mpc_imagref (r), sum, 3, MPC_RND_IM (rnd));
49 
50   mpfr_clear (rea_reb);
51   mpfr_clear (rea_imb);
52   mpfr_clear (ima_reb);
53   mpfr_clear (ima_imb);
54 
55   return MPC_INEX(inex_re, inex_im);
56 }
57 
58 /* The algorithm is as follows:
59    - in a first pass, we use the target precision + some extra bits
60    - if it fails, we add the number of cancelled bits when adding
61      Re(a*b) and Re(c) [similarly for the imaginary part]
62    - it is fails again, we call the mpc_fma_naive function, which also
63      deals with the special cases */
64 int
mpc_fma(mpc_ptr r,mpc_srcptr a,mpc_srcptr b,mpc_srcptr c,mpc_rnd_t rnd)65 mpc_fma (mpc_ptr r, mpc_srcptr a, mpc_srcptr b, mpc_srcptr c, mpc_rnd_t rnd)
66 {
67   mpc_t ab;
68   mpfr_prec_t pre, pim, wpre, wpim;
69   mpfr_exp_t diffre, diffim;
70   int i, inex = 0, okre = 0, okim = 0;
71 
72   if (mpc_fin_p (a) == 0 || mpc_fin_p (b) == 0 || mpc_fin_p (c) == 0)
73     return mpc_fma_naive (r, a, b, c, rnd);
74 
75   pre = mpfr_get_prec (mpc_realref(r));
76   pim = mpfr_get_prec (mpc_imagref(r));
77   wpre = pre + mpc_ceil_log2 (pre) + 10;
78   wpim = pim + mpc_ceil_log2 (pim) + 10;
79   mpc_init3 (ab, wpre, wpim);
80   for (i = 0; i < 2; ++i)
81     {
82       mpc_mul (ab, a, b, MPC_RNDZZ);
83       if (mpfr_zero_p (mpc_realref(ab)) || mpfr_zero_p (mpc_imagref(ab)))
84         break;
85       diffre = mpfr_get_exp (mpc_realref(ab));
86       diffim = mpfr_get_exp (mpc_imagref(ab));
87       mpc_add (ab, ab, c, MPC_RNDZZ);
88       if (mpfr_zero_p (mpc_realref(ab)) || mpfr_zero_p (mpc_imagref(ab)))
89         break;
90       diffre -= mpfr_get_exp (mpc_realref(ab));
91       diffim -= mpfr_get_exp (mpc_imagref(ab));
92       diffre = (diffre > 0 ? diffre + 1 : 1);
93       diffim = (diffim > 0 ? diffim + 1 : 1);
94       okre = diffre > (mpfr_exp_t) wpre ? 0 : mpfr_can_round (mpc_realref(ab),
95                                  wpre - diffre, MPFR_RNDN, MPFR_RNDZ,
96                                  pre + (MPC_RND_RE (rnd) == MPFR_RNDN));
97       okim = diffim > (mpfr_exp_t) wpim ? 0 : mpfr_can_round (mpc_imagref(ab),
98                                  wpim - diffim, MPFR_RNDN, MPFR_RNDZ,
99                                  pim + (MPC_RND_IM (rnd) == MPFR_RNDN));
100       if (okre && okim)
101         {
102           inex = mpc_set (r, ab, rnd);
103           break;
104         }
105       if (i == 1)
106         break;
107       if (okre == 0 && diffre > 1)
108         wpre += diffre;
109       if (okim == 0 && diffim > 1)
110         wpim += diffim;
111       mpfr_set_prec (mpc_realref(ab), wpre);
112       mpfr_set_prec (mpc_imagref(ab), wpim);
113     }
114   mpc_clear (ab);
115 
116   return (okre && okim) ? inex : mpc_fma_naive (r, a, b, c, rnd);
117 }
118