1 /* Haswell 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 64
32 #define GMP_LIMB_BYTES 8
33
34 /* 3600-4000 MHz Intel Xeon E3-1271v3 Haswell */
35 /* FFT tuning limit = 467,964,359 */
36 /* Generated by tuneup.c, 2019-10-18, gcc 8.3 */
37
38 #define MOD_1_NORM_THRESHOLD 0 /* always */
39 #define MOD_1_UNNORM_THRESHOLD 0 /* always */
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD 4
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD 3
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 9
43 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 26
44 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 9
45 #define USE_PREINV_DIVREM_1 1 /* native */
46 #define DIV_QR_1_NORM_THRESHOLD 1
47 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
48 #define DIV_QR_2_PI2_THRESHOLD 9
49 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
50 #define BMOD_1_TO_MOD_1_THRESHOLD 25
51
52 #define DIV_1_VS_MUL_1_PERCENT 427
53
54 #define MUL_TOOM22_THRESHOLD 20
55 #define MUL_TOOM33_THRESHOLD 74
56 #define MUL_TOOM44_THRESHOLD 195
57 #define MUL_TOOM6H_THRESHOLD 276
58 #define MUL_TOOM8H_THRESHOLD 381
59
60 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 120
61 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 139
62 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 128
63 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 129
64 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 170
65
66 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
67 #define SQR_TOOM2_THRESHOLD 32
68 #define SQR_TOOM3_THRESHOLD 117
69 #define SQR_TOOM4_THRESHOLD 315
70 #define SQR_TOOM6_THRESHOLD 414
71 #define SQR_TOOM8_THRESHOLD 0 /* always */
72
73 #define MULMID_TOOM42_THRESHOLD 42
74
75 #define MULMOD_BNM1_THRESHOLD 13
76 #define SQRMOD_BNM1_THRESHOLD 17
77
78 #define MUL_FFT_MODF_THRESHOLD 376 /* k = 5 */
79 #define MUL_FFT_TABLE3 \
80 { { 376, 5}, { 21, 6}, { 11, 5}, { 23, 6}, \
81 { 21, 7}, { 11, 6}, { 25, 7}, { 13, 6}, \
82 { 27, 7}, { 21, 8}, { 11, 7}, { 25, 8}, \
83 { 13, 7}, { 28, 8}, { 15, 7}, { 31, 8}, \
84 { 17, 7}, { 35, 8}, { 21, 9}, { 11, 8}, \
85 { 27, 9}, { 15, 8}, { 35, 9}, { 19, 8}, \
86 { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \
87 { 15, 9}, { 39,10}, { 23, 9}, { 51,11}, \
88 { 15,10}, { 31, 9}, { 71,10}, { 39, 9}, \
89 { 83,10}, { 47, 9}, { 95,11}, { 31,10}, \
90 { 79,11}, { 47,10}, { 95,12}, { 31,11}, \
91 { 63,10}, { 127, 9}, { 255,10}, { 135,11}, \
92 { 79,10}, { 167,11}, { 95,10}, { 191, 9}, \
93 { 383,11}, { 111,12}, { 63, 8}, { 1023,11}, \
94 { 143,10}, { 287, 9}, { 575,10}, { 303,11}, \
95 { 159,10}, { 319,12}, { 95,11}, { 191,10}, \
96 { 383,11}, { 207,13}, { 63,12}, { 127,11}, \
97 { 255,10}, { 543,11}, { 287,10}, { 575,11}, \
98 { 303,10}, { 607,12}, { 159,11}, { 319,10}, \
99 { 639,11}, { 335,10}, { 671,11}, { 351,10}, \
100 { 703,11}, { 367,10}, { 735,11}, { 383,10}, \
101 { 767,11}, { 415,10}, { 831,11}, { 447,10}, \
102 { 895,11}, { 479,13}, { 127,11}, { 543,10}, \
103 { 1087,12}, { 287,11}, { 607,10}, { 1215,12}, \
104 { 319,11}, { 671,12}, { 351,11}, { 735,13}, \
105 { 191,12}, { 383,11}, { 767,12}, { 415,11}, \
106 { 831,12}, { 447,11}, { 895,12}, { 479,11}, \
107 { 959,14}, { 127,12}, { 543,11}, { 1087,12}, \
108 { 607,11}, { 1215,10}, { 2431,12}, { 671,11}, \
109 { 1343,12}, { 703,11}, { 1407,12}, { 735,13}, \
110 { 383,12}, { 831,13}, { 447,12}, { 959,13}, \
111 { 511,12}, { 1087,11}, { 2175,13}, { 575,12}, \
112 { 1215,11}, { 2431,13}, { 639,12}, { 1343,13}, \
113 { 703,12}, { 1407,14}, { 383,13}, { 767,12}, \
114 { 1535,13}, { 831,12}, { 1727,13}, { 959,12}, \
115 { 1919,14}, { 511,13}, { 1023,12}, { 2047,13}, \
116 { 1087,12}, { 2175,13}, { 1215,14}, { 639,13}, \
117 { 1343,12}, { 2687,13}, { 1407,12}, { 2815,13}, \
118 { 1471,14}, { 767,13}, { 1599,12}, { 3199,13}, \
119 { 1727,14}, { 895,13}, { 1791,12}, { 3583,13}, \
120 { 1919,15}, { 511,14}, { 1023,13}, { 2175,14}, \
121 { 1151,13}, { 2431,12}, { 4863,14}, { 1279,13}, \
122 { 2687,14}, { 1407,13}, { 2815,15}, { 767,14}, \
123 { 1535,13}, { 3199,14}, { 1663,13}, { 3455,12}, \
124 { 6911,14}, { 1791,13}, { 3583,14}, { 1919,16}, \
125 { 511,15}, { 1023,14}, { 2175,13}, { 4351,14}, \
126 { 2431,13}, { 4863,15}, { 1279,14}, { 2943,13}, \
127 { 5887,12}, { 11775,15}, { 1535,14}, { 3455,13}, \
128 { 6911,15}, { 1791,14}, { 3839,13}, { 7679,16}, \
129 { 1023,15}, { 2047,14}, { 4351,15}, { 2303,14}, \
130 { 4863,15}, { 2815,14}, { 5887,13}, { 11775,16}, \
131 { 1535,15}, { 3327,14}, { 6911,15}, { 3839,14}, \
132 { 7679,17}, { 1023,16}, { 2047,15}, { 4863,16}, \
133 { 2559,15}, { 5887,14}, { 11775,16}, { 3071,15}, \
134 { 6911,16}, { 3583,15}, { 7679,14}, { 15359,15}, \
135 { 7935,17}, { 2047,16}, { 4095,15}, { 8447,16}, \
136 { 4607,15}, { 9983,14}, { 19967,16}, { 5631,15}, \
137 { 11775,17}, { 3071,16}, { 65536,17}, { 131072,18}, \
138 { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \
139 {4194304,23}, {8388608,24} }
140 #define MUL_FFT_TABLE3_SIZE 238
141 #define MUL_FFT_THRESHOLD 4736
142
143 #define SQR_FFT_MODF_THRESHOLD 368 /* k = 5 */
144 #define SQR_FFT_TABLE3 \
145 { { 368, 5}, { 21, 6}, { 11, 5}, { 23, 6}, \
146 { 25, 7}, { 13, 6}, { 27, 7}, { 21, 8}, \
147 { 11, 7}, { 25, 8}, { 13, 7}, { 28, 8}, \
148 { 15, 7}, { 31, 8}, { 21, 9}, { 11, 8}, \
149 { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \
150 { 41, 9}, { 23, 8}, { 47, 9}, { 27,10}, \
151 { 15, 9}, { 39,10}, { 23, 9}, { 51,11}, \
152 { 15,10}, { 31, 9}, { 67,10}, { 39, 9}, \
153 { 79,10}, { 55,11}, { 31,10}, { 79,11}, \
154 { 47,10}, { 95,12}, { 31,11}, { 63,10}, \
155 { 127, 9}, { 255,11}, { 79,10}, { 159, 9}, \
156 { 319,11}, { 95,10}, { 191,12}, { 63,11}, \
157 { 127,10}, { 255, 9}, { 511,10}, { 271,11}, \
158 { 143,10}, { 287, 9}, { 575,10}, { 303, 9}, \
159 { 607,11}, { 159,10}, { 319, 6}, { 5631, 7}, \
160 { 2943, 6}, { 5887, 8}, { 1535,11}, { 207,10}, \
161 { 415,11}, { 223,10}, { 447,11}, { 239,10}, \
162 { 479,12}, { 127,11}, { 255,10}, { 511,11}, \
163 { 271,10}, { 543,11}, { 287,10}, { 575,11}, \
164 { 303,10}, { 607,11}, { 319,10}, { 639,11}, \
165 { 335,10}, { 671,11}, { 351,10}, { 703,11}, \
166 { 367,10}, { 735,11}, { 383,10}, { 767,11}, \
167 { 415,10}, { 831,11}, { 447,10}, { 895,11}, \
168 { 479,13}, { 127,11}, { 511,10}, { 1023,11}, \
169 { 543,10}, { 1087,12}, { 287,11}, { 607,10}, \
170 { 1215,11}, { 671,12}, { 351,11}, { 735,12}, \
171 { 383,11}, { 767,12}, { 415,11}, { 831,12}, \
172 { 447,11}, { 895,12}, { 479,11}, { 959,14}, \
173 { 127,12}, { 511,11}, { 1023,12}, { 543,11}, \
174 { 1087,12}, { 607,11}, { 1215,12}, { 735,13}, \
175 { 383,12}, { 831,13}, { 447,12}, { 959,13}, \
176 { 511,12}, { 1087,13}, { 575,12}, { 1151,13}, \
177 { 639,12}, { 1279,13}, { 703,12}, { 1407,11}, \
178 { 2815,14}, { 383,13}, { 767,12}, { 1535,13}, \
179 { 831,12}, { 1727,11}, { 3455,13}, { 959,14}, \
180 { 511,13}, { 1087,12}, { 2175,13}, { 1215,14}, \
181 { 639,13}, { 1279,12}, { 2559,13}, { 1343,12}, \
182 { 2687,13}, { 1407,12}, { 2815,13}, { 1471,14}, \
183 { 767,13}, { 1599,12}, { 3199,13}, { 1727,14}, \
184 { 895,13}, { 1791,12}, { 3583,13}, { 1919,15}, \
185 { 511,14}, { 1023,13}, { 2175,14}, { 1151,13}, \
186 { 2303,12}, { 4607,13}, { 2431,12}, { 4863,14}, \
187 { 1279,13}, { 2687,14}, { 1407,13}, { 2815,15}, \
188 { 767,14}, { 1535,13}, { 3199,14}, { 1663,13}, \
189 { 3455,12}, { 6911,14}, { 1791,13}, { 3583,14}, \
190 { 1919,16}, { 511,15}, { 1023,14}, { 2431,13}, \
191 { 4863,15}, { 1279,14}, { 2943,13}, { 5887,15}, \
192 { 1535,14}, { 3455,13}, { 6911,15}, { 1791,14}, \
193 { 3839,13}, { 7679,16}, { 1023,15}, { 2047,14}, \
194 { 4351,15}, { 2303,14}, { 4863,15}, { 2815,14}, \
195 { 5887,16}, { 1535,15}, { 3071,14}, { 6143,15}, \
196 { 3327,14}, { 6911,15}, { 3839,14}, { 7679,17}, \
197 { 1023,16}, { 2047,15}, { 4863,16}, { 2559,15}, \
198 { 5887,14}, { 11775,16}, { 3071,15}, { 6911,16}, \
199 { 3583,15}, { 7679,14}, { 15359,15}, { 7935,17}, \
200 { 2047,16}, { 4095,15}, { 8191,16}, { 4607,15}, \
201 { 9983,14}, { 19967,16}, { 5631,15}, { 11775,17}, \
202 { 3071,16}, { 65536,17}, { 131072,18}, { 262144,19}, \
203 { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \
204 {8388608,24} }
205 #define SQR_FFT_TABLE3_SIZE 237
206 #define SQR_FFT_THRESHOLD 3264
207
208 #define MULLO_BASECASE_THRESHOLD 0 /* always */
209 #define MULLO_DC_THRESHOLD 68
210 #define MULLO_MUL_N_THRESHOLD 8967
211 #define SQRLO_BASECASE_THRESHOLD 11
212 #define SQRLO_DC_THRESHOLD 80
213 #define SQRLO_SQR_THRESHOLD 6481
214
215 #define DC_DIV_QR_THRESHOLD 58
216 #define DC_DIVAPPR_Q_THRESHOLD 182
217 #define DC_BDIV_QR_THRESHOLD 60
218 #define DC_BDIV_Q_THRESHOLD 123
219
220 #define INV_MULMOD_BNM1_THRESHOLD 38
221 #define INV_NEWTON_THRESHOLD 179
222 #define INV_APPR_THRESHOLD 182
223
224 #define BINV_NEWTON_THRESHOLD 230
225 #define REDC_1_TO_REDC_2_THRESHOLD 48
226 #define REDC_2_TO_REDC_N_THRESHOLD 63
227
228 #define MU_DIV_QR_THRESHOLD 1470
229 #define MU_DIVAPPR_Q_THRESHOLD 1528
230 #define MUPI_DIV_QR_THRESHOLD 82
231 #define MU_BDIV_QR_THRESHOLD 1334
232 #define MU_BDIV_Q_THRESHOLD 1506
233
234 #define POWM_SEC_TABLE 1,22,194,473,1297,2698
235
236 #define GET_STR_DC_THRESHOLD 12
237 #define GET_STR_PRECOMPUTE_THRESHOLD 19
238 #define SET_STR_DC_THRESHOLD 1391
239 #define SET_STR_PRECOMPUTE_THRESHOLD 2654
240
241 #define FAC_DSC_THRESHOLD 562
242 #define FAC_ODD_THRESHOLD 0 /* always */
243
244 #define MATRIX22_STRASSEN_THRESHOLD 15
245 #define HGCD2_DIV1_METHOD 5 /* 3.49% faster than 3 */
246 #define HGCD_THRESHOLD 96
247 #define HGCD_APPR_THRESHOLD 92
248 #define HGCD_REDUCE_THRESHOLD 2681
249 #define GCD_DC_THRESHOLD 501
250 #define GCDEXT_DC_THRESHOLD 365
251 #define JACOBI_BASE_METHOD 1 /* 23.87% faster than 4 */
252
253 /* Tuneup completed successfully, took 238360 seconds */