1 /* POWER7 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 /* 3720 MHz POWER7/SMT4 */
35 /* FFT tuning limit = 0.5 M */
36 /* Generated by tuneup.c, 2019-10-02, gcc 4.8 */
37
38 #define MOD_1_NORM_THRESHOLD 0 /* always */
39 #define MOD_1_UNNORM_THRESHOLD 0 /* always */
40 #define MOD_1N_TO_MOD_1_1_THRESHOLD 6
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD 4
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 8
43 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 16
44 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 13
45 #define USE_PREINV_DIVREM_1 0
46 /* From gcc110.osuosl.org, 2023-07-27 */
47 #define DIV_QR_1N_PI1_METHOD 3 /* 8.45% faster than 4 */
48 #define DIV_QR_1_NORM_THRESHOLD 1
49 #define DIV_QR_1_UNNORM_THRESHOLD 1
50 #define DIV_QR_2_PI2_THRESHOLD MP_SIZE_T_MAX /* never */
51 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
52 #define BMOD_1_TO_MOD_1_THRESHOLD 27
53
54 #define DIV_1_VS_MUL_1_PERCENT 341
55
56 #define MUL_TOOM22_THRESHOLD 22
57 #define MUL_TOOM33_THRESHOLD 71
58 #define MUL_TOOM44_THRESHOLD 196
59 #define MUL_TOOM6H_THRESHOLD 298
60 #define MUL_TOOM8H_THRESHOLD 406
61
62 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 81
63 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 140
64 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 132
65 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 139
66 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 120
67
68 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
69 #define SQR_TOOM2_THRESHOLD 32
70 #define SQR_TOOM3_THRESHOLD 105
71 #define SQR_TOOM4_THRESHOLD 190
72 #define SQR_TOOM6_THRESHOLD 318
73 #define SQR_TOOM8_THRESHOLD 547
74
75 #define MULMID_TOOM42_THRESHOLD 56
76
77 #define MULMOD_BNM1_THRESHOLD 18
78 #define SQRMOD_BNM1_THRESHOLD 20
79
80 #define MUL_FFT_MODF_THRESHOLD 436 /* k = 5 */
81 #define MUL_FFT_TABLE3 \
82 { { 436, 5}, { 21, 6}, { 21, 7}, { 11, 6}, \
83 { 23, 7}, { 12, 6}, { 25, 7}, { 21, 8}, \
84 { 11, 7}, { 25, 8}, { 13, 7}, { 28, 8}, \
85 { 15, 7}, { 33, 8}, { 17, 7}, { 35, 8}, \
86 { 19, 7}, { 39, 8}, { 21, 9}, { 11, 8}, \
87 { 29, 9}, { 15, 8}, { 35, 9}, { 19, 8}, \
88 { 41, 9}, { 23, 8}, { 49, 9}, { 27,10}, \
89 { 15, 9}, { 31, 8}, { 63, 9}, { 43,10}, \
90 { 23, 9}, { 55,11}, { 15,10}, { 31, 9}, \
91 { 67,10}, { 39, 9}, { 79,10}, { 47, 9}, \
92 { 95,10}, { 55,11}, { 31,10}, { 63, 9}, \
93 { 127,10}, { 79,11}, { 47,10}, { 103,12}, \
94 { 31,11}, { 63,10}, { 135,11}, { 79,10}, \
95 { 159,11}, { 95,10}, { 191, 9}, { 383,11}, \
96 { 111,12}, { 63,11}, { 127,10}, { 255, 9}, \
97 { 511,11}, { 143,10}, { 287, 9}, { 575,11}, \
98 { 159,10}, { 319,12}, { 95,11}, { 191,10}, \
99 { 383, 9}, { 767,11}, { 207,10}, { 415,13}, \
100 { 8192,14}, { 16384,15}, { 32768,16}, { 65536,17}, \
101 { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \
102 {2097152,22}, {4194304,23}, {8388608,24} }
103 #define MUL_FFT_TABLE3_SIZE 83
104 #define MUL_FFT_THRESHOLD 4736
105
106 #define SQR_FFT_MODF_THRESHOLD 368 /* k = 5 */
107 #define SQR_FFT_TABLE3 \
108 { { 368, 5}, { 19, 6}, { 10, 5}, { 21, 6}, \
109 { 21, 7}, { 11, 6}, { 23, 7}, { 12, 6}, \
110 { 25, 7}, { 13, 6}, { 27, 7}, { 25, 8}, \
111 { 13, 7}, { 28, 8}, { 15, 7}, { 32, 8}, \
112 { 17, 7}, { 35, 8}, { 19, 7}, { 39, 8}, \
113 { 21, 9}, { 11, 8}, { 29, 9}, { 15, 8}, \
114 { 35, 9}, { 19, 8}, { 41, 9}, { 23, 8}, \
115 { 47, 9}, { 27,10}, { 15, 9}, { 31, 8}, \
116 { 63, 9}, { 39,10}, { 23, 9}, { 51,11}, \
117 { 15,10}, { 31, 9}, { 67,10}, { 39, 9}, \
118 { 79,10}, { 47, 9}, { 95,10}, { 55,11}, \
119 { 31,10}, { 79,11}, { 47,10}, { 95,12}, \
120 { 31,11}, { 63,10}, { 135,11}, { 79,10}, \
121 { 159, 9}, { 319,11}, { 95,10}, { 191, 9}, \
122 { 383,11}, { 111,12}, { 63,11}, { 127,10}, \
123 { 255, 9}, { 511,11}, { 143,10}, { 287, 9}, \
124 { 575,10}, { 303,11}, { 159,10}, { 319, 9}, \
125 { 639,12}, { 95,11}, { 191,10}, { 383, 9}, \
126 { 767,13}, { 8192,14}, { 16384,15}, { 32768,16}, \
127 { 65536,17}, { 131072,18}, { 262144,19}, { 524288,20}, \
128 {1048576,21}, {2097152,22}, {4194304,23}, {8388608,24} }
129 #define SQR_FFT_TABLE3_SIZE 84
130 #define SQR_FFT_THRESHOLD 3264
131
132 #define MULLO_BASECASE_THRESHOLD 3
133 #define MULLO_DC_THRESHOLD 35
134 #define MULLO_MUL_N_THRESHOLD 9449
135 #define SQRLO_BASECASE_THRESHOLD 3
136 #define SQRLO_DC_THRESHOLD 119
137 #define SQRLO_SQR_THRESHOLD 6440
138
139 #define DC_DIV_QR_THRESHOLD 33
140 #define DC_DIVAPPR_Q_THRESHOLD 124
141 #define DC_BDIV_QR_THRESHOLD 62
142 #define DC_BDIV_Q_THRESHOLD 144
143
144 #define INV_MULMOD_BNM1_THRESHOLD 67
145 #define INV_NEWTON_THRESHOLD 123
146 #define INV_APPR_THRESHOLD 123
147
148 #define BINV_NEWTON_THRESHOLD 284
149 #define REDC_1_TO_REDC_2_THRESHOLD 18
150 #define REDC_2_TO_REDC_N_THRESHOLD 109
151
152 #define MU_DIV_QR_THRESHOLD 1387
153 #define MU_DIVAPPR_Q_THRESHOLD 1334
154 #define MUPI_DIV_QR_THRESHOLD 50
155 #define MU_BDIV_QR_THRESHOLD 1308
156 #define MU_BDIV_Q_THRESHOLD 1499
157
158 #define POWM_SEC_TABLE 1,23,121,579,642
159
160 #define GET_STR_DC_THRESHOLD 11
161 #define GET_STR_PRECOMPUTE_THRESHOLD 18
162 #define SET_STR_DC_THRESHOLD 1562
163 #define SET_STR_PRECOMPUTE_THRESHOLD 3100
164
165 #define FAC_DSC_THRESHOLD 774
166 #define FAC_ODD_THRESHOLD 25
167
168 #define MATRIX22_STRASSEN_THRESHOLD 18
169 #define HGCD2_DIV1_METHOD 5 /* 3.27% faster than 3 */
170 #define HGCD_THRESHOLD 118
171 #define HGCD_APPR_THRESHOLD 150
172 #define HGCD_REDUCE_THRESHOLD 3014
173 #define GCD_DC_THRESHOLD 386
174 #define GCDEXT_DC_THRESHOLD 365
175 #define JACOBI_BASE_METHOD 4 /* 27.64% faster than 1 */