xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/coreisbr/gmp-mparam.h (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* x86/coreisbr 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 /* 3400-3800 MHz Intel Xeon E3-1270 Sandy Bridge */
35 /* FFT tuning limit = 67,000,000 */
36 /* Generated by tuneup.c, 2019-10-24, gcc 8.3 */
37 
38 #define MOD_1_NORM_THRESHOLD                28
39 #define MOD_1_UNNORM_THRESHOLD              26
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD          9
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD          7
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        11
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      4
45 #define USE_PREINV_DIVREM_1                  1  /* native */
46 #define DIV_QR_1N_PI1_METHOD                 2  /* 88.29% faster than 1 */
47 #define DIV_QR_1_NORM_THRESHOLD             21
48 #define DIV_QR_1_UNNORM_THRESHOLD        MP_SIZE_T_MAX  /* never */
49 #define DIV_QR_2_PI2_THRESHOLD              14
50 #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
51 #define BMOD_1_TO_MOD_1_THRESHOLD           20
52 
53 #define DIV_1_VS_MUL_1_PERCENT             297
54 
55 #define MUL_TOOM22_THRESHOLD                32
56 #define MUL_TOOM33_THRESHOLD               105
57 #define MUL_TOOM44_THRESHOLD               190
58 #define MUL_TOOM6H_THRESHOLD               294
59 #define MUL_TOOM8H_THRESHOLD               478
60 
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD     109
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     144
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD     116
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD     129
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     160
66 
67 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68 #define SQR_TOOM2_THRESHOLD                 48
69 #define SQR_TOOM3_THRESHOLD                163
70 #define SQR_TOOM4_THRESHOLD                250
71 #define SQR_TOOM6_THRESHOLD                354
72 #define SQR_TOOM8_THRESHOLD                502
73 
74 #define MULMID_TOOM42_THRESHOLD             98
75 
76 #define MULMOD_BNM1_THRESHOLD               19
77 #define SQRMOD_BNM1_THRESHOLD               23
78 
79 #define MUL_FFT_MODF_THRESHOLD             666  /* k = 5 */
80 #define MUL_FFT_TABLE3                                      \
81   { {    666, 5}, {     28, 6}, {     15, 5}, {     31, 6}, \
82     {     28, 7}, {     15, 6}, {     33, 7}, {     17, 6}, \
83     {     36, 7}, {     19, 6}, {     39, 7}, {     23, 6}, \
84     {     47, 7}, {     29, 8}, {     15, 7}, {     35, 8}, \
85     {     19, 7}, {     41, 8}, {     23, 7}, {     49, 8}, \
86     {     27, 7}, {     55, 8}, {     31, 7}, {     63, 8}, \
87     {     43, 9}, {     23, 8}, {     55, 9}, {     31, 8}, \
88     {     71, 9}, {     39, 8}, {     79, 9}, {     47, 8}, \
89     {     99, 9}, {     55,10}, {     31, 9}, {     79,10}, \
90     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
91     {    135,10}, {     79, 9}, {    159,10}, {     95, 9}, \
92     {    191,11}, {     63,10}, {    159, 7}, {   1343, 8}, \
93     {    703, 9}, {    367, 8}, {    735, 9}, {    383,10}, \
94     {    207,11}, {    127,10}, {    255, 9}, {    511,10}, \
95     {    271, 9}, {    543,10}, {    287,11}, {    159,10}, \
96     {    319, 9}, {    639,10}, {    335,11}, {    191,10}, \
97     {    383, 9}, {    767,11}, {    223,12}, {    127,11}, \
98     {    255,10}, {    543,11}, {    287,10}, {    607, 9}, \
99     {   1215,11}, {    319,10}, {    671,12}, {    191,11}, \
100     {    383,10}, {    799,13}, {    127,12}, {    255,11}, \
101     {    511,10}, {   1023,11}, {    543,10}, {   1087,11}, \
102     {    607,10}, {   1215,12}, {    319,11}, {    671,10}, \
103     {   1343,11}, {    735,10}, {   1471,12}, {    383,11}, \
104     {    799,10}, {   1599,11}, {    863,12}, {    447,11}, \
105     {    959,13}, {    255,12}, {    511,11}, {   1087,12}, \
106     {    575,11}, {   1215,12}, {    639,11}, {   1343,12}, \
107     {    703,11}, {   1471,13}, {    383,12}, {    767,11}, \
108     {   1599,12}, {    831,11}, {   1727,12}, {    959,14}, \
109     {    255,13}, {    511,12}, {   1087,11}, {   2239,12}, \
110     {   1215,13}, {    639,12}, {   1471,11}, {   2943,13}, \
111     {    767,12}, {   1727,11}, {   3455,13}, {    895,12}, \
112     {   1983,14}, {    511,13}, {   1023,12}, {   2239,13}, \
113     {   1151,12}, {   2495,13}, {   1279,12}, {   2623,13}, \
114     {   1407,12}, {   2943,14}, {    767,13}, {   1535,12}, \
115     {   3071,13}, {   1663,12}, {   3455,13}, {   1919,15}, \
116     {    511,14}, {   1023,13}, {   2175,12}, {   4479,13}, \
117     {   2431,14}, {   1279,13}, {   2943,12}, {   5887,14}, \
118     {   1535,13}, {   3455,14}, {   1791,13}, {   3967,15}, \
119     {   1023,14}, {   2047,13}, {   4479,14}, {   2303,13}, \
120     {   4991,12}, {   9983,14}, {   2815,13}, {   5887,15}, \
121     {   1535,14}, {   3839,13}, {   7679,16} }
122 #define MUL_FFT_TABLE3_SIZE 163
123 #define MUL_FFT_THRESHOLD                 7552
124 
125 #define SQR_FFT_MODF_THRESHOLD             570  /* k = 5 */
126 #define SQR_FFT_TABLE3                                      \
127   { {    570, 5}, {     28, 6}, {     15, 5}, {     32, 6}, \
128     {     17, 5}, {     35, 6}, {     29, 7}, {     15, 6}, \
129     {     33, 7}, {     17, 6}, {     36, 7}, {     19, 6}, \
130     {     40, 7}, {     23, 6}, {     47, 7}, {     29, 8}, \
131     {     15, 7}, {     35, 8}, {     19, 7}, {     43, 8}, \
132     {     23, 7}, {     49, 8}, {     27, 7}, {     55, 8}, \
133     {     31, 7}, {     63, 8}, {     43, 9}, {     23, 8}, \
134     {     55, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
135     {     79, 9}, {     47, 8}, {     95, 9}, {     55,10}, \
136     {     31, 9}, {     79,10}, {     47, 9}, {     95,11}, \
137     {     31,10}, {     63, 9}, {    135,10}, {     79, 9}, \
138     {    159,10}, {     95,11}, {     63,10}, {    159,11}, \
139     {     95,10}, {    191,12}, {     63, 8}, {   1023, 9}, \
140     {    543,11}, {    159,10}, {    319, 9}, {    639,10}, \
141     {    335,11}, {    191,10}, {    383, 9}, {    767,10}, \
142     {    399, 9}, {    799,12}, {    127,11}, {    255,10}, \
143     {    511, 9}, {   1023,10}, {    543,11}, {    287,10}, \
144     {    607,11}, {    319,10}, {    671,11}, {    351,12}, \
145     {    191,11}, {    383,10}, {    799,13}, {    127,12}, \
146     {    255,11}, {    511,10}, {   1023,11}, {    543,10}, \
147     {   1087,11}, {    607,12}, {    319,11}, {    671,10}, \
148     {   1343,11}, {    735,10}, {   1471,12}, {    383,11}, \
149     {    799,10}, {   1599,11}, {    863,12}, {    447,11}, \
150     {    991,13}, {    255,12}, {    511,11}, {   1087,12}, \
151     {    575,11}, {   1215,12}, {    639,11}, {   1343,12}, \
152     {    703,11}, {   1471,13}, {    383,12}, {    767,11}, \
153     {   1599,12}, {    831,11}, {   1727,12}, {    959,11}, \
154     {   1919,14}, {    255,13}, {    511,12}, {   1023,11}, \
155     {   2047,12}, {   1087,11}, {   2239,12}, {   1215,11}, \
156     {   2431,13}, {    639,12}, {   1471,13}, {    767,12}, \
157     {   1727,13}, {    895,12}, {   1983,14}, {    511,13}, \
158     {   1023,12}, {   2239,13}, {   1151,12}, {   2495,13}, \
159     {   1279,12}, {   2623,13}, {   1407,12}, {   2943,14}, \
160     {    767,13}, {   1663,12}, {   3455,13}, {   1919,12}, \
161     {   3967,15}, {    511,14}, {   1023,13}, {   2175,12}, \
162     {   4479,13}, {   2431,12}, {   4863,14}, {   1279,13}, \
163     {   2943,12}, {   5887,14}, {   1535,13}, {   3455,14}, \
164     {   1791,13}, {   3967,15}, {   1023,14}, {   2047,13}, \
165     {   4479,14}, {   2303,13}, {   4991,12}, {   9983,14}, \
166     {   2559,13}, {   5119,14}, {   2815,13}, {   5887,15}, \
167     {   1535,14}, {   3839,13}, {   7679,16} }
168 #define SQR_FFT_TABLE3_SIZE 163
169 #define SQR_FFT_THRESHOLD                 5760
170 
171 #define MULLO_BASECASE_THRESHOLD            16
172 #define MULLO_DC_THRESHOLD                  46
173 #define MULLO_MUL_N_THRESHOLD            14281
174 #define SQRLO_BASECASE_THRESHOLD             0  /* always */
175 #define SQRLO_DC_THRESHOLD                 159
176 #define SQRLO_SQR_THRESHOLD              11317
177 
178 #define DC_DIV_QR_THRESHOLD                 47
179 #define DC_DIVAPPR_Q_THRESHOLD             191
180 #define DC_BDIV_QR_THRESHOLD               107
181 #define DC_BDIV_Q_THRESHOLD                232
182 
183 #define INV_MULMOD_BNM1_THRESHOLD           62
184 #define INV_NEWTON_THRESHOLD               181
185 #define INV_APPR_THRESHOLD                 182
186 
187 #define BINV_NEWTON_THRESHOLD              378
188 #define REDC_1_TO_REDC_N_THRESHOLD          91
189 
190 #define MU_DIV_QR_THRESHOLD               1858
191 #define MU_DIVAPPR_Q_THRESHOLD            1858
192 #define MUPI_DIV_QR_THRESHOLD               77
193 #define MU_BDIV_QR_THRESHOLD              1830
194 #define MU_BDIV_Q_THRESHOLD               2166
195 
196 #define POWM_SEC_TABLE  1,16,126,428,1442
197 
198 #define GET_STR_DC_THRESHOLD                10
199 #define GET_STR_PRECOMPUTE_THRESHOLD        16
200 #define SET_STR_DC_THRESHOLD               418
201 #define SET_STR_PRECOMPUTE_THRESHOLD      1104
202 
203 #define FAC_DSC_THRESHOLD                  149
204 #define FAC_ODD_THRESHOLD                   34
205 
206 #define MATRIX22_STRASSEN_THRESHOLD         21
207 #define HGCD2_DIV1_METHOD                    1  /* 5.54% faster than 4 */
208 #define HGCD_THRESHOLD                      66
209 #define HGCD_APPR_THRESHOLD                135
210 #define HGCD_REDUCE_THRESHOLD             4284
211 #define GCD_DC_THRESHOLD                   642
212 #define GCDEXT_DC_THRESHOLD                465
213 #define JACOBI_BASE_METHOD                   3  /* 14.76% faster than 4 */
214 
215 /* Tuneup completed successfully, took 44241 seconds */
216