xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/p6/sse2/gmp-mparam.h (revision a8c74629f602faa0ccf8a463757d7baf858bbf3a)
1 /* Intel P6/sse2 gmp-mparam.h -- Compiler/machine parameter header file.
2 
3 Copyright 1991, 1993, 1994, 1999-2003, 2008-2010 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 
32 #define GMP_LIMB_BITS 32
33 #define GMP_LIMB_BYTES 4
34 
35 
36 /* NOTE: In a fat binary build SQR_TOOM2_THRESHOLD here cannot be more than the
37    value in mpn/x86/p6/gmp-mparam.h.  The latter is used as a hard limit in
38    mpn/x86/p6/sqr_basecase.asm.  */
39 
40 
41 /* 1867 MHz P6 model 13 */
42 
43 #define MOD_1_NORM_THRESHOLD                 4
44 #define MOD_1_UNNORM_THRESHOLD               4
45 #define MOD_1N_TO_MOD_1_1_THRESHOLD          5
46 #define MOD_1U_TO_MOD_1_1_THRESHOLD          4
47 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        11
48 #define MOD_1_2_TO_MOD_1_4_THRESHOLD         0  /* never mpn_mod_1s_2p */
49 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD      8
50 #define USE_PREINV_DIVREM_1                  1  /* native */
51 #define DIV_QR_2_PI2_THRESHOLD           MP_SIZE_T_MAX  /* never */
52 #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
53 #define BMOD_1_TO_MOD_1_THRESHOLD           21
54 
55 #define MUL_TOOM22_THRESHOLD                20
56 #define MUL_TOOM33_THRESHOLD                77
57 #define MUL_TOOM44_THRESHOLD               169
58 #define MUL_TOOM6H_THRESHOLD               246
59 #define MUL_TOOM8H_THRESHOLD               381
60 
61 #define MUL_TOOM32_TO_TOOM43_THRESHOLD      73
62 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     114
63 #define MUL_TOOM42_TO_TOOM53_THRESHOLD      97
64 #define MUL_TOOM42_TO_TOOM63_THRESHOLD      80
65 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     106
66 
67 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68 #define SQR_TOOM2_THRESHOLD                 30
69 #define SQR_TOOM3_THRESHOLD                101
70 #define SQR_TOOM4_THRESHOLD                154
71 #define SQR_TOOM6_THRESHOLD                222
72 #define SQR_TOOM8_THRESHOLD                527
73 
74 #define MULMID_TOOM42_THRESHOLD             58
75 
76 #define MULMOD_BNM1_THRESHOLD               13
77 #define SQRMOD_BNM1_THRESHOLD               17
78 
79 #define MUL_FFT_MODF_THRESHOLD             690  /* k = 5 */
80 #define MUL_FFT_TABLE3                                      \
81   { {    565, 5}, {     25, 6}, {     13, 5}, {     27, 6}, \
82     {     25, 7}, {     13, 6}, {     28, 7}, {     15, 6}, \
83     {     31, 7}, {     17, 6}, {     35, 7}, {     27, 8}, \
84     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
85     {     23, 7}, {     47, 8}, {     27, 9}, {     15, 8}, \
86     {     31, 7}, {     63, 8}, {     39, 9}, {     23, 5}, \
87     {    383, 4}, {    991, 5}, {    511, 6}, {    267, 7}, \
88     {    157, 8}, {     91, 9}, {     47, 8}, {    111, 9}, \
89     {     63, 8}, {    127, 9}, {     79,10}, {     47, 9}, \
90     {     95,11}, {     31,10}, {     63, 9}, {    135,10}, \
91     {     79, 9}, {    159,10}, {     95,11}, {     63,10}, \
92     {    143, 9}, {    287,10}, {    159,11}, {     95,10}, \
93     {    191,12}, {     63,11}, {    127,10}, {    255, 9}, \
94     {    511,10}, {    271, 9}, {    543,10}, {    287,11}, \
95     {    159,10}, {    335, 9}, {    671,11}, {    191,10}, \
96     {    383, 9}, {    767,10}, {    399, 9}, {    799,10}, \
97     {    415,11}, {    223,12}, {    127,11}, {    255,10}, \
98     {    543, 9}, {   1087,11}, {    287,10}, {    607,11}, \
99     {    319,10}, {    671,12}, {    191,11}, {    383,10}, \
100     {    799,11}, {    415,10}, {    831,13}, {    127,12}, \
101     {    255,11}, {    543,10}, {   1087,11}, {    607,10}, \
102     {   1215,12}, {    319,11}, {    671,10}, {   1343,11}, \
103     {    735,10}, {   1471,12}, {    383,11}, {    799,10}, \
104     {   1599,11}, {    863,12}, {    447,11}, {    959,13}, \
105     {    255,12}, {    511,11}, {   1087,12}, {    575,11}, \
106     {   1215,12}, {    639,11}, {   1343,12}, {    703,11}, \
107     {   1471,13}, {    383,12}, {    831,11}, {   1727,12}, \
108     {    959,14}, {    255,13}, {    511,12}, {   1215,13}, \
109     {    639,12}, {   1471,11}, {   2943,13}, {    767,12}, \
110     {   1727,13}, {    895,12}, {   1919,14}, {    511,13}, \
111     {   1023,12}, {   2111,13}, {   1151,12}, {   2431,13}, \
112     {   1407,12}, {   2815,14}, {    767,13}, {   1663,12}, \
113     {   3455,13}, {   8192,14}, {  16384,15}, {  32768,16} }
114 #define MUL_FFT_TABLE3_SIZE 132
115 #define MUL_FFT_THRESHOLD                 7424
116 
117 #define SQR_FFT_MODF_THRESHOLD             565  /* k = 5 */
118 #define SQR_FFT_TABLE3                                      \
119   { {    472, 5}, {     25, 6}, {     13, 5}, {     27, 6}, \
120     {     25, 7}, {     13, 6}, {     27, 7}, {     15, 6}, \
121     {     31, 7}, {     17, 6}, {     35, 7}, {     27, 8}, \
122     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
123     {     23, 7}, {     49, 8}, {     27, 9}, {     15, 8}, \
124     {     39, 9}, {     23, 8}, {     51,10}, {     15, 9}, \
125     {     31, 8}, {     63, 4}, {   1023, 8}, {     67, 9}, \
126     {     39, 5}, {    639, 4}, {   1471, 6}, {    383, 7}, \
127     {    209, 8}, {    119, 9}, {     63, 7}, {    255, 8}, \
128     {    139, 9}, {     71, 8}, {    143, 9}, {     79,10}, \
129     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
130     {    135,10}, {     79, 9}, {    159, 8}, {    319, 9}, \
131     {    167,10}, {     95,11}, {     63,10}, {    143, 9}, \
132     {    287,10}, {    159,11}, {     95,10}, {    191,12}, \
133     {     63,11}, {    127,10}, {    255, 9}, {    543, 8}, \
134     {   1087,10}, {    287, 9}, {    575,11}, {    159,10}, \
135     {    319, 9}, {    639,10}, {    335, 9}, {    671,10}, \
136     {    351, 9}, {    703,11}, {    191,10}, {    383, 9}, \
137     {    767,10}, {    399, 9}, {    799,10}, {    415, 9}, \
138     {    831,11}, {    223,12}, {    127,11}, {    255,10}, \
139     {    543, 9}, {   1087,11}, {    287,10}, {    607, 9}, \
140     {   1215,11}, {    319,10}, {    671, 9}, {   1343,11}, \
141     {    351,10}, {    703,12}, {    191,11}, {    383,10}, \
142     {    799,11}, {    415,10}, {    831,13}, {    127,12}, \
143     {    255,11}, {    543,10}, {   1087,11}, {    607,12}, \
144     {    319,11}, {    671,10}, {   1343,11}, {    735,12}, \
145     {    383,11}, {    799,10}, {   1599,11}, {    863,12}, \
146     {    447,11}, {    959,13}, {    255,12}, {    511,11}, \
147     {   1087,12}, {    575,11}, {   1215,12}, {    639,11}, \
148     {   1343,12}, {    703,11}, {   1471,13}, {    383,12}, \
149     {    767,11}, {   1599,12}, {    831,11}, {   1727,12}, \
150     {    959,14}, {    255,13}, {    511,12}, {   1215,13}, \
151     {    639,12}, {   1471,13}, {    767,12}, {   1727,13}, \
152     {    895,12}, {   1919,14}, {    511,13}, {   1023,12}, \
153     {   2111,13}, {   1151,12}, {   2431,13}, {   1407,14}, \
154     {    767,13}, {   1663,12}, {   3455,13}, {   8192,14}, \
155     {  16384,15}, {  32768,16} }
156 #define SQR_FFT_TABLE3_SIZE 146
157 #define SQR_FFT_THRESHOLD                 5760
158 
159 #define MULLO_BASECASE_THRESHOLD             0  /* always */
160 #define MULLO_DC_THRESHOLD                  31
161 #define MULLO_MUL_N_THRESHOLD            13463
162 #define SQRLO_BASECASE_THRESHOLD             0  /* always */
163 #define SQRLO_DC_THRESHOLD                 100
164 #define SQRLO_SQR_THRESHOLD               9236
165 
166 #define DC_DIV_QR_THRESHOLD                 25
167 #define DC_DIVAPPR_Q_THRESHOLD              55
168 #define DC_BDIV_QR_THRESHOLD                60
169 #define DC_BDIV_Q_THRESHOLD                132
170 
171 #define INV_MULMOD_BNM1_THRESHOLD           38
172 #define INV_NEWTON_THRESHOLD                65
173 #define INV_APPR_THRESHOLD                  65
174 
175 #define BINV_NEWTON_THRESHOLD              252
176 #define REDC_1_TO_REDC_N_THRESHOLD          62
177 
178 #define MU_DIV_QR_THRESHOLD               1164
179 #define MU_DIVAPPR_Q_THRESHOLD             748
180 #define MUPI_DIV_QR_THRESHOLD               38
181 #define MU_BDIV_QR_THRESHOLD              1360
182 #define MU_BDIV_Q_THRESHOLD               1470
183 
184 #define POWM_SEC_TABLE  2,23,258,879,2246
185 
186 #define GET_STR_DC_THRESHOLD                13
187 #define GET_STR_PRECOMPUTE_THRESHOLD        25
188 #define SET_STR_DC_THRESHOLD               582
189 #define SET_STR_PRECOMPUTE_THRESHOLD      1118
190 
191 #define FAC_DSC_THRESHOLD                  178
192 #define FAC_ODD_THRESHOLD                   34
193 
194 #define MATRIX22_STRASSEN_THRESHOLD         17
195 #define HGCD_THRESHOLD                      69
196 #define HGCD_APPR_THRESHOLD                112
197 #define HGCD_REDUCE_THRESHOLD             3389
198 #define GCD_DC_THRESHOLD                   386
199 #define GCDEXT_DC_THRESHOLD                303
200 #define JACOBI_BASE_METHOD                   1
201