1 /* AMD Zen3 gmp-mparam.h -- Compiler/machine parameter header file.
2
3 Copyright 2021 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 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 /* 3800-4700 MHz Vermeer */
39 /* FFT tuning limit = 10,000,000 */
40 /* Generated by tuneup.c, 2021-01-01, gcc 9.3 */
41
42 #define MOD_1_NORM_THRESHOLD 64
43 #define MOD_1_UNNORM_THRESHOLD 85
44 #define MOD_1N_TO_MOD_1_1_THRESHOLD 5
45 #define MOD_1U_TO_MOD_1_1_THRESHOLD 6
46 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 11
47 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 35
48 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 9
49 #define USE_PREINV_DIVREM_1 1 /* native */
50 #define DIV_QR_1_NORM_THRESHOLD 9
51 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
52 #define DIV_QR_2_PI2_THRESHOLD 15
53 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
54 #define BMOD_1_TO_MOD_1_THRESHOLD 18
55
56 #define DIV_1_VS_MUL_1_PERCENT 398
57
58 #define MUL_TOOM22_THRESHOLD 20
59 #define MUL_TOOM33_THRESHOLD 89
60 #define MUL_TOOM44_THRESHOLD 130
61 #define MUL_TOOM6H_THRESHOLD 303
62 #define MUL_TOOM8H_THRESHOLD 418
63
64 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 89
65 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 107
66 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 87
67 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 94
68 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 130
69
70 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
71 #define SQR_TOOM2_THRESHOLD 32
72 #define SQR_TOOM3_THRESHOLD 109
73 #define SQR_TOOM4_THRESHOLD 336
74 #define SQR_TOOM6_THRESHOLD 414
75 #define SQR_TOOM8_THRESHOLD 592
76
77 #define MULMID_TOOM42_THRESHOLD 38
78
79 #define MULMOD_BNM1_THRESHOLD 14
80 #define SQRMOD_BNM1_THRESHOLD 17
81
82 #define MUL_FFT_MODF_THRESHOLD 332 /* k = 5 */
83 #define MUL_FFT_TABLE3 \
84 { { 332, 5}, { 19, 6}, { 10, 5}, { 21, 6}, \
85 { 21, 7}, { 11, 6}, { 23, 7}, { 12, 6}, \
86 { 25, 7}, { 21, 8}, { 11, 7}, { 25, 8}, \
87 { 13, 7}, { 27, 8}, { 15, 7}, { 31, 8}, \
88 { 17, 7}, { 35, 8}, { 21, 9}, { 11, 8}, \
89 { 27, 9}, { 15, 8}, { 35, 9}, { 19, 8}, \
90 { 39, 9}, { 23, 8}, { 47, 9}, { 27,10}, \
91 { 15, 9}, { 31, 8}, { 63, 9}, { 35, 8}, \
92 { 73, 9}, { 39,10}, { 23, 9}, { 51,11}, \
93 { 15,10}, { 31, 9}, { 71,10}, { 39, 9}, \
94 { 79,10}, { 55,11}, { 31,10}, { 79,11}, \
95 { 47,10}, { 95,12}, { 31,11}, { 63,10}, \
96 { 127, 9}, { 255,10}, { 135,11}, { 79,10}, \
97 { 159, 9}, { 319, 8}, { 639,10}, { 175, 9}, \
98 { 351,11}, { 95,10}, { 191, 9}, { 383,10}, \
99 { 207,11}, { 111,10}, { 223,12}, { 63,11}, \
100 { 127,10}, { 255, 9}, { 511,10}, { 271, 9}, \
101 { 543,11}, { 143,10}, { 287, 9}, { 575,10}, \
102 { 303, 9}, { 607,11}, { 159,10}, { 319, 9}, \
103 { 639,11}, { 175,10}, { 351,12}, { 95,11}, \
104 { 191,10}, { 383,11}, { 207,10}, { 415,11}, \
105 { 223,13}, { 63,12}, { 127,11}, { 255,10}, \
106 { 511,11}, { 271,10}, { 543,11}, { 287,10}, \
107 { 575,11}, { 303,10}, { 607,12}, { 159,11}, \
108 { 319,10}, { 639,11}, { 351,10}, { 703,12}, \
109 { 191,11}, { 415,12}, { 223,11}, { 479,13}, \
110 { 127,12}, { 255,11}, { 543,10}, { 1087,12}, \
111 { 287,11}, { 575,10}, { 1151,11}, { 607,12}, \
112 { 319,11}, { 639,12}, { 351,11}, { 703,13}, \
113 { 191,12}, { 383,11}, { 767,12}, { 415,11}, \
114 { 831,10}, { 1663,12}, { 447,11}, { 895,12}, \
115 { 479,14}, { 127,13}, { 255,12}, { 511,11}, \
116 { 1023,12}, { 543,11}, { 1087,12}, { 575,11}, \
117 { 1151,12}, { 607,13}, { 319,12}, { 639,11}, \
118 { 1279,12}, { 671,11}, { 1343,10}, { 2687,12}, \
119 { 703,13}, { 383,12}, { 767,11}, { 1535,12}, \
120 { 831,11}, { 1663,13}, { 447,12}, { 895,11}, \
121 { 1791,12}, { 959,11}, { 1919,10}, { 3839,14}, \
122 { 255,13}, { 511,12}, { 1087,13}, { 575,12}, \
123 { 1151,11}, { 2303,12}, { 1215,11}, { 2431,13}, \
124 { 639,12}, { 1279, 8}, { 24063,10}, { 6399,11}, \
125 { 3327,13}, { 895,12}, { 1791,13}, { 959,12}, \
126 { 1919,11}, { 3839,15}, { 255,14}, { 511,13}, \
127 { 1087,12}, { 2175,13}, { 1215,12}, { 2431,14}, \
128 { 16384,15}, { 32768,16}, { 65536,17}, { 131072,18}, \
129 { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \
130 {4194304,23}, {8388608,24} }
131 #define MUL_FFT_TABLE3_SIZE 186
132 #define MUL_FFT_THRESHOLD 3264
133
134 #define SQR_FFT_MODF_THRESHOLD 340 /* k = 5 */
135 #define SQR_FFT_TABLE3 \
136 { { 340, 5}, { 19, 6}, { 10, 5}, { 21, 6}, \
137 { 11, 5}, { 23, 6}, { 25, 7}, { 13, 6}, \
138 { 27, 7}, { 25, 8}, { 13, 7}, { 28, 8}, \
139 { 15, 7}, { 31, 8}, { 21, 9}, { 11, 8}, \
140 { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \
141 { 43, 9}, { 23, 8}, { 47, 9}, { 27,10}, \
142 { 15, 9}, { 31, 8}, { 63, 9}, { 43,10}, \
143 { 23, 9}, { 47,11}, { 15,10}, { 31, 9}, \
144 { 67,10}, { 39, 9}, { 83,10}, { 47, 9}, \
145 { 95,10}, { 55,11}, { 31,10}, { 79,11}, \
146 { 47,10}, { 95,12}, { 31,11}, { 63,10}, \
147 { 127, 9}, { 255,10}, { 135,11}, { 79,10}, \
148 { 159, 9}, { 319,11}, { 95,10}, { 191, 9}, \
149 { 383,12}, { 63,11}, { 127,10}, { 255, 9}, \
150 { 511,10}, { 271,11}, { 143,10}, { 287, 9}, \
151 { 575,10}, { 303, 9}, { 607,11}, { 159,10}, \
152 { 319, 9}, { 639,12}, { 95,11}, { 191,10}, \
153 { 383,13}, { 63,12}, { 127,11}, { 255,10}, \
154 { 511,11}, { 271,10}, { 543,11}, { 287,10}, \
155 { 575,11}, { 303,10}, { 607,12}, { 159,11}, \
156 { 319,10}, { 639,11}, { 335,10}, { 671,11}, \
157 { 351,10}, { 703,12}, { 191,11}, { 383,10}, \
158 { 767,11}, { 415,12}, { 223,11}, { 447,13}, \
159 { 127,12}, { 255,11}, { 543,12}, { 287,11}, \
160 { 607,12}, { 319,11}, { 671,12}, { 351,11}, \
161 { 703,13}, { 191,12}, { 383,11}, { 767,12}, \
162 { 415,11}, { 831,12}, { 479,14}, { 127,13}, \
163 { 255,12}, { 511,11}, { 1023,12}, { 607,11}, \
164 { 1215,13}, { 319,12}, { 671,11}, { 1343,12}, \
165 { 735,13}, { 383,12}, { 799,11}, { 1599,10}, \
166 { 3199,12}, { 831,13}, { 447,12}, { 895,11}, \
167 { 1791,12}, { 959,11}, { 1919,13}, { 511,12}, \
168 { 1087,13}, { 575,12}, { 1215,13}, { 639,12}, \
169 { 1343,13}, { 703,12}, { 1407,14}, { 383,13}, \
170 { 767,12}, { 1599,13}, { 831,12}, { 1663,13}, \
171 { 895,12}, { 1791,13}, { 959,12}, { 1919,15}, \
172 { 255,14}, { 511,13}, { 1023, 9}, { 17919,10}, \
173 { 9727,12}, { 4096,13}, { 8192,14}, { 16384,15}, \
174 { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \
175 { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \
176 {8388608,24} }
177 #define SQR_FFT_TABLE3_SIZE 161
178 #define SQR_FFT_THRESHOLD 2624
179
180 #define MULLO_BASECASE_THRESHOLD 0 /* always */
181 #define MULLO_DC_THRESHOLD 31
182 #define MULLO_MUL_N_THRESHOLD 6440
183 #define SQRLO_BASECASE_THRESHOLD 9
184 #define SQRLO_DC_THRESHOLD 129
185 #define SQRLO_SQR_THRESHOLD 5103
186
187 #define DC_DIV_QR_THRESHOLD 19
188 #define DC_DIVAPPR_Q_THRESHOLD 123
189 #define DC_BDIV_QR_THRESHOLD 79
190 #define DC_BDIV_Q_THRESHOLD 154
191
192 #define INV_MULMOD_BNM1_THRESHOLD 42
193 #define INV_NEWTON_THRESHOLD 107
194 #define INV_APPR_THRESHOLD 107
195
196 #define BINV_NEWTON_THRESHOLD 312
197 #define REDC_1_TO_REDC_N_THRESHOLD 77
198
199 #define MU_DIV_QR_THRESHOLD 1142
200 #define MU_DIVAPPR_Q_THRESHOLD 1258
201 #define MUPI_DIV_QR_THRESHOLD 30
202 #define MU_BDIV_QR_THRESHOLD 1120
203 #define MU_BDIV_Q_THRESHOLD 1394
204
205 #define POWM_SEC_TABLE 6,19,203,579,2245
206
207 #define GET_STR_DC_THRESHOLD 10
208 #define GET_STR_PRECOMPUTE_THRESHOLD 18
209 #define SET_STR_DC_THRESHOLD 115
210 #define SET_STR_PRECOMPUTE_THRESHOLD 1941
211
212 #define FAC_DSC_THRESHOLD 182
213 #define FAC_ODD_THRESHOLD 44
214
215 #define MATRIX22_STRASSEN_THRESHOLD 15
216 #define HGCD2_DIV1_METHOD 1 /* 13.04% faster than 3 */
217 #define HGCD_THRESHOLD 65
218 #define HGCD_APPR_THRESHOLD 50
219 #define HGCD_REDUCE_THRESHOLD 2121
220 #define GCD_DC_THRESHOLD 293
221 #define GCDEXT_DC_THRESHOLD 186
222 #define JACOBI_BASE_METHOD 1 /* 12.79% faster than 3 */