1 /*
2 Copyright (C) 2021 Fredrik Johansson
3
4 This file is part of Arb.
5
6 Arb is free software: you can redistribute it and/or modify it under
7 the terms of the GNU Lesser General Public License (LGPL) as published
8 by the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version. See <http://www.gnu.org/licenses/>.
10 */
11
12 #include "arb.h"
13
14 static void
arf_shallow_set_uiui(arf_t res,ulong vhi,ulong vlo)15 arf_shallow_set_uiui(arf_t res, ulong vhi, ulong vlo)
16 {
17 unsigned int bc;
18
19 if (vhi == 0)
20 {
21 if (vlo == 0)
22 {
23 ARF_XSIZE(res) = 0;
24 ARF_EXP(res) = ARF_EXP_ZERO;
25 }
26 else
27 {
28 count_leading_zeros(bc, vlo);
29 ARF_EXP(res) = FLINT_BITS - bc;
30 ARF_NOPTR_D(res)[0] = vlo << bc;
31 ARF_XSIZE(res) = ARF_MAKE_XSIZE(1, 0);
32 }
33 }
34 else if (vlo == 0)
35 {
36 count_leading_zeros(bc, vhi);
37 ARF_EXP(res) = 2 * FLINT_BITS - bc;
38 ARF_NOPTR_D(res)[0] = vhi << bc;
39 ARF_XSIZE(res) = ARF_MAKE_XSIZE(1, 0);
40 }
41 else
42 {
43 count_leading_zeros(bc, vhi);
44 ARF_EXP(res) = 2 * FLINT_BITS - bc;
45 ARF_NOPTR_D(res)[0] = vlo << bc;
46 if (bc == 0)
47 ARF_NOPTR_D(res)[1] = vhi;
48 else
49 ARF_NOPTR_D(res)[1] = (vhi << bc) | (vlo >> (FLINT_BITS - bc));
50 ARF_XSIZE(res) = ARF_MAKE_XSIZE(2, 0);
51 }
52 }
53
54 void
arb_dot_uiui(arb_t res,const arb_t initial,int subtract,arb_srcptr x,slong xstep,const ulong * y,slong ystep,slong len,slong prec)55 arb_dot_uiui(arb_t res, const arb_t initial, int subtract, arb_srcptr x, slong xstep, const ulong * y, slong ystep, slong len, slong prec)
56 {
57 arb_ptr t;
58 slong i;
59 ulong vhi, vlo;
60 TMP_INIT;
61
62 /* todo: fast fma and fmma (len=2) code */
63 if (len <= 1)
64 {
65 if (initial == NULL)
66 {
67 if (len <= 0)
68 arb_zero(res);
69 else
70 {
71 arf_t t;
72 arf_shallow_set_uiui(t, y[1], y[0]);
73 arb_mul_arf(res, x, t, prec);
74 if (subtract)
75 arb_neg(res, res);
76 }
77 return;
78 }
79 else if (len <= 0)
80 {
81 arb_set_round(res, initial, prec);
82 return;
83 }
84 }
85
86 TMP_START;
87 t = TMP_ALLOC(sizeof(arb_struct) * len);
88
89 for (i = 0; i < len; i++)
90 {
91 vlo = y[2 * i * ystep];
92 vhi = y[2 * i * ystep + 1];
93
94 arf_shallow_set_uiui(arb_midref(t + i), vhi, vlo);
95
96 MAG_EXP(arb_radref(t + i)) = 0;
97 MAG_MAN(arb_radref(t + i)) = 0;
98 }
99
100 arb_dot(res, initial, subtract, x, xstep, t, 1, len, prec);
101
102 TMP_END;
103 }
104
105