1 /* x86/coreihwl gmp-mparam.h -- Compiler/machine parameter header file.
2
3 Copyright 2019 Free Software Foundation, Inc.
4
5 This file is part of the GNU MP Library.
6
7 The GNU MP Library is free software; you can redistribute it and/or modify
8 it under the terms of either:
9
10 * the GNU Lesser General Public License as published by the Free
11 Software Foundation; either version 3 of the License, or (at your
12 option) any later version.
13
14 or
15
16 * the GNU General Public License as published by the Free Software
17 Foundation; either version 2 of the License, or (at your option) any
18 later version.
19
20 or both in parallel, as here.
21
22 The GNU MP Library is distributed in the hope that it will be useful, but
23 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
24 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
25 for more details.
26
27 You should have received copies of the GNU General Public License and the
28 GNU Lesser General Public License along with the GNU MP Library. If not,
29 see https://www.gnu.org/licenses/. */
30
31 #define GMP_LIMB_BITS 32
32 #define GMP_LIMB_BYTES 4
33
34 /* 3600-4000 MHz Intel Xeon E3-1271v3 Haswell */
35 /* FFT tuning limit = 67,000,000 */
36 /* Generated by tuneup.c, 2019-10-21, gcc 8.3 */
37
38 #define MOD_1_NORM_THRESHOLD 17
39 #define MOD_1_UNNORM_THRESHOLD 17
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD 10
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD 9
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 0 /* never mpn_mod_1_1p */
43 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 10
44 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 5
45 #define USE_PREINV_DIVREM_1 1 /* native */
46 /* From sky.gmplib.org, 2023-07-20 */
47 #define DIV_QR_1N_PI1_METHOD 3 /* 5.86% faster than 1 */
48 #define DIV_QR_1_NORM_THRESHOLD 13
49 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
50 #define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */
51 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
52 #define BMOD_1_TO_MOD_1_THRESHOLD 21
53
54 #define DIV_1_VS_MUL_1_PERCENT 296
55
56 #define MUL_TOOM22_THRESHOLD 28
57 #define MUL_TOOM33_THRESHOLD 108
58 #define MUL_TOOM44_THRESHOLD 232
59 #define MUL_TOOM6H_THRESHOLD 306
60 #define MUL_TOOM8H_THRESHOLD 478
61
62 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 109
63 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 183
64 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 113
65 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 113
66 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 136
67
68 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
69 #define SQR_TOOM2_THRESHOLD 44
70 #define SQR_TOOM3_THRESHOLD 141
71 #define SQR_TOOM4_THRESHOLD 384
72 #define SQR_TOOM6_THRESHOLD 517
73 #define SQR_TOOM8_THRESHOLD 698
74
75 #define MULMID_TOOM42_THRESHOLD 98
76
77 #define MULMOD_BNM1_THRESHOLD 20
78 #define SQRMOD_BNM1_THRESHOLD 23
79
80 #define MUL_FFT_MODF_THRESHOLD 565 /* k = 5 */
81 #define MUL_FFT_TABLE3 \
82 { { 565, 5}, { 28, 6}, { 15, 5}, { 31, 6}, \
83 { 16, 5}, { 33, 6}, { 29, 7}, { 15, 6}, \
84 { 33, 7}, { 17, 6}, { 36, 7}, { 19, 6}, \
85 { 39, 7}, { 21, 6}, { 43, 7}, { 23, 6}, \
86 { 47, 7}, { 29, 8}, { 15, 7}, { 35, 8}, \
87 { 19, 7}, { 43, 8}, { 23, 7}, { 47, 8}, \
88 { 27, 7}, { 55, 9}, { 15, 8}, { 31, 7}, \
89 { 63, 8}, { 43, 9}, { 23, 8}, { 55, 9}, \
90 { 31, 8}, { 71, 9}, { 39, 8}, { 83, 9}, \
91 { 47, 8}, { 95, 9}, { 55,10}, { 31, 9}, \
92 { 79,10}, { 47, 9}, { 95,11}, { 31,10}, \
93 { 63, 9}, { 135,10}, { 79, 9}, { 159,10}, \
94 { 95, 9}, { 191,10}, { 111,11}, { 63,10}, \
95 { 143, 9}, { 287,10}, { 159,11}, { 95,10}, \
96 { 191, 6}, { 3199, 7}, { 1727, 9}, { 447,10}, \
97 { 239, 9}, { 479,10}, { 287,11}, { 159,10}, \
98 { 319, 9}, { 639,10}, { 335, 9}, { 671,11}, \
99 { 191,10}, { 399, 9}, { 799,10}, { 415,11}, \
100 { 223,12}, { 127,11}, { 255,10}, { 511, 9}, \
101 { 1023,10}, { 527,11}, { 287,10}, { 607,11}, \
102 { 319,10}, { 671,12}, { 191,11}, { 383,10}, \
103 { 799,11}, { 415,13}, { 127,12}, { 255,11}, \
104 { 511,10}, { 1023,11}, { 543,10}, { 1087,11}, \
105 { 607,12}, { 319,11}, { 671,10}, { 1343,11}, \
106 { 735,12}, { 383,11}, { 799,10}, { 1599,11}, \
107 { 863,12}, { 447,11}, { 991,13}, { 255,12}, \
108 { 511,11}, { 1087,12}, { 575,11}, { 1215,12}, \
109 { 639,11}, { 1343,12}, { 703,11}, { 1407,13}, \
110 { 383,12}, { 767,11}, { 1599,12}, { 831,11}, \
111 { 1727,12}, { 959,11}, { 1919,14}, { 255,13}, \
112 { 511,12}, { 1087,11}, { 2239,12}, { 1215,13}, \
113 { 639,12}, { 1471,13}, { 767,12}, { 1727,13}, \
114 { 895,12}, { 1919,14}, { 511,13}, { 1023,12}, \
115 { 2239,13}, { 1151,12}, { 2431,13}, { 1279,12}, \
116 { 2623,13}, { 1407,14}, { 767,13}, { 1663,12}, \
117 { 3455,13}, { 1919,15}, { 511,14}, { 1023,13}, \
118 { 2175,12}, { 4479,13}, { 2431,14}, { 1279,13}, \
119 { 2943,12}, { 5887,14}, { 1535,13}, { 3455,14}, \
120 { 1791,13}, { 3967,15}, { 1023,14}, { 2047,13}, \
121 { 4479,14}, { 2303,13}, { 4991,14}, { 2559,13}, \
122 { 5375,14}, { 2815,13}, { 5887,15}, { 1535,14}, \
123 { 3839,16} }
124 #define MUL_FFT_TABLE3_SIZE 165
125 #define MUL_FFT_THRESHOLD 7808
126
127 #define SQR_FFT_MODF_THRESHOLD 560 /* k = 5 */
128 #define SQR_FFT_TABLE3 \
129 { { 560, 5}, { 29, 6}, { 15, 5}, { 31, 6}, \
130 { 16, 5}, { 33, 6}, { 17, 5}, { 36, 6}, \
131 { 29, 7}, { 15, 6}, { 33, 7}, { 17, 6}, \
132 { 36, 7}, { 19, 6}, { 40, 7}, { 21, 6}, \
133 { 43, 7}, { 23, 6}, { 47, 7}, { 29, 8}, \
134 { 15, 7}, { 35, 8}, { 19, 7}, { 43, 8}, \
135 { 23, 7}, { 49, 8}, { 27, 7}, { 55, 9}, \
136 { 15, 8}, { 31, 7}, { 63, 8}, { 43, 9}, \
137 { 23, 8}, { 55,10}, { 15, 9}, { 31, 8}, \
138 { 67, 9}, { 39, 8}, { 79, 9}, { 47, 8}, \
139 { 95, 9}, { 55,10}, { 31, 9}, { 79,10}, \
140 { 47, 9}, { 95,11}, { 31,10}, { 63, 9}, \
141 { 135,10}, { 79, 9}, { 159,10}, { 95,11}, \
142 { 63,10}, { 143, 9}, { 287,10}, { 159,11}, \
143 { 95,12}, { 63,11}, { 127, 9}, { 511, 5}, \
144 { 8959, 7}, { 2431, 8}, { 1247, 7}, { 2495, 8}, \
145 { 1279, 9}, { 671,10}, { 367,11}, { 191,10}, \
146 { 399, 9}, { 799,10}, { 415,12}, { 127,11}, \
147 { 255,10}, { 527,11}, { 287,10}, { 607,11}, \
148 { 319,10}, { 671,11}, { 351,10}, { 703,12}, \
149 { 191,11}, { 383,10}, { 799,11}, { 415,10}, \
150 { 831,13}, { 127,11}, { 543,10}, { 1119,11}, \
151 { 607,12}, { 319,11}, { 671,10}, { 1343,11}, \
152 { 735,12}, { 383,11}, { 863,12}, { 447,11}, \
153 { 991,12}, { 511,11}, { 1119,12}, { 575,11}, \
154 { 1215,12}, { 639,11}, { 1343,12}, { 703,13}, \
155 { 383,12}, { 767,11}, { 1599,12}, { 831,11}, \
156 { 1727,12}, { 959,11}, { 1983,13}, { 511,12}, \
157 { 1087,11}, { 2239,12}, { 1215,13}, { 639,12}, \
158 { 1471,13}, { 767,12}, { 1727,13}, { 895,12}, \
159 { 1983,14}, { 511,13}, { 1023,12}, { 2239,13}, \
160 { 1151,12}, { 2495,13}, { 1279,12}, { 2623,13}, \
161 { 1407,14}, { 767,13}, { 1663,12}, { 3455,13}, \
162 { 1919,15}, { 511,14}, { 1023,13}, { 2175,12}, \
163 { 4479,13}, { 2431,14}, { 1279,13}, { 2943,12}, \
164 { 5887,14}, { 1535,13}, { 3455,14}, { 1791,13}, \
165 { 3967,15}, { 1023,14}, { 2047,13}, { 4479,14}, \
166 { 2303,13}, { 4991,12}, { 9983,14}, { 2559,13}, \
167 { 5119,14}, { 2815,13}, { 5887,15}, { 1535,14}, \
168 { 3327,13}, { 6911,14}, { 3839,16} }
169 #define SQR_FFT_TABLE3_SIZE 159
170 #define SQR_FFT_THRESHOLD 5568
171
172 #define MULLO_BASECASE_THRESHOLD 17
173 #define MULLO_DC_THRESHOLD 40
174 #define MULLO_MUL_N_THRESHOLD 14281
175 #define SQRLO_BASECASE_THRESHOLD 0 /* always */
176 #define SQRLO_DC_THRESHOLD 141
177 #define SQRLO_SQR_THRESHOLD 10821
178
179 #define DC_DIV_QR_THRESHOLD 30
180 #define DC_DIVAPPR_Q_THRESHOLD 190
181 #define DC_BDIV_QR_THRESHOLD 67
182 #define DC_BDIV_Q_THRESHOLD 254
183
184 #define INV_MULMOD_BNM1_THRESHOLD 54
185 #define INV_NEWTON_THRESHOLD 157
186 #define INV_APPR_THRESHOLD 163
187
188 #define BINV_NEWTON_THRESHOLD 236
189 #define REDC_1_TO_REDC_N_THRESHOLD 79
190
191 #define MU_DIV_QR_THRESHOLD 1895
192 #define MU_DIVAPPR_Q_THRESHOLD 1718
193 #define MUPI_DIV_QR_THRESHOLD 54
194 #define MU_BDIV_QR_THRESHOLD 1589
195 #define MU_BDIV_Q_THRESHOLD 1898
196
197 #define POWM_SEC_TABLE 1,16,95,480,1442
198
199 #define GET_STR_DC_THRESHOLD 10
200 #define GET_STR_PRECOMPUTE_THRESHOLD 16
201 #define SET_STR_DC_THRESHOLD 372
202 #define SET_STR_PRECOMPUTE_THRESHOLD 1037
203
204 #define FAC_DSC_THRESHOLD 141
205 #define FAC_ODD_THRESHOLD 34
206
207 #define MATRIX22_STRASSEN_THRESHOLD 21
208 #define HGCD2_DIV1_METHOD 3 /* 6.26% faster than 1 */
209 #define HGCD_THRESHOLD 70
210 #define HGCD_APPR_THRESHOLD 129
211 #define HGCD_REDUCE_THRESHOLD 3664
212 #define GCD_DC_THRESHOLD 573
213 #define GCDEXT_DC_THRESHOLD 483
214 #define JACOBI_BASE_METHOD 1 /* 27.01% faster than 3 */
215
216 /* Tuneup completed successfully, took 35232 seconds */