1 /* AMD K8 gmp-mparam.h -- Compiler/machine parameter header file.
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 #if 0
35 #undef mpn_sublsh_n
36 #define mpn_sublsh_n(rp,up,vp,n,c) \
37 (((rp) == (up)) ? mpn_submul_1 (rp, vp, n, CNST_LIMB(1) << (c)) \
38 : MPN(mpn_sublsh_n)(rp,up,vp,n,c))
39 #endif
40
41 /* 2500 MHz K8 Brisbane */
42 /* FFT tuning limit = 115,768,433 */
43 /* Generated by tuneup.c, 2019-10-18, gcc 8.3 */
44
45 #define MOD_1_NORM_THRESHOLD 0 /* always */
46 #define MOD_1_UNNORM_THRESHOLD 0 /* always */
47 #define MOD_1N_TO_MOD_1_1_THRESHOLD 5
48 #define MOD_1U_TO_MOD_1_1_THRESHOLD 2
49 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 14
50 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 35
51 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 9
52 #define USE_PREINV_DIVREM_1 1 /* native */
53 #define DIV_QR_1_NORM_THRESHOLD 1
54 #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */
55 #define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */
56 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
57 #define BMOD_1_TO_MOD_1_THRESHOLD 16
58
59 #define DIV_1_VS_MUL_1_PERCENT 309
60
61 #define MUL_TOOM22_THRESHOLD 28
62 #define MUL_TOOM33_THRESHOLD 81
63 #define MUL_TOOM44_THRESHOLD 232
64 #define MUL_TOOM6H_THRESHOLD 324
65 #define MUL_TOOM8H_THRESHOLD 478
66
67 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 97
68 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 153
69 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 154
70 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 160
71 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 226
72
73 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
74 #define SQR_TOOM2_THRESHOLD 34
75 #define SQR_TOOM3_THRESHOLD 114
76 #define SQR_TOOM4_THRESHOLD 336
77 #define SQR_TOOM6_THRESHOLD 430
78 #define SQR_TOOM8_THRESHOLD 0 /* always */
79
80 #define MULMID_TOOM42_THRESHOLD 36
81
82 #define MULMOD_BNM1_THRESHOLD 17
83 #define SQRMOD_BNM1_THRESHOLD 19
84
85 #define MUL_FFT_MODF_THRESHOLD 654 /* k = 5 */
86 #define MUL_FFT_TABLE3 \
87 { { 654, 5}, { 21, 6}, { 11, 5}, { 23, 6}, \
88 { 12, 5}, { 25, 6}, { 13, 5}, { 27, 6}, \
89 { 27, 7}, { 14, 6}, { 29, 7}, { 15, 6}, \
90 { 31, 7}, { 29, 8}, { 15, 7}, { 32, 8}, \
91 { 17, 7}, { 37, 8}, { 19, 7}, { 39, 8}, \
92 { 21, 7}, { 44, 8}, { 23, 7}, { 47, 8}, \
93 { 25, 7}, { 51, 8}, { 31, 7}, { 63, 8}, \
94 { 37, 9}, { 19, 8}, { 43, 9}, { 23, 8}, \
95 { 53, 9}, { 27, 8}, { 57, 9}, { 31, 8}, \
96 { 67, 9}, { 35, 8}, { 71, 9}, { 39, 8}, \
97 { 81, 9}, { 43,10}, { 23, 9}, { 55, 8}, \
98 { 111,10}, { 31, 9}, { 71,10}, { 39, 9}, \
99 { 87,10}, { 47, 9}, { 99,10}, { 55, 9}, \
100 { 111,11}, { 31,10}, { 63, 9}, { 131,10}, \
101 { 71, 9}, { 147,10}, { 87,11}, { 47,10}, \
102 { 111,11}, { 63,10}, { 143,11}, { 79,10}, \
103 { 167,11}, { 95,10}, { 199,11}, { 111,12}, \
104 { 63,11}, { 143,10}, { 287,11}, { 159,12}, \
105 { 95,11}, { 191,10}, { 383,11}, { 207,10}, \
106 { 415,13}, { 63,12}, { 127,11}, { 255,10}, \
107 { 511,11}, { 271,10}, { 543,11}, { 287,12}, \
108 { 159,11}, { 319,10}, { 639,11}, { 335,10}, \
109 { 671,11}, { 351,12}, { 191,11}, { 415,12}, \
110 { 223,11}, { 447,13}, { 127,12}, { 255,11}, \
111 { 543,12}, { 287,11}, { 575,10}, { 1151,11}, \
112 { 607,12}, { 319,11}, { 671,12}, { 351,11}, \
113 { 703,13}, { 191,12}, { 383,11}, { 767,12}, \
114 { 415,11}, { 831,12}, { 447,11}, { 895,12}, \
115 { 479,14}, { 127,13}, { 255,12}, { 543,11}, \
116 { 1087,12}, { 575,11}, { 1151,12}, { 607,13}, \
117 { 319,12}, { 735,13}, { 383,12}, { 831,13}, \
118 { 447,12}, { 959,14}, { 255,13}, { 511,12}, \
119 { 1087,13}, { 575,12}, { 1215,13}, { 639,12}, \
120 { 1279,13}, { 703,12}, { 1407,14}, { 383,13}, \
121 { 767,12}, { 1535,13}, { 831,12}, { 1663,13}, \
122 { 959,15}, { 255,14}, { 511,13}, { 1215,14}, \
123 { 639,13}, { 1471,14}, { 767,13}, { 1663,14}, \
124 { 895,13}, { 1855,15}, { 511,14}, { 1023,13}, \
125 { 2047,14}, { 1151,13}, { 2367,14}, { 1407,15}, \
126 { 767,14}, { 1791,16}, { 511,15}, { 1023,14}, \
127 { 2303,15}, { 1279,14}, { 2687,15}, { 1535,14}, \
128 { 3199,15}, { 1791,16}, { 1023,15}, { 2047,14}, \
129 { 4223,15}, { 2303,14}, { 4735,15}, { 2559,16}, \
130 { 1535,15}, { 3071,14}, { 6271,15}, { 3327,17}, \
131 { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \
132 {2097152,22}, {4194304,23}, {8388608,24} }
133 #define MUL_FFT_TABLE3_SIZE 183
134 #define MUL_FFT_THRESHOLD 11520
135
136 #define SQR_FFT_MODF_THRESHOLD 540 /* k = 5 */
137 #define SQR_FFT_TABLE3 \
138 { { 540, 5}, { 21, 6}, { 11, 5}, { 23, 6}, \
139 { 12, 5}, { 25, 6}, { 13, 5}, { 27, 6}, \
140 { 16, 5}, { 33, 6}, { 29, 7}, { 15, 6}, \
141 { 31, 7}, { 16, 6}, { 33, 7}, { 33, 8}, \
142 { 17, 7}, { 37, 8}, { 19, 7}, { 39, 8}, \
143 { 21, 7}, { 43, 8}, { 23, 7}, { 47, 8}, \
144 { 25, 7}, { 51, 8}, { 29, 9}, { 15, 8}, \
145 { 37, 9}, { 19, 8}, { 43, 9}, { 23, 8}, \
146 { 51, 9}, { 27, 8}, { 55, 9}, { 31, 8}, \
147 { 65, 9}, { 35, 8}, { 71, 9}, { 43,10}, \
148 { 23, 9}, { 55,10}, { 31, 9}, { 71,10}, \
149 { 39, 9}, { 83,10}, { 47, 9}, { 99,10}, \
150 { 55, 9}, { 111,11}, { 31,10}, { 63, 9}, \
151 { 127,10}, { 87,11}, { 47,10}, { 111,12}, \
152 { 31,11}, { 63,10}, { 143,11}, { 79,10}, \
153 { 167,11}, { 95,10}, { 191,11}, { 111,12}, \
154 { 63,11}, { 127, 9}, { 511,11}, { 143,10}, \
155 { 287, 9}, { 575,11}, { 159,12}, { 95,11}, \
156 { 191,10}, { 383, 9}, { 767,11}, { 207,10}, \
157 { 415,13}, { 63,12}, { 127,10}, { 511, 9}, \
158 { 1023,11}, { 271,10}, { 543, 9}, { 1087,11}, \
159 { 287,10}, { 575,12}, { 159,11}, { 319,10}, \
160 { 639,11}, { 335,10}, { 671,11}, { 351,10}, \
161 { 703,12}, { 191,11}, { 383,10}, { 767,11}, \
162 { 415,10}, { 831,12}, { 223,11}, { 447,13}, \
163 { 127,11}, { 511,10}, { 1023,11}, { 543,10}, \
164 { 1087,12}, { 287,11}, { 575,10}, { 1151,11}, \
165 { 607,12}, { 319,11}, { 639,10}, { 1279,11}, \
166 { 671,12}, { 351,11}, { 703,13}, { 191,12}, \
167 { 383,11}, { 767,12}, { 415,11}, { 831,12}, \
168 { 447,11}, { 895,14}, { 127,12}, { 511,11}, \
169 { 1023,12}, { 543,11}, { 1087,12}, { 575,11}, \
170 { 1151,12}, { 607,11}, { 1215,13}, { 319,12}, \
171 { 639,11}, { 1279,12}, { 671,11}, { 1343,12}, \
172 { 703,11}, { 1407,12}, { 735,13}, { 383,12}, \
173 { 767,11}, { 1535,12}, { 831,13}, { 447,12}, \
174 { 959,13}, { 511,12}, { 1087,13}, { 575,12}, \
175 { 1215,13}, { 639,12}, { 1343,13}, { 703,12}, \
176 { 1407,14}, { 383,13}, { 767,12}, { 1535,13}, \
177 { 831,12}, { 1663,13}, { 895,12}, { 1791,13}, \
178 { 959,14}, { 511,13}, { 1215,14}, { 639,13}, \
179 { 1471,14}, { 767,13}, { 1663,14}, { 895,13}, \
180 { 1791,15}, { 511,14}, { 1023,13}, { 2111,14}, \
181 { 1151,13}, { 2303,14}, { 1407,15}, { 767,14}, \
182 { 1791,16}, { 511,15}, { 1023,14}, { 2303,15}, \
183 { 1279,14}, { 2687,15}, { 1535,14}, { 3199,15}, \
184 { 1791,16}, { 1023,15}, { 2047,14}, { 4223,15}, \
185 { 2303,14}, { 4863,15}, { 2559,16}, { 1535,15}, \
186 { 3071,14}, { 6271,15}, { 3327,17}, { 131072,18}, \
187 { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \
188 {4194304,23}, {8388608,24} }
189 #define SQR_FFT_TABLE3_SIZE 202
190 #define SQR_FFT_THRESHOLD 7296
191
192 #define MULLO_BASECASE_THRESHOLD 0 /* always */
193 #define MULLO_DC_THRESHOLD 61
194 #define MULLO_MUL_N_THRESHOLD 22239
195 #define SQRLO_BASECASE_THRESHOLD 8
196 #define SQRLO_DC_THRESHOLD 0 /* never mpn_sqrlo_basecase */
197 #define SQRLO_SQR_THRESHOLD 14281
198
199 #define DC_DIV_QR_THRESHOLD 47
200 #define DC_DIVAPPR_Q_THRESHOLD 266
201 #define DC_BDIV_QR_THRESHOLD 38
202 #define DC_BDIV_Q_THRESHOLD 104
203
204 #define INV_MULMOD_BNM1_THRESHOLD 54
205 #define INV_NEWTON_THRESHOLD 252
206 #define INV_APPR_THRESHOLD 250
207
208 #define BINV_NEWTON_THRESHOLD 258
209 #define REDC_1_TO_REDC_2_THRESHOLD 35
210 #define REDC_2_TO_REDC_N_THRESHOLD 79
211
212 #define MU_DIV_QR_THRESHOLD 2089
213 #define MU_DIVAPPR_Q_THRESHOLD 1895
214 #define MUPI_DIV_QR_THRESHOLD 99
215 #define MU_BDIV_QR_THRESHOLD 1787
216 #define MU_BDIV_Q_THRESHOLD 1895
217
218 #define POWM_SEC_TABLE 1,16,194,960,2825
219
220 #define GET_STR_DC_THRESHOLD 16
221 #define GET_STR_PRECOMPUTE_THRESHOLD 26
222 #define SET_STR_DC_THRESHOLD 248
223 #define SET_STR_PRECOMPUTE_THRESHOLD 1747
224
225 #define FAC_DSC_THRESHOLD 1240
226 #define FAC_ODD_THRESHOLD 27
227
228 #define MATRIX22_STRASSEN_THRESHOLD 21
229 #define HGCD2_DIV1_METHOD 3 /* 4.10% faster than 5 */
230 #define HGCD_THRESHOLD 141
231 #define HGCD_APPR_THRESHOLD 181
232 #define HGCD_REDUCE_THRESHOLD 4633
233 #define GCD_DC_THRESHOLD 622
234 #define GCDEXT_DC_THRESHOLD 496
235 #define JACOBI_BASE_METHOD 1 /* 0.97% faster than 3 */
236
237 /* Tuneup completed successfully, took 131832 seconds */