LMMP
4.3.0
Lamina High-Precision Arithmetic Library
Toggle main menu visibility
载入中...
搜索中...
未找到
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
#if defined(__SIZEOF_INT128__)
27
28
mp_limb_t
lmmp_sub_nc_
(
mp_ptr
dst,
mp_srcptr
numa,
mp_srcptr
numb
,
mp_size_t
n
,
mp_limb_t
c) {
29
/*
30
a - b - c = a + ~b + (1 - c), 借位补码化后与 add_nc_ 完全同构,
31
借用编译器对 3 项 128 位加法的并行 add/adc 分解,消除减法借位
32
提取(shr+neg)在迭代间串行链上的开销。~b 须先读入局部量:
33
表达式内直接 ~numb[i] 会被 clang 向量化(成对的 movdqu/pxor),
34
破坏进位链模式匹配,实测劣化约 2 倍。
35
*/
36
mp_size_t
i = 0;
37
mp_limb_t
cy = 1 - c;
38
39
for
(; i + 4 <=
n
; i += 4) {
40
mp_limb_t
nb0 =
~numb
[i + 0], nb1 =
~numb
[i + 1], nb2 =
~numb
[i + 2], nb3 =
~numb
[i + 3];
41
__uint128_t s0 = (__uint128_t)numa[i + 0] + nb0 + cy;
42
__uint128_t s1 = (__uint128_t)numa[i + 1] + nb1 + (
mp_limb_t
)(s0 >> 64);
43
__uint128_t s2 = (__uint128_t)numa[i + 2] + nb2 + (
mp_limb_t
)(s1 >> 64);
44
__uint128_t s3 = (__uint128_t)numa[i + 3] + nb3 + (
mp_limb_t
)(s2 >> 64);
45
dst[i + 0] = (
mp_limb_t
)s0;
46
dst[i + 1] = (
mp_limb_t
)s1;
47
dst[i + 2] = (
mp_limb_t
)s2;
48
dst[i + 3] = (
mp_limb_t
)s3;
49
cy = (
mp_limb_t
)(s3 >> 64);
50
}
51
for
(; i <
n
; i++) {
52
mp_limb_t
nb =
~numb
[i];
53
__uint128_t
s
= (__uint128_t)numa[i] + nb + cy;
54
dst[i] = (
mp_limb_t
)
s
;
55
cy = (
mp_limb_t
)(
s
>> 64);
56
}
57
58
return
1 - cy;
59
}
60
61
mp_limb_t
lmmp_sub_n_
(
mp_ptr
dst,
mp_srcptr
numa,
mp_srcptr
numb
,
mp_size_t
n
) {
62
return
lmmp_sub_nc_
(dst, numa,
numb
,
n
, 0);
63
}
64
65
#else
/* !__SIZEOF_INT128__ 朴素参考实现 */
66
67
mp_limb_t
lmmp_sub_nc_
(
mp_ptr
dst,
mp_srcptr
numa,
mp_srcptr
numb
,
mp_size_t
n
,
mp_limb_t
c) {
68
mp_size_t
i = 0;
69
mp_limb_t
cy = c;
70
71
for
(; i <
n
; i++) {
72
mp_limb_t
a, b;
73
a = numa[i];
74
b =
numb
[i];
75
b += cy;
76
cy = (b < cy);
77
cy += (a < b);
78
dst[i] = a - b;
79
}
80
81
return
cy;
82
}
83
84
mp_limb_t
lmmp_sub_n_
(
mp_ptr
dst,
mp_srcptr
numa,
mp_srcptr
numb
,
mp_size_t
n
) {
85
mp_size_t
i = 0;
86
mp_limb_t
cy = 0;
87
88
for
(; i <
n
; i++) {
89
mp_limb_t
a, b;
90
a = numa[i];
91
b =
numb
[i];
92
b += cy;
93
cy = (b < cy);
94
cy += (a < b);
95
dst[i] = a - b;
96
}
97
98
return
cy;
99
}
100
101
#endif
/* __SIZEOF_INT128__ */
mp_ptr
mp_limb_t * mp_ptr
定义
lmmp.h:96
mp_size_t
uint64_t mp_size_t
定义
lmmp.h:93
mp_srcptr
const mp_limb_t * mp_srcptr
定义
lmmp.h:97
mp_limb_t
uint64_t mp_limb_t
定义
lmmp.h:92
lmmpn.h
lmmp_sub_n_
mp_limb_t lmmp_sub_n_(mp_ptr dst, mp_srcptr numa, mp_srcptr numb, mp_size_t n)
无借位的n位减法 [dst,n] = [numa,n] - [numb,n]
定义
sub_n.c:84
lmmp_sub_nc_
mp_limb_t lmmp_sub_nc_(mp_ptr dst, mp_srcptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t c)
带借位的n位减法 [dst,n] = [numa,n] - [numb,n] - c
定义
sub_n.c:67
numb
#define numb
s
#define s
n
#define n
src
lmmp
lmmpn
arch
generic
sub_n.c
生成于
, 为 LMMP使用
1.18.0