LMMP 4.3.0
Lamina High-Precision Arithmetic Library
载入中...
搜索中...
未找到
mul_1.c 文件参考
#include "impl/longlong.h"
#include "lmmp/lmmpn.h"
mul_1.c 的引用(Include)关系图:

浏览该文件的源代码.

宏定义

#define LMMP_SUBMUL_1_STEP(k)

函数

mp_limb_t lmmp_addmul_1_ (mp_ptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t b)
 乘以单limb并累加操作 [numa,n] += [numb,n] * b
mp_limb_t lmmp_addmul_2_ (mp_ptr numa, mp_srcptr numb, mp_size_t n, mp_srcptr bp)
 乘以双limb并累加 [numa,n] += [numb,n] * [bp,2]
mp_limb_t lmmp_mul_1_ (mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_limb_t x)
 Copyright (C) 2026 HJimmyK (Jericho Knox)
void lmmp_mullo_basecase_ (mp_ptr restrict dst, mp_srcptr restrict numa, mp_srcptr restrict numb, mp_size_t n)
void lmmp_sqrlo_basecase_ (mp_ptr restrict dst, mp_srcptr restrict numa, mp_size_t n)
mp_limb_t lmmp_submul_1_ (mp_ptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t b)
 乘以单limb并累减操作 [numa,n] -= [numb,n] * b

宏定义说明

◆ LMMP_SUBMUL_1_STEP

#define LMMP_SUBMUL_1_STEP ( k)
值:
{ \
__uint128_t P = (__uint128_t)numb[i + (k)] * b; \
mp_limb_t o = numa[i + (k)]; \
mp_limb_t u = o - (mp_limb_t)P; \
mp_limb_t r = u - d; \
d = (mp_limb_t)(P >> 64) + (u > o) + (r > u); \
numa[i + (k)] = r; \
}
#define k
uint64_t mp_limb_t
定义 lmmp.h:92
#define numb

被这些函数引用 lmmp_submul_1_().

函数说明

◆ lmmp_addmul_1_()

mp_limb_t lmmp_addmul_1_ ( mp_ptr numa,
mp_srcptr numb,
mp_size_t n,
mp_limb_t b )

乘以单limb并累加操作 [numa,n] += [numb,n] * b

参数
numa被加数指针(结果也存储在此)
numb乘数指针
nlimb长度
b乘数
警告
n>0, eqsep(numa,numb)), numa!=NULL, numb!=NULL
返回
运算后的进位limb值

在文件 mul_1.c 第 76 行定义.

76 {
77 mp_limb_t cl = 0;
78 mp_size_t i = 0;
79
80 for (; i + 4 <= n; i += 4) {
81 __uint128_t a0 = (__uint128_t)numb[i + 0] * b + numa[i + 0] + cl;
82 __uint128_t a1 = (__uint128_t)numb[i + 1] * b + numa[i + 1] + (mp_limb_t)(a0 >> 64);
83 __uint128_t a2 = (__uint128_t)numb[i + 2] * b + numa[i + 2] + (mp_limb_t)(a1 >> 64);
84 __uint128_t a3 = (__uint128_t)numb[i + 3] * b + numa[i + 3] + (mp_limb_t)(a2 >> 64);
85 numa[i + 0] = (mp_limb_t)a0;
86 numa[i + 1] = (mp_limb_t)a1;
87 numa[i + 2] = (mp_limb_t)a2;
88 numa[i + 3] = (mp_limb_t)a3;
89 cl = (mp_limb_t)(a3 >> 64);
90 }
91 for (; i < n; i++) {
92 __uint128_t a = (__uint128_t)numb[i] * b + numa[i] + cl;
93 numa[i] = (mp_limb_t)a;
94 cl = (mp_limb_t)(a >> 64);
95 }
96 return cl;
97}
uint64_t mp_size_t
定义 lmmp.h:93
#define a0
#define a1
#define a2
#define a3
#define n

引用了 a0, a1, a2, a3, lmmp_addmul_1_(), n , 以及 numb.

被这些函数引用 bench_addmul_1(), check_addmul_2(), gcdext_row_qscalar_(), lc1_(), lmmp_addmul_1_(), lmmp_cbrt_(), lmmp_cbrt_divide_(), lmmp_divexact_basecase_(), lmmp_hgcd_matrix_mul_u_(), lmmp_mont_redc_bc_(), lmmp_mul_toom62_(), lmmp_mul_toom62_cache_(), lmmp_mul_toom62_cache_init_(), lmmp_mullo_basecase_(), lmmp_shlnot_(), lmmp_sqrlo_basecase_(), lmmp_sqrt_(), lmmp_sqrt_divide_(), lmmp_toom_eval_pm2exp_(), lmmp_toom_eval_pm2rexp_(), lmmp_toom_interp12_(), lmmp_toom_interp16_(), lmmp_toom_interp7_(), TEST_CASE(), TEST_CASE(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图:

◆ lmmp_addmul_2_()

mp_limb_t lmmp_addmul_2_ ( mp_ptr numa,
mp_srcptr numb,
mp_size_t n,
mp_srcptr bp )

乘以双limb并累加 [numa,n] += [numb,n] * [bp,2]

参数
numa被加数指针(前 n limb 就地累加结果,第 n+1 limb 直接覆写)
numb乘数数组指针
nlimb长度
bp双limb乘数(2 个limb)
警告
n>0, eqsep(numa,numb), sep(numa,bp), numb!=NULL, bp!=NULL
返回
第 n+1 limb(最高溢出 limb)
注解
第 n limb 写入 numa[n](覆写而非累加该处旧值)

在文件 mul_1.c 第 99 行定义.

99 {
100 mp_limb_t b0 = bp[0], b1 = bp[1];
101 mp_limb_t alo = 0, ahi = 0;
102 mp_size_t i = 0;
103
104 /* 双 limb 滚动累加器 alo*B^i + ahi*B^(i+1):t 收纳本位(含 numa[i] 与 b0 行),
105 u 收纳上位(b1 行 + t 进位),两项各 < B^2 无 u128 回绕 */
106 for (; i < n; i++) {
107 __uint128_t t = (__uint128_t)numb[i] * b0 + numa[i] + alo;
108 __uint128_t u = (__uint128_t)numb[i] * b1 + ahi + (mp_limb_t)(t >> 64);
109 numa[i] = (mp_limb_t)t;
110 alo = (mp_limb_t)u;
111 ahi = (mp_limb_t)(u >> 64);
112 }
113 numa[n] = alo;
114 return ahi;
115}
#define b0
#define b1
#define t

引用了 b0, b1, lmmp_addmul_2_(), n, numb , 以及 t.

被这些函数引用 check_addmul_2(), lmmp_addmul_2_(), lmmp_divexact_basecase_(), lmmp_mont_redc_bc_() , 以及 lmmp_shlnot_().

函数调用图:
这是这个函数的调用关系图:

◆ lmmp_mul_1_()

mp_limb_t lmmp_mul_1_ ( mp_ptr dst,
mp_srcptr numa,
mp_size_t na,
mp_limb_t x )

Copyright (C) 2026 HJimmyK (Jericho Knox)

乘以单limb操作 [dst,na] = [numa,na] * x

This file is part of LMMP.

LMMP is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version.

This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License along with this library. If not, see https://www.gnu.org/licenses/.

SPDX-License-Identifier: LGPL-3.0-or-later

Copyright (C) 2026 HJimmyK (Jericho Knox)

参数
dst结果输出指针
numa被乘数指针
na操作数的 limb 长度
x单个limb乘数
警告
na>0, eqsep(dst,numa), dst!=NULL, numa!=NULL
返回
运算后的进位limb值

在文件 mul_1.c 第 27 行定义.

27 {
28 /*
29 链式 (a_{k-1} >> 64) 进位提取使 GCC/Clang 均生成 add/adc 融合序列,
30 单 limb 链长约 2c,展开宽度 W 可将链摊薄为 (W+1)/W c:
31 - clang 会把 4 路链式源码回卷成 2 路(实测慢约 7~12%),6 路是其
32 能保持的最小宽度;
33 - gcc 恰好相反,4 路最优、6 路慢约 4%。
34 故按编译器选择展开宽度。
35 */
36 mp_limb_t cl = 0;
37 mp_size_t i = 0;
38
39#if defined(__clang__)
40 for (; i + 6 <= na; i += 6) {
41 __uint128_t a0 = (__uint128_t)numa[i + 0] * x + cl;
42 __uint128_t a1 = (__uint128_t)numa[i + 1] * x + (mp_limb_t)(a0 >> 64);
43 __uint128_t a2 = (__uint128_t)numa[i + 2] * x + (mp_limb_t)(a1 >> 64);
44 __uint128_t a3 = (__uint128_t)numa[i + 3] * x + (mp_limb_t)(a2 >> 64);
45 __uint128_t a4 = (__uint128_t)numa[i + 4] * x + (mp_limb_t)(a3 >> 64);
46 __uint128_t a5 = (__uint128_t)numa[i + 5] * x + (mp_limb_t)(a4 >> 64);
47 dst[i + 0] = (mp_limb_t)a0;
48 dst[i + 1] = (mp_limb_t)a1;
49 dst[i + 2] = (mp_limb_t)a2;
50 dst[i + 3] = (mp_limb_t)a3;
51 dst[i + 4] = (mp_limb_t)a4;
52 dst[i + 5] = (mp_limb_t)a5;
53 cl = (mp_limb_t)(a5 >> 64);
54 }
55#else
56 for (; i + 4 <= na; i += 4) {
57 __uint128_t a0 = (__uint128_t)numa[i + 0] * x + cl;
58 __uint128_t a1 = (__uint128_t)numa[i + 1] * x + (mp_limb_t)(a0 >> 64);
59 __uint128_t a2 = (__uint128_t)numa[i + 2] * x + (mp_limb_t)(a1 >> 64);
60 __uint128_t a3 = (__uint128_t)numa[i + 3] * x + (mp_limb_t)(a2 >> 64);
61 dst[i + 0] = (mp_limb_t)a0;
62 dst[i + 1] = (mp_limb_t)a1;
63 dst[i + 2] = (mp_limb_t)a2;
64 dst[i + 3] = (mp_limb_t)a3;
65 cl = (mp_limb_t)(a3 >> 64);
66 }
67#endif
68 for (; i < na; i++) {
69 __uint128_t a = (__uint128_t)numa[i] * x + cl;
70 dst[i] = (mp_limb_t)a;
71 cl = (mp_limb_t)(a >> 64);
72 }
73 return cl;
74}
#define a4
#define a5

引用了 a0, a1, a2, a3, a4, a5 , 以及 lmmp_mul_1_().

被这些函数引用 bench_mul_1(), cbrt_floor_ok(), chain_mul_small(), check_pow1_laws(), gcdext_pick_s_(), ipn_sdiv_mul1_(), lc1_(), lmmp_3pow_1_(), lmmp_cbrt_(), lmmp_cbrt_6_fast_(), lmmp_cbrt_divide_(), lmmp_cbrt_knorm_(), lmmp_elem_mul_ulong_(), lmmp_from_str_(), lmmp_from_str_basecase_(), lmmp_gcdext_1_(), lmmp_gcdext_hgcd_(), lmmp_gcdext_lehmer_(), lmmp_hgcd_matrix_mul_u_(), lmmp_hgcd_step_(), lmmp_lehmer_mul_(), lmmp_mul_1_(), lmmp_mullo_basecase_(), lmmp_nthroot_(), lmmp_pow_(), lmmp_shlnot_(), lmmp_sqrlo_basecase_(), lmmp_to_str_(), lmmp_u32_pow_1_(), lmmp_u64_pow_1_(), lmmp_u8_pow_1_(), ref_value(), run_gcdext_pair(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE(), TEST_CASE() , 以及 verify_inverse().

函数调用图:
这是这个函数的调用关系图:

◆ lmmp_mullo_basecase_()

void lmmp_mullo_basecase_ ( mp_ptr restrict dst,
mp_srcptr restrict numa,
mp_srcptr restrict numb,
mp_size_t n )

在文件 mul_1.c 第 162 行定义.

162 {
163 mp_limb_t h;
164 h = numa[0] * numb[n - 1];
165 if (n != 1) {
166 mp_size_t i;
168
169 b0 = *numb++;
170 h += numa[n - 1] * b0 + lmmp_mul_1_(dst, numa, n - 1, b0);
171 dst++;
172
173 for (i = n - 2; i > 0; i--) {
174 b0 = *numb++;
175 h += numa[i] * b0 + lmmp_addmul_1_(dst, numa, i, b0);
176 dst++;
177 }
178 }
179
180 dst[0] = h;
181}
mp_limb_t lmmp_addmul_1_(mp_ptr numa, mp_srcptr numb, mp_size_t n, mp_limb_t b)
乘以单limb并累加操作 [numa,n] += [numb,n] * b
定义 mul_1.c:76
mp_limb_t lmmp_mul_1_(mp_ptr dst, mp_srcptr numa, mp_size_t na, mp_limb_t x)
乘以单limb操作 [dst,na] = [numa,na] * x
定义 mul_1.c:27

引用了 b0, lmmp_addmul_1_(), lmmp_mul_1_(), lmmp_mullo_basecase_(), n, numb , 以及 restrict.

函数调用图:

◆ lmmp_sqrlo_basecase_()

void lmmp_sqrlo_basecase_ ( mp_ptr restrict dst,
mp_srcptr restrict numa,
mp_size_t n )

在文件 mul_1.c 第 191 行定义.

191 {
192 lmmp_param_assert(n > 0);
193 mp_size_t j;
194
195 dst[0] = 0;
196 if (n > 1)
197 (void)lmmp_mul_1_(dst + 1, numa + 1, n - 1, numa[0]);
198 for (j = 1; 2 * j + 2 <= n; j++)
199 (void)lmmp_addmul_1_(dst + 2 * j + 1, numa + j + 1, n - 2 * j - 1, numa[j]);
200
201 // dst = 2*dst + 对角平方,截断于 n。分步累加防止 u128 回绕:
202 // a^2 + 2*d + cl 总和可达 B^2+B-1 > 2^128-1(a=d=B-1 极端值)
203 __uint128_t cl = 0;
204 for (j = 0; 2 * j < n; j++) {
205 __uint128_t p = (__uint128_t)numa[j] * numa[j];
206 __uint128_t v = (mp_limb_t)p + cl + ((__uint128_t)dst[2 * j] << 1); // ≤ 4B 无回绕
207 dst[2 * j] = (mp_limb_t)v;
208 cl = (v >> 64) + (p >> 64); // ≤ B+1
209 if (2 * j + 1 < n) {
210 v = ((__uint128_t)dst[2 * j + 1] << 1) + cl;
211 dst[2 * j + 1] = (mp_limb_t)v;
212 cl = v >> 64; // ≤ 2
213 }
214 }
215}
#define lmmp_param_assert(x)
定义 lmmp.h:509

引用了 lmmp_addmul_1_(), lmmp_mul_1_(), lmmp_param_assert, lmmp_sqrlo_basecase_(), n , 以及 restrict.

函数调用图:

◆ lmmp_submul_1_()

mp_limb_t lmmp_submul_1_ ( mp_ptr numa,
mp_srcptr numb,
mp_size_t n,
mp_limb_t b )

乘以单limb并累减操作 [numa,n] -= [numb,n] * b

参数
numa被减数指针(结果也存储在此)
numb乘数指针
nlimb长度
b乘数
警告
n>0, eqsep(numa,numb)), numa!=NULL, numb!=NULL
返回
运算后的借位limb值

在文件 mul_1.c 第 131 行定义.

131 {
132 mp_limb_t d = 0;
133 mp_size_t i = 0;
134
135#define LMMP_SUBMUL_1_STEP(k) \
136 { \
137 __uint128_t P = (__uint128_t)numb[i + (k)] * b; \
138 mp_limb_t o = numa[i + (k)]; \
139 mp_limb_t u = o - (mp_limb_t)P; \
140 mp_limb_t r = u - d; \
141 d = (mp_limb_t)(P >> 64) + (u > o) + (r > u); \
142 numa[i + (k)] = r; \
143 }
144
145 for (; i + 8 <= n; i += 8) {
154 }
155 for (; i < n; i++) {
157 }
158#undef LMMP_SUBMUL_1_STEP
159 return d;
160}
#define LMMP_SUBMUL_1_STEP(k)

引用了 lmmp_submul_1_(), LMMP_SUBMUL_1_STEP, n , 以及 numb.

被这些函数引用 bench_submul_1(), ipn_sdiv_mul1_(), lc1_(), lmmp_cbrt_(), lmmp_cbrt_divide_(), lmmp_div_(), lmmp_div_basecase_(), lmmp_hgcd_step_(), lmmp_shlnot_(), lmmp_sqrt_(), lmmp_submul_1_(), lmmp_toom_interp12_(), lmmp_toom_interp16_(), lmmp_toom_interp7_(), TEST_CASE(), TEST_CASE(), TEST_CASE() , 以及 TEST_CASE().

函数调用图:
这是这个函数的调用关系图: