xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/p6/sse2/gmp-mparam.h (revision 37afb7eb6895c833050f8bfb1d1bb2f99f332539)
1 /* Intel P6/sse2 gmp-mparam.h -- Compiler/machine parameter header file.
2 
3 Copyright 1991, 1993, 1994, 1999, 2000, 2001, 2002, 2003, 2008, 2009, 2010 Free
4 Software Foundation, Inc.
5 
6 This file is part of the GNU MP Library.
7 
8 The GNU MP Library is free software; you can redistribute it and/or modify
9 it under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or (at your
11 option) any later version.
12 
13 The GNU MP Library is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
15 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
16 License for more details.
17 
18 You should have received a copy of the GNU Lesser General Public License
19 along with the GNU MP Library.  If not, see http://www.gnu.org/licenses/.  */
20 
21 
22 #define GMP_LIMB_BITS 32
23 #define BYTES_PER_MP_LIMB 4
24 
25 
26 /* NOTE: In a fat binary build SQR_TOOM2_THRESHOLD here cannot be more than the
27    value in mpn/x86/p6/gmp-mparam.h.  The latter is used as a hard limit in
28    mpn/x86/p6/sqr_basecase.asm.  */
29 
30 
31 /* 1867 MHz P6 model 13 */
32 
33 #define MOD_1_NORM_THRESHOLD                 4
34 #define MOD_1_UNNORM_THRESHOLD               4
35 #define MOD_1N_TO_MOD_1_1_THRESHOLD          5
36 #define MOD_1U_TO_MOD_1_1_THRESHOLD          4
37 #define MOD_1_1_TO_MOD_1_2_THRESHOLD        11
38 #define MOD_1_2_TO_MOD_1_4_THRESHOLD         0  /* never mpn_mod_1s_2p */
39 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD      8
40 #define USE_PREINV_DIVREM_1                  1  /* native */
41 #define DIV_QR_2_PI2_THRESHOLD           MP_SIZE_T_MAX  /* never */
42 #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
43 #define BMOD_1_TO_MOD_1_THRESHOLD           21
44 
45 #define MUL_TOOM22_THRESHOLD                20
46 #define MUL_TOOM33_THRESHOLD                77
47 #define MUL_TOOM44_THRESHOLD               169
48 #define MUL_TOOM6H_THRESHOLD               246
49 #define MUL_TOOM8H_THRESHOLD               381
50 
51 #define MUL_TOOM32_TO_TOOM43_THRESHOLD      73
52 #define MUL_TOOM32_TO_TOOM53_THRESHOLD     114
53 #define MUL_TOOM42_TO_TOOM53_THRESHOLD      97
54 #define MUL_TOOM42_TO_TOOM63_THRESHOLD      80
55 #define MUL_TOOM43_TO_TOOM54_THRESHOLD     106
56 
57 #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
58 #define SQR_TOOM2_THRESHOLD                 30
59 #define SQR_TOOM3_THRESHOLD                101
60 #define SQR_TOOM4_THRESHOLD                154
61 #define SQR_TOOM6_THRESHOLD                222
62 #define SQR_TOOM8_THRESHOLD                527
63 
64 #define MULMID_TOOM42_THRESHOLD             58
65 
66 #define MULMOD_BNM1_THRESHOLD               13
67 #define SQRMOD_BNM1_THRESHOLD               17
68 
69 #define MUL_FFT_MODF_THRESHOLD             690  /* k = 5 */
70 #define MUL_FFT_TABLE3                                      \
71   { {    565, 5}, {     25, 6}, {     13, 5}, {     27, 6}, \
72     {     25, 7}, {     13, 6}, {     28, 7}, {     15, 6}, \
73     {     31, 7}, {     17, 6}, {     35, 7}, {     27, 8}, \
74     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
75     {     23, 7}, {     47, 8}, {     27, 9}, {     15, 8}, \
76     {     31, 7}, {     63, 8}, {     39, 9}, {     23, 5}, \
77     {    383, 4}, {    991, 5}, {    511, 6}, {    267, 7}, \
78     {    157, 8}, {     91, 9}, {     47, 8}, {    111, 9}, \
79     {     63, 8}, {    127, 9}, {     79,10}, {     47, 9}, \
80     {     95,11}, {     31,10}, {     63, 9}, {    135,10}, \
81     {     79, 9}, {    159,10}, {     95,11}, {     63,10}, \
82     {    143, 9}, {    287,10}, {    159,11}, {     95,10}, \
83     {    191,12}, {     63,11}, {    127,10}, {    255, 9}, \
84     {    511,10}, {    271, 9}, {    543,10}, {    287,11}, \
85     {    159,10}, {    335, 9}, {    671,11}, {    191,10}, \
86     {    383, 9}, {    767,10}, {    399, 9}, {    799,10}, \
87     {    415,11}, {    223,12}, {    127,11}, {    255,10}, \
88     {    543, 9}, {   1087,11}, {    287,10}, {    607,11}, \
89     {    319,10}, {    671,12}, {    191,11}, {    383,10}, \
90     {    799,11}, {    415,10}, {    831,13}, {    127,12}, \
91     {    255,11}, {    543,10}, {   1087,11}, {    607,10}, \
92     {   1215,12}, {    319,11}, {    671,10}, {   1343,11}, \
93     {    735,10}, {   1471,12}, {    383,11}, {    799,10}, \
94     {   1599,11}, {    863,12}, {    447,11}, {    959,13}, \
95     {    255,12}, {    511,11}, {   1087,12}, {    575,11}, \
96     {   1215,12}, {    639,11}, {   1343,12}, {    703,11}, \
97     {   1471,13}, {    383,12}, {    831,11}, {   1727,12}, \
98     {    959,14}, {    255,13}, {    511,12}, {   1215,13}, \
99     {    639,12}, {   1471,11}, {   2943,13}, {    767,12}, \
100     {   1727,13}, {    895,12}, {   1919,14}, {    511,13}, \
101     {   1023,12}, {   2111,13}, {   1151,12}, {   2431,13}, \
102     {   1407,12}, {   2815,14}, {    767,13}, {   1663,12}, \
103     {   3455,13}, {   8192,14}, {  16384,15}, {  32768,16} }
104 #define MUL_FFT_TABLE3_SIZE 132
105 #define MUL_FFT_THRESHOLD                 7424
106 
107 #define SQR_FFT_MODF_THRESHOLD             565  /* k = 5 */
108 #define SQR_FFT_TABLE3                                      \
109   { {    472, 5}, {     25, 6}, {     13, 5}, {     27, 6}, \
110     {     25, 7}, {     13, 6}, {     27, 7}, {     15, 6}, \
111     {     31, 7}, {     17, 6}, {     35, 7}, {     27, 8}, \
112     {     15, 7}, {     35, 8}, {     19, 7}, {     41, 8}, \
113     {     23, 7}, {     49, 8}, {     27, 9}, {     15, 8}, \
114     {     39, 9}, {     23, 8}, {     51,10}, {     15, 9}, \
115     {     31, 8}, {     63, 4}, {   1023, 8}, {     67, 9}, \
116     {     39, 5}, {    639, 4}, {   1471, 6}, {    383, 7}, \
117     {    209, 8}, {    119, 9}, {     63, 7}, {    255, 8}, \
118     {    139, 9}, {     71, 8}, {    143, 9}, {     79,10}, \
119     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
120     {    135,10}, {     79, 9}, {    159, 8}, {    319, 9}, \
121     {    167,10}, {     95,11}, {     63,10}, {    143, 9}, \
122     {    287,10}, {    159,11}, {     95,10}, {    191,12}, \
123     {     63,11}, {    127,10}, {    255, 9}, {    543, 8}, \
124     {   1087,10}, {    287, 9}, {    575,11}, {    159,10}, \
125     {    319, 9}, {    639,10}, {    335, 9}, {    671,10}, \
126     {    351, 9}, {    703,11}, {    191,10}, {    383, 9}, \
127     {    767,10}, {    399, 9}, {    799,10}, {    415, 9}, \
128     {    831,11}, {    223,12}, {    127,11}, {    255,10}, \
129     {    543, 9}, {   1087,11}, {    287,10}, {    607, 9}, \
130     {   1215,11}, {    319,10}, {    671, 9}, {   1343,11}, \
131     {    351,10}, {    703,12}, {    191,11}, {    383,10}, \
132     {    799,11}, {    415,10}, {    831,13}, {    127,12}, \
133     {    255,11}, {    543,10}, {   1087,11}, {    607,12}, \
134     {    319,11}, {    671,10}, {   1343,11}, {    735,12}, \
135     {    383,11}, {    799,10}, {   1599,11}, {    863,12}, \
136     {    447,11}, {    959,13}, {    255,12}, {    511,11}, \
137     {   1087,12}, {    575,11}, {   1215,12}, {    639,11}, \
138     {   1343,12}, {    703,11}, {   1471,13}, {    383,12}, \
139     {    767,11}, {   1599,12}, {    831,11}, {   1727,12}, \
140     {    959,14}, {    255,13}, {    511,12}, {   1215,13}, \
141     {    639,12}, {   1471,13}, {    767,12}, {   1727,13}, \
142     {    895,12}, {   1919,14}, {    511,13}, {   1023,12}, \
143     {   2111,13}, {   1151,12}, {   2431,13}, {   1407,14}, \
144     {    767,13}, {   1663,12}, {   3455,13}, {   8192,14}, \
145     {  16384,15}, {  32768,16} }
146 #define SQR_FFT_TABLE3_SIZE 146
147 #define SQR_FFT_THRESHOLD                 5760
148 
149 #define MULLO_BASECASE_THRESHOLD             0  /* always */
150 #define MULLO_DC_THRESHOLD                  31
151 #define MULLO_MUL_N_THRESHOLD            13463
152 
153 #define DC_DIV_QR_THRESHOLD                 25
154 #define DC_DIVAPPR_Q_THRESHOLD              55
155 #define DC_BDIV_QR_THRESHOLD                60
156 #define DC_BDIV_Q_THRESHOLD                132
157 
158 #define INV_MULMOD_BNM1_THRESHOLD           38
159 #define INV_NEWTON_THRESHOLD                65
160 #define INV_APPR_THRESHOLD                  65
161 
162 #define BINV_NEWTON_THRESHOLD              252
163 #define REDC_1_TO_REDC_N_THRESHOLD          62
164 
165 #define MU_DIV_QR_THRESHOLD               1164
166 #define MU_DIVAPPR_Q_THRESHOLD             748
167 #define MUPI_DIV_QR_THRESHOLD               38
168 #define MU_BDIV_QR_THRESHOLD              1360
169 #define MU_BDIV_Q_THRESHOLD               1470
170 
171 #define POWM_SEC_TABLE  2,23,258,879,2246
172 
173 #define MATRIX22_STRASSEN_THRESHOLD         17
174 #define HGCD_THRESHOLD                      69
175 #define HGCD_APPR_THRESHOLD                112
176 #define HGCD_REDUCE_THRESHOLD             3389
177 #define GCD_DC_THRESHOLD                   386
178 #define GCDEXT_DC_THRESHOLD                303
179 #define JACOBI_BASE_METHOD                   1
180 
181 #define GET_STR_DC_THRESHOLD                13
182 #define GET_STR_PRECOMPUTE_THRESHOLD        25
183 #define SET_STR_DC_THRESHOLD               582
184 #define SET_STR_PRECOMPUTE_THRESHOLD      1118
185 
186 #define FAC_DSC_THRESHOLD                  178
187 #define FAC_ODD_THRESHOLD                   34
188