1/* POWER9 gmp-mparam.h -- Compiler/machine parameter header file. 2 3Copyright 2019 Free Software Foundation, Inc. 4 5This file is part of the GNU MP Library. 6 7The GNU MP Library is free software; you can redistribute it and/or modify 8it 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 14or 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 20or both in parallel, as here. 21 22The GNU MP Library is distributed in the hope that it will be useful, but 23WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 24or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 25for more details. 26 27You should have received copies of the GNU General Public License and the 28GNU Lesser General Public License along with the GNU MP Library. If not, 29see https://www.gnu.org/licenses/. */ 30 31#define GMP_LIMB_BITS 64 32#define GMP_LIMB_BYTES 8 33 34/* 2200MHz POWER9 */ 35/* FFT tuning limit = 221,245,838 */ 36/* Generated by tuneup.c, 2019-10-29, gcc 8.3 */ 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 8 41#define MOD_1U_TO_MOD_1_1_THRESHOLD 5 42#define MOD_1_1_TO_MOD_1_2_THRESHOLD 7 43#define MOD_1_2_TO_MOD_1_4_THRESHOLD 44 44#define PREINV_MOD_1_TO_MOD_1_THRESHOLD 11 45#define USE_PREINV_DIVREM_1 0 46#define DIV_QR_1N_PI1_METHOD 2 /* 19.28% faster than 1 */ 47#define DIV_QR_1_NORM_THRESHOLD 3 48#define DIV_QR_1_UNNORM_THRESHOLD 2 49#define DIV_QR_2_PI2_THRESHOLD 7 50#define DIVEXACT_1_THRESHOLD 0 /* always (native) */ 51#define BMOD_1_TO_MOD_1_THRESHOLD 33 52 53#define DIV_1_VS_MUL_1_PERCENT 365 54 55#define MUL_TOOM22_THRESHOLD 34 56#define MUL_TOOM33_THRESHOLD 109 57#define MUL_TOOM44_THRESHOLD 458 58#define MUL_TOOM6H_THRESHOLD 517 59#define MUL_TOOM8H_THRESHOLD 608 60 61#define MUL_TOOM32_TO_TOOM43_THRESHOLD 113 62#define MUL_TOOM32_TO_TOOM53_THRESHOLD 292 63#define MUL_TOOM42_TO_TOOM53_THRESHOLD 204 64#define MUL_TOOM42_TO_TOOM63_THRESHOLD 211 65#define MUL_TOOM43_TO_TOOM54_THRESHOLD 178 66 67#define SQR_BASECASE_THRESHOLD 0 /* always (native) */ 68#define SQR_TOOM2_THRESHOLD 46 69#define SQR_TOOM3_THRESHOLD 158 70#define SQR_TOOM4_THRESHOLD 674 71#define SQR_TOOM6_THRESHOLD 0 /* always */ 72#define SQR_TOOM8_THRESHOLD 898 73 74#define MULMID_TOOM42_THRESHOLD 70 75 76#define MULMOD_BNM1_THRESHOLD 17 77#define SQRMOD_BNM1_THRESHOLD 25 78 79#define MUL_FFT_MODF_THRESHOLD 404 /* k = 5 */ 80#define MUL_FFT_TABLE3 \ 81 { { 404, 5}, { 23, 6}, { 12, 5}, { 25, 6}, \ 82 { 13, 5}, { 27, 6}, { 27, 7}, { 14, 6}, \ 83 { 29, 7}, { 15, 6}, { 31, 7}, { 25, 8}, \ 84 { 13, 7}, { 28, 8}, { 15, 7}, { 32, 8}, \ 85 { 17, 7}, { 35, 8}, { 27, 9}, { 15, 8}, \ 86 { 35, 9}, { 19, 8}, { 39, 9}, { 23, 8}, \ 87 { 47, 9}, { 27,10}, { 15, 9}, { 31, 8}, \ 88 { 63, 9}, { 35, 8}, { 71, 9}, { 39,10}, \ 89 { 23, 9}, { 51,11}, { 15,10}, { 31, 9}, \ 90 { 71,10}, { 39, 9}, { 83,10}, { 47, 9}, \ 91 { 95,10}, { 55,11}, { 31,10}, { 79,11}, \ 92 { 47,10}, { 95,12}, { 31,11}, { 63,10}, \ 93 { 135,11}, { 79,10}, { 159,11}, { 95,12}, \ 94 { 63,11}, { 127,10}, { 255, 9}, { 511,11}, \ 95 { 143,10}, { 287, 9}, { 575,10}, { 303,11}, \ 96 { 159,12}, { 95,11}, { 191,13}, { 63,12}, \ 97 { 127,11}, { 255,10}, { 511,11}, { 271,10}, \ 98 { 543,11}, { 287,10}, { 575,11}, { 303,12}, \ 99 { 159,11}, { 319,10}, { 639,11}, { 335,10}, \ 100 { 671,11}, { 351,10}, { 703,11}, { 367,10}, \ 101 { 735,12}, { 191,11}, { 383,10}, { 767,11}, \ 102 { 415,10}, { 831,12}, { 223,11}, { 447,10}, \ 103 { 895,11}, { 479,13}, { 127,12}, { 255,11}, \ 104 { 511,10}, { 1023,11}, { 543,12}, { 287,11}, \ 105 { 575,10}, { 1151,11}, { 607,12}, { 319,11}, \ 106 { 639,10}, { 1279,11}, { 671,12}, { 351,11}, \ 107 { 703,10}, { 1407,11}, { 735,13}, { 191,12}, \ 108 { 383,11}, { 767,10}, { 1535,11}, { 799,12}, \ 109 { 415,11}, { 831,10}, { 1663,11}, { 863,12}, \ 110 { 447,11}, { 895,12}, { 479,14}, { 127,13}, \ 111 { 255,12}, { 511,11}, { 1023,12}, { 543,11}, \ 112 { 1087,12}, { 575,11}, { 1151,12}, { 607,13}, \ 113 { 319,12}, { 639,11}, { 1279,12}, { 671,11}, \ 114 { 1343,12}, { 703,11}, { 1407,12}, { 735,11}, \ 115 { 1471,13}, { 383,12}, { 767,11}, { 1535,12}, \ 116 { 799,11}, { 1599,12}, { 831,11}, { 1663,13}, \ 117 { 447,12}, { 895,11}, { 1791,12}, { 959,14}, \ 118 { 255,13}, { 511,12}, { 1087,11}, { 2175,13}, \ 119 { 575,12}, { 1215,13}, { 639,12}, { 1343,13}, \ 120 { 703,12}, { 1471,14}, { 383,13}, { 767,12}, \ 121 { 1599,13}, { 831,12}, { 1727,13}, { 895,11}, \ 122 { 3583,12}, { 1919,15}, { 255,14}, { 511,13}, \ 123 { 1087,12}, { 2175,13}, { 1215,14}, { 639,13}, \ 124 { 1343,12}, { 2687,13}, { 1471,14}, { 767,13}, \ 125 { 1599,12}, { 3199,13}, { 1727,14}, { 895,13}, \ 126 { 1919,15}, { 511,14}, { 1023,13}, { 2175,14}, \ 127 { 1151,13}, { 2431,12}, { 4863,14}, { 1279,13}, \ 128 { 2687,14}, { 1407,13}, { 2943,15}, { 767,14}, \ 129 { 1535,13}, { 3199,14}, { 1663,13}, { 3455,12}, \ 130 { 6911,14}, { 1919,16}, { 511,15}, { 1023,14}, \ 131 { 2175,13}, { 4479,14}, { 2431,13}, { 4863,15}, \ 132 { 1279,14}, { 2943,13}, { 5887,15}, { 1535,14}, \ 133 { 3455,13}, { 6911,15}, { 1791,14}, { 3839,13}, \ 134 { 7679,16}, { 1023,15}, { 2047,14}, { 4351,15}, \ 135 { 2303,14}, { 4863,15}, { 2815,14}, { 5887,16}, \ 136 { 1535,15}, { 3327,14}, { 6911,15}, { 3839,14}, \ 137 { 7679,17}, { 1023,16}, { 2047,15}, { 4351,14}, \ 138 { 8959,15}, { 4863,16}, { 2559,15}, { 5887,14}, \ 139 { 11775,16}, { 3071,15}, { 32768,16}, { 65536,17}, \ 140 { 131072,18}, { 262144,19}, { 524288,20}, {1048576,21}, \ 141 {2097152,22}, {4194304,23}, {8388608,24} } 142#define MUL_FFT_TABLE3_SIZE 243 143#define MUL_FFT_THRESHOLD 3712 144 145#define SQR_FFT_MODF_THRESHOLD 404 /* k = 5 */ 146#define SQR_FFT_TABLE3 \ 147 { { 404, 5}, { 25, 6}, { 13, 5}, { 27, 6}, \ 148 { 29, 7}, { 15, 6}, { 31, 7}, { 25, 8}, \ 149 { 13, 7}, { 28, 8}, { 15, 7}, { 32, 8}, \ 150 { 17, 7}, { 35, 8}, { 29, 9}, { 15, 8}, \ 151 { 35, 9}, { 19, 8}, { 41, 9}, { 23, 8}, \ 152 { 47, 9}, { 27,10}, { 15, 9}, { 39,10}, \ 153 { 23, 9}, { 51,11}, { 15,10}, { 31, 9}, \ 154 { 71,10}, { 39, 9}, { 83,10}, { 47, 9}, \ 155 { 95,10}, { 55,11}, { 31,10}, { 79,11}, \ 156 { 47,10}, { 95,12}, { 31,11}, { 63,10}, \ 157 { 127, 9}, { 255,10}, { 135,11}, { 79,10}, \ 158 { 159,11}, { 95,12}, { 63,11}, { 127,10}, \ 159 { 255, 9}, { 511,10}, { 271,11}, { 143,10}, \ 160 { 287, 9}, { 575,10}, { 303,11}, { 159,12}, \ 161 { 95,13}, { 63,12}, { 127,11}, { 255,10}, \ 162 { 511,11}, { 271,10}, { 543,11}, { 287,10}, \ 163 { 575,11}, { 303,12}, { 159,11}, { 319,10}, \ 164 { 639,11}, { 335,10}, { 671,11}, { 351,10}, \ 165 { 703,11}, { 367,10}, { 735,12}, { 191,11}, \ 166 { 383,10}, { 767,11}, { 415,12}, { 223,11}, \ 167 { 447,10}, { 895,13}, { 127,12}, { 255,11}, \ 168 { 511,10}, { 1023,11}, { 543,12}, { 287,11}, \ 169 { 575,10}, { 1151,11}, { 607,12}, { 319,11}, \ 170 { 671,12}, { 351,11}, { 703,10}, { 1407,11}, \ 171 { 735,13}, { 191,12}, { 383,11}, { 767,10}, \ 172 { 1535,12}, { 415,11}, { 831,12}, { 447,11}, \ 173 { 895,12}, { 479,14}, { 127,13}, { 255,12}, \ 174 { 511,11}, { 1023,12}, { 543,11}, { 1087,12}, \ 175 { 575,11}, { 1151,12}, { 607,13}, { 319,12}, \ 176 { 639,11}, { 1279,12}, { 671,11}, { 1343,12}, \ 177 { 703,11}, { 1407,12}, { 735,13}, { 383,12}, \ 178 { 767,11}, { 1535,12}, { 799,11}, { 1599,12}, \ 179 { 831,13}, { 447,12}, { 895,11}, { 1791,12}, \ 180 { 959,14}, { 255,13}, { 511,12}, { 1023,11}, \ 181 { 2047,12}, { 1087,13}, { 575,12}, { 1215,13}, \ 182 { 639,12}, { 1343,13}, { 703,12}, { 1407,14}, \ 183 { 383,13}, { 767,12}, { 1599,13}, { 831,12}, \ 184 { 1727,13}, { 895,12}, { 1791,13}, { 959,15}, \ 185 { 255,14}, { 511,13}, { 1023,12}, { 2047,13}, \ 186 { 1087,12}, { 2175,13}, { 1215,14}, { 639,13}, \ 187 { 1343,12}, { 2687,13}, { 1471,14}, { 767,13}, \ 188 { 1599,12}, { 3199,13}, { 1727,14}, { 895,13}, \ 189 { 1919,15}, { 511,14}, { 1023,13}, { 2175,14}, \ 190 { 1151,13}, { 2431,12}, { 4863,14}, { 1279,13}, \ 191 { 2687,14}, { 1407,13}, { 2815,15}, { 767,14}, \ 192 { 1535,13}, { 3199,14}, { 1663,13}, { 3455,14}, \ 193 { 1919,16}, { 511,15}, { 1023,14}, { 2175,13}, \ 194 { 4479,14}, { 2431,13}, { 4863,15}, { 1279,14}, \ 195 { 2943,13}, { 5887,15}, { 1535,14}, { 3455,13}, \ 196 { 6911,15}, { 1791,14}, { 3839,16}, { 1023,15}, \ 197 { 2047,14}, { 4479,15}, { 2303,14}, { 4863,15}, \ 198 { 2559,14}, { 5119,15}, { 2815,14}, { 5887,16}, \ 199 { 1535,15}, { 3327,14}, { 6911,15}, { 3839,17}, \ 200 { 1023,16}, { 2047,15}, { 4351,14}, { 8959,15}, \ 201 { 4863,16}, { 2559,15}, { 5887,14}, { 11775,16}, \ 202 { 3071,15}, { 32768,16}, { 65536,17}, { 131072,18}, \ 203 { 262144,19}, { 524288,20}, {1048576,21}, {2097152,22}, \ 204 {4194304,23}, {8388608,24} } 205#define SQR_FFT_TABLE3_SIZE 230 206#define SQR_FFT_THRESHOLD 3264 207 208#define MULLO_BASECASE_THRESHOLD 3 209#define MULLO_DC_THRESHOLD 39 210#define MULLO_MUL_N_THRESHOLD 7246 211#define SQRLO_BASECASE_THRESHOLD 6 212#define SQRLO_DC_THRESHOLD 40 213#define SQRLO_SQR_THRESHOLD 6440 214 215#define DC_DIV_QR_THRESHOLD 30 216#define DC_DIVAPPR_Q_THRESHOLD 88 217#define DC_BDIV_QR_THRESHOLD 35 218#define DC_BDIV_Q_THRESHOLD 62 219 220#define INV_MULMOD_BNM1_THRESHOLD 79 221#define INV_NEWTON_THRESHOLD 11 222#define INV_APPR_THRESHOLD 11 223 224#define BINV_NEWTON_THRESHOLD 264 225#define REDC_1_TO_REDC_2_THRESHOLD 8 226#define REDC_2_TO_REDC_N_THRESHOLD 79 227 228#define MU_DIV_QR_THRESHOLD 1442 229#define MU_DIVAPPR_Q_THRESHOLD 1470 230#define MUPI_DIV_QR_THRESHOLD 0 /* always */ 231#define MU_BDIV_QR_THRESHOLD 1470 232#define MU_BDIV_Q_THRESHOLD 1652 233 234#define POWM_SEC_TABLE 1,16,151,839 235 236#define GET_STR_DC_THRESHOLD 7 237#define GET_STR_PRECOMPUTE_THRESHOLD 15 238#define SET_STR_DC_THRESHOLD 406 239#define SET_STR_PRECOMPUTE_THRESHOLD 885 240 241#define FAC_DSC_THRESHOLD 179 242#define FAC_ODD_THRESHOLD 53 243 244#define MATRIX22_STRASSEN_THRESHOLD 19 245#define HGCD2_DIV1_METHOD 1 /* 9.10% faster than 3 */ 246#define HGCD_THRESHOLD 45 247#define HGCD_APPR_THRESHOLD 50 248#define HGCD_REDUCE_THRESHOLD 2479 249#define GCD_DC_THRESHOLD 321 250#define GCDEXT_DC_THRESHOLD 258 251#define JACOBI_BASE_METHOD 4 /* 15.45% faster than 1 */ 252 253/* Tuneup completed successfully, took 179422 seconds */ 254