LMMP 4.3.0
Lamina High-Precision Arithmetic Library
载入中...
搜索中...
未找到
sqr_toom8.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 "impl/mul_hard.h"
24#include "mparam.h"
25#include "lmmpn/toom_interp.h"
26
27
28/* sqr_toom8 递归可达的最大子问题规模(FFT 接管前) */
29#define SQR_TOOM8_MAX ((SQR_FFT_THRESHOLD + 8 * 2 - 1 + 7) / 8)
30#define MAYBE_sqr_basecase (SQR_TOOM8_THRESHOLD < 8 * SQR_TOOM22_THRESHOLD)
31#define MAYBE_sqr_above_basecase (SQR_TOOM8_MAX >= SQR_TOOM22_THRESHOLD)
32#define MAYBE_sqr_toom2 (SQR_TOOM8_THRESHOLD < 8 * SQR_TOOM33_THRESHOLD)
33#define MAYBE_sqr_above_toom2 (SQR_TOOM8_MAX >= SQR_TOOM33_THRESHOLD)
34#define MAYBE_sqr_toom3 (SQR_TOOM8_THRESHOLD < 8 * SQR_TOOM44_THRESHOLD)
35#define MAYBE_sqr_above_toom3 (SQR_TOOM8_MAX >= SQR_TOOM44_THRESHOLD)
36#define MAYBE_sqr_toom4 (SQR_TOOM8_THRESHOLD < 8 * SQR_TOOM6_THRESHOLD)
37#define MAYBE_sqr_above_toom4 (SQR_TOOM8_MAX >= SQR_TOOM6_THRESHOLD)
38#define MAYBE_sqr_above_toom6 (SQR_TOOM8_MAX >= SQR_TOOM8_THRESHOLD)
39
40#define lmmp_sqr_(dst, numa, n) \
41 if (MAYBE_sqr_basecase && (!MAYBE_sqr_above_basecase || (n) < SQR_TOOM22_THRESHOLD)) \
42 lmmp_sqr_hard_n_((dst), (numa), (n)); \
43 else if (MAYBE_sqr_toom2 && (!MAYBE_sqr_above_toom2 || (n) < SQR_TOOM33_THRESHOLD)) \
44 lmmp_sqr_toom2_((dst), (numa), (n)); \
45 else if (MAYBE_sqr_toom3 && (!MAYBE_sqr_above_toom3 || (n) < SQR_TOOM44_THRESHOLD)) \
46 lmmp_sqr_toom3_((dst), (numa), (n)); \
47 else if (MAYBE_sqr_toom4 && (!MAYBE_sqr_above_toom4 || (n) < SQR_TOOM6_THRESHOLD)) \
48 lmmp_sqr_toom4_((dst), (numa), (n)); \
49 else if (!MAYBE_sqr_above_toom6 || (n) < SQR_TOOM8_THRESHOLD) \
50 lmmp_sqr_toom6_((dst), (numa), (n)); \
51 else \
52 lmmp_sqr_toom8_((dst), (numa), (n))
53
54/*
55Evaluate in: +8,-8, +4,-4, +2,-2, +1,-1, +1/2,-1/2, +1/4,-1/4, +1/8,-1/8, 0
56(平方无 inf 点:(-v)^2 = v^2,符号恒正)
57*/
58
60 lmmp_param_assert(na > 0);
61 lmmp_param_assert(dst != NULL);
62 lmmp_param_assert(numa != NULL);
63 lmmp_param_assert(na >= 51); /* 契约排除 s=1 退化点 na=50/57(16 点插值需 2s>3) */
64 lmmp_param_assert(na != 57);
65 mp_size_t n, s;
66
67 n = 1 + ((na - 1) >> 3);
68 s = na - 7 * n;
69
70 lmmp_debug_assert(0 < s && s <= n);
71 /* 契约排除 na=50、57(na>=50 中仅此两点 s=1,不满足 2s>3) */
72 lmmp_debug_assert(s + s > 3);
73
74#define r6 (dst + 3 * n) /* 3n+1 */
75#define r4 (dst + 7 * n) /* 3n+1 */
76#define r2 (dst + 11 * n) /* 3n+1 */
77#define r0 (dst + 15 * n) /* 2s */
78#define r7 (scratch) /* 3n+1 */
79#define r5 (scratch + 3 * n + 1) /* 3n+1 */
80#define r3 (scratch + 6 * n + 2) /* 3n+1 */
81#define r1 (scratch + 9 * n + 3) /* 3n+1 */
82#define v0 (dst + 11 * n) /* n+1 */
83#define v2 (dst + 13 * n + 2) /* n+1 */
84#define wse (scratch + 12 * n + 4) /* 3n+1 */
85
86 /* scratch 总需求 15n+6 limbs;TALLOC 双轨(>32KiB 落池/堆) */
89
90 /* ±1/8 */
91 lmmp_toom_eval_pm2rexp_(v2, v0, 7, numa, n, s, 3, dst);
92 lmmp_sqr_(dst, v0, n + 1);
93 lmmp_sqr_(r7, v2, n + 1);
94 lmmp_toom_couple_handling_(r7, 2 * n + 1, dst, 0, n, 3, 0);
95
96 /* ±1/4 */
97 lmmp_toom_eval_pm2rexp_(v2, v0, 7, numa, n, s, 2, dst);
98 lmmp_sqr_(dst, v0, n + 1);
99 lmmp_sqr_(r5, v2, n + 1);
100 lmmp_toom_couple_handling_(r5, 2 * n + 1, dst, 0, n, 2, 0);
101
102 /* ±2 */
103 lmmp_toom_eval_pm2_(v2, v0, 7, numa, n, s, dst);
104 lmmp_sqr_(dst, v0, n + 1);
105 lmmp_sqr_(r3, v2, n + 1);
106 lmmp_toom_couple_handling_(r3, 2 * n + 1, dst, 0, n, 1, 2);
107
108 /* ±8 */
109 lmmp_toom_eval_pm2exp_(v2, v0, 7, numa, n, s, 3, dst);
110 lmmp_sqr_(dst, v0, n + 1);
111 lmmp_sqr_(r1, v2, n + 1);
112 lmmp_toom_couple_handling_(r1, 2 * n + 1, dst, 0, n, 3, 6);
113
114 /* ±1/2 */
115 lmmp_toom_eval_pm2rexp_(v2, v0, 7, numa, n, s, 1, dst);
116 lmmp_sqr_(dst, v0, n + 1);
117 lmmp_sqr_(r6, v2, n + 1);
118 lmmp_toom_couple_handling_(r6, 2 * n + 1, dst, 0, n, 1, 0);
119
120 /* ±1 */
121 lmmp_toom_eval_pm1_(v2, v0, 7, numa, n, s, dst);
122 lmmp_sqr_(dst, v0, n + 1);
123 lmmp_sqr_(r4, v2, n + 1);
124 lmmp_toom_couple_handling_(r4, 2 * n + 1, dst, 0, n, 0, 0);
125
126 /* ±4 */
127 lmmp_toom_eval_pm2exp_(v2, v0, 7, numa, n, s, 2, dst);
128 lmmp_sqr_(dst, v0, n + 1);
129 lmmp_sqr_(r2, v2, n + 1);
130 lmmp_toom_couple_handling_(r2, 2 * n + 1, dst, 0, n, 2, 4);
131
132#undef v0
133#undef v2
134
135 /* A(0)^2 */
136 lmmp_sqr_(dst, numa, n);
137
138 lmmp_toom_interp16_(dst, r1, r3, r5, r7, n, 2 * s, 0, wse);
139
140#undef r0
141#undef r1
142#undef r2
143#undef r3
144#undef r4
145#undef r5
146#undef r6
147#undef r7
148
149 TEMP_FREE;
150}
151
152#undef lmmp_sqr_
153#undef MAYBE_sqr_basecase
154#undef MAYBE_sqr_above_basecase
155#undef MAYBE_sqr_toom2
156#undef MAYBE_sqr_above_toom2
157#undef MAYBE_sqr_toom3
158#undef MAYBE_sqr_above_toom3
159#undef MAYBE_sqr_toom4
160#undef MAYBE_sqr_above_toom4
161#undef MAYBE_sqr_above_toom6
162#undef SQR_TOOM8_MAX
#define lmmp_sqr_
mp_limb_t * mp_ptr
定义 lmmp.h:96
uint64_t mp_size_t
定义 lmmp.h:93
#define lmmp_debug_assert(x)
定义 lmmp.h:498
const mp_limb_t * mp_srcptr
定义 lmmp.h:97
uint64_t mp_limb_t
定义 lmmp.h:92
#define lmmp_param_assert(x)
定义 lmmp.h:509
void lmmp_sqr_toom8_(mp_ptr dst, mp_srcptr numa, mp_size_t na)
Toom-8平方运算 [dst,2*na] = [numa,na]^2 (评估点 ±8,±4,±2,±1,±1/2,±1/4,±1/8,0,无inf点)
#define v0
#define r2
#define r1
#define r3
#define v2
#define s
#define n
#define scratch
#define wse
#define r4
#define r5
#define r6
#define r7
void lmmp_toom_couple_handling_(mp_ptr pp, mp_size_t n, mp_ptr np, int nsign, mp_size_t off, int ps, int ns)
Copyright (C) 2026 HJimmyK (Jericho Knox)
int lmmp_toom_eval_pm2exp_(mp_ptr xp2, mp_ptr xm2, unsigned k, mp_srcptr xp, mp_size_t n, mp_size_t hn, unsigned shift, mp_ptr tp)
通用高阶 Toom 求值:k次多项式在 x = +2^shift 和 x = -2^shift 处求值 (toom6h 的 ±4 / toom8h 的 ±4、±8 点)
int lmmp_toom_eval_pm2_(mp_ptr xp2, mp_ptr xm2, unsigned k, mp_srcptr xp, mp_size_t n, mp_size_t hn, mp_ptr tp)
通用高阶 Toom 求值:k次多项式在 x = +2 和 x = -2 处求值
int lmmp_toom_eval_pm2rexp_(mp_ptr rp, mp_ptr rm, unsigned q, mp_srcptr ap, mp_size_t n, mp_size_t t, unsigned s, mp_ptr ws)
通用高阶 Toom 求值:q次多项式在 x = +1/2^s 和 x = -1/2^s 处求值, 结果放大 2^(s*q)(toom6h 的 ±1/2、±1/4 / toom8h 的 ±1/2、±1/4...
int lmmp_toom_eval_pm1_(mp_ptr xp1, mp_ptr xm1, unsigned k, mp_srcptr xp, mp_size_t n, mp_size_t hn, mp_ptr tp)
通用高阶 Toom 求值:k次多项式在 x = +1 和 x = -1 处求值
void lmmp_toom_interp16_(mp_ptr dst, mp_ptr r1, mp_ptr r3, mp_ptr r5, mp_ptr r7, mp_size_t n, mp_size_t spt, int half, mp_ptr tp)
Toom插值计算(16点插值):用于 Toom-8.5(toom8h/toom8 平方)乘法 评估点:inf(仅 half), ±8, ±4, ±2, ±1, ±1/2,...
#define restrict
Copyright (C) 2026 HJimmyK (Jericho Knox)
#define TEMP_DECL
#define TEMP_FREE
#define TALLOC_TYPE(n, type)