1 /* Pentium 4-64 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 /* These routines exists for all x86_64 chips, but they are slower on Pentium4
35 than separate add/sub and shift. Make sure they are not really used. */
36 #undef HAVE_NATIVE_mpn_rsblsh1_n
37 #undef HAVE_NATIVE_mpn_rsblsh2_n
38 #undef HAVE_NATIVE_mpn_addlsh_n
39 #undef HAVE_NATIVE_mpn_rsblsh_n
40
41 /* 3400 MHz Pentium4 Nocona / 1024 Kibyte L2 cache */
42 /* FFT tuning limit = 107,095,964 */
43 /* Generated by tuneup.c, 2019-11-09, gcc 8.3 */
44
45 #define MOD_1_NORM_THRESHOLD 0 /* always */
46 #define MOD_1_UNNORM_THRESHOLD 0 /* always */
47 #define MOD_1N_TO_MOD_1_1_THRESHOLD 5
48 #define MOD_1U_TO_MOD_1_1_THRESHOLD 3
49 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 14
50 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 32
51 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 11
52 #define USE_PREINV_DIVREM_1 1 /* native */
53 #define DIV_QR_1_NORM_THRESHOLD 2
54 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
55 #define DIV_QR_2_PI2_THRESHOLD 12
56 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
57 #define BMOD_1_TO_MOD_1_THRESHOLD 20
58
59 #define DIV_1_VS_MUL_1_PERCENT 228
60
61 #define MUL_TOOM22_THRESHOLD 12
62 #define MUL_TOOM33_THRESHOLD 81
63 #define MUL_TOOM44_THRESHOLD 130
64 #define MUL_TOOM6H_THRESHOLD 173
65 #define MUL_TOOM8H_THRESHOLD 430
66
67 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 81
68 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 91
69 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 89
70 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 88
71 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 112
72
73 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
74 #define SQR_TOOM2_THRESHOLD 18
75 #define SQR_TOOM3_THRESHOLD 113
76 #define SQR_TOOM4_THRESHOLD 202
77 #define SQR_TOOM6_THRESHOLD 238
78 #define SQR_TOOM8_THRESHOLD 430
79
80 #define MULMID_TOOM42_THRESHOLD 20
81
82 #define MULMOD_BNM1_THRESHOLD 9
83 #define SQRMOD_BNM1_THRESHOLD 11
84
85 #define MUL_FFT_MODF_THRESHOLD 236 /* k = 5 */
86 #define MUL_FFT_TABLE3 \
87 { { 236, 5}, { 11, 6}, { 6, 5}, { 13, 6}, \
88 { 9, 5}, { 19, 6}, { 17, 7}, { 9, 6}, \
89 { 19, 7}, { 10, 6}, { 21, 7}, { 11, 6}, \
90 { 23, 7}, { 13, 8}, { 7, 7}, { 17, 8}, \
91 { 9, 7}, { 21, 8}, { 11, 7}, { 23, 8}, \
92 { 13, 9}, { 7, 8}, { 15, 7}, { 31, 8}, \
93 { 21, 9}, { 11, 8}, { 27,10}, { 7, 9}, \
94 { 15, 8}, { 33, 9}, { 19, 8}, { 39, 9}, \
95 { 23, 8}, { 47, 9}, { 27,10}, { 15, 9}, \
96 { 39,10}, { 23, 9}, { 51,11}, { 15,10}, \
97 { 31, 9}, { 67,10}, { 39, 9}, { 83,10}, \
98 { 47, 9}, { 95,10}, { 55,11}, { 31,10}, \
99 { 63, 9}, { 127, 8}, { 255,10}, { 71, 9}, \
100 { 143, 8}, { 287,10}, { 79,11}, { 47,10}, \
101 { 95, 9}, { 191,12}, { 31,11}, { 63,10}, \
102 { 127, 9}, { 255,10}, { 143, 9}, { 287,11}, \
103 { 79,10}, { 159, 9}, { 319,10}, { 175,11}, \
104 { 95,10}, { 191, 9}, { 383,10}, { 223,12}, \
105 { 63,11}, { 127,10}, { 255,11}, { 143,10}, \
106 { 287,11}, { 159,10}, { 319,11}, { 175,12}, \
107 { 95,11}, { 191,10}, { 383,11}, { 223,13}, \
108 { 63,12}, { 127,11}, { 255,10}, { 511,11}, \
109 { 287,10}, { 575,12}, { 159,11}, { 351,12}, \
110 { 191,11}, { 383,12}, { 223,11}, { 447,13}, \
111 { 127,12}, { 255,11}, { 511,12}, { 287,11}, \
112 { 575,10}, { 1151,12}, { 351,13}, { 191,12}, \
113 { 415,11}, { 831,10}, { 1663,12}, { 447,14}, \
114 { 127,13}, { 255,12}, { 511,11}, { 1023,12}, \
115 { 543,11}, { 1087,10}, { 2175,12}, { 575,11}, \
116 { 1151,13}, { 319,12}, { 639,11}, { 1279,12}, \
117 { 671,11}, { 1343,12}, { 703,13}, { 383,12}, \
118 { 767,11}, { 1535,12}, { 831,11}, { 1663,13}, \
119 { 447,14}, { 255,13}, { 511,12}, { 1023,11}, \
120 { 2047,12}, { 1087,11}, { 2175,13}, { 575,12}, \
121 { 1151,11}, { 2303,12}, { 1215,11}, { 2431,10}, \
122 { 4863,13}, { 639,12}, { 1279,11}, { 2559,12}, \
123 { 1343,13}, { 703,14}, { 383,13}, { 767,12}, \
124 { 1535,13}, { 831,12}, { 1663,15}, { 255,14}, \
125 { 511,13}, { 1023,12}, { 2047,13}, { 1087,12}, \
126 { 2175,13}, { 1151,12}, { 2303,13}, { 1215,12}, \
127 { 2431,11}, { 4863,14}, { 639,13}, { 1279,12}, \
128 { 2559,13}, { 1343,12}, { 2687,13}, { 1407,12}, \
129 { 2815,13}, { 1471,14}, { 767,13}, { 1663,14}, \
130 { 895,13}, { 1791,12}, { 3583,13}, { 1919,12}, \
131 { 3839,15}, { 511,14}, { 1023,13}, { 2175,14}, \
132 { 1151,13}, { 2303,12}, { 4607,13}, { 2431,12}, \
133 { 4863,14}, { 1279,13}, { 2687,14}, { 1407,13}, \
134 { 2815,15}, { 767,14}, { 1791,13}, { 3583,14}, \
135 { 1919,13}, { 3839,16}, { 511,15}, { 1023,14}, \
136 { 2175,13}, { 4351,14}, { 2303,13}, { 4607,14}, \
137 { 2431,13}, { 4863,15}, { 1279,14}, { 2943,13}, \
138 { 5887,15}, { 1535,14}, { 3199,15}, { 1791,14}, \
139 { 3839,13}, { 7679,16}, { 1023,15}, { 2047,14}, \
140 { 4351,15}, { 2303,14}, { 4863,15}, { 2815,14}, \
141 { 5887,16}, { 1535,15}, { 3071,14}, { 6143,15}, \
142 { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \
143 { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \
144 {8388608,24} }
145 #define MUL_FFT_TABLE3_SIZE 229
146 #define MUL_FFT_THRESHOLD 2752
147
148 #define SQR_FFT_MODF_THRESHOLD 240 /* k = 5 */
149 #define SQR_FFT_TABLE3 \
150 { { 240, 5}, { 11, 6}, { 6, 5}, { 13, 6}, \
151 { 9, 5}, { 19, 6}, { 17, 7}, { 9, 6}, \
152 { 23, 7}, { 12, 6}, { 25, 7}, { 13, 8}, \
153 { 7, 7}, { 17, 8}, { 9, 7}, { 21, 8}, \
154 { 11, 7}, { 24, 8}, { 13, 9}, { 7, 8}, \
155 { 15, 7}, { 31, 8}, { 21, 9}, { 11, 8}, \
156 { 27,10}, { 7, 9}, { 15, 8}, { 33, 9}, \
157 { 19, 8}, { 39, 9}, { 27,10}, { 15, 9}, \
158 { 39,10}, { 23, 9}, { 47,11}, { 15,10}, \
159 { 31, 9}, { 63,10}, { 39, 9}, { 79,10}, \
160 { 55,11}, { 31,10}, { 63, 9}, { 127, 8}, \
161 { 255,10}, { 71, 9}, { 143, 8}, { 287,10}, \
162 { 79,11}, { 47,10}, { 95, 9}, { 191,12}, \
163 { 31,11}, { 63,10}, { 127, 9}, { 255,10}, \
164 { 143, 9}, { 287,11}, { 79,10}, { 159, 9}, \
165 { 319,10}, { 175, 9}, { 351,11}, { 95,10}, \
166 { 191, 9}, { 383,10}, { 207, 9}, { 415,10}, \
167 { 223,12}, { 63,11}, { 127,10}, { 255,11}, \
168 { 143,10}, { 287,11}, { 159,10}, { 319,11}, \
169 { 175,10}, { 351,12}, { 95,11}, { 191,10}, \
170 { 383,11}, { 207,10}, { 415,11}, { 223,13}, \
171 { 63,12}, { 127,11}, { 255,10}, { 511,11}, \
172 { 287,10}, { 575,12}, { 159,11}, { 319,10}, \
173 { 639,11}, { 351,12}, { 191,11}, { 383,10}, \
174 { 767,12}, { 223,11}, { 447,13}, { 127,12}, \
175 { 255,11}, { 511,12}, { 287,11}, { 575,10}, \
176 { 1151,12}, { 319,11}, { 639,12}, { 351,13}, \
177 { 191,12}, { 383,11}, { 767,12}, { 415,11}, \
178 { 831,12}, { 447,14}, { 127,13}, { 255,12}, \
179 { 511,11}, { 1023,12}, { 543,11}, { 1087,12}, \
180 { 575,11}, { 1151,13}, { 319,12}, { 639,11}, \
181 { 1279,12}, { 671,11}, { 1343,13}, { 383,12}, \
182 { 767,11}, { 1535,12}, { 831,13}, { 447,14}, \
183 { 255,13}, { 511,12}, { 1023,11}, { 2047,12}, \
184 { 1087,13}, { 575,12}, { 1151,11}, { 2303,12}, \
185 { 1215,11}, { 2431,10}, { 4863,13}, { 639,12}, \
186 { 1279,11}, { 2559,12}, { 1343,11}, { 2687,14}, \
187 { 383,13}, { 767,12}, { 1535,13}, { 831,12}, \
188 { 1663,15}, { 255,14}, { 511,13}, { 1023,12}, \
189 { 2047,13}, { 1087,12}, { 2175,13}, { 1151,12}, \
190 { 2303,13}, { 1215,12}, { 2431,11}, { 4863,14}, \
191 { 639,13}, { 1279,12}, { 2559,13}, { 1343,12}, \
192 { 2687,13}, { 1407,12}, { 2815,13}, { 1471,14}, \
193 { 767,13}, { 1663,14}, { 895,13}, { 1791,12}, \
194 { 3583,13}, { 1919,12}, { 3839,15}, { 511,14}, \
195 { 1023,13}, { 2175,14}, { 1151,13}, { 2303,12}, \
196 { 4607,13}, { 2431,12}, { 4863,14}, { 1279,13}, \
197 { 2687,14}, { 1407,13}, { 2943,15}, { 767,14}, \
198 { 1663,13}, { 3327,14}, { 1791,13}, { 3583,14}, \
199 { 1919,13}, { 3839,16}, { 511,15}, { 1023,14}, \
200 { 2175,13}, { 4351,14}, { 2303,13}, { 4607,14}, \
201 { 2431,13}, { 4863,15}, { 1279,14}, { 2815,13}, \
202 { 5631,14}, { 2943,13}, { 5887,15}, { 1535,14}, \
203 { 3327,15}, { 1791,14}, { 3839,13}, { 7679,16}, \
204 { 1023,15}, { 2047,14}, { 4351,15}, { 2303,14}, \
205 { 4863,15}, { 2815,14}, { 5887,16}, { 1535,15}, \
206 { 3071,14}, { 6143,15}, { 32768,16}, { 65536,17}, \
207 { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \
208 {2097152,22}, {4194304,23}, {8388608,24} }
209 #define SQR_FFT_TABLE3_SIZE 235
210 #define SQR_FFT_THRESHOLD 2368
211
212 #define MULLO_BASECASE_THRESHOLD 0 /* always */
213 #define MULLO_DC_THRESHOLD 45
214 #define MULLO_MUL_N_THRESHOLD 5397
215 #define SQRLO_BASECASE_THRESHOLD 6
216 #define SQRLO_DC_THRESHOLD 46
217 #define SQRLO_SQR_THRESHOLD 4658
218
219 #define DC_DIV_QR_THRESHOLD 36
220 #define DC_DIVAPPR_Q_THRESHOLD 95
221 #define DC_BDIV_QR_THRESHOLD 35
222 #define DC_BDIV_Q_THRESHOLD 47
223
224 #define INV_MULMOD_BNM1_THRESHOLD 22
225 #define INV_NEWTON_THRESHOLD 178
226 #define INV_APPR_THRESHOLD 116
227
228 #define BINV_NEWTON_THRESHOLD 206
229 #define REDC_1_TO_REDC_2_THRESHOLD 24
230 #define REDC_2_TO_REDC_N_THRESHOLD 50
231
232 #define MU_DIV_QR_THRESHOLD 979
233 #define MU_DIVAPPR_Q_THRESHOLD 979
234 #define MUPI_DIV_QR_THRESHOLD 97
235 #define MU_BDIV_QR_THRESHOLD 762
236 #define MU_BDIV_Q_THRESHOLD 942
237
238 #define POWM_SEC_TABLE 7,34,114,523,1486
239
240 #define GET_STR_DC_THRESHOLD 13
241 #define GET_STR_PRECOMPUTE_THRESHOLD 25
242 #define SET_STR_DC_THRESHOLD 381
243 #define SET_STR_PRECOMPUTE_THRESHOLD 1659
244
245 #define FAC_DSC_THRESHOLD 969
246 #define FAC_ODD_THRESHOLD 0 /* always */
247
248 #define MATRIX22_STRASSEN_THRESHOLD 29
249 #define HGCD2_DIV1_METHOD 3 /* 2.03% faster than 5 */
250 #define HGCD_THRESHOLD 92
251 #define HGCD_APPR_THRESHOLD 95
252 #define HGCD_REDUCE_THRESHOLD 1815
253 #define GCD_DC_THRESHOLD 195
254 #define GCDEXT_DC_THRESHOLD 233
255 #define JACOBI_BASE_METHOD 4 /* 17.06% faster than 1 */
256
257 /* Tuneup completed successfully, took 297016 seconds */