1*0957b409SSimon J. Gerraty /*
2*0957b409SSimon J. Gerraty * Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
3*0957b409SSimon J. Gerraty *
4*0957b409SSimon J. Gerraty * Permission is hereby granted, free of charge, to any person obtaining
5*0957b409SSimon J. Gerraty * a copy of this software and associated documentation files (the
6*0957b409SSimon J. Gerraty * "Software"), to deal in the Software without restriction, including
7*0957b409SSimon J. Gerraty * without limitation the rights to use, copy, modify, merge, publish,
8*0957b409SSimon J. Gerraty * distribute, sublicense, and/or sell copies of the Software, and to
9*0957b409SSimon J. Gerraty * permit persons to whom the Software is furnished to do so, subject to
10*0957b409SSimon J. Gerraty * the following conditions:
11*0957b409SSimon J. Gerraty *
12*0957b409SSimon J. Gerraty * The above copyright notice and this permission notice shall be
13*0957b409SSimon J. Gerraty * included in all copies or substantial portions of the Software.
14*0957b409SSimon J. Gerraty *
15*0957b409SSimon J. Gerraty * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16*0957b409SSimon J. Gerraty * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17*0957b409SSimon J. Gerraty * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18*0957b409SSimon J. Gerraty * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19*0957b409SSimon J. Gerraty * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20*0957b409SSimon J. Gerraty * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21*0957b409SSimon J. Gerraty * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22*0957b409SSimon J. Gerraty * SOFTWARE.
23*0957b409SSimon J. Gerraty */
24*0957b409SSimon J. Gerraty
25*0957b409SSimon J. Gerraty #include "inner.h"
26*0957b409SSimon J. Gerraty
27*0957b409SSimon J. Gerraty /* see inner.h */
28*0957b409SSimon J. Gerraty void
br_i31_mulacc(uint32_t * d,const uint32_t * a,const uint32_t * b)29*0957b409SSimon J. Gerraty br_i31_mulacc(uint32_t *d, const uint32_t *a, const uint32_t *b)
30*0957b409SSimon J. Gerraty {
31*0957b409SSimon J. Gerraty size_t alen, blen, u;
32*0957b409SSimon J. Gerraty uint32_t dl, dh;
33*0957b409SSimon J. Gerraty
34*0957b409SSimon J. Gerraty alen = (a[0] + 31) >> 5;
35*0957b409SSimon J. Gerraty blen = (b[0] + 31) >> 5;
36*0957b409SSimon J. Gerraty
37*0957b409SSimon J. Gerraty /*
38*0957b409SSimon J. Gerraty * We want to add the two bit lengths, but these are encoded,
39*0957b409SSimon J. Gerraty * which requires some extra care.
40*0957b409SSimon J. Gerraty */
41*0957b409SSimon J. Gerraty dl = (a[0] & 31) + (b[0] & 31);
42*0957b409SSimon J. Gerraty dh = (a[0] >> 5) + (b[0] >> 5);
43*0957b409SSimon J. Gerraty d[0] = (dh << 5) + dl + (~(uint32_t)(dl - 31) >> 31);
44*0957b409SSimon J. Gerraty
45*0957b409SSimon J. Gerraty for (u = 0; u < blen; u ++) {
46*0957b409SSimon J. Gerraty uint32_t f;
47*0957b409SSimon J. Gerraty size_t v;
48*0957b409SSimon J. Gerraty
49*0957b409SSimon J. Gerraty /*
50*0957b409SSimon J. Gerraty * Carry always fits on 31 bits; we want to keep it in a
51*0957b409SSimon J. Gerraty * 32-bit register on 32-bit architectures (on a 64-bit
52*0957b409SSimon J. Gerraty * architecture, cast down from 64 to 32 bits means
53*0957b409SSimon J. Gerraty * clearing the high bits, which is not free; on a 32-bit
54*0957b409SSimon J. Gerraty * architecture, the same operation really means ignoring
55*0957b409SSimon J. Gerraty * the top register, which has negative or zero cost).
56*0957b409SSimon J. Gerraty */
57*0957b409SSimon J. Gerraty #if BR_64
58*0957b409SSimon J. Gerraty uint64_t cc;
59*0957b409SSimon J. Gerraty #else
60*0957b409SSimon J. Gerraty uint32_t cc;
61*0957b409SSimon J. Gerraty #endif
62*0957b409SSimon J. Gerraty
63*0957b409SSimon J. Gerraty f = b[1 + u];
64*0957b409SSimon J. Gerraty cc = 0;
65*0957b409SSimon J. Gerraty for (v = 0; v < alen; v ++) {
66*0957b409SSimon J. Gerraty uint64_t z;
67*0957b409SSimon J. Gerraty
68*0957b409SSimon J. Gerraty z = (uint64_t)d[1 + u + v] + MUL31(f, a[1 + v]) + cc;
69*0957b409SSimon J. Gerraty cc = z >> 31;
70*0957b409SSimon J. Gerraty d[1 + u + v] = (uint32_t)z & 0x7FFFFFFF;
71*0957b409SSimon J. Gerraty }
72*0957b409SSimon J. Gerraty d[1 + u + alen] = (uint32_t)cc;
73*0957b409SSimon J. Gerraty }
74*0957b409SSimon J. Gerraty }
75