xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/bt2/gmp-mparam.h (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1 /* x86/bobcat 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 /* NOTE: In a fat binary build SQR_TOOM2_THRESHOLD here cannot be greater than
35    the value in mpn/x86/k7/gmp-mparam.h.  The latter is used as a hard limit in
36    k7/sqr_basecase.asm.  */
37 
38 /* 2050 MHz AMD Jaguar/Kabini */
39 /* FFT tuning limit = 67,000,000 */
40 /* Generated by tuneup.c, 2019-10-24, gcc 8.3 */
41 
42 #define MOD_1_NORM_THRESHOLD                 4
43 #define MOD_1_UNNORM_THRESHOLD               6
44 #define MOD_1N_TO_MOD_1_1_THRESHOLD          4
45 #define MOD_1U_TO_MOD_1_1_THRESHOLD          4
46 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        18
47 #define MOD_1_2_TO_MOD_1_4_THRESHOLD         0  /* never mpn_mod_1s_2p */
48 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD      9
49 #define USE_PREINV_DIVREM_1                  1  /* native */
50 #define DIV_QR_1N_PI1_METHOD                 1  /* 47.53% faster than 2 */
51 #define DIV_QR_1_NORM_THRESHOLD          MP_SIZE_T_MAX  /* never */
52 #define DIV_QR_1_UNNORM_THRESHOLD        MP_SIZE_T_MAX  /* never */
53 #define DIV_QR_2_PI2_THRESHOLD           MP_SIZE_T_MAX  /* never */
54 #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
55 #define BMOD_1_TO_MOD_1_THRESHOLD           27
56 
57 #define DIV_1_VS_MUL_1_PERCENT             243
58 
59 #define MUL_TOOM22_THRESHOLD                32
60 #define MUL_TOOM33_THRESHOLD                90
61 #define MUL_TOOM44_THRESHOLD               154
62 #define MUL_TOOM6H_THRESHOLD               286
63 #define MUL_TOOM8H_THRESHOLD               478
64 
65 #define MUL_TOOM32_TO_TOOM43_THRESHOLD      97
66 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     152
67 #define MUL_TOOM42_TO_TOOM53_THRESHOLD     103
68 #define MUL_TOOM42_TO_TOOM63_THRESHOLD     113
69 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     154
70 
71 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
72 #define SQR_TOOM2_THRESHOLD                 38
73 #define SQR_TOOM3_THRESHOLD                126
74 #define SQR_TOOM4_THRESHOLD                220
75 #define SQR_TOOM6_THRESHOLD                318
76 #define SQR_TOOM8_THRESHOLD                502
77 
78 #define MULMID_TOOM42_THRESHOLD             68
79 
80 #define MULMOD_BNM1_THRESHOLD               19
81 #define SQRMOD_BNM1_THRESHOLD               25
82 
83 #define MUL_FFT_MODF_THRESHOLD             570  /* k = 5 */
84 #define MUL_FFT_TABLE3                                      \
85   { {    570, 5}, {     25, 6}, {     13, 5}, {     27, 6}, \
86     {     15, 5}, {     31, 6}, {     28, 7}, {     15, 6}, \
87     {     33, 7}, {     17, 6}, {     35, 7}, {     19, 6}, \
88     {     39, 7}, {     23, 6}, {     47, 7}, {     27, 8}, \
89     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
90     {     23, 7}, {     49, 8}, {     31, 7}, {     63, 8}, \
91     {     39, 9}, {     23, 8}, {     55, 9}, {     31, 8}, \
92     {     67, 9}, {     39, 8}, {     79, 9}, {     47, 8}, \
93     {     95, 9}, {     55,10}, {     31, 9}, {     79,10}, \
94     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
95     {    135,10}, {     79, 9}, {    159,10}, {     95,11}, \
96     {     63,10}, {    159,11}, {     95,10}, {    191,12}, \
97     {     63,11}, {    127,10}, {    255, 9}, {    511,10}, \
98     {    271, 9}, {    543,10}, {    287,11}, {    159,10}, \
99     {    319, 9}, {    639,10}, {    335, 9}, {    671,11}, \
100     {    191,10}, {    383, 9}, {    767,10}, {    399, 9}, \
101     {    799,10}, {    415,11}, {    223,12}, {    127,11}, \
102     {    255,10}, {    543,11}, {    287,10}, {    607, 9}, \
103     {   1215,11}, {    319,10}, {    671,11}, {    351,12}, \
104     {    191,11}, {    383,10}, {    799,11}, {    415,13}, \
105     {    127,12}, {    255,11}, {    543,10}, {   1087,11}, \
106     {    607,10}, {   1215,12}, {    319,11}, {    671,10}, \
107     {   1343,11}, {    735,10}, {   1471,12}, {    383,11}, \
108     {    799,10}, {   1599,11}, {    863,12}, {    447,11}, \
109     {    991,13}, {    255,12}, {    511,11}, {   1087,12}, \
110     {    575,11}, {   1215,12}, {    639,11}, {   1343,12}, \
111     {    703,11}, {   1471,13}, {    383,12}, {    767,11}, \
112     {   1599,12}, {    831,11}, {   1727,12}, {    959,14}, \
113     {    255,13}, {    511,12}, {   1215,13}, {    639,12}, \
114     {   1471,13}, {    767,12}, {   1727,13}, {    895,12}, \
115     {   1919,14}, {    511,13}, {   1023,12}, {   2111,13}, \
116     {   1151,12}, {   2431,13}, {   1407,14}, {    767,13}, \
117     {   1663,12}, {   3455,13}, {   1919,15}, {    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}, {   5887,15}, {   1535,14}, \
123     {   3839,16} }
124 #define MUL_FFT_TABLE3_SIZE 153
125 #define MUL_FFT_THRESHOLD                 5760
126 
127 #define SQR_FFT_MODF_THRESHOLD             530  /* k = 5 */
128 #define SQR_FFT_TABLE3                                      \
129   { {    530, 5}, {     27, 6}, {     15, 5}, {     31, 6}, \
130     {     28, 7}, {     15, 6}, {     33, 7}, {     17, 6}, \
131     {     35, 7}, {     19, 6}, {     39, 7}, {     23, 6}, \
132     {     47, 7}, {     29, 8}, {     15, 7}, {     35, 8}, \
133     {     19, 7}, {     41, 8}, {     23, 7}, {     49, 8}, \
134     {     31, 7}, {     63, 8}, {     39, 9}, {     23, 8}, \
135     {     55, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
136     {     79, 9}, {     47, 8}, {     95, 9}, {     55,10}, \
137     {     31, 9}, {     79,10}, {     47, 9}, {     95,11}, \
138     {     31,10}, {     63, 9}, {    135,10}, {     95,11}, \
139     {     63,10}, {    143, 9}, {    287,10}, {    159,11}, \
140     {     95,12}, {     63,11}, {    127,10}, {    255, 9}, \
141     {    511,10}, {    271, 9}, {    543,10}, {    287,11}, \
142     {    159,10}, {    319, 9}, {    639,10}, {    335, 9}, \
143     {    671,10}, {    351,11}, {    191,10}, {    383, 9}, \
144     {    767,10}, {    399, 9}, {    799,12}, {    127,11}, \
145     {    255,10}, {    543,11}, {    287,10}, {    607, 9}, \
146     {   1215,11}, {    319,10}, {    671,11}, {    351,12}, \
147     {    191,11}, {    383,10}, {    799,11}, {    415,10}, \
148     {    831,13}, {    127,12}, {    255,11}, {    543,10}, \
149     {   1087,11}, {    607,10}, {   1215,12}, {    319,11}, \
150     {    671,10}, {   1343,11}, {    735,10}, {   1471,12}, \
151     {    383,11}, {    799,10}, {   1599,11}, {    863,12}, \
152     {    447,11}, {    991,13}, {    255,12}, {    511,11}, \
153     {   1087,12}, {    575,11}, {   1215,12}, {    639,11}, \
154     {   1343,12}, {    703,11}, {   1471,13}, {    383,12}, \
155     {    767,11}, {   1599,12}, {    831,11}, {   1727,12}, \
156     {    959,11}, {   1919,14}, {    255,13}, {    511,12}, \
157     {   1215,13}, {    639,12}, {   1471,13}, {    767,12}, \
158     {   1727,13}, {    895,12}, {   1919,14}, {    511,13}, \
159     {   1023,12}, {   2111,13}, {   1151,12}, {   2495,13}, \
160     {   1407,14}, {    767,13}, {   1663,12}, {   3455,13}, \
161     {   1919,15}, {    511,14}, {   1023,13}, {   2175,12}, \
162     {   4479,13}, {   2431,14}, {   1279,13}, {   2943,12}, \
163     {   5887,14}, {   1535,13}, {   3455,14}, {   1791,13}, \
164     {   3967,15}, {   1023,14}, {   2047,13}, {   4479,14}, \
165     {   2303,13}, {   4991,12}, {   9983,14}, {   2815,13}, \
166     {   5887,15}, {   1535,14}, {   3839,16} }
167 #define SQR_FFT_TABLE3_SIZE 151
168 #define SQR_FFT_THRESHOLD                 4736
169 
170 #define MULLO_BASECASE_THRESHOLD             8
171 #define MULLO_DC_THRESHOLD                  44
172 #define MULLO_MUL_N_THRESHOLD            11278
173 #define SQRLO_BASECASE_THRESHOLD            13
174 #define SQRLO_DC_THRESHOLD                  62
175 #define SQRLO_SQR_THRESHOLD               8907
176 
177 #define DC_DIV_QR_THRESHOLD                 79
178 #define DC_DIVAPPR_Q_THRESHOLD             228
179 #define DC_BDIV_QR_THRESHOLD                75
180 #define DC_BDIV_Q_THRESHOLD                136
181 
182 #define INV_MULMOD_BNM1_THRESHOLD           90
183 #define INV_NEWTON_THRESHOLD               260
184 #define INV_APPR_THRESHOLD                 236
185 
186 #define BINV_NEWTON_THRESHOLD              294
187 #define REDC_1_TO_REDC_N_THRESHOLD          80
188 
189 #define MU_DIV_QR_THRESHOLD               1787
190 #define MU_DIVAPPR_Q_THRESHOLD            1718
191 #define MUPI_DIV_QR_THRESHOLD              118
192 #define MU_BDIV_QR_THRESHOLD              1442
193 #define MU_BDIV_Q_THRESHOLD               1652
194 
195 #define POWM_SEC_TABLE  1,16,96,615,865,1442
196 
197 #define GET_STR_DC_THRESHOLD                16
198 #define GET_STR_PRECOMPUTE_THRESHOLD        27
199 #define SET_STR_DC_THRESHOLD               252
200 #define SET_STR_PRECOMPUTE_THRESHOLD       638
201 
202 #define FAC_DSC_THRESHOLD                  141
203 #define FAC_ODD_THRESHOLD                   39
204 
205 #define MATRIX22_STRASSEN_THRESHOLD         19
206 #define HGCD2_DIV1_METHOD                    1  /* 13.65% faster than 3 */
207 #define HGCD_THRESHOLD                      81
208 #define HGCD_APPR_THRESHOLD                 66
209 #define HGCD_REDUCE_THRESHOLD             3389
210 #define GCD_DC_THRESHOLD                   531
211 #define GCDEXT_DC_THRESHOLD                345
212 #define JACOBI_BASE_METHOD                   1  /* 0.84% faster than 4 */
213 
214 /* Tuneup completed successfully, took 103818 seconds */
215