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_i32_montymul(uint32_t * d,const uint32_t * x,const uint32_t * y,const uint32_t * m,uint32_t m0i)29*0957b409SSimon J. Gerraty br_i32_montymul(uint32_t *d, const uint32_t *x, const uint32_t *y,
30*0957b409SSimon J. Gerraty const uint32_t *m, uint32_t m0i)
31*0957b409SSimon J. Gerraty {
32*0957b409SSimon J. Gerraty size_t len, u, v;
33*0957b409SSimon J. Gerraty uint64_t dh;
34*0957b409SSimon J. Gerraty
35*0957b409SSimon J. Gerraty len = (m[0] + 31) >> 5;
36*0957b409SSimon J. Gerraty br_i32_zero(d, m[0]);
37*0957b409SSimon J. Gerraty dh = 0;
38*0957b409SSimon J. Gerraty for (u = 0; u < len; u ++) {
39*0957b409SSimon J. Gerraty uint32_t f, xu;
40*0957b409SSimon J. Gerraty uint64_t r1, r2, zh;
41*0957b409SSimon J. Gerraty
42*0957b409SSimon J. Gerraty xu = x[u + 1];
43*0957b409SSimon J. Gerraty f = (d[1] + x[u + 1] * y[1]) * m0i;
44*0957b409SSimon J. Gerraty r1 = 0;
45*0957b409SSimon J. Gerraty r2 = 0;
46*0957b409SSimon J. Gerraty for (v = 0; v < len; v ++) {
47*0957b409SSimon J. Gerraty uint64_t z;
48*0957b409SSimon J. Gerraty uint32_t t;
49*0957b409SSimon J. Gerraty
50*0957b409SSimon J. Gerraty z = (uint64_t)d[v + 1] + MUL(xu, y[v + 1]) + r1;
51*0957b409SSimon J. Gerraty r1 = z >> 32;
52*0957b409SSimon J. Gerraty t = (uint32_t)z;
53*0957b409SSimon J. Gerraty z = (uint64_t)t + MUL(f, m[v + 1]) + r2;
54*0957b409SSimon J. Gerraty r2 = z >> 32;
55*0957b409SSimon J. Gerraty if (v != 0) {
56*0957b409SSimon J. Gerraty d[v] = (uint32_t)z;
57*0957b409SSimon J. Gerraty }
58*0957b409SSimon J. Gerraty }
59*0957b409SSimon J. Gerraty zh = dh + r1 + r2;
60*0957b409SSimon J. Gerraty d[len] = (uint32_t)zh;
61*0957b409SSimon J. Gerraty dh = zh >> 32;
62*0957b409SSimon J. Gerraty }
63*0957b409SSimon J. Gerraty
64*0957b409SSimon J. Gerraty /*
65*0957b409SSimon J. Gerraty * d[] may still be greater than m[] at that point; notably, the
66*0957b409SSimon J. Gerraty * 'dh' word may be non-zero.
67*0957b409SSimon J. Gerraty */
68*0957b409SSimon J. Gerraty br_i32_sub(d, m, NEQ(dh, 0) | NOT(br_i32_sub(d, m, 0)));
69*0957b409SSimon J. Gerraty }
70