1 /* x86/skylake 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 E3-1270v5 Skylake */
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 15
39 #define MOD_1_UNNORM_THRESHOLD 16
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD 10
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD 8
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 9
45 #define USE_PREINV_DIVREM_1 1 /* native */
46 #define DIV_QR_1N_PI1_METHOD 1 /* 5.63% faster than 2 */
47 #define DIV_QR_1_NORM_THRESHOLD 12
48 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
49 #define DIV_QR_2_PI2_THRESHOLD 17
50 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
51 #define BMOD_1_TO_MOD_1_THRESHOLD 18
52
53 #define DIV_1_VS_MUL_1_PERCENT 348
54
55 #define MUL_TOOM22_THRESHOLD 24
56 #define MUL_TOOM33_THRESHOLD 81
57 #define MUL_TOOM44_THRESHOLD 208
58 #define MUL_TOOM6H_THRESHOLD 303
59 #define MUL_TOOM8H_THRESHOLD 454
60
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 81
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 149
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 137
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 145
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 196
66
67 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
68 #define SQR_TOOM2_THRESHOLD 40
69 #define SQR_TOOM3_THRESHOLD 129
70 #define SQR_TOOM4_THRESHOLD 220
71 #define SQR_TOOM6_THRESHOLD 354
72 #define SQR_TOOM8_THRESHOLD 608
73
74 #define MULMID_TOOM42_THRESHOLD 72
75
76 #define MULMOD_BNM1_THRESHOLD 17
77 #define SQRMOD_BNM1_THRESHOLD 21
78
79 #define MUL_FFT_MODF_THRESHOLD 530 /* k = 5 */
80 #define MUL_FFT_TABLE3 \
81 { { 530, 5}, { 29, 6}, { 15, 5}, { 31, 6}, \
82 { 29, 7}, { 15, 6}, { 33, 7}, { 17, 6}, \
83 { 36, 7}, { 19, 6}, { 39, 7}, { 21, 6}, \
84 { 43, 7}, { 23, 6}, { 47, 7}, { 29, 8}, \
85 { 15, 7}, { 35, 8}, { 19, 7}, { 43, 8}, \
86 { 23, 7}, { 49, 8}, { 27, 7}, { 55, 9}, \
87 { 15, 8}, { 31, 7}, { 63, 8}, { 43, 9}, \
88 { 23, 8}, { 51, 9}, { 31, 8}, { 67, 9}, \
89 { 39, 8}, { 83, 9}, { 47, 8}, { 95, 9}, \
90 { 55,10}, { 31, 9}, { 79,10}, { 47, 9}, \
91 { 95,11}, { 31,10}, { 63, 9}, { 135,10}, \
92 { 79, 9}, { 159,10}, { 95, 9}, { 191,10}, \
93 { 111,11}, { 63,10}, { 143, 9}, { 287,10}, \
94 { 159,11}, { 95,10}, { 191,12}, { 63,11}, \
95 { 127,10}, { 255, 9}, { 511,10}, { 271, 9}, \
96 { 543,10}, { 287,11}, { 159,10}, { 351,11}, \
97 { 191,10}, { 415,12}, { 127,11}, { 255,10}, \
98 { 543,11}, { 287,10}, { 607,11}, { 319,10}, \
99 { 671,11}, { 351,12}, { 191,11}, { 383,10}, \
100 { 799,11}, { 415,13}, { 127,12}, { 255,11}, \
101 { 543,10}, { 1087,11}, { 607,12}, { 319,11}, \
102 { 671,10}, { 1343,11}, { 735,12}, { 383,11}, \
103 { 799,10}, { 1599,11}, { 863,12}, { 447,11}, \
104 { 959,13}, { 255,12}, { 511,11}, { 1087,12}, \
105 { 575,11}, { 1215,12}, { 639,11}, { 1343,12}, \
106 { 703,13}, { 383,12}, { 767,11}, { 1599,12}, \
107 { 831,11}, { 1727,12}, { 959,14}, { 255,13}, \
108 { 511,12}, { 1087,11}, { 2239,12}, { 1215,11}, \
109 { 2431,13}, { 639,12}, { 1471,13}, { 767,12}, \
110 { 1727,13}, { 895,12}, { 1919,14}, { 511,13}, \
111 { 1023,12}, { 2239,13}, { 1151,12}, { 2431,13}, \
112 { 1279,12}, { 2623,13}, { 1407,12}, { 2815,14}, \
113 { 767,13}, { 1663,12}, { 3455,13}, { 1919,15}, \
114 { 511,14}, { 1023,13}, { 2175,12}, { 4479,13}, \
115 { 2431,14}, { 1279,13}, { 2943,12}, { 5887,14}, \
116 { 1535,13}, { 3455,14}, { 1791,13}, { 3967,15}, \
117 { 1023,14}, { 2047,13}, { 4479,14}, { 2303,13}, \
118 { 4991,12}, { 9983,14}, { 2815,13}, { 5887,15}, \
119 { 1535,14}, { 3839,16} }
120 #define MUL_FFT_TABLE3_SIZE 154
121 #define MUL_FFT_THRESHOLD 6784
122
123 #define SQR_FFT_MODF_THRESHOLD 460 /* k = 5 */
124 #define SQR_FFT_TABLE3 \
125 { { 460, 5}, { 29, 6}, { 15, 5}, { 31, 6}, \
126 { 29, 7}, { 15, 6}, { 33, 7}, { 17, 6}, \
127 { 36, 7}, { 19, 6}, { 39, 7}, { 29, 8}, \
128 { 15, 7}, { 35, 8}, { 19, 7}, { 41, 8}, \
129 { 23, 7}, { 49, 8}, { 27, 7}, { 55, 9}, \
130 { 15, 8}, { 31, 7}, { 63, 8}, { 43, 9}, \
131 { 23, 8}, { 55,10}, { 15, 9}, { 31, 8}, \
132 { 67, 9}, { 39, 8}, { 79, 9}, { 47, 8}, \
133 { 95, 9}, { 55,10}, { 31, 9}, { 79,10}, \
134 { 47, 9}, { 95,11}, { 31,10}, { 63, 9}, \
135 { 135,10}, { 79, 9}, { 159,10}, { 95,11}, \
136 { 63,10}, { 127, 9}, { 255,10}, { 143, 9}, \
137 { 287,10}, { 159,11}, { 95,12}, { 63,11}, \
138 { 127,10}, { 271, 9}, { 543,10}, { 287,11}, \
139 { 159,10}, { 319, 9}, { 639,10}, { 351,11}, \
140 { 191,10}, { 415,12}, { 127,11}, { 255,10}, \
141 { 543,11}, { 287,10}, { 575,11}, { 319,10}, \
142 { 671,11}, { 351,10}, { 703,12}, { 191,11}, \
143 { 383,10}, { 799,11}, { 415,10}, { 831,13}, \
144 { 127,12}, { 255,11}, { 511,10}, { 1023,11}, \
145 { 543,10}, { 1087,11}, { 607,12}, { 319,11}, \
146 { 671,10}, { 1343,11}, { 735,12}, { 383,11}, \
147 { 799,10}, { 1599,11}, { 863,12}, { 447,11}, \
148 { 927,13}, { 255,12}, { 511,11}, { 1087,12}, \
149 { 575,11}, { 1215,12}, { 639,11}, { 1343,12}, \
150 { 703,11}, { 1407,13}, { 383,12}, { 767,11}, \
151 { 1599,12}, { 831,11}, { 1727,12}, { 895,11}, \
152 { 1791,14}, { 255,13}, { 511,12}, { 1087,11}, \
153 { 2239,12}, { 1215,13}, { 639,12}, { 1471,13}, \
154 { 767,12}, { 1727,13}, { 895,12}, { 1919,14}, \
155 { 511,13}, { 1023,12}, { 2239,13}, { 1151,12}, \
156 { 2431,13}, { 1279,12}, { 2623,13}, { 1407,12}, \
157 { 2815,14}, { 767,13}, { 1663,12}, { 3455,13}, \
158 { 1919,15}, { 511,14}, { 1023,13}, { 2175,12}, \
159 { 4479,13}, { 2431,14}, { 1279,13}, { 2943,12}, \
160 { 5887,14}, { 1535,13}, { 3455,14}, { 1791,13}, \
161 { 3967,15}, { 1023,14}, { 2047,13}, { 4479,14}, \
162 { 2303,13}, { 4991,12}, { 9983,14}, { 2815,13}, \
163 { 5887,15}, { 1535,14}, { 3839,16} }
164 #define SQR_FFT_TABLE3_SIZE 155
165 #define SQR_FFT_THRESHOLD 5568
166
167 #define MULLO_BASECASE_THRESHOLD 0 /* always */
168 #define MULLO_DC_THRESHOLD 68
169 #define MULLO_MUL_N_THRESHOLD 13555
170 #define SQRLO_BASECASE_THRESHOLD 0 /* always */
171 #define SQRLO_DC_THRESHOLD 117
172 #define SQRLO_SQR_THRESHOLD 10988
173
174 #define DC_DIV_QR_THRESHOLD 42
175 #define DC_DIVAPPR_Q_THRESHOLD 163
176 #define DC_BDIV_QR_THRESHOLD 66
177 #define DC_BDIV_Q_THRESHOLD 160
178
179 #define INV_MULMOD_BNM1_THRESHOLD 46
180 #define INV_NEWTON_THRESHOLD 165
181 #define INV_APPR_THRESHOLD 157
182
183 #define BINV_NEWTON_THRESHOLD 300
184 #define REDC_1_TO_REDC_N_THRESHOLD 68
185
186 #define MU_DIV_QR_THRESHOLD 1718
187 #define MU_DIVAPPR_Q_THRESHOLD 1685
188 #define MUPI_DIV_QR_THRESHOLD 62
189 #define MU_BDIV_QR_THRESHOLD 1589
190 #define MU_BDIV_Q_THRESHOLD 1830
191
192 #define POWM_SEC_TABLE 1,17,129,547,1317
193
194 #define GET_STR_DC_THRESHOLD 10
195 #define GET_STR_PRECOMPUTE_THRESHOLD 16
196 #define SET_STR_DC_THRESHOLD 354
197 #define SET_STR_PRECOMPUTE_THRESHOLD 860
198
199 #define FAC_DSC_THRESHOLD 141
200 #define FAC_ODD_THRESHOLD 34
201
202 #define MATRIX22_STRASSEN_THRESHOLD 20
203 #define HGCD2_DIV1_METHOD 5 /* 1.04% faster than 3 */
204 #define HGCD_THRESHOLD 114
205 #define HGCD_APPR_THRESHOLD 132
206 #define HGCD_REDUCE_THRESHOLD 3524
207 #define GCD_DC_THRESHOLD 474
208 #define GCDEXT_DC_THRESHOLD 379
209 #define JACOBI_BASE_METHOD 1 /* 27.39% faster than 4 */
210
211 /* Tuneup completed successfully, took 31721 seconds */