1 /* 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 64
32 #define GMP_LIMB_BYTES 8
33
34 /* Disable use of slow functions. FIXME: We should disable lib inclusion. */
35 #undef HAVE_NATIVE_mpn_mul_2
36 #undef HAVE_NATIVE_mpn_addmul_2
37
38 /* 3600-4000 MHz Intel Xeon E3-1270v5 Skylake */
39 /* FFT tuning limit = 465,990,371 */
40 /* Generated by tuneup.c, 2019-10-18, gcc 8.3 */
41
42 #define MOD_1_NORM_THRESHOLD 0 /* always */
43 #define MOD_1_UNNORM_THRESHOLD 0 /* always */
44 #define MOD_1N_TO_MOD_1_1_THRESHOLD 4
45 #define MOD_1U_TO_MOD_1_1_THRESHOLD 3
46 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 13
47 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 32
48 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 9
49 #define USE_PREINV_DIVREM_1 1 /* native */
50 #define DIV_QR_1_NORM_THRESHOLD 1
51 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
52 #define DIV_QR_2_PI2_THRESHOLD 41
53 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
54 #define BMOD_1_TO_MOD_1_THRESHOLD 20
55
56 #define DIV_1_VS_MUL_1_PERCENT 473
57
58 #define MUL_TOOM22_THRESHOLD 26
59 #define MUL_TOOM33_THRESHOLD 73
60 #define MUL_TOOM44_THRESHOLD 208
61 #define MUL_TOOM6H_THRESHOLD 300
62 #define MUL_TOOM8H_THRESHOLD 406
63
64 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 73
65 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 153
66 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 137
67 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 151
68 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 106
69
70 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
71 #define SQR_TOOM2_THRESHOLD 32
72 #define SQR_TOOM3_THRESHOLD 117
73 #define SQR_TOOM4_THRESHOLD 336
74 #define SQR_TOOM6_THRESHOLD 426
75 #define SQR_TOOM8_THRESHOLD 547
76
77 #define MULMID_TOOM42_THRESHOLD 46
78
79 #define MULMOD_BNM1_THRESHOLD 15
80 #define SQRMOD_BNM1_THRESHOLD 17
81
82 #define MUL_FFT_MODF_THRESHOLD 404 /* k = 5 */
83 #define MUL_FFT_TABLE3 \
84 { { 404, 5}, { 21, 6}, { 11, 5}, { 23, 6}, \
85 { 12, 5}, { 25, 6}, { 28, 7}, { 15, 6}, \
86 { 31, 7}, { 21, 8}, { 11, 7}, { 25, 8}, \
87 { 13, 7}, { 28, 8}, { 15, 7}, { 32, 8}, \
88 { 17, 7}, { 35, 8}, { 19, 7}, { 39, 8}, \
89 { 21, 9}, { 11, 8}, { 23, 7}, { 47, 8}, \
90 { 27, 9}, { 15, 8}, { 35, 9}, { 19, 8}, \
91 { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \
92 { 15, 9}, { 39, 8}, { 79, 9}, { 43,10}, \
93 { 23, 9}, { 55,11}, { 15,10}, { 31, 9}, \
94 { 71,10}, { 39, 9}, { 83,10}, { 47, 9}, \
95 { 99,10}, { 55,11}, { 31,10}, { 79,11}, \
96 { 47,10}, { 103,12}, { 31,11}, { 63,10}, \
97 { 135,11}, { 79,10}, { 167,11}, { 95,10}, \
98 { 191, 9}, { 383,10}, { 199,11}, { 111,12}, \
99 { 63,11}, { 127,10}, { 255, 9}, { 511,10}, \
100 { 271, 9}, { 543,11}, { 143,10}, { 287, 9}, \
101 { 575,11}, { 159,12}, { 95,11}, { 191,10}, \
102 { 383,13}, { 63,12}, { 127,11}, { 255,10}, \
103 { 511,11}, { 271,10}, { 543,11}, { 287,10}, \
104 { 575,11}, { 303,12}, { 159,11}, { 319,10}, \
105 { 639,11}, { 335,10}, { 671,11}, { 351,10}, \
106 { 703,12}, { 191,11}, { 383,10}, { 767,11}, \
107 { 415,12}, { 223,11}, { 447,13}, { 127,12}, \
108 { 255,11}, { 543,12}, { 287,11}, { 607,12}, \
109 { 319,11}, { 671,12}, { 351,11}, { 703,13}, \
110 { 191,12}, { 383,11}, { 767,12}, { 415,11}, \
111 { 831,12}, { 479,14}, { 127,13}, { 255,12}, \
112 { 543,11}, { 1087,12}, { 607,13}, { 319,12}, \
113 { 671,11}, { 1343,12}, { 703,13}, { 383,12}, \
114 { 831,13}, { 447,12}, { 959,13}, { 511,12}, \
115 { 1087,13}, { 575,12}, { 1151,13}, { 639,12}, \
116 { 1343,13}, { 703,14}, { 383,13}, { 767,12}, \
117 { 1535,13}, { 831,12}, { 1727,13}, { 959,14}, \
118 { 511,13}, { 1087,12}, { 2175,13}, { 1151,14}, \
119 { 639,13}, { 1343,12}, { 2687,13}, { 1407,14}, \
120 { 767,13}, { 1599,12}, { 3199,13}, { 1663,14}, \
121 { 895,13}, { 1791,15}, { 511,14}, { 1023,13}, \
122 { 2175,14}, { 1151,13}, { 2431,12}, { 4863,14}, \
123 { 1279,13}, { 2687,14}, { 1407,13}, { 2815,15}, \
124 { 767,14}, { 1535,13}, { 3199,14}, { 1663,13}, \
125 { 3455,12}, { 6911,14}, { 1791,16}, { 511,15}, \
126 { 1023,14}, { 2175,13}, { 4351,14}, { 2431,13}, \
127 { 4863,15}, { 1279,14}, { 2943,13}, { 5887,15}, \
128 { 1535,14}, { 3455,13}, { 6911,15}, { 1791,14}, \
129 { 3839,13}, { 7679,16}, { 1023,15}, { 2047,14}, \
130 { 4223,15}, { 2303,14}, { 4863,15}, { 2559,14}, \
131 { 5247,15}, { 2815,14}, { 5887,16}, { 1535,15}, \
132 { 3327,14}, { 6911,15}, { 3839,14}, { 7679,17}, \
133 { 1023,16}, { 2047,15}, { 4351,14}, { 8703,15}, \
134 { 4863,16}, { 2559,15}, { 5887,14}, { 11775,16}, \
135 { 3071,15}, { 6911,16}, { 3583,15}, { 7679,14}, \
136 { 15359,15}, { 7935,17}, { 2047,16}, { 4095,15}, \
137 { 8703,16}, { 4607,15}, { 9983,14}, { 19967,16}, \
138 { 5631,15}, { 11775,17}, { 3071,16}, { 65536,17}, \
139 { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \
140 {2097152,22}, {4194304,23}, {8388608,24} }
141 #define MUL_FFT_TABLE3_SIZE 227
142 #define MUL_FFT_THRESHOLD 6272
143
144 #define SQR_FFT_MODF_THRESHOLD 400 /* k = 5 */
145 #define SQR_FFT_TABLE3 \
146 { { 400, 5}, { 25, 6}, { 13, 5}, { 27, 6}, \
147 { 29, 7}, { 15, 6}, { 31, 7}, { 28, 8}, \
148 { 15, 7}, { 32, 8}, { 17, 7}, { 35, 8}, \
149 { 19, 7}, { 39, 8}, { 27, 9}, { 15, 8}, \
150 { 33, 9}, { 19, 8}, { 39, 9}, { 23, 8}, \
151 { 47, 9}, { 27,10}, { 15, 9}, { 39,10}, \
152 { 23, 9}, { 51,11}, { 15,10}, { 31, 9}, \
153 { 67,10}, { 39, 9}, { 79,10}, { 55,11}, \
154 { 31,10}, { 79,11}, { 47,10}, { 95,12}, \
155 { 31,11}, { 63,10}, { 135,11}, { 79,10}, \
156 { 159, 9}, { 319,11}, { 95,12}, { 63,11}, \
157 { 127,10}, { 255, 9}, { 511,10}, { 271,11}, \
158 { 143,10}, { 287, 9}, { 575,10}, { 303,11}, \
159 { 159,10}, { 319,12}, { 95,10}, { 383,13}, \
160 { 63,12}, { 127,11}, { 255,10}, { 511,11}, \
161 { 271,10}, { 543,11}, { 287,10}, { 575,11}, \
162 { 303,10}, { 607,12}, { 159,11}, { 319,10}, \
163 { 639,11}, { 335,10}, { 671,11}, { 351,10}, \
164 { 703,11}, { 367,10}, { 735,11}, { 383,10}, \
165 { 767,11}, { 415,10}, { 831,12}, { 223,11}, \
166 { 479,13}, { 127,12}, { 255,11}, { 543,12}, \
167 { 287,11}, { 607,12}, { 319,11}, { 671,12}, \
168 { 351,11}, { 735,12}, { 383,11}, { 799,12}, \
169 { 415,11}, { 831,12}, { 479,14}, { 127,13}, \
170 { 255,12}, { 511,11}, { 1023,12}, { 607,13}, \
171 { 319,12}, { 735,13}, { 383,12}, { 831,13}, \
172 { 447,12}, { 959,13}, { 511,12}, { 1023,13}, \
173 { 575,12}, { 1151,13}, { 639,12}, { 1279,13}, \
174 { 703,12}, { 1407,14}, { 383,13}, { 767,12}, \
175 { 1535,13}, { 831,12}, { 1727,13}, { 895,12}, \
176 { 1791,13}, { 959,14}, { 511,13}, { 1087,12}, \
177 { 2175,13}, { 1151,14}, { 639,13}, { 1343,12}, \
178 { 2687,13}, { 1407,14}, { 767,13}, { 1599,12}, \
179 { 3199,13}, { 1663,14}, { 895,13}, { 1791,15}, \
180 { 511,14}, { 1023,13}, { 2175,14}, { 1151,13}, \
181 { 2431,12}, { 4863,14}, { 1279,13}, { 2687,14}, \
182 { 1407,15}, { 767,14}, { 1535,13}, { 3199,14}, \
183 { 1663,13}, { 3455,14}, { 1791,16}, { 511,15}, \
184 { 1023,14}, { 2431,13}, { 4863,15}, { 1279,14}, \
185 { 2943,13}, { 5887,15}, { 1535,14}, { 3455,15}, \
186 { 1791,14}, { 3839,16}, { 1023,15}, { 2047,14}, \
187 { 4223,15}, { 2303,14}, { 4863,15}, { 2559,14}, \
188 { 5119,15}, { 2815,14}, { 5887,16}, { 1535,15}, \
189 { 3071,14}, { 6143,15}, { 3327,14}, { 6911,15}, \
190 { 3839,17}, { 1023,16}, { 2047,15}, { 4863,16}, \
191 { 2559,15}, { 5887,14}, { 11775,16}, { 3071,15}, \
192 { 6911,16}, { 3583,15}, { 7679,14}, { 15359,17}, \
193 { 2047,16}, { 4095,15}, { 8191,16}, { 4607,15}, \
194 { 9983,14}, { 19967,16}, { 5631,15}, { 11775,17}, \
195 { 3071,16}, { 65536,17}, { 131072,18}, { 262144,19}, \
196 { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \
197 {8388608,24} }
198 #define SQR_FFT_TABLE3_SIZE 205
199 #define SQR_FFT_THRESHOLD 4224
200
201 #define MULLO_BASECASE_THRESHOLD 0 /* always */
202 #define MULLO_DC_THRESHOLD 79
203 #define MULLO_MUL_N_THRESHOLD 11278
204 #define SQRLO_BASECASE_THRESHOLD 10
205 #define SQRLO_DC_THRESHOLD 109
206 #define SQRLO_SQR_THRESHOLD 8207
207
208 #define DC_DIV_QR_THRESHOLD 55
209 #define DC_DIVAPPR_Q_THRESHOLD 179
210 #define DC_BDIV_QR_THRESHOLD 82
211 #define DC_BDIV_Q_THRESHOLD 166
212
213 #define INV_MULMOD_BNM1_THRESHOLD 50
214 #define INV_NEWTON_THRESHOLD 170
215 #define INV_APPR_THRESHOLD 171
216
217 #define BINV_NEWTON_THRESHOLD 294
218 #define REDC_1_TO_REDC_2_THRESHOLD 33
219 #define REDC_2_TO_REDC_N_THRESHOLD 59
220
221 #define MU_DIV_QR_THRESHOLD 1528
222 #define MU_DIVAPPR_Q_THRESHOLD 1589
223 #define MUPI_DIV_QR_THRESHOLD 62
224 #define MU_BDIV_QR_THRESHOLD 1470
225 #define MU_BDIV_Q_THRESHOLD 1597
226
227 #define POWM_SEC_TABLE 2,8,191,452,904
228
229 #define GET_STR_DC_THRESHOLD 12
230 #define GET_STR_PRECOMPUTE_THRESHOLD 19
231 #define SET_STR_DC_THRESHOLD 898
232 #define SET_STR_PRECOMPUTE_THRESHOLD 1670
233
234 #define FAC_DSC_THRESHOLD 474
235 #define FAC_ODD_THRESHOLD 0 /* always */
236
237 #define MATRIX22_STRASSEN_THRESHOLD 16
238 #define HGCD2_DIV1_METHOD 5 /* 3.85% faster than 3 */
239 #define HGCD_THRESHOLD 64
240 #define HGCD_APPR_THRESHOLD 60
241 #define HGCD_REDUCE_THRESHOLD 2681
242 #define GCD_DC_THRESHOLD 618
243 #define GCDEXT_DC_THRESHOLD 321
244 #define JACOBI_BASE_METHOD 1 /* 12.01% faster than 4 */
245
246 /* Tuneup completed successfully, took 213784 seconds */