xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/core2/gmp-mparam.h (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* x86/core2 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 /* 3000 MHz Penryn */
35 /* FFT tuning limit = 67,000,000 */
36 /* Generated by tuneup.c, 2019-10-20, gcc 8.3 */
37 
38 #define MOD_1_NORM_THRESHOLD             MP_SIZE_T_MAX  /* never */
39 #define MOD_1_UNNORM_THRESHOLD           MP_SIZE_T_MAX  /* never */
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD          9
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD          8
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD         9
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                 2  /* 22.20% faster than 1 */
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           18
52 
53 #define DIV_1_VS_MUL_1_PERCENT             277
54 
55 #define MUL_TOOM22_THRESHOLD                24
56 #define MUL_TOOM33_THRESHOLD                93
57 #define MUL_TOOM44_THRESHOLD               136
58 #define MUL_TOOM6H_THRESHOLD               300
59 #define MUL_TOOM8H_THRESHOLD               478
60 
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD      91
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     153
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD      93
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD      94
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     130
66 
67 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68 #define SQR_TOOM2_THRESHOLD                 34
69 #define SQR_TOOM3_THRESHOLD                117
70 #define SQR_TOOM4_THRESHOLD                184
71 #define SQR_TOOM6_THRESHOLD                262
72 #define SQR_TOOM8_THRESHOLD                597
73 
74 #define MULMID_TOOM42_THRESHOLD             70
75 
76 #define MULMOD_BNM1_THRESHOLD               17
77 #define SQRMOD_BNM1_THRESHOLD               25
78 
79 #define MUL_FFT_MODF_THRESHOLD             505  /* k = 5 */
80 #define MUL_FFT_TABLE3                                      \
81   { {    505, 5}, {     28, 6}, {     15, 5}, {     31, 6}, \
82     {     29, 7}, {     15, 6}, {     32, 7}, {     17, 6}, \
83     {     35, 7}, {     19, 6}, {     39, 7}, {     29, 8}, \
84     {     15, 7}, {     35, 8}, {     19, 7}, {     43, 8}, \
85     {     23, 7}, {     47, 8}, {     27, 9}, {     15, 8}, \
86     {     31, 7}, {     63, 8}, {     43, 9}, {     23, 8}, \
87     {     55, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
88     {     79, 9}, {     55,10}, {     31, 9}, {     79,10}, \
89     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
90     {    135,10}, {     79, 9}, {    159,10}, {     95,11}, \
91     {     63, 9}, {    255,10}, {    159,11}, {     95,10}, \
92     {    191,12}, {     63,11}, {    127,10}, {    271, 9}, \
93     {    543,10}, {    287,11}, {    159,10}, {    335, 9}, \
94     {    671,10}, {    351,11}, {    191,10}, {    399, 9}, \
95     {    799,11}, {    223,12}, {    127,11}, {    255,10}, \
96     {    543,11}, {    287,10}, {    607,11}, {    319,10}, \
97     {    671,11}, {    351,12}, {    191,11}, {    383,10}, \
98     {    799,11}, {    415,13}, {    127,12}, {    255,11}, \
99     {    543,10}, {   1087,11}, {    607,12}, {    319,11}, \
100     {    671,10}, {   1343,11}, {    735,10}, {   1471,12}, \
101     {    383,11}, {    799,10}, {   1599,11}, {    863,12}, \
102     {    447,11}, {    959,13}, {    255,12}, {    511,11}, \
103     {   1087,12}, {    575,11}, {   1215,12}, {    639,11}, \
104     {   1343,12}, {    703,11}, {   1471,13}, {    383,12}, \
105     {    767,11}, {   1599,12}, {    831,11}, {   1727,12}, \
106     {    959,14}, {    255,13}, {    511,12}, {   1087,11}, \
107     {   2239,12}, {   1215,13}, {    639,12}, {   1471,11}, \
108     {   2943,13}, {    767,12}, {   1727,13}, {    895,12}, \
109     {   1919,14}, {    511,13}, {   1023,12}, {   2239,13}, \
110     {   1151,12}, {   2431,13}, {   1407,12}, {   2943,14}, \
111     {    767,13}, {   1663,12}, {   3455,13}, {   1919,15}, \
112     {    511,14}, {   1023,13}, {   2175,12}, {   4479,13}, \
113     {   2431,14}, {   1279,13}, {   2943,12}, {   5887,14}, \
114     {   1535,13}, {   3455,14}, {   1791,13}, {   3967,12}, \
115     {   7935,15}, {   1023,14}, {   2047,13}, {   4479,14}, \
116     {   2303,13}, {   4991,12}, {   9983,14}, {   2815,13}, \
117     {   5887,15}, {   1535,14}, {   3839,16} }
118 #define MUL_FFT_TABLE3_SIZE 147
119 #define MUL_FFT_THRESHOLD                 6784
120 
121 #define SQR_FFT_MODF_THRESHOLD             464  /* k = 5 */
122 #define SQR_FFT_TABLE3                                      \
123   { {    464, 5}, {     28, 6}, {     15, 5}, {     31, 6}, \
124     {     29, 7}, {     15, 6}, {     32, 7}, {     17, 6}, \
125     {     35, 7}, {     19, 6}, {     39, 7}, {     29, 8}, \
126     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
127     {     23, 7}, {     49, 8}, {     27, 9}, {     15, 8}, \
128     {     39, 9}, {     23, 8}, {     51,10}, {     15, 9}, \
129     {     31, 8}, {     67, 9}, {     39, 8}, {     79, 9}, \
130     {     47, 8}, {     95, 9}, {     55,10}, {     31, 9}, \
131     {     79,10}, {     47, 9}, {     95,11}, {     31,10}, \
132     {     63, 9}, {    127,10}, {     79, 9}, {    159,10}, \
133     {     95,11}, {     63,10}, {    127, 9}, {    255,10}, \
134     {    143, 9}, {    287, 5}, {   4863, 6}, {   2495, 7}, \
135     {   1343, 8}, {    703, 9}, {    367,12}, {     63,11}, \
136     {    127,10}, {    303,11}, {    159,10}, {    319, 9}, \
137     {    639,10}, {    335, 9}, {    671,10}, {    351, 9}, \
138     {    703,10}, {    367,11}, {    191,10}, {    383, 9}, \
139     {    767,10}, {    399, 9}, {    799,10}, {    415, 9}, \
140     {    831,12}, {    127,11}, {    255,10}, {    543,11}, \
141     {    287,10}, {    607,11}, {    319,10}, {    671,11}, \
142     {    351,10}, {    703,12}, {    191,11}, {    383,10}, \
143     {    799,11}, {    415,10}, {    863,13}, {    127,12}, \
144     {    255,11}, {    543,10}, {   1087,11}, {    607,12}, \
145     {    319,11}, {    671,10}, {   1343,11}, {    735,10}, \
146     {   1471,12}, {    383,11}, {    799,10}, {   1599,11}, \
147     {    863,12}, {    447,11}, {    959,13}, {    255,12}, \
148     {    511,11}, {   1087,12}, {    575,11}, {   1215,12}, \
149     {    639,11}, {   1343,12}, {    703,11}, {   1407,13}, \
150     {    383,12}, {    767,11}, {   1599,12}, {    831,11}, \
151     {   1727,12}, {    959,14}, {    255,13}, {    511,12}, \
152     {   1215,13}, {    639,12}, {   1471,11}, {   2943,13}, \
153     {    767,12}, {   1727,13}, {    895,12}, {   1919,14}, \
154     {    511,13}, {   1023,12}, {   2111,13}, {   1151,12}, \
155     {   2431,13}, {   1407,12}, {   2943,14}, {    767,13}, \
156     {   1663,12}, {   3455,13}, {   1919,15}, {    511,14}, \
157     {   1023,13}, {   2175,12}, {   4479,13}, {   2431,14}, \
158     {   1279,13}, {   2943,12}, {   5887,14}, {   1535,13}, \
159     {   3455,14}, {   1791,13}, {   3967,15}, {   1023,14}, \
160     {   2047,13}, {   4479,14}, {   2303,13}, {   4991,12}, \
161     {   9983,14}, {   2815,13}, {   5887,15}, {   1535,14}, \
162     {   3839,16} }
163 #define SQR_FFT_TABLE3_SIZE 157
164 #define SQR_FFT_THRESHOLD                 5312
165 
166 #define MULLO_BASECASE_THRESHOLD             0  /* always */
167 #define MULLO_DC_THRESHOLD                  36
168 #define MULLO_MUL_N_THRESHOLD            13463
169 #define SQRLO_BASECASE_THRESHOLD             0  /* always */
170 #define SQRLO_DC_THRESHOLD                 140
171 #define SQRLO_SQR_THRESHOLD              10393
172 
173 #define DC_DIV_QR_THRESHOLD                 32
174 #define DC_DIVAPPR_Q_THRESHOLD             116
175 #define DC_BDIV_QR_THRESHOLD                76
176 #define DC_BDIV_Q_THRESHOLD                180
177 
178 #define INV_MULMOD_BNM1_THRESHOLD           46
179 #define INV_NEWTON_THRESHOLD               138
180 #define INV_APPR_THRESHOLD                 123
181 
182 #define BINV_NEWTON_THRESHOLD              306
183 #define REDC_1_TO_REDC_N_THRESHOLD          82
184 
185 #define MU_DIV_QR_THRESHOLD               1499
186 #define MU_DIVAPPR_Q_THRESHOLD            1442
187 #define MUPI_DIV_QR_THRESHOLD               63
188 #define MU_BDIV_QR_THRESHOLD              1442
189 #define MU_BDIV_Q_THRESHOLD               1589
190 
191 #define POWM_SEC_TABLE  1,22,66,428,1035
192 
193 #define GET_STR_DC_THRESHOLD                13
194 #define GET_STR_PRECOMPUTE_THRESHOLD        18
195 #define SET_STR_DC_THRESHOLD               732
196 #define SET_STR_PRECOMPUTE_THRESHOLD      1118
197 
198 #define FAC_DSC_THRESHOLD                  115
199 #define FAC_ODD_THRESHOLD                   50
200 
201 #define MATRIX22_STRASSEN_THRESHOLD         25
202 #define HGCD2_DIV1_METHOD                    1  /* 5.78% faster than 3 */
203 #define HGCD_THRESHOLD                     121
204 #define HGCD_APPR_THRESHOLD                151
205 #define HGCD_REDUCE_THRESHOLD             3259
206 #define GCD_DC_THRESHOLD                   368
207 #define GCDEXT_DC_THRESHOLD                306
208 #define JACOBI_BASE_METHOD                   4  /* 14.19% faster than 1 */
209 
210 /* Tuneup completed successfully, took 67142 seconds */
211