1/* AMD Bobcat gmp-mparam.h -- Compiler/machine parameter header file.
2
3Copyright 2019 Free Software Foundation, Inc.
4
5This file is part of the GNU MP Library.
6
7The GNU MP Library is free software; you can redistribute it and/or modify
8it 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
14or
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
20or both in parallel, as here.
21
22The GNU MP Library is distributed in the hope that it will be useful, but
23WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
24or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
25for more details.
26
27You should have received copies of the GNU General Public License and the
28GNU Lesser General Public License along with the GNU MP Library.  If not,
29see https://www.gnu.org/licenses/.  */
30
31#define GMP_LIMB_BITS 64
32#define GMP_LIMB_BYTES 8
33
34/* Disable use of slow functions.  FIXME: We should disable lib inclusion.  */
35#undef HAVE_NATIVE_mpn_mul_2
36#undef HAVE_NATIVE_mpn_addmul_2
37
38/* 1600 MHz AMD Bobcat/Zacate */
39/* FFT tuning limit = 110,472,704 */
40/* Generated by tuneup.c, 2019-10-12, gcc 8.3 */
41
42#define MOD_1_NORM_THRESHOLD                 0  /* always */
43#define MOD_1_UNNORM_THRESHOLD               0  /* always */
44#define MOD_1N_TO_MOD_1_1_THRESHOLD          7
45#define MOD_1U_TO_MOD_1_1_THRESHOLD          3
46#define MOD_1_1_TO_MOD_1_2_THRESHOLD        31
47#define MOD_1_2_TO_MOD_1_4_THRESHOLD        71
48#define PREINV_MOD_1_TO_MOD_1_THRESHOLD     14
49#define USE_PREINV_DIVREM_1                  1  /* native */
50#define DIV_QR_1_NORM_THRESHOLD              1
51#define DIV_QR_1_UNNORM_THRESHOLD        MP_SIZE_T_MAX  /* never */
52#define DIV_QR_2_PI2_THRESHOLD           MP_SIZE_T_MAX  /* never */
53#define DIVEXACT_1_THRESHOLD                 0  /* always (native) */
54#define BMOD_1_TO_MOD_1_THRESHOLD           20
55
56#define DIV_1_VS_MUL_1_PERCENT             270
57
58#define MUL_TOOM22_THRESHOLD                24
59#define MUL_TOOM33_THRESHOLD                66
60#define MUL_TOOM44_THRESHOLD               190
61#define MUL_TOOM6H_THRESHOLD               274
62#define MUL_TOOM8H_THRESHOLD               381
63
64#define MUL_TOOM32_TO_TOOM43_THRESHOLD     129
65#define MUL_TOOM32_TO_TOOM53_THRESHOLD     138
66#define MUL_TOOM42_TO_TOOM53_THRESHOLD     127
67#define MUL_TOOM42_TO_TOOM63_THRESHOLD     131
68#define MUL_TOOM43_TO_TOOM54_THRESHOLD     100
69
70#define SQR_BASECASE_THRESHOLD               0  /* always (native) */
71#define SQR_TOOM2_THRESHOLD                 30
72#define SQR_TOOM3_THRESHOLD                101
73#define SQR_TOOM4_THRESHOLD                278
74#define SQR_TOOM6_THRESHOLD                372
75#define SQR_TOOM8_THRESHOLD                478
76
77#define MULMID_TOOM42_THRESHOLD             22
78
79#define MULMOD_BNM1_THRESHOLD               11
80#define SQRMOD_BNM1_THRESHOLD               13
81
82#define MUL_FFT_MODF_THRESHOLD             444  /* k = 5 */
83#define MUL_FFT_TABLE3                                      \
84  { {    444, 5}, {     19, 6}, {     10, 5}, {     21, 6}, \
85    {     11, 5}, {     23, 6}, {     21, 7}, {     11, 6}, \
86    {     25, 7}, {     13, 6}, {     27, 7}, {     21, 8}, \
87    {     11, 7}, {     25, 8}, {     13, 7}, {     28, 8}, \
88    {     15, 7}, {     31, 8}, {     17, 7}, {     35, 8}, \
89    {     19, 7}, {     39, 8}, {     27, 9}, {     15, 8}, \
90    {     35, 9}, {     19, 8}, {     41, 9}, {     23, 8}, \
91    {     49, 9}, {     27,10}, {     15, 9}, {     39,10}, \
92    {     23, 9}, {     55,11}, {     15,10}, {     31, 9}, \
93    {     71,10}, {     39, 9}, {     83, 5}, {   1343, 4}, \
94    {   2687, 5}, {   1407, 6}, {    735, 7}, {    415, 8}, \
95    {    223,10}, {     79,11}, {     47,10}, {    103,12}, \
96    {     31,11}, {     63,10}, {    135,11}, {     79,10}, \
97    {    167,11}, {     95,10}, {    191,11}, {    111,12}, \
98    {     63,11}, {    127,10}, {    255,11}, {    143,10}, \
99    {    287, 9}, {    575,11}, {    159,12}, {     95,11}, \
100    {    191,10}, {    383,11}, {    207,10}, {    415,13}, \
101    {     63,12}, {    127,11}, {    255,10}, {    511,11}, \
102    {    271,10}, {    543,11}, {    287,10}, {    575,12}, \
103    {    159,11}, {    319,10}, {    639,11}, {    351,10}, \
104    {    703,12}, {    191,11}, {    383,10}, {    767,11}, \
105    {    415,12}, {    223,13}, {    127,12}, {    255,11}, \
106    {    543,12}, {    287,11}, {    607,12}, {    319,11}, \
107    {    671,12}, {    351,11}, {    703,13}, {    191,12}, \
108    {    383,11}, {    767,12}, {    415,11}, {    831,12}, \
109    {    447,14}, {    127,13}, {    255,12}, {    607,13}, \
110    {    319,12}, {    703,13}, {    383,12}, {    831,13}, \
111    {    447,12}, {    959,14}, {    255,13}, {    511,12}, \
112    {   1023,13}, {    575,12}, {   1151,13}, {    703,14}, \
113    {    383,13}, {    831,12}, {   1663,13}, {    959,15}, \
114    {    255,14}, {    511,13}, {   1087,12}, {   2175,13}, \
115    {   1151,14}, {    639,13}, {   1343,12}, {   2687,13}, \
116    {   1407,14}, {    767,13}, {   1599,12}, {   3199,13}, \
117    {   1663,14}, {    895,15}, {    511,14}, {   1023,13}, \
118    {   2175,14}, {   1151,13}, {   2431,12}, {   4863,14}, \
119    {   1279,13}, {   2687,14}, {   1407,15}, {    767,14}, \
120    {   1535,13}, {   3199,14}, {   1663,13}, {   3455,16}, \
121    {    511,15}, {   1023,14}, {   2175,13}, {   4479,14}, \
122    {   2431,13}, {   4863,15}, {   1279,14}, {   2943,13}, \
123    {   5887,15}, {   1535,14}, {   3455,13}, {   6911,15}, \
124    {   1791,14}, {   3839,16}, {   1023,15}, {   2047,14}, \
125    {   4479,15}, {   2303,14}, {   4991,15}, {   2559,14}, \
126    {   5247,15}, {   2815,14}, {   5887,16}, {   1535,15}, \
127    {   3327,14}, {  16384,15}, {  32768,16}, {  65536,17}, \
128    { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \
129    {2097152,22}, {4194304,23}, {8388608,24} }
130#define MUL_FFT_TABLE3_SIZE 183
131#define MUL_FFT_THRESHOLD                 5760
132
133#define SQR_FFT_MODF_THRESHOLD             380  /* k = 5 */
134#define SQR_FFT_TABLE3                                      \
135  { {    380, 5}, {     21, 6}, {     11, 5}, {     23, 6}, \
136    {     25, 7}, {     25, 8}, {     13, 7}, {     27, 8}, \
137    {     15, 7}, {     31, 8}, {     17, 7}, {     35, 8}, \
138    {     19, 7}, {     39, 8}, {     27, 9}, {     15, 8}, \
139    {     35, 9}, {     19, 8}, {     41, 9}, {     23, 8}, \
140    {     47, 9}, {     27,10}, {     15, 9}, {     31, 8}, \
141    {     63, 9}, {     39,10}, {     23, 9}, {     51,11}, \
142    {     15,10}, {     31, 9}, {     67,10}, {     39, 9}, \
143    {     79,10}, {     47, 9}, {     95,10}, {     55,11}, \
144    {     31,10}, {     63, 6}, {   1087, 7}, {    575, 8}, \
145    {    303, 9}, {    159,10}, {    103,12}, {     31,11}, \
146    {     63,10}, {    127, 9}, {    255,10}, {    135,11}, \
147    {     79,10}, {    159, 9}, {    319,11}, {     95,10}, \
148    {    191, 9}, {    383,11}, {    111,12}, {     63,11}, \
149    {    127,10}, {    255, 9}, {    511,10}, {    271,11}, \
150    {    143,10}, {    287, 9}, {    575,10}, {    303,11}, \
151    {    159,10}, {    319,12}, {     95,11}, {    191,10}, \
152    {    383,11}, {    207,13}, {     63,12}, {    127,11}, \
153    {    255,10}, {    511,11}, {    271,10}, {    543,11}, \
154    {    287,10}, {    575,11}, {    303,12}, {    159,11}, \
155    {    319,10}, {    639,11}, {    335,10}, {    671,11}, \
156    {    351,10}, {    703,12}, {    191,11}, {    383,10}, \
157    {    767,11}, {    415,12}, {    223,11}, {    447,13}, \
158    {    127,12}, {    255,11}, {    543,12}, {    287,11}, \
159    {    607,12}, {    319,11}, {    671,12}, {    351,11}, \
160    {    703,13}, {    191,12}, {    383,11}, {    767,12}, \
161    {    415,11}, {    831,12}, {    479,14}, {    127,13}, \
162    {    255,12}, {    607,13}, {    319,12}, {    703,13}, \
163    {    383,12}, {    831,13}, {    447,12}, {    895,14}, \
164    {    255,13}, {    511,12}, {   1023,13}, {    703,14}, \
165    {    383,13}, {    831,12}, {   1663,13}, {    895,15}, \
166    {    255,14}, {    511,13}, {   1087,12}, {   2175,13}, \
167    {   1151,14}, {    639,13}, {   1343,12}, {   2687,13}, \
168    {   1407,14}, {    767,13}, {   1599,12}, {   3199,13}, \
169    {   1663,14}, {    895,15}, {    511,14}, {   1023,13}, \
170    {   2175,14}, {   1151,13}, {   2431,12}, {   4863,14}, \
171    {   1279,13}, {   2687,14}, {   1407,15}, {    767,14}, \
172    {   1535,13}, {   3199,14}, {   1663,13}, {   3455,16}, \
173    {    511,15}, {   1023,14}, {   2175,13}, {   4351,14}, \
174    {   2431,13}, {   4863,15}, {   1279,14}, {   2943,13}, \
175    {   5887,15}, {   1535,14}, {   3455,15}, {   1791,14}, \
176    {   3839,16}, {   1023,15}, {   2047,14}, {   4479,15}, \
177    {   2303,14}, {   4863,15}, {   2559,14}, {   5247,15}, \
178    {   2815,14}, {   5887,16}, {   1535,15}, {   3327,14}, \
179    {  16384,15}, {  32768,16}, {  65536,17}, { 131072,18}, \
180    { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \
181    {4194304,23}, {8388608,24} }
182#define SQR_FFT_TABLE3_SIZE 186
183#define SQR_FFT_THRESHOLD                 3712
184
185#define MULLO_BASECASE_THRESHOLD             0  /* always */
186#define MULLO_DC_THRESHOLD                  42
187#define MULLO_MUL_N_THRESHOLD            10950
188#define SQRLO_BASECASE_THRESHOLD             7
189#define SQRLO_DC_THRESHOLD                 100
190#define SQRLO_SQR_THRESHOLD               7293
191
192#define DC_DIV_QR_THRESHOLD                 70
193#define DC_DIVAPPR_Q_THRESHOLD             204
194#define DC_BDIV_QR_THRESHOLD                59
195#define DC_BDIV_Q_THRESHOLD                148
196
197#define INV_MULMOD_BNM1_THRESHOLD           46
198#define INV_NEWTON_THRESHOLD               246
199#define INV_APPR_THRESHOLD                 236
200
201#define BINV_NEWTON_THRESHOLD              252
202#define REDC_1_TO_REDC_2_THRESHOLD          67
203#define REDC_2_TO_REDC_N_THRESHOLD           0  /* always */
204
205#define MU_DIV_QR_THRESHOLD               1589
206#define MU_DIVAPPR_Q_THRESHOLD            1589
207#define MUPI_DIV_QR_THRESHOLD              108
208#define MU_BDIV_QR_THRESHOLD              1442
209#define MU_BDIV_Q_THRESHOLD               1470
210
211#define POWM_SEC_TABLE  1,16,194,960,1603,1811,2499
212
213#define GET_STR_DC_THRESHOLD                20
214#define GET_STR_PRECOMPUTE_THRESHOLD        34
215#define SET_STR_DC_THRESHOLD               345
216#define SET_STR_PRECOMPUTE_THRESHOLD      1787
217
218#define FAC_DSC_THRESHOLD                  781
219#define FAC_ODD_THRESHOLD                  104
220
221#define MATRIX22_STRASSEN_THRESHOLD         17
222#define HGCD2_DIV1_METHOD                    3  /* 3.20% faster than 5 */
223#define HGCD_THRESHOLD                     110
224#define HGCD_APPR_THRESHOLD                 50
225#define HGCD_REDUCE_THRESHOLD             2681
226#define GCD_DC_THRESHOLD                   474
227#define GCDEXT_DC_THRESHOLD                293
228#define JACOBI_BASE_METHOD                   2  /* 9.38% faster than 1 */
229
230/* Tuneup completed successfully, took 358881 seconds */
231