xref: /netbsd-src/external/lgpl3/gmp/dist/mpn/x86/silvermont/gmp-mparam.h (revision 72c7faa4dbb41dbb0238d6b4a109da0d4b236dd4)
1*72c7faa4Smrg /* Intel Silvermont/32 gmp-mparam.h -- Compiler/machine parameter header file.
2*72c7faa4Smrg 
3*72c7faa4Smrg Copyright 2019 Free Software Foundation, Inc.
4*72c7faa4Smrg 
5*72c7faa4Smrg This file is part of the GNU MP Library.
6*72c7faa4Smrg 
7*72c7faa4Smrg The GNU MP Library is free software; you can redistribute it and/or modify
8*72c7faa4Smrg it under the terms of either:
9*72c7faa4Smrg 
10*72c7faa4Smrg   * the GNU Lesser General Public License as published by the Free
11*72c7faa4Smrg     Software Foundation; either version 3 of the License, or (at your
12*72c7faa4Smrg     option) any later version.
13*72c7faa4Smrg 
14*72c7faa4Smrg or
15*72c7faa4Smrg 
16*72c7faa4Smrg   * the GNU General Public License as published by the Free Software
17*72c7faa4Smrg     Foundation; either version 2 of the License, or (at your option) any
18*72c7faa4Smrg     later version.
19*72c7faa4Smrg 
20*72c7faa4Smrg or both in parallel, as here.
21*72c7faa4Smrg 
22*72c7faa4Smrg The GNU MP Library is distributed in the hope that it will be useful, but
23*72c7faa4Smrg WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
24*72c7faa4Smrg or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
25*72c7faa4Smrg for more details.
26*72c7faa4Smrg 
27*72c7faa4Smrg You should have received copies of the GNU General Public License and the
28*72c7faa4Smrg GNU Lesser General Public License along with the GNU MP Library.  If not,
29*72c7faa4Smrg see https://www.gnu.org/licenses/.  */
30*72c7faa4Smrg 
31*72c7faa4Smrg #define GMP_LIMB_BITS 32
32*72c7faa4Smrg #define GMP_LIMB_BYTES 4
33*72c7faa4Smrg 
34*72c7faa4Smrg /* 2400 MHz Intel Atom C2758 Silvermont/Rangeley */
35*72c7faa4Smrg /* FFT tuning limit = 67,000,000 */
36*72c7faa4Smrg /* Generated by tuneup.c, 2019-10-30, gcc 8.3 */
37*72c7faa4Smrg 
38*72c7faa4Smrg #define MOD_1_NORM_THRESHOLD                 3
39*72c7faa4Smrg #define MOD_1_UNNORM_THRESHOLD               5
40*72c7faa4Smrg #define MOD_1N_TO_MOD_1_1_THRESHOLD          9
41*72c7faa4Smrg #define MOD_1U_TO_MOD_1_1_THRESHOLD          5
42*72c7faa4Smrg #define MOD_1_1_TO_MOD_1_2_THRESHOLD        11
43*72c7faa4Smrg #define MOD_1_2_TO_MOD_1_4_THRESHOLD         0  /* never mpn_mod_1s_2p */
44*72c7faa4Smrg #define PREINV_MOD_1_TO_MOD_1_THRESHOLD     16
45*72c7faa4Smrg #define USE_PREINV_DIVREM_1                  1  /* native */
46*72c7faa4Smrg #define DIV_QR_1N_PI1_METHOD                 1  /* 64.62% faster than 2 */
47*72c7faa4Smrg #define DIV_QR_1_NORM_THRESHOLD          MP_SIZE_T_MAX  /* never */
48*72c7faa4Smrg #define DIV_QR_1_UNNORM_THRESHOLD        MP_SIZE_T_MAX  /* never */
49*72c7faa4Smrg #define DIV_QR_2_PI2_THRESHOLD           MP_SIZE_T_MAX  /* never */
50*72c7faa4Smrg #define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
51*72c7faa4Smrg #define BMOD_1_TO_MOD_1_THRESHOLD           32
52*72c7faa4Smrg 
53*72c7faa4Smrg #define DIV_1_VS_MUL_1_PERCENT             204
54*72c7faa4Smrg 
55*72c7faa4Smrg #define MUL_TOOM22_THRESHOLD                26
56*72c7faa4Smrg #define MUL_TOOM33_THRESHOLD               105
57*72c7faa4Smrg #define MUL_TOOM44_THRESHOLD               236
58*72c7faa4Smrg #define MUL_TOOM6H_THRESHOLD               351
59*72c7faa4Smrg #define MUL_TOOM8H_THRESHOLD               502
60*72c7faa4Smrg 
61*72c7faa4Smrg #define MUL_TOOM32_TO_TOOM43_THRESHOLD     105
62*72c7faa4Smrg #define MUL_TOOM32_TO_TOOM53_THRESHOLD     163
63*72c7faa4Smrg #define MUL_TOOM42_TO_TOOM53_THRESHOLD     137
64*72c7faa4Smrg #define MUL_TOOM42_TO_TOOM63_THRESHOLD     174
65*72c7faa4Smrg #define MUL_TOOM43_TO_TOOM54_THRESHOLD     215
66*72c7faa4Smrg 
67*72c7faa4Smrg #define SQR_BASECASE_THRESHOLD               0  /* always (native) */
68*72c7faa4Smrg #define SQR_TOOM2_THRESHOLD                 36
69*72c7faa4Smrg #define SQR_TOOM3_THRESHOLD                138
70*72c7faa4Smrg #define SQR_TOOM4_THRESHOLD                360
71*72c7faa4Smrg #define SQR_TOOM6_THRESHOLD                494
72*72c7faa4Smrg #define SQR_TOOM8_THRESHOLD                620
73*72c7faa4Smrg 
74*72c7faa4Smrg #define MULMID_TOOM42_THRESHOLD             58
75*72c7faa4Smrg 
76*72c7faa4Smrg #define MULMOD_BNM1_THRESHOLD               15
77*72c7faa4Smrg #define SQRMOD_BNM1_THRESHOLD               19
78*72c7faa4Smrg 
79*72c7faa4Smrg #define MUL_FFT_MODF_THRESHOLD             460  /* k = 5 */
80*72c7faa4Smrg #define MUL_FFT_TABLE3                                      \
81*72c7faa4Smrg   { {    460, 5}, {     21, 6}, {     11, 5}, {     23, 6}, \
82*72c7faa4Smrg     {     12, 5}, {     25, 6}, {     13, 5}, {     27, 6}, \
83*72c7faa4Smrg     {     21, 7}, {     11, 6}, {     25, 7}, {     13, 6}, \
84*72c7faa4Smrg     {     27, 7}, {     15, 6}, {     31, 7}, {     17, 6}, \
85*72c7faa4Smrg     {     35, 7}, {     19, 6}, {     39, 7}, {     21, 8}, \
86*72c7faa4Smrg     {     11, 7}, {     27, 8}, {     15, 7}, {     35, 8}, \
87*72c7faa4Smrg     {     19, 7}, {     41, 8}, {     23, 7}, {     47, 8}, \
88*72c7faa4Smrg     {     27, 9}, {     15, 8}, {     31, 7}, {     63, 8}, \
89*72c7faa4Smrg     {     39, 9}, {     23, 8}, {     51,10}, {     15, 9}, \
90*72c7faa4Smrg     {     31, 8}, {     67, 9}, {     39, 8}, {     79, 9}, \
91*72c7faa4Smrg     {     47, 8}, {     95,10}, {     31, 9}, {     79,10}, \
92*72c7faa4Smrg     {     47, 9}, {     95,11}, {     31,10}, {     63, 9}, \
93*72c7faa4Smrg     {    135,10}, {     79, 9}, {    159,10}, {     95,11}, \
94*72c7faa4Smrg     {     63,10}, {    127, 9}, {    255,10}, {    143, 9}, \
95*72c7faa4Smrg     {    287,10}, {    159,11}, {     95,10}, {    191,12}, \
96*72c7faa4Smrg     {     63,11}, {    127,10}, {    255, 9}, {    511,10}, \
97*72c7faa4Smrg     {    271, 9}, {    543,10}, {    287, 9}, {    575,11}, \
98*72c7faa4Smrg     {    159,10}, {    319, 9}, {    639,10}, {    335, 9}, \
99*72c7faa4Smrg     {    671,10}, {    351, 9}, {    703,11}, {    191,10}, \
100*72c7faa4Smrg     {    383, 9}, {    767,10}, {    415, 9}, {    831,12}, \
101*72c7faa4Smrg     {    127,11}, {    255,10}, {    543, 9}, {   1087,11}, \
102*72c7faa4Smrg     {    287,10}, {    607, 9}, {   1215,11}, {    319,10}, \
103*72c7faa4Smrg     {    671,11}, {    351,10}, {    735,12}, {    191,11}, \
104*72c7faa4Smrg     {    383,10}, {    767,11}, {    415,10}, {    831,13}, \
105*72c7faa4Smrg     {    127,12}, {    255,11}, {    543,10}, {   1087,11}, \
106*72c7faa4Smrg     {    607,10}, {   1215,12}, {    319,11}, {    671,10}, \
107*72c7faa4Smrg     {   1343,11}, {    735,10}, {   1471,12}, {    383,11}, \
108*72c7faa4Smrg     {    863,10}, {   1727,12}, {    447,11}, {    959,13}, \
109*72c7faa4Smrg     {    255,12}, {    511,11}, {   1087,12}, {    575,11}, \
110*72c7faa4Smrg     {   1215,10}, {   2431,12}, {    639,11}, {   1343,12}, \
111*72c7faa4Smrg     {    703,11}, {   1471,13}, {    383,12}, {    767,11}, \
112*72c7faa4Smrg     {   1535,12}, {    831,11}, {   1727,10}, {   3455,12}, \
113*72c7faa4Smrg     {    959,11}, {   1919,14}, {    255,13}, {    511,12}, \
114*72c7faa4Smrg     {   1215,11}, {   2431,13}, {    639,12}, {   1471,11}, \
115*72c7faa4Smrg     {   2943,13}, {    767,12}, {   1727,11}, {   3455,13}, \
116*72c7faa4Smrg     {    895,12}, {   1919,14}, {    511,13}, {   1023,12}, \
117*72c7faa4Smrg     {   2111,13}, {   1151,12}, {   2431,13}, {   1407,12}, \
118*72c7faa4Smrg     {   2943,14}, {    767,13}, {   1663,12}, {   3455,13}, \
119*72c7faa4Smrg     {   1919,15}, {    511,14}, {   1023,13}, {   2175,12}, \
120*72c7faa4Smrg     {   4479,13}, {   2431,14}, {   1279,13}, {   2943,12}, \
121*72c7faa4Smrg     {   5887,14}, {   1535,13}, {   3455,14}, {   1791,13}, \
122*72c7faa4Smrg     {   3967,12}, {   7935,15}, {   1023,14}, {   2047,13}, \
123*72c7faa4Smrg     {   4479,14}, {   2303,13}, {   4991,12}, {   9983,14}, \
124*72c7faa4Smrg     {   2815,13}, {   5887,15}, {   1535,14}, {   3839,13}, \
125*72c7faa4Smrg     {   7935,16} }
126*72c7faa4Smrg #define MUL_FFT_TABLE3_SIZE 177
127*72c7faa4Smrg #define MUL_FFT_THRESHOLD                 4544
128*72c7faa4Smrg 
129*72c7faa4Smrg #define SQR_FFT_MODF_THRESHOLD             400  /* k = 5 */
130*72c7faa4Smrg #define SQR_FFT_TABLE3                                      \
131*72c7faa4Smrg   { {    400, 5}, {     21, 6}, {     11, 5}, {     23, 6}, \
132*72c7faa4Smrg     {     12, 5}, {     25, 6}, {     13, 5}, {     28, 6}, \
133*72c7faa4Smrg     {     21, 7}, {     11, 6}, {     25, 7}, {     13, 6}, \
134*72c7faa4Smrg     {     28, 7}, {     15, 6}, {     32, 7}, {     17, 6}, \
135*72c7faa4Smrg     {     35, 7}, {     19, 6}, {     39, 7}, {     21, 8}, \
136*72c7faa4Smrg     {     11, 7}, {     27, 8}, {     15, 7}, {     35, 8}, \
137*72c7faa4Smrg     {     19, 7}, {     41, 8}, {     23, 7}, {     47, 8}, \
138*72c7faa4Smrg     {     27, 9}, {     15, 8}, {     31, 7}, {     63, 8}, \
139*72c7faa4Smrg     {     39, 9}, {     23, 8}, {     51,10}, {     15, 9}, \
140*72c7faa4Smrg     {     31, 8}, {     67, 9}, {     39, 8}, {     79, 9}, \
141*72c7faa4Smrg     {     55,10}, {     31, 9}, {     79,10}, {     47, 9}, \
142*72c7faa4Smrg     {     95,11}, {     31,10}, {     63, 9}, {    127,10}, \
143*72c7faa4Smrg     {     79, 9}, {    159,10}, {     95,11}, {     63,10}, \
144*72c7faa4Smrg     {    127, 9}, {    255,10}, {    143, 9}, {    287, 8}, \
145*72c7faa4Smrg     {    575,10}, {    159,11}, {     95,12}, {     63,11}, \
146*72c7faa4Smrg     {    127,10}, {    255, 9}, {    511,10}, {    271, 9}, \
147*72c7faa4Smrg     {    543,10}, {    287, 9}, {    575,11}, {    159,10}, \
148*72c7faa4Smrg     {    319, 9}, {    639,10}, {    335, 9}, {    671,10}, \
149*72c7faa4Smrg     {    351, 9}, {    735,11}, {    191,10}, {    383, 9}, \
150*72c7faa4Smrg     {    799,10}, {    415, 9}, {    831,11}, {    223,12}, \
151*72c7faa4Smrg     {    127,11}, {    255,10}, {    543, 9}, {   1087,11}, \
152*72c7faa4Smrg     {    287,10}, {    607, 9}, {   1215,11}, {    319,10}, \
153*72c7faa4Smrg     {    671,11}, {    351,10}, {    735, 9}, {   1471,12}, \
154*72c7faa4Smrg     {    191,11}, {    383,10}, {    799,11}, {    415,10}, \
155*72c7faa4Smrg     {    863,13}, {    127,12}, {    255,11}, {    543,10}, \
156*72c7faa4Smrg     {   1087,11}, {    607,10}, {   1215,12}, {    319,11}, \
157*72c7faa4Smrg     {    671,10}, {   1343,11}, {    735,10}, {   1471,12}, \
158*72c7faa4Smrg     {    383,11}, {    863,10}, {   1727,12}, {    447,11}, \
159*72c7faa4Smrg     {    959,13}, {    255,12}, {    511,11}, {   1087,12}, \
160*72c7faa4Smrg     {    575,11}, {   1215,10}, {   2431,12}, {    639,11}, \
161*72c7faa4Smrg     {   1343,12}, {    703,11}, {   1471,13}, {    383,12}, \
162*72c7faa4Smrg     {    767,11}, {   1535,12}, {    831,11}, {   1727,12}, \
163*72c7faa4Smrg     {    959,14}, {    255,13}, {    511,12}, {   1215,11}, \
164*72c7faa4Smrg     {   2431,13}, {    639,12}, {   1471,11}, {   2943,13}, \
165*72c7faa4Smrg     {    767,12}, {   1727,11}, {   3455,13}, {    895,12}, \
166*72c7faa4Smrg     {   1919,14}, {    511,13}, {   1023,12}, {   2111,13}, \
167*72c7faa4Smrg     {   1151,12}, {   2431,13}, {   1407,12}, {   2943,14}, \
168*72c7faa4Smrg     {    767,13}, {   1663,12}, {   3455,13}, {   1919,15}, \
169*72c7faa4Smrg     {    511,14}, {   1023,13}, {   2175,12}, {   4479,13}, \
170*72c7faa4Smrg     {   2431,14}, {   1279,13}, {   2943,12}, {   5887,14}, \
171*72c7faa4Smrg     {   1535,13}, {   3455,14}, {   1791,13}, {   3967,15}, \
172*72c7faa4Smrg     {   1023,14}, {   2047,13}, {   4479,14}, {   2303,13}, \
173*72c7faa4Smrg     {   4991,12}, {   9983,14}, {   2815,13}, {   5887,15}, \
174*72c7faa4Smrg     {   1535,14}, {   3839,13}, {   7679,16} }
175*72c7faa4Smrg #define SQR_FFT_TABLE3_SIZE 175
176*72c7faa4Smrg #define SQR_FFT_THRESHOLD                 3712
177*72c7faa4Smrg 
178*72c7faa4Smrg #define MULLO_BASECASE_THRESHOLD             0  /* always */
179*72c7faa4Smrg #define MULLO_DC_THRESHOLD                  56
180*72c7faa4Smrg #define MULLO_MUL_N_THRESHOLD             8907
181*72c7faa4Smrg #define SQRLO_BASECASE_THRESHOLD             0  /* always */
182*72c7faa4Smrg #define SQRLO_DC_THRESHOLD                 137
183*72c7faa4Smrg #define SQRLO_SQR_THRESHOLD               7373
184*72c7faa4Smrg 
185*72c7faa4Smrg #define DC_DIV_QR_THRESHOLD                 76
186*72c7faa4Smrg #define DC_DIVAPPR_Q_THRESHOLD             336
187*72c7faa4Smrg #define DC_BDIV_QR_THRESHOLD                66
188*72c7faa4Smrg #define DC_BDIV_Q_THRESHOLD                218
189*72c7faa4Smrg 
190*72c7faa4Smrg #define INV_MULMOD_BNM1_THRESHOLD           50
191*72c7faa4Smrg #define INV_NEWTON_THRESHOLD               345
192*72c7faa4Smrg #define INV_APPR_THRESHOLD                 342
193*72c7faa4Smrg 
194*72c7faa4Smrg #define BINV_NEWTON_THRESHOLD              366
195*72c7faa4Smrg #define REDC_1_TO_REDC_N_THRESHOLD          91
196*72c7faa4Smrg 
197*72c7faa4Smrg #define MU_DIV_QR_THRESHOLD               1652
198*72c7faa4Smrg #define MU_DIVAPPR_Q_THRESHOLD            1858
199*72c7faa4Smrg #define MUPI_DIV_QR_THRESHOLD              171
200*72c7faa4Smrg #define MU_BDIV_QR_THRESHOLD              1442
201*72c7faa4Smrg #define MU_BDIV_Q_THRESHOLD               1830
202*72c7faa4Smrg 
203*72c7faa4Smrg #define POWM_SEC_TABLE  3,17,102,404,1185
204*72c7faa4Smrg 
205*72c7faa4Smrg #define GET_STR_DC_THRESHOLD                14
206*72c7faa4Smrg #define GET_STR_PRECOMPUTE_THRESHOLD        21
207*72c7faa4Smrg #define SET_STR_DC_THRESHOLD               272
208*72c7faa4Smrg #define SET_STR_PRECOMPUTE_THRESHOLD       788
209*72c7faa4Smrg 
210*72c7faa4Smrg #define FAC_DSC_THRESHOLD                  132
211*72c7faa4Smrg #define FAC_ODD_THRESHOLD                   34
212*72c7faa4Smrg 
213*72c7faa4Smrg #define MATRIX22_STRASSEN_THRESHOLD         19
214*72c7faa4Smrg #define HGCD2_DIV1_METHOD                    1  /* 0.59% faster than 3 */
215*72c7faa4Smrg #define HGCD_THRESHOLD                     142
216*72c7faa4Smrg #define HGCD_APPR_THRESHOLD                181
217*72c7faa4Smrg #define HGCD_REDUCE_THRESHOLD             2681
218*72c7faa4Smrg #define GCD_DC_THRESHOLD                   492
219*72c7faa4Smrg #define GCDEXT_DC_THRESHOLD                365
220*72c7faa4Smrg #define JACOBI_BASE_METHOD                   1  /* 0.41% faster than 2 */
221*72c7faa4Smrg 
222*72c7faa4Smrg /* Tuneup completed successfully, took 147027 seconds */
223