1 /* POWER8 gmp-mparam.h -- Compiler/machine parameter header file.
2
3 Copyright 2017 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 /* 4150 MHz POWER8/SMT4 */
35 /* FFT tuning limit = 0.5 M */
36 /* Generated by tuneup.c, 2019-09-24, gcc 7.2 */
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 7
41 #define MOD_1U_TO_MOD_1_1_THRESHOLD 6
42 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 11
43 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 22
44 #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 10
45 #define USE_PREINV_DIVREM_1 0
46 /* From gcc112.osuosl.org, 2023-07-27 */
47 #define DIV_QR_1N_PI1_METHOD 3 /* 13.00% faster than 4 */
48 #define DIV_QR_1_NORM_THRESHOLD 2
49 #define DIV_QR_1_UNNORM_THRESHOLD 1
50 #define DIV_QR_2_PI2_THRESHOLD 9
51 #define DIVEXACT_1_THRESHOLD 0 /* always (native) */
52 #define BMOD_1_TO_MOD_1_THRESHOLD 34
53
54 #define DIV_1_VS_MUL_1_PERCENT 276
55
56 #define MUL_TOOM22_THRESHOLD 18
57 #define MUL_TOOM33_THRESHOLD 73
58 #define MUL_TOOM44_THRESHOLD 195
59 #define MUL_TOOM6H_THRESHOLD 278
60 #define MUL_TOOM8H_THRESHOLD 406
61
62 #define MUL_TOOM32_TO_TOOM43_THRESHOLD 73
63 #define MUL_TOOM32_TO_TOOM53_THRESHOLD 131
64 #define MUL_TOOM42_TO_TOOM53_THRESHOLD 121
65 #define MUL_TOOM42_TO_TOOM63_THRESHOLD 138
66 #define MUL_TOOM43_TO_TOOM54_THRESHOLD 106
67
68 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */
69 #define SQR_TOOM2_THRESHOLD 32
70 #define SQR_TOOM3_THRESHOLD 97
71 #define SQR_TOOM4_THRESHOLD 178
72 #define SQR_TOOM6_THRESHOLD 303
73 #define SQR_TOOM8_THRESHOLD 454
74
75 #define MULMID_TOOM42_THRESHOLD 42
76
77 #define MULMOD_BNM1_THRESHOLD 15
78 #define SQRMOD_BNM1_THRESHOLD 19
79
80 #define MUL_FFT_MODF_THRESHOLD 404 /* k = 5 */
81 #define MUL_FFT_TABLE3 \
82 { { 404, 5}, { 19, 6}, { 10, 5}, { 21, 6}, \
83 { 21, 7}, { 11, 6}, { 23, 7}, { 12, 6}, \
84 { 25, 7}, { 21, 8}, { 11, 7}, { 25, 8}, \
85 { 13, 7}, { 27, 8}, { 15, 7}, { 31, 8}, \
86 { 17, 7}, { 35, 8}, { 19, 7}, { 39, 8}, \
87 { 21, 9}, { 11, 8}, { 27, 9}, { 15, 8}, \
88 { 35, 9}, { 19, 8}, { 41, 9}, { 23, 8}, \
89 { 47, 9}, { 27,10}, { 15, 9}, { 39,10}, \
90 { 23, 9}, { 51,11}, { 15,10}, { 31, 9}, \
91 { 63,10}, { 39, 9}, { 79,10}, { 47, 9}, \
92 { 95,11}, { 31,10}, { 63, 9}, { 131,10}, \
93 { 79,11}, { 47,10}, { 95,12}, { 31,11}, \
94 { 63,10}, { 127, 9}, { 255,10}, { 135,11}, \
95 { 79,10}, { 159,11}, { 95, 8}, { 767, 7}, \
96 { 1599,11}, { 111,12}, { 63,11}, { 127,10}, \
97 { 255, 9}, { 511,11}, { 143,10}, { 287, 9}, \
98 { 575,11}, { 159,12}, { 95,11}, { 191,10}, \
99 { 383,13}, { 8192,14}, { 16384,15}, { 32768,16}, \
100 { 65536,17}, { 131072,18}, { 262144,19}, { 524288,20}, \
101 {1048576,21}, {2097152,22}, {4194304,23}, {8388608,24} }
102 #define MUL_FFT_TABLE3_SIZE 80
103 #define MUL_FFT_THRESHOLD 4736
104
105 #define SQR_FFT_MODF_THRESHOLD 340 /* k = 5 */
106 #define SQR_FFT_TABLE3 \
107 { { 340, 5}, { 19, 6}, { 10, 5}, { 21, 6}, \
108 { 21, 7}, { 11, 6}, { 23, 7}, { 21, 8}, \
109 { 11, 7}, { 25, 8}, { 13, 7}, { 27, 8}, \
110 { 15, 7}, { 31, 8}, { 21, 9}, { 11, 8}, \
111 { 27, 9}, { 15, 8}, { 33, 9}, { 19, 8}, \
112 { 41, 9}, { 23, 8}, { 47, 9}, { 27,10}, \
113 { 15, 9}, { 39,10}, { 23, 9}, { 47,11}, \
114 { 15,10}, { 31, 9}, { 67,10}, { 39, 9}, \
115 { 79,10}, { 47,11}, { 31,10}, { 79,11}, \
116 { 47,10}, { 95,12}, { 31,11}, { 63,10}, \
117 { 127, 9}, { 255,11}, { 79, 9}, { 319,11}, \
118 { 95,10}, { 191,12}, { 63,11}, { 127,10}, \
119 { 255, 9}, { 511,10}, { 271,11}, { 143,10}, \
120 { 287, 9}, { 575,10}, { 303, 9}, { 607,10}, \
121 { 319,12}, { 95,11}, { 191,10}, { 383,13}, \
122 { 8192,14}, { 16384,15}, { 32768,16}, { 65536,17}, \
123 { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \
124 {2097152,22}, {4194304,23}, {8388608,24} }
125 #define SQR_FFT_TABLE3_SIZE 71
126 #define SQR_FFT_THRESHOLD 3264
127
128 #define MULLO_BASECASE_THRESHOLD 3
129 #define MULLO_DC_THRESHOLD 33
130 #define MULLO_MUL_N_THRESHOLD 9174
131 #define SQRLO_BASECASE_THRESHOLD 0 /* always */
132 #define SQRLO_DC_THRESHOLD 114
133 #define SQRLO_SQR_THRESHOLD 6461
134
135 #define DC_DIV_QR_THRESHOLD 38
136 #define DC_DIVAPPR_Q_THRESHOLD 158
137 #define DC_BDIV_QR_THRESHOLD 48
138 #define DC_BDIV_Q_THRESHOLD 112
139
140 #define INV_MULMOD_BNM1_THRESHOLD 74
141 #define INV_NEWTON_THRESHOLD 132
142 #define INV_APPR_THRESHOLD 131
143
144 #define BINV_NEWTON_THRESHOLD 278
145 #define REDC_1_TO_REDC_2_THRESHOLD 56
146 #define REDC_2_TO_REDC_N_THRESHOLD 0 /* always */
147
148 #define MU_DIV_QR_THRESHOLD 1142
149 #define MU_DIVAPPR_Q_THRESHOLD 1142
150 #define MUPI_DIV_QR_THRESHOLD 46
151 #define MU_BDIV_QR_THRESHOLD 1142
152 #define MU_BDIV_Q_THRESHOLD 1470
153
154 #define POWM_SEC_TABLE 3,19,117,672,1867
155
156 #define GET_STR_DC_THRESHOLD 11
157 #define GET_STR_PRECOMPUTE_THRESHOLD 18
158 #define SET_STR_DC_THRESHOLD 608
159 #define SET_STR_PRECOMPUTE_THRESHOLD 2405
160
161 #define FAC_DSC_THRESHOLD 164
162 #define FAC_ODD_THRESHOLD 0 /* always */
163
164 #define MATRIX22_STRASSEN_THRESHOLD 14
165 #define HGCD2_DIV1_METHOD 1 /* 6.88% faster than 3 */
166 #define HGCD_THRESHOLD 114
167 #define HGCD_APPR_THRESHOLD 118
168 #define HGCD_REDUCE_THRESHOLD 2205
169 #define GCD_DC_THRESHOLD 440
170 #define GCDEXT_DC_THRESHOLD 345
171 #define JACOBI_BASE_METHOD 1 /* 0.74% faster than 4 */