1 /* gmp-mparam.h -- Compiler/machine parameter header file for a53.
2
3 Copyright 2019 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 /* 1536 MHz Cortex-A53 */
35 /* FFT tuning limit = 21,583,800 */
36 /* Generated by tuneup.c, 2019-10-22, gcc 5.4 */
37
38 #define DIVREM_1_NORM_THRESHOLD 3
39 #define DIVREM_1_UNNORM_THRESHOLD 4
40 #define MOD_1_1P_METHOD 2 /* 4.84% faster than 1 */
41 #define MOD_1_NORM_THRESHOLD 3
42 #define MOD_1_UNNORM_THRESHOLD 4
43 #define MOD_1N_TO_MOD_1_1_THRESHOLD 7
44 #define MOD_1U_TO_MOD_1_1_THRESHOLD 7
45 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 12
46 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 18
47 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 22
48 #define USE_PREINV_DIVREM_1 1
49 #define DIV_QR_1N_PI1_METHOD 1 /* 39.05% faster than 2 */
50 #define DIV_QR_1_NORM_THRESHOLD 21
51 #define DIV_QR_1_UNNORM_THRESHOLD 21
52 #define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */
53 #define DIVEXACT_1_THRESHOLD 0 /* always */
54 #define BMOD_1_TO_MOD_1_THRESHOLD 38
55
56 #define DIV_1_VS_MUL_1_PERCENT 161
57
58 #define MUL_TOOM22_THRESHOLD 14
59 #define MUL_TOOM33_THRESHOLD 49
60 #define MUL_TOOM44_THRESHOLD 73
61 #define MUL_TOOM6H_THRESHOLD 173
62 #define MUL_TOOM8H_THRESHOLD 236
63
64 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 81
65 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 77
66 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 81
67 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 88
68 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 65
69
70 #define SQR_BASECASE_THRESHOLD 0 /* always */
71 #define SQR_TOOM2_THRESHOLD 18
72 #define SQR_TOOM3_THRESHOLD 68
73 #define SQR_TOOM4_THRESHOLD 183
74 #define SQR_TOOM6_THRESHOLD 230
75 #define SQR_TOOM8_THRESHOLD 357
76
77 #define MULMID_TOOM42_THRESHOLD 23
78
79 #define MULMOD_BNM1_THRESHOLD 9
80 #define SQRMOD_BNM1_THRESHOLD 11
81
82 #define MUL_FFT_MODF_THRESHOLD 316 /* k = 5 */
83 #define MUL_FFT_TABLE3 \
84 { { 316, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \
85 { 13, 7}, { 7, 6}, { 15, 7}, { 8, 6}, \
86 { 17, 7}, { 9, 6}, { 19, 7}, { 17, 8}, \
87 { 9, 7}, { 20, 8}, { 11, 7}, { 23, 8}, \
88 { 13, 9}, { 7, 8}, { 19, 9}, { 11, 8}, \
89 { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \
90 { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \
91 { 15, 9}, { 39,10}, { 23, 9}, { 51,11}, \
92 { 15,10}, { 31, 9}, { 71,10}, { 39, 9}, \
93 { 83,10}, { 47, 9}, { 99,10}, { 55,11}, \
94 { 31,10}, { 63, 8}, { 255,10}, { 71, 8}, \
95 { 287,10}, { 79, 9}, { 159, 8}, { 319,10}, \
96 { 87,11}, { 47,10}, { 95, 9}, { 191, 8}, \
97 { 383,10}, { 103, 9}, { 207, 8}, { 415,10}, \
98 { 111, 9}, { 223,12}, { 31,11}, { 63, 9}, \
99 { 255, 8}, { 511,10}, { 135, 9}, { 287, 8}, \
100 { 575,11}, { 79,10}, { 159, 9}, { 319, 8}, \
101 { 639,10}, { 175, 9}, { 351, 8}, { 703,11}, \
102 { 95,10}, { 191, 9}, { 383, 8}, { 767,10}, \
103 { 207, 9}, { 415, 8}, { 831,10}, { 223, 9}, \
104 { 447,12}, { 63,10}, { 255, 9}, { 511, 8}, \
105 { 1023, 9}, { 543,10}, { 287, 9}, { 575, 8}, \
106 { 1151,11}, { 159,10}, { 319, 9}, { 639,11}, \
107 { 175,10}, { 351, 9}, { 703, 8}, { 1407,12}, \
108 { 95,11}, { 191,10}, { 383, 9}, { 767,11}, \
109 { 207,10}, { 415, 9}, { 831,11}, { 223,10}, \
110 { 447,13}, { 63,11}, { 255,10}, { 543,11}, \
111 { 287,10}, { 575, 9}, { 1151,12}, { 159,11}, \
112 { 319,10}, { 639,11}, { 351,10}, { 703, 9}, \
113 { 1407, 8}, { 2815,12}, { 191,11}, { 383,10}, \
114 { 767,11}, { 415,10}, { 831,12}, { 223,11}, \
115 { 447,10}, { 895,11}, { 479,10}, { 959, 9}, \
116 { 1919,12}, { 255,11}, { 511,10}, { 1023,11}, \
117 { 543,10}, { 1087,12}, { 287,11}, { 575,10}, \
118 { 1151,12}, { 319,11}, { 639,12}, { 351,11}, \
119 { 703,10}, { 1407, 9}, { 2815,13}, { 191,12}, \
120 { 383,11}, { 767,12}, { 415,11}, { 831,10}, \
121 { 1663,12}, { 447,11}, { 895,10}, { 1791,12}, \
122 { 479,11}, { 959,13}, { 255,12}, { 511,11}, \
123 { 1023,12}, { 543,11}, { 1087,12}, { 575,11}, \
124 { 1151,13}, { 319,12}, { 703,11}, { 1407,10}, \
125 { 2815,13}, { 383,12}, { 831,11}, { 1663,13}, \
126 { 447,12}, { 895,11}, { 1791,12}, { 959,11}, \
127 { 1919,14}, { 255,13}, { 511,12}, { 1087,13}, \
128 { 575,12}, { 1151,13}, { 703,12}, { 1407,11}, \
129 { 2815,14}, { 383,13}, { 831,12}, { 1663,13}, \
130 { 895,12}, { 1791,13}, { 959,12}, { 1919,15}, \
131 { 255,14}, { 511,13}, { 1087,12}, { 2175,13}, \
132 { 1215,14}, { 639,13}, { 1407,12}, { 2815,14}, \
133 { 767,13}, { 1663,14}, { 895,13}, { 1919,12}, \
134 { 3839,15}, { 511,14}, { 1023,13}, { 2175,14}, \
135 { 1151,13}, { 2431,12}, { 4863,14}, { 16384,15}, \
136 { 32768,16}, { 65536,17}, { 131072,18}, { 262144,19}, \
137 { 524288,20}, {1048576,21}, {2097152,22}, {4194304,23}, \
138 {8388608,24} }
139 #define MUL_FFT_TABLE3_SIZE 217
140 #define MUL_FFT_THRESHOLD 3200
141
142 #define SQR_FFT_MODF_THRESHOLD 276 /* k = 5 */
143 #define SQR_FFT_TABLE3 \
144 { { 276, 5}, { 13, 6}, { 7, 5}, { 15, 6}, \
145 { 17, 7}, { 17, 8}, { 9, 7}, { 20, 8}, \
146 { 11, 7}, { 23, 8}, { 13, 9}, { 7, 8}, \
147 { 15, 7}, { 31, 8}, { 19, 9}, { 11, 8}, \
148 { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \
149 { 39, 9}, { 23, 8}, { 47, 9}, { 27,10}, \
150 { 15, 9}, { 39,10}, { 23, 9}, { 47,11}, \
151 { 15,10}, { 31, 9}, { 67,10}, { 39, 9}, \
152 { 79,10}, { 47, 9}, { 95, 8}, { 191,10}, \
153 { 55,11}, { 31,10}, { 63, 8}, { 255,10}, \
154 { 71, 9}, { 143, 8}, { 287,10}, { 79, 9}, \
155 { 159,11}, { 47,10}, { 95, 9}, { 191, 8}, \
156 { 383, 7}, { 767,10}, { 103,12}, { 31,11}, \
157 { 63, 9}, { 255, 8}, { 511, 7}, { 1023,10}, \
158 { 143, 9}, { 287,11}, { 79,10}, { 159, 9}, \
159 { 319, 8}, { 639,10}, { 175, 9}, { 351, 8}, \
160 { 703,11}, { 95,10}, { 191, 9}, { 383, 8}, \
161 { 767,10}, { 207, 9}, { 415, 8}, { 831,10}, \
162 { 223, 9}, { 447,12}, { 63,10}, { 255, 9}, \
163 { 511, 8}, { 1023,11}, { 143,10}, { 287, 9}, \
164 { 575, 8}, { 1151,11}, { 159,10}, { 319, 9}, \
165 { 639,11}, { 175,10}, { 351, 9}, { 703,12}, \
166 { 95,11}, { 191,10}, { 383, 9}, { 767,11}, \
167 { 207,10}, { 415, 9}, { 831,11}, { 223,10}, \
168 { 447,13}, { 63,11}, { 255,10}, { 511, 9}, \
169 { 1023,11}, { 287,10}, { 575, 9}, { 1151,12}, \
170 { 159,11}, { 319,10}, { 639,11}, { 351,10}, \
171 { 703, 9}, { 1407,12}, { 191,11}, { 383,10}, \
172 { 767,11}, { 415,10}, { 831,12}, { 223,11}, \
173 { 447,10}, { 895,11}, { 479,10}, { 959,12}, \
174 { 255,11}, { 511,10}, { 1023,12}, { 287,11}, \
175 { 575,10}, { 1151,12}, { 319,11}, { 639,12}, \
176 { 351,11}, { 703,10}, { 1407,13}, { 191,12}, \
177 { 383,11}, { 767,12}, { 415,11}, { 831,10}, \
178 { 1663,12}, { 447,11}, { 895,12}, { 479,11}, \
179 { 959,10}, { 1919,13}, { 255,12}, { 511,11}, \
180 { 1023,12}, { 543,11}, { 1087,12}, { 575,11}, \
181 { 1151,13}, { 319,12}, { 703,11}, { 1407,10}, \
182 { 2815,13}, { 383,12}, { 831,11}, { 1663,13}, \
183 { 447,12}, { 895,11}, { 1791,12}, { 959,14}, \
184 { 255,13}, { 511,12}, { 1087,13}, { 575,12}, \
185 { 1151,13}, { 703,12}, { 1407,11}, { 2815,14}, \
186 { 383,13}, { 831,12}, { 1663,13}, { 895,12}, \
187 { 1791,13}, { 959,12}, { 1919,15}, { 255,14}, \
188 { 511,13}, { 1087,12}, { 2175,13}, { 1151,14}, \
189 { 639,13}, { 1407,12}, { 2815,14}, { 767,13}, \
190 { 1663,14}, { 895,13}, { 1919,12}, { 3839,15}, \
191 { 511,14}, { 1023,13}, { 2175,14}, { 1151,13}, \
192 { 2431,12}, { 4863,14}, { 16384,15}, { 32768,16}, \
193 { 65536,17}, { 131072,18}, { 262144,19}, { 524288,20}, \
194 {1048576,21}, {2097152,22}, {4194304,23}, {8388608,24} }
195 #define SQR_FFT_TABLE3_SIZE 204
196 #define SQR_FFT_THRESHOLD 2688
197
198 #define MULLO_BASECASE_THRESHOLD 0 /* always */
199 #define MULLO_DC_THRESHOLD 38
200 #define MULLO_MUL_N_THRESHOLD 6253
201 #define SQRLO_BASECASE_THRESHOLD 4
202 #define SQRLO_DC_THRESHOLD 67
203 #define SQRLO_SQR_THRESHOLD 5240
204
205 #define DC_DIV_QR_THRESHOLD 43
206 #define DC_DIVAPPR_Q_THRESHOLD 155
207 #define DC_BDIV_QR_THRESHOLD 39
208 #define DC_BDIV_Q_THRESHOLD 89
209
210 #define INV_MULMOD_BNM1_THRESHOLD 34
211 #define INV_NEWTON_THRESHOLD 163
212 #define INV_APPR_THRESHOLD 161
213
214 #define BINV_NEWTON_THRESHOLD 196
215 #define REDC_1_TO_REDC_N_THRESHOLD 43
216
217 #define MU_DIV_QR_THRESHOLD 998
218 #define MU_DIVAPPR_Q_THRESHOLD 998
219 #define MUPI_DIV_QR_THRESHOLD 91
220 #define MU_BDIV_QR_THRESHOLD 807
221 #define MU_BDIV_Q_THRESHOLD 924
222
223 #define POWM_SEC_TABLE 6,30,125,579,1730
224
225 #define GET_STR_DC_THRESHOLD 15
226 #define GET_STR_PRECOMPUTE_THRESHOLD 30
227 #define SET_STR_DC_THRESHOLD 802
228 #define SET_STR_PRECOMPUTE_THRESHOLD 1815
229
230 #define FAC_DSC_THRESHOLD 258
231 #define FAC_ODD_THRESHOLD 24
232
233 #define MATRIX22_STRASSEN_THRESHOLD 10
234 #define HGCD2_DIV1_METHOD 1 /* 7.05% faster than 3 */
235 #define HGCD_THRESHOLD 107
236 #define HGCD_APPR_THRESHOLD 112
237 #define HGCD_REDUCE_THRESHOLD 1679
238 #define GCD_DC_THRESHOLD 324
239 #define GCDEXT_DC_THRESHOLD 242
240 #define JACOBI_BASE_METHOD 4 /* 22.41% faster than 1 */
241
242 /* Tuneup completed successfully, took 66624 seconds */