LMMP 4.3.0
Lamina High-Precision Arithmetic Library
载入中...
搜索中...
未找到
add_n_sub_n.c
浏览该文件的文档.
1/**
2 * Copyright (C) 2026 HJimmyK (Jericho Knox)
3 *
4 * This file is part of LMMP.
5 *
6 * LMMP is free software: you can redistribute it and/or modify it
7 * under the terms of the GNU Lesser General Public License as
8 * published by the Free Software Foundation; either version 3 of
9 * the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library. If not, see
18 * <https://www.gnu.org/licenses/>.
19 *
20 * SPDX-License-Identifier: LGPL-3.0-or-later
21 */
22
23#include "lmmp/lmmpn.h"
24
25/*
26 汇编不可用时的通用 C 实现(eqsep 重叠安全:每组加载先于同组存储)。
27 汇编构建使用 asm/{x64,arm64}/add_n_sub_n.S。
28 差链与 sub_n_ 同样补码化: d = a + ~b + k, k0 = 1, 借位 = 1 - k,
29 借用编译器对 3 项 128 位加法的并行 add/adc 分解;~b 须先读入局部量,
30 避免被 clang 向量化破坏进位链模式匹配。
31*/
32
34 mp_size_t i = 0;
35 mp_limb_t cy = 0;
36 mp_limb_t k = 1;
37
38 for (; i + 4 <= n; i += 4) {
39 mp_limb_t a0 = numa[i + 0], a1 = numa[i + 1], a2 = numa[i + 2], a3 = numa[i + 3];
40 mp_limb_t b0 = numb[i + 0], b1 = numb[i + 1], b2 = numb[i + 2], b3 = numb[i + 3];
41 mp_limb_t nb0 = ~b0, nb1 = ~b1, nb2 = ~b2, nb3 = ~b3;
42
43 __uint128_t s0 = (__uint128_t)a0 + b0 + cy;
44 __uint128_t s1 = (__uint128_t)a1 + b1 + (mp_limb_t)(s0 >> 64);
45 __uint128_t s2 = (__uint128_t)a2 + b2 + (mp_limb_t)(s1 >> 64);
46 __uint128_t s3 = (__uint128_t)a3 + b3 + (mp_limb_t)(s2 >> 64);
47 cy = (mp_limb_t)(s3 >> 64);
48
49 __uint128_t t0 = (__uint128_t)a0 + nb0 + k;
50 __uint128_t t1 = (__uint128_t)a1 + nb1 + (mp_limb_t)(t0 >> 64);
51 __uint128_t t2 = (__uint128_t)a2 + nb2 + (mp_limb_t)(t1 >> 64);
52 __uint128_t t3 = (__uint128_t)a3 + nb3 + (mp_limb_t)(t2 >> 64);
53 k = (mp_limb_t)(t3 >> 64);
54
55 dsta[i + 0] = (mp_limb_t)s0;
56 dsta[i + 1] = (mp_limb_t)s1;
57 dsta[i + 2] = (mp_limb_t)s2;
58 dsta[i + 3] = (mp_limb_t)s3;
59 dstb[i + 0] = (mp_limb_t)t0;
60 dstb[i + 1] = (mp_limb_t)t1;
61 dstb[i + 2] = (mp_limb_t)t2;
62 dstb[i + 3] = (mp_limb_t)t3;
63 }
64 for (; i < n; i++) {
65 mp_limb_t a = numa[i], b = numb[i], nb = ~b;
66 __uint128_t s = (__uint128_t)a + b + cy;
67 __uint128_t t = (__uint128_t)a + nb + k;
68 dsta[i] = (mp_limb_t)s;
69 cy = (mp_limb_t)(s >> 64);
70 dstb[i] = (mp_limb_t)t;
71 k = (mp_limb_t)(t >> 64);
72 }
73
74 return 2 * cy + (1 - k);
75}
#define k
mp_limb_t * mp_ptr
定义 lmmp.h:96
uint64_t mp_size_t
定义 lmmp.h:93
const mp_limb_t * mp_srcptr
定义 lmmp.h:97
uint64_t mp_limb_t
定义 lmmp.h:92
mp_limb_t lmmp_add_n_sub_n_(mp_ptr dsta, mp_ptr dstb, mp_srcptr numa, mp_srcptr numb, mp_size_t n)
同时执行n位加法和减法 ([dsta,n],[dstb,n]) = ([numa,n]+[numb,n],[numa,n]-[numb,n])
#define b0
#define b1
#define a0
#define a1
#define a2
#define b2
#define t
#define a3
#define numb
#define s
#define n
#define b3