xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/pentium4/sse2/gmp-mparam.h (revision 122b5006ee1bd67145794b4cde92f4fe4781a5ec)
1 /* Intel Pentium-4 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 /* 2600 MHz P4 Northwood */
35 /* FFT tuning limit = 23,700,309 */
36 /* Generated by tuneup.c, 2019-11-09, gcc 8.2 */
37 
38 #define MOD_1_NORM_THRESHOLD                 5
39 #define MOD_1_UNNORM_THRESHOLD              14
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD          7
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD          5
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        13
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      7
45 #define USE_PREINV_DIVREM_1                  1  /* native */
46 #define DIV_QR_1N_PI1_METHOD                 2  /* 4.36% faster than 1 */
47 #define DIV_QR_1_NORM_THRESHOLD             16
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           21
52 
53 #define DIV_1_VS_MUL_1_PERCENT             358
54 
55 #define MUL_TOOM22_THRESHOLD                26
56 #define MUL_TOOM33_THRESHOLD               101
57 #define MUL_TOOM44_THRESHOLD               284
58 #define MUL_TOOM6H_THRESHOLD               406
59 #define MUL_TOOM8H_THRESHOLD               592
60 
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD     101
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     191
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD     189
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD     195
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     151
66 
67 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68 #define SQR_TOOM2_THRESHOLD                 51
69 #define SQR_TOOM3_THRESHOLD                163
70 #define SQR_TOOM4_THRESHOLD                254
71 #define SQR_TOOM6_THRESHOLD                614
72 #define SQR_TOOM8_THRESHOLD                842
73 
74 #define MULMID_TOOM42_THRESHOLD             58
75 
76 #define MULMOD_BNM1_THRESHOLD               19
77 #define SQRMOD_BNM1_THRESHOLD               23
78 
79 #define MUL_FFT_MODF_THRESHOLD             824  /* k = 5 */
80 #define MUL_FFT_TABLE3                                      \
81   { {    824, 5}, {     29, 6}, {     15, 5}, {     33, 6}, \
82     {     17, 5}, {     36, 6}, {     19, 5}, {     39, 6}, \
83     {     29, 7}, {     15, 6}, {     32, 7}, {     17, 6}, \
84     {     36, 7}, {     19, 6}, {     39, 7}, {     21, 6}, \
85     {     43, 7}, {     23, 6}, {     48, 7}, {     29, 8}, \
86     {     15, 7}, {     37, 8}, {     19, 7}, {     43, 8}, \
87     {     23, 7}, {     49, 8}, {     27, 7}, {     55, 8}, \
88     {     31, 7}, {     63, 8}, {     43, 9}, {     23, 8}, \
89     {     55, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
90     {     79, 9}, {     47, 8}, {     99, 9}, {     55,10}, \
91     {     31, 9}, {     79,10}, {     47, 9}, {    103,11}, \
92     {     31,10}, {     63, 9}, {    143,10}, {     79, 9}, \
93     {    167,10}, {     95, 9}, {    191,10}, {    111,11}, \
94     {     63,10}, {    127, 9}, {    255,10}, {    159,11}, \
95     {     95,10}, {    191,12}, {     63,11}, {    127,10}, \
96     {    271,11}, {    159,10}, {    319, 9}, {    639,10}, \
97     {    335,11}, {    191,10}, {    383, 9}, {    799,10}, \
98     {    415,11}, {    223,12}, {    127,11}, {    255,10}, \
99     {    527,11}, {    287,10}, {    607, 9}, {   1215,11}, \
100     {    319,10}, {    671,12}, {    191,11}, {    383,10}, \
101     {    799,11}, {    415,10}, {    863,13}, {    127,12}, \
102     {    255,11}, {    543,10}, {   1119, 9}, {   2239,11}, \
103     {    607,10}, {   1215,12}, {    319,11}, {    671,10}, \
104     {   1343,11}, {    735,10}, {   1471, 9}, {   2943,12}, \
105     {    383,11}, {    799,10}, {   1599,11}, {    863,12}, \
106     {    447,11}, {    927,10}, {   1855,11}, {    959,13}, \
107     {    255,12}, {    511,11}, {   1119,12}, {    575,11}, \
108     {   1215,10}, {   2431,11}, {   1247,12}, {    639,11}, \
109     {   1343,12}, {    703,11}, {   1471,10}, {   2943,13}, \
110     {    383,12}, {    767,11}, {   1599,12}, {    831,11}, \
111     {   1727,10}, {   3455,12}, {    895,14}, {    255,13}, \
112     {    511,12}, {   1087,11}, {   2239,10}, {   4479,12}, \
113     {   1215,11}, {   2431,13}, {    639,12}, {   1343,11}, \
114     {   2687,12}, {   1471,11}, {   2943,13}, {    767,12}, \
115     {   1727,11}, {   3455,13}, {    895,12}, {   1983,14}, \
116     {    511,13}, {   1023,12}, {   2239,13}, {   1151,12}, \
117     {   2495,11}, {   4991,13}, {   1407,12}, {   2943,14}, \
118     {    767,13}, {   1535,12}, {   3135,13}, {   1663,12}, \
119     {   3455,13}, {   1919,12}, {   3967,15}, {    511,14}, \
120     {   1023,13}, {   2175,12}, {   4479,13}, {   2431,12}, \
121     {   4991,14}, {   1279,13}, {   2687,12}, {   5503,13}, \
122     {   8192,14}, {  16384,15}, {  32768,16} }
123 #define MUL_FFT_TABLE3_SIZE 167
124 #define MUL_FFT_THRESHOLD                 7808
125 
126 #define SQR_FFT_MODF_THRESHOLD             560  /* k = 5 */
127 #define SQR_FFT_TABLE3                                      \
128   { {    560, 5}, {     33, 6}, {     17, 5}, {     35, 6}, \
129     {     33, 7}, {     17, 6}, {     36, 7}, {     19, 6}, \
130     {     39, 7}, {     35, 8}, {     19, 7}, {     43, 8}, \
131     {     23, 7}, {     47, 8}, {     27, 7}, {     55, 8}, \
132     {     31, 7}, {     63, 8}, {     43, 9}, {     23, 8}, \
133     {     55, 9}, {     31, 8}, {     67, 9}, {     39, 8}, \
134     {     79, 9}, {     47, 8}, {     95, 9}, {     55,10}, \
135     {     31, 9}, {     79,10}, {     47, 9}, {     95,11}, \
136     {     31,10}, {     63, 9}, {    135,10}, {     79, 9}, \
137     {    159,10}, {    111,11}, {     63,10}, {    143, 9}, \
138     {    287,10}, {    159,11}, {     95,10}, {    191,12}, \
139     {     63,11}, {    127, 9}, {    511, 8}, {   1023, 9}, \
140     {    527,11}, {    159,10}, {    319, 9}, {    639,10}, \
141     {    351,11}, {    191,10}, {    431,11}, {    223,12}, \
142     {    127,11}, {    255,10}, {    543,11}, {    287,10}, \
143     {    607, 9}, {   1215,11}, {    319,10}, {    639,11}, \
144     {    351,12}, {    191,11}, {    383,10}, {    767,11}, \
145     {    415,10}, {    831,13}, {    127,12}, {    255,11}, \
146     {    543,10}, {   1119,11}, {    607,12}, {    319,11}, \
147     {    671,10}, {   1343,11}, {    735,12}, {    383,11}, \
148     {    799,10}, {   1599,11}, {    863,12}, {    447,11}, \
149     {    927,10}, {   1855,11}, {    991,13}, {    255,12}, \
150     {    511,11}, {   1055,10}, {   2111,11}, {   1087,12}, \
151     {    575,11}, {   1215,10}, {   2431,12}, {    639,11}, \
152     {   1343,12}, {    703,11}, {   1407,13}, {    383,12}, \
153     {    767,11}, {   1599,12}, {    831,11}, {   1727,10}, \
154     {   3455,12}, {    895,11}, {   1855,12}, {    959,14}, \
155     {    255,13}, {    511,12}, {   1087,11}, {   2239,12}, \
156     {   1215,11}, {   2431,13}, {    639,12}, {   1471,11}, \
157     {   2943,13}, {    767,12}, {   1727,11}, {   3455,13}, \
158     {    895,12}, {   1983,14}, {    511,13}, {   1023,12}, \
159     {   2239,13}, {   1151,12}, {   2495,11}, {   4991,13}, \
160     {   1279,12}, {   2623,13}, {   1407,12}, {   2943,14}, \
161     {    767,13}, {   1663,12}, {   3455,13}, {   1919,12}, \
162     {   3839,15}, {    511,14}, {   1023,13}, {   2175,12}, \
163     {   4479,13}, {   2431,12}, {   4991,14}, {   1279,13}, \
164     {   2687,12}, {   5503,13}, {   8192,14}, {  16384,15}, \
165     {  32768,16} }
166 #define SQR_FFT_TABLE3_SIZE 149
167 #define SQR_FFT_THRESHOLD                 4800
168 
169 #define MULLO_BASECASE_THRESHOLD            12
170 #define MULLO_DC_THRESHOLD                  44
171 #define MULLO_MUL_N_THRESHOLD            14281
172 #define SQRLO_BASECASE_THRESHOLD            13
173 #define SQRLO_DC_THRESHOLD                  42
174 #define SQRLO_SQR_THRESHOLD               9449
175 
176 #define DC_DIV_QR_THRESHOLD                 38
177 #define DC_DIVAPPR_Q_THRESHOLD             105
178 #define DC_BDIV_QR_THRESHOLD                52
179 #define DC_BDIV_Q_THRESHOLD                 83
180 
181 #define INV_MULMOD_BNM1_THRESHOLD           50
182 #define INV_NEWTON_THRESHOLD               158
183 #define INV_APPR_THRESHOLD                 118
184 
185 #define BINV_NEWTON_THRESHOLD              342
186 #define REDC_1_TO_REDC_N_THRESHOLD          67
187 
188 #define MU_DIV_QR_THRESHOLD               2130
189 #define MU_DIVAPPR_Q_THRESHOLD            1895
190 #define MUPI_DIV_QR_THRESHOLD               60
191 #define MU_BDIV_QR_THRESHOLD              1652
192 #define MU_BDIV_Q_THRESHOLD               2089
193 
194 #define POWM_SEC_TABLE  1,22,96,446,723,1378
195 
196 #define GET_STR_DC_THRESHOLD                13
197 #define GET_STR_PRECOMPUTE_THRESHOLD        20
198 #define SET_STR_DC_THRESHOLD               298
199 #define SET_STR_PRECOMPUTE_THRESHOLD       960
200 
201 #define FAC_DSC_THRESHOLD                  212
202 #define FAC_ODD_THRESHOLD                   71
203 
204 #define MATRIX22_STRASSEN_THRESHOLD         26
205 #define HGCD2_DIV1_METHOD                    3  /* 0.68% faster than 1 */
206 #define HGCD_THRESHOLD                      80
207 #define HGCD_APPR_THRESHOLD                138
208 #define HGCD_REDUCE_THRESHOLD             4455
209 #define GCD_DC_THRESHOLD                   365
210 #define GCDEXT_DC_THRESHOLD                245
211 #define JACOBI_BASE_METHOD                   4  /* 23.41% faster than 1 */
212 
213 /* Tuneup completed successfully, took 63807 seconds */
214