xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/bd1/gmp-mparam.h (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* AMD bd1 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 /* 3600-3800 MHz Bulldozer Zambezi */
35 /* FFT tuning limit = 67,000,000 */
36 /* Generated by tuneup.c, 2019-10-27, gcc 8.3 */
37 
38 #define MOD_1_NORM_THRESHOLD                 0  /* always */
39 #define MOD_1_UNNORM_THRESHOLD               0  /* always */
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD          7
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD          4
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        15
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     13
45 #define USE_PREINV_DIVREM_1                  1  /* native */
46 #define DIV_QR_1N_PI1_METHOD                 1  /* 59.59% faster than 2 */
47 #define DIV_QR_1_NORM_THRESHOLD              5
48 #define DIV_QR_1_UNNORM_THRESHOLD        MP_SIZE_T_MAX  /* never */
49 #define DIV_QR_2_PI2_THRESHOLD           MP_SIZE_T_MAX  /* never */
50 #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
51 #define BMOD_1_TO_MOD_1_THRESHOLD           27
52 
53 #define DIV_1_VS_MUL_1_PERCENT             245
54 
55 #define MUL_TOOM22_THRESHOLD                32
56 #define MUL_TOOM33_THRESHOLD                89
57 #define MUL_TOOM44_THRESHOLD               154
58 #define MUL_TOOM6H_THRESHOLD               230
59 #define MUL_TOOM8H_THRESHOLD               351
60 
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD      89
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     110
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD     101
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD     111
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     130
66 
67 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68 #define SQR_TOOM2_THRESHOLD                 46
69 #define SQR_TOOM3_THRESHOLD                 87
70 #define SQR_TOOM4_THRESHOLD                216
71 #define SQR_TOOM6_THRESHOLD                294
72 #define SQR_TOOM8_THRESHOLD                442
73 
74 #define MULMID_TOOM42_THRESHOLD             50
75 
76 #define MULMOD_BNM1_THRESHOLD               22
77 #define SQRMOD_BNM1_THRESHOLD               26
78 
79 #define MUL_FFT_MODF_THRESHOLD             636  /* k = 5 */
80 #define MUL_FFT_TABLE3                                      \
81   { {    636, 5}, {     28, 6}, {     15, 5}, {     31, 6}, \
82     {     28, 7}, {     15, 6}, {     33, 7}, {     17, 6}, \
83     {     35, 7}, {     19, 6}, {     39, 7}, {     23, 6}, \
84     {     47, 7}, {     29, 8}, {     15, 7}, {     35, 8}, \
85     {     19, 7}, {     41, 8}, {     23, 7}, {     47, 8}, \
86     {     27, 7}, {     55, 8}, {     31, 7}, {     63, 8}, \
87     {     43, 9}, {     23, 8}, {     55, 9}, {     31, 8}, \
88     {     67, 9}, {     39, 8}, {     79, 9}, {     47, 8}, \
89     {     95, 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, 7}, {   1023, 8}, {    543,11}, \
93     {     95,10}, {    191,12}, {     63,11}, {    127,10}, \
94     {    255, 9}, {    511,10}, {    271,11}, {    159,10}, \
95     {    319, 9}, {    639,10}, {    335,11}, {    191,10}, \
96     {    399,11}, {    223,12}, {    127,11}, {    255,10}, \
97     {    543,11}, {    287,10}, {    607,11}, {    319,10}, \
98     {    639,12}, {    191,11}, {    383,10}, {    799,11}, \
99     {    415,13}, {    127,12}, {    255,11}, {    543,10}, \
100     {   1087,11}, {    607,12}, {    319,11}, {    671,10}, \
101     {   1343,11}, {    735,12}, {    383,11}, {    799,10}, \
102     {   1599,11}, {    863,12}, {    447,11}, {    895,13}, \
103     {    255,12}, {    511,11}, {   1087,12}, {    575,11}, \
104     {   1215,12}, {    639,11}, {   1343,12}, {    703,11}, \
105     {   1471,13}, {    383,12}, {    767,11}, {   1599,12}, \
106     {    831,11}, {   1727,12}, {    895,14}, {    255,13}, \
107     {    511,12}, {   1087,11}, {   2239,10}, {   4479,12}, \
108     {   1215,13}, {    639,12}, {   1471,11}, {   2943,13}, \
109     {    767,12}, {   1727,11}, {   3455,13}, {    895,12}, \
110     {   1919,14}, {    511,13}, {   1023,12}, {   2239,11}, \
111     {   4479,13}, {   1151,12}, {   2495,11}, {   4991,13}, \
112     {   1279,12}, {   2623,13}, {   1407,12}, {   2943,14}, \
113     {    767,13}, {   1663,12}, {   3455,13}, {   1919,15}, \
114     {    511,14}, {   1023,13}, {   2175,12}, {   4479,13}, \
115     {   2431,12}, {   4991,14}, {   1279,13}, {   2943,12}, \
116     {   5887,14}, {   1535,13}, {   3455,14}, {   1791,13}, \
117     {   3967,12}, {   7935,15}, {   1023,14}, {   2047,13}, \
118     {   4479,14}, {   2303,13}, {   4991,12}, {   9983,14}, \
119     {   2815,13}, {   5887,15}, {   1535,14}, {   3327,13}, \
120     {   6911,14}, {   3839,13}, {   7935,16} }
121 #define MUL_FFT_TABLE3_SIZE 159
122 #define MUL_FFT_THRESHOLD                 6784
123 
124 #define SQR_FFT_MODF_THRESHOLD             565  /* k = 5 */
125 #define SQR_FFT_TABLE3                                      \
126   { {    565, 5}, {     29, 6}, {     15, 5}, {     32, 6}, \
127     {     17, 5}, {     35, 6}, {     29, 7}, {     15, 6}, \
128     {     33, 7}, {     17, 6}, {     35, 7}, {     19, 6}, \
129     {     39, 7}, {     23, 6}, {     47, 7}, {     35, 8}, \
130     {     19, 7}, {     41, 8}, {     23, 7}, {     49, 8}, \
131     {     27, 7}, {     55, 8}, {     43, 9}, {     23, 8}, \
132     {     55, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
133     {     79, 9}, {     47, 8}, {     95, 9}, {     55,10}, \
134     {     31, 9}, {     79,10}, {     47, 9}, {     95,11}, \
135     {     31,10}, {     63, 9}, {    135,10}, {     79, 9}, \
136     {    159,10}, {     95,11}, {     63,10}, {    159,11}, \
137     {    127,10}, {    255, 9}, {    511,10}, {    271, 9}, \
138     {    543,11}, {    159,10}, {    319, 9}, {    639,10}, \
139     {    335, 9}, {    671,11}, {    191,10}, {    415,11}, \
140     {    223,12}, {    127,11}, {    255,10}, {    543,11}, \
141     {    287,10}, {    607,11}, {    319,10}, {    671,12}, \
142     {    191,11}, {    383,10}, {    799,11}, {    415,10}, \
143     {    831,13}, {    127,12}, {    255,11}, {    543,10}, \
144     {   1087,11}, {    607,12}, {    319,11}, {    671,10}, \
145     {   1343,11}, {    735,12}, {    383,11}, {    863,12}, \
146     {    447,11}, {    959,13}, {    255,12}, {    511,11}, \
147     {   1087,12}, {    575,11}, {   1215,12}, {    639,11}, \
148     {   1343,12}, {    703,13}, {    383,12}, {    767,11}, \
149     {   1535,12}, {    831,11}, {   1727,12}, {    895,11}, \
150     {   1791,12}, {    959,14}, {    255,13}, {    511,12}, \
151     {   1087,11}, {   2239,10}, {   4479,12}, {   1215,13}, \
152     {    639,12}, {   1471,11}, {   2943,13}, {    767,12}, \
153     {   1727,13}, {    895,12}, {   1919,14}, {    511,13}, \
154     {   1023,12}, {   2239,11}, {   4479,13}, {   1151,12}, \
155     {   2495,11}, {   4991,13}, {   1279,12}, {   2623,13}, \
156     {   1407,12}, {   2943,14}, {    767,13}, {   1663,12}, \
157     {   3455,13}, {   1919,15}, {    511,14}, {   1023,13}, \
158     {   2175,12}, {   4479,13}, {   2431,12}, {   4991,14}, \
159     {   1279,13}, {   2943,12}, {   5887,14}, {   1535,13}, \
160     {   3455,14}, {   1791,13}, {   3967,15}, {   1023,14}, \
161     {   2047,13}, {   4479,14}, {   2303,13}, {   4991,12}, \
162     {   9983,14}, {   2815,13}, {   5887,15}, {   1535,14}, \
163     {   3327,13}, {   6783,14}, {   3839,13}, {   7679,16} }
164 #define SQR_FFT_TABLE3_SIZE 152
165 #define SQR_FFT_THRESHOLD                 5760
166 
167 #define MULLO_BASECASE_THRESHOLD             3
168 #define MULLO_DC_THRESHOLD                  31
169 #define MULLO_MUL_N_THRESHOLD            13463
170 #define SQRLO_BASECASE_THRESHOLD             0  /* always */
171 #define SQRLO_DC_THRESHOLD                  33
172 #define SQRLO_SQR_THRESHOLD              11278
173 
174 #define DC_DIV_QR_THRESHOLD                 52
175 #define DC_DIVAPPR_Q_THRESHOLD             198
176 #define DC_BDIV_QR_THRESHOLD                48
177 #define DC_BDIV_Q_THRESHOLD                126
178 
179 #define INV_MULMOD_BNM1_THRESHOLD           82
180 #define INV_NEWTON_THRESHOLD               212
181 #define INV_APPR_THRESHOLD                 202
182 
183 #define BINV_NEWTON_THRESHOLD              238
184 #define REDC_1_TO_REDC_N_THRESHOLD          55
185 
186 #define MU_DIV_QR_THRESHOLD               1652
187 #define MU_DIVAPPR_Q_THRESHOLD            1528
188 #define MUPI_DIV_QR_THRESHOLD              110
189 #define MU_BDIV_QR_THRESHOLD              1442
190 #define MU_BDIV_Q_THRESHOLD               1528
191 
192 #define POWM_SEC_TABLE  1,20,96,386,1221,2698
193 
194 #define GET_STR_DC_THRESHOLD                11
195 #define GET_STR_PRECOMPUTE_THRESHOLD        21
196 #define SET_STR_DC_THRESHOLD               100
197 #define SET_STR_PRECOMPUTE_THRESHOLD       762
198 
199 #define FAC_DSC_THRESHOLD                  118
200 #define FAC_ODD_THRESHOLD                   34
201 
202 #define MATRIX22_STRASSEN_THRESHOLD         16
203 #define HGCD2_DIV1_METHOD                    4  /* 1.22% faster than 3 */
204 #define HGCD_THRESHOLD                      67
205 #define HGCD_APPR_THRESHOLD                150
206 #define HGCD_REDUCE_THRESHOLD             3389
207 #define GCD_DC_THRESHOLD                   483
208 #define GCDEXT_DC_THRESHOLD                345
209 #define JACOBI_BASE_METHOD                   4  /* 5.07% faster than 1 */
210 
211 /* Tuneup completed successfully, took 65358 seconds */
212