xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/coreinhm/gmp-mparam.h (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* x86/coreinhm 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 /* 2933-3200 MHz Intel Xeon X3470 Nehalem */
35 /* FFT tuning limit = 67,000,000 */
36 /* Generated by tuneup.c, 2019-10-23, gcc 8.3 */
37 
38 #define MOD_1_NORM_THRESHOLD                36
39 #define MOD_1_UNNORM_THRESHOLD              40
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD          8
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD          7
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        12
43 #define MOD_1_2_TO_MOD_1_4_THRESHOLD         0  /* never mpn_mod_1s_2p */
44 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD      3
45 #define USE_PREINV_DIVREM_1                  1  /* native */
46 #define DIV_QR_1N_PI1_METHOD                 1  /* 42.59% faster than 2 */
47 #define DIV_QR_1_NORM_THRESHOLD          MP_SIZE_T_MAX  /* never */
48 #define DIV_QR_1_UNNORM_THRESHOLD        MP_SIZE_T_MAX  /* never */
49 #define DIV_QR_2_PI2_THRESHOLD               9
50 #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
51 #define BMOD_1_TO_MOD_1_THRESHOLD           17
52 
53 #define DIV_1_VS_MUL_1_PERCENT             288
54 
55 #define MUL_TOOM22_THRESHOLD                24
56 #define MUL_TOOM33_THRESHOLD                93
57 #define MUL_TOOM44_THRESHOLD               214
58 #define MUL_TOOM6H_THRESHOLD               306
59 #define MUL_TOOM8H_THRESHOLD               430
60 
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD      93
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     134
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD     145
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD      94
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     118
66 
67 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68 #define SQR_TOOM2_THRESHOLD                 38
69 #define SQR_TOOM3_THRESHOLD                133
70 #define SQR_TOOM4_THRESHOLD                212
71 #define SQR_TOOM6_THRESHOLD                318
72 #define SQR_TOOM8_THRESHOLD                620
73 
74 #define MULMID_TOOM42_THRESHOLD             68
75 
76 #define MULMOD_BNM1_THRESHOLD               17
77 #define SQRMOD_BNM1_THRESHOLD               23
78 
79 #define MUL_FFT_MODF_THRESHOLD             595  /* k = 5 */
80 #define MUL_FFT_TABLE3                                      \
81   { {    595, 5}, {     28, 6}, {     15, 5}, {     31, 6}, \
82     {     17, 5}, {     35, 6}, {     28, 7}, {     15, 6}, \
83     {     33, 7}, {     17, 6}, {     36, 7}, {     19, 6}, \
84     {     39, 7}, {     23, 6}, {     47, 7}, {     27, 8}, \
85     {     15, 7}, {     35, 8}, {     19, 7}, {     43, 8}, \
86     {     23, 7}, {     49, 8}, {     27, 9}, {     15, 8}, \
87     {     31, 7}, {     63, 8}, {     43, 9}, {     23, 8}, \
88     {     51, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
89     {     79, 9}, {     47, 8}, {     99, 9}, {     55,10}, \
90     {     31, 9}, {     63, 8}, {    127, 9}, {     79,10}, \
91     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
92     {    135,10}, {     79, 9}, {    159,10}, {     95, 9}, \
93     {    191,11}, {     63, 9}, {    255,10}, {    159,11}, \
94     {     95,10}, {    191,12}, {     63,11}, {    127,10}, \
95     {    255, 9}, {    511,10}, {    271, 9}, {    543,11}, \
96     {    159,10}, {    335,11}, {    191,10}, {    383, 9}, \
97     {    767,10}, {    399,12}, {    127,11}, {    255,10}, \
98     {    511, 9}, {   1023,10}, {    543, 9}, {   1087,11}, \
99     {    287,10}, {    607,11}, {    319,10}, {    671,12}, \
100     {    191,11}, {    383,10}, {    767,13}, {    127,12}, \
101     {    255,11}, {    511,10}, {   1023,11}, {    543,10}, \
102     {   1119,11}, {    607,12}, {    319,11}, {    671,10}, \
103     {   1343,11}, {    735,10}, {   1471,12}, {    383,11}, \
104     {    799,10}, {   1599,11}, {    863,10}, {   1727,12}, \
105     {    447,11}, {    959,13}, {    255,12}, {    511,11}, \
106     {   1119,12}, {    575,11}, {   1215,10}, {   2431,12}, \
107     {    639,11}, {   1343,12}, {    703,11}, {   1471,10}, \
108     {   2943,13}, {    383,12}, {    767,11}, {   1599,12}, \
109     {    831,11}, {   1727,10}, {   3455,12}, {    959,14}, \
110     {    255,13}, {    511,12}, {   1087,11}, {   2239,10}, \
111     {   4479,12}, {   1215,11}, {   2431,13}, {    639,12}, \
112     {   1471,11}, {   2943,13}, {    767,12}, {   1727,11}, \
113     {   3455,13}, {    895,12}, {   1983,14}, {    511,13}, \
114     {   1023,12}, {   2239,11}, {   4479,13}, {   1151,12}, \
115     {   2431,13}, {   1279,12}, {   2559,13}, {   1407,12}, \
116     {   2943,11}, {   5887,14}, {    767,13}, {   1663,12}, \
117     {   3455,13}, {   1919,12}, {   3839,15}, {    511,14}, \
118     {   1023,13}, {   2175,12}, {   4479,13}, {   2431,14}, \
119     {   1279,13}, {   2943,12}, {   5887,14}, {   1535,13}, \
120     {   3455,14}, {   1791,13}, {   3967,15}, {   1023,14}, \
121     {   2047,13}, {   4479,14}, {   2303,13}, {   4991,12}, \
122     {   9983,14}, {   2815,13}, {   6015,15}, {   1535,14}, \
123     {   3839,13}, {   7679,16} }
124 #define MUL_FFT_TABLE3_SIZE 170
125 #define MUL_FFT_THRESHOLD                 6784
126 
127 #define SQR_FFT_MODF_THRESHOLD             525  /* k = 5 */
128 #define SQR_FFT_TABLE3                                      \
129   { {    525, 5}, {     29, 6}, {     15, 5}, {     33, 6}, \
130     {     17, 5}, {     35, 6}, {     29, 7}, {     15, 6}, \
131     {     33, 7}, {     17, 6}, {     35, 7}, {     19, 6}, \
132     {     39, 7}, {     23, 6}, {     47, 7}, {     29, 8}, \
133     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
134     {     23, 7}, {     49, 8}, {     27, 7}, {     55, 9}, \
135     {     15, 8}, {     31, 7}, {     63, 8}, {     39, 9}, \
136     {     23, 8}, {     55, 9}, {     31, 8}, {     67, 9}, \
137     {     39, 8}, {     79, 9}, {     47, 8}, {     95, 9}, \
138     {     55,10}, {     31, 9}, {     79,10}, {     47, 9}, \
139     {     95,11}, {     31,10}, {     63, 9}, {    135,10}, \
140     {     79, 9}, {    159,10}, {     95,11}, {     63,10}, \
141     {    143, 9}, {    287,10}, {    159, 6}, {   2687, 7}, \
142     {   1407, 9}, {    367, 8}, {    735, 9}, {    383,10}, \
143     {    207, 9}, {    415,11}, {    127,10}, {    271, 9}, \
144     {    543,10}, {    287,11}, {    159,10}, {    319, 9}, \
145     {    639,10}, {    335, 9}, {    671,10}, {    351,11}, \
146     {    191,10}, {    383, 9}, {    767,10}, {    399, 9}, \
147     {    799,10}, {    415,12}, {    127,11}, {    255,10}, \
148     {    511, 9}, {   1023,10}, {    543,11}, {    287,10}, \
149     {    607,11}, {    319,10}, {    671,11}, {    351,12}, \
150     {    191,11}, {    383,10}, {    799,11}, {    415,13}, \
151     {    127,12}, {    255,11}, {    511,10}, {   1023,11}, \
152     {    543,10}, {   1087,11}, {    607,10}, {   1215,12}, \
153     {    319,11}, {    671,10}, {   1343,11}, {    735,10}, \
154     {   1471,12}, {    383,11}, {    799,10}, {   1599,11}, \
155     {    863,10}, {   1727,12}, {    447,11}, {    991,10}, \
156     {   1983,13}, {    255,12}, {    511,11}, {   1119,12}, \
157     {    575,11}, {   1215,10}, {   2431,12}, {    639,11}, \
158     {   1343,12}, {    703,11}, {   1471,13}, {    383,12}, \
159     {    767,11}, {   1599,12}, {    831,11}, {   1727,10}, \
160     {   3455,12}, {    895,11}, {   1791,12}, {    959,11}, \
161     {   1983,14}, {    255,13}, {    511,12}, {   1023,11}, \
162     {   2047,12}, {   1087,11}, {   2239,12}, {   1215,11}, \
163     {   2431,13}, {    639,12}, {   1471,11}, {   2943,13}, \
164     {    767,12}, {   1727,11}, {   3455,13}, {    895,12}, \
165     {   1983,11}, {   3967,14}, {    511,13}, {   1023,12}, \
166     {   2239,13}, {   1151,12}, {   2495,13}, {   1279,12}, \
167     {   2623,13}, {   1407,12}, {   2943,14}, {    767,13}, \
168     {   1663,12}, {   3455,13}, {   1919,12}, {   3967,15}, \
169     {    511,14}, {   1023,13}, {   2175,12}, {   4479,13}, \
170     {   2431,12}, {   4863,14}, {   1279,13}, {   2943,12}, \
171     {   5887,14}, {   1535,13}, {   3455,14}, {   1791,13}, \
172     {   3967,12}, {   7935,15}, {   1023,14}, {   2047,13}, \
173     {   4479,14}, {   2303,13}, {   4991,12}, {   9983,14}, \
174     {   2815,13}, {   5887,15}, {   1535,14}, {   3327,13}, \
175     {   6655,14}, {   3839,13}, {   7935,16} }
176 #define SQR_FFT_TABLE3_SIZE 187
177 #define SQR_FFT_THRESHOLD                 5312
178 
179 #define MULLO_BASECASE_THRESHOLD             0  /* always */
180 #define MULLO_DC_THRESHOLD                  43
181 #define MULLO_MUL_N_THRESHOLD            13463
182 #define SQRLO_BASECASE_THRESHOLD             9
183 #define SQRLO_DC_THRESHOLD                  42
184 #define SQRLO_SQR_THRESHOLD              10323
185 
186 #define DC_DIV_QR_THRESHOLD                 43
187 #define DC_DIVAPPR_Q_THRESHOLD             132
188 #define DC_BDIV_QR_THRESHOLD                83
189 #define DC_BDIV_Q_THRESHOLD                130
190 
191 #define INV_MULMOD_BNM1_THRESHOLD           46
192 #define INV_NEWTON_THRESHOLD               189
193 #define INV_APPR_THRESHOLD                 167
194 
195 #define BINV_NEWTON_THRESHOLD              372
196 #define REDC_1_TO_REDC_N_THRESHOLD          83
197 
198 #define MU_DIV_QR_THRESHOLD               1589
199 #define MU_DIVAPPR_Q_THRESHOLD            1589
200 #define MUPI_DIV_QR_THRESHOLD               97
201 #define MU_BDIV_QR_THRESHOLD              1589
202 #define MU_BDIV_Q_THRESHOLD               1718
203 
204 #define POWM_SEC_TABLE  1,28,96,473,803
205 
206 #define GET_STR_DC_THRESHOLD                12
207 #define GET_STR_PRECOMPUTE_THRESHOLD        16
208 #define SET_STR_DC_THRESHOLD               145
209 #define SET_STR_PRECOMPUTE_THRESHOLD       419
210 
211 #define FAC_DSC_THRESHOLD                  114
212 #define FAC_ODD_THRESHOLD                   57
213 
214 #define MATRIX22_STRASSEN_THRESHOLD         20
215 #define HGCD2_DIV1_METHOD                    1  /* 1.03% faster than 3 */
216 #define HGCD_THRESHOLD                     117
217 #define HGCD_APPR_THRESHOLD                137
218 #define HGCD_REDUCE_THRESHOLD             3524
219 #define GCD_DC_THRESHOLD                   389
220 #define GCDEXT_DC_THRESHOLD                318
221 #define JACOBI_BASE_METHOD                   4  /* 6.10% faster than 1 */
222 
223 /* Tuneup completed successfully, took 67994 seconds */
224