1*671ea119Smrg /* x86/coreibwl gmp-mparam.h -- Compiler/machine parameter header file. 2*671ea119Smrg 3*671ea119Smrg Copyright 2019 Free Software Foundation, Inc. 4*671ea119Smrg 5*671ea119Smrg This file is part of the GNU MP Library. 6*671ea119Smrg 7*671ea119Smrg The GNU MP Library is free software; you can redistribute it and/or modify 8*671ea119Smrg it under the terms of either: 9*671ea119Smrg 10*671ea119Smrg * the GNU Lesser General Public License as published by the Free 11*671ea119Smrg Software Foundation; either version 3 of the License, or (at your 12*671ea119Smrg option) any later version. 13*671ea119Smrg 14*671ea119Smrg or 15*671ea119Smrg 16*671ea119Smrg * the GNU General Public License as published by the Free Software 17*671ea119Smrg Foundation; either version 2 of the License, or (at your option) any 18*671ea119Smrg later version. 19*671ea119Smrg 20*671ea119Smrg or both in parallel, as here. 21*671ea119Smrg 22*671ea119Smrg The GNU MP Library is distributed in the hope that it will be useful, but 23*671ea119Smrg WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 24*671ea119Smrg or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 25*671ea119Smrg for more details. 26*671ea119Smrg 27*671ea119Smrg You should have received copies of the GNU General Public License and the 28*671ea119Smrg GNU Lesser General Public License along with the GNU MP Library. If not, 29*671ea119Smrg see https://www.gnu.org/licenses/. */ 30*671ea119Smrg 31*671ea119Smrg #define GMP_LIMB_BITS 32 32*671ea119Smrg #define GMP_LIMB_BYTES 4 33*671ea119Smrg 34*671ea119Smrg /* 3400-3800 MHz Intel Xeon E3-1285Lv4 Broadwell */ 35*671ea119Smrg /* FFT tuning limit = 67,000,000 */ 36*671ea119Smrg /* Generated by tuneup.c, 2019-10-20, gcc 8.3 */ 37*671ea119Smrg 38*671ea119Smrg #define MOD_1_NORM_THRESHOLD 15 39*671ea119Smrg #define MOD_1_UNNORM_THRESHOLD 16 40*671ea119Smrg #define MOD_1N_TO_MOD_1_1_THRESHOLD 10 41*671ea119Smrg #define MOD_1U_TO_MOD_1_1_THRESHOLD 8 42*671ea119Smrg #define MOD_1_1_TO_MOD_1_2_THRESHOLD 0 /* never mpn_mod_1_1p */ 43*671ea119Smrg #define MOD_1_2_TO_MOD_1_4_THRESHOLD 10 44*671ea119Smrg #define PREINV_MOD_1_TO_MOD_1_THRESHOLD 11 45*671ea119Smrg #define USE_PREINV_DIVREM_1 1 /* native */ 46*671ea119Smrg #define DIV_QR_1N_PI1_METHOD 1 /* 21.34% faster than 2 */ 47*671ea119Smrg #define DIV_QR_1_NORM_THRESHOLD 14 48*671ea119Smrg #define DIV_QR_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */ 49*671ea119Smrg #define DIV_QR_2_PI2_THRESHOLD 29 50*671ea119Smrg #define DIVEXACT_1_THRESHOLD 0 /* always (native) */ 51*671ea119Smrg #define BMOD_1_TO_MOD_1_THRESHOLD 19 52*671ea119Smrg 53*671ea119Smrg #define DIV_1_VS_MUL_1_PERCENT 295 54*671ea119Smrg 55*671ea119Smrg #define MUL_TOOM22_THRESHOLD 26 56*671ea119Smrg #define MUL_TOOM33_THRESHOLD 97 57*671ea119Smrg #define MUL_TOOM44_THRESHOLD 220 58*671ea119Smrg #define MUL_TOOM6H_THRESHOLD 306 59*671ea119Smrg #define MUL_TOOM8H_THRESHOLD 454 60*671ea119Smrg 61*671ea119Smrg #define MUL_TOOM32_TO_TOOM43_THRESHOLD 93 62*671ea119Smrg #define MUL_TOOM32_TO_TOOM53_THRESHOLD 153 63*671ea119Smrg #define MUL_TOOM42_TO_TOOM53_THRESHOLD 154 64*671ea119Smrg #define MUL_TOOM42_TO_TOOM63_THRESHOLD 169 65*671ea119Smrg #define MUL_TOOM43_TO_TOOM54_THRESHOLD 136 66*671ea119Smrg 67*671ea119Smrg #define SQR_BASECASE_THRESHOLD 0 /* always (native) */ 68*671ea119Smrg #define SQR_TOOM2_THRESHOLD 44 69*671ea119Smrg #define SQR_TOOM3_THRESHOLD 134 70*671ea119Smrg #define SQR_TOOM4_THRESHOLD 242 71*671ea119Smrg #define SQR_TOOM6_THRESHOLD 342 72*671ea119Smrg #define SQR_TOOM8_THRESHOLD 502 73*671ea119Smrg 74*671ea119Smrg #define MULMID_TOOM42_THRESHOLD 98 75*671ea119Smrg 76*671ea119Smrg #define MULMOD_BNM1_THRESHOLD 20 77*671ea119Smrg #define SQRMOD_BNM1_THRESHOLD 23 78*671ea119Smrg 79*671ea119Smrg #define MUL_FFT_MODF_THRESHOLD 540 /* k = 5 */ 80*671ea119Smrg #define MUL_FFT_TABLE3 \ 81*671ea119Smrg { { 540, 5}, { 29, 6}, { 15, 5}, { 31, 6}, \ 82*671ea119Smrg { 16, 5}, { 33, 6}, { 17, 5}, { 36, 6}, \ 83*671ea119Smrg { 25, 7}, { 13, 6}, { 29, 7}, { 15, 6}, \ 84*671ea119Smrg { 33, 7}, { 17, 6}, { 36, 7}, { 19, 6}, \ 85*671ea119Smrg { 39, 7}, { 21, 6}, { 43, 7}, { 23, 6}, \ 86*671ea119Smrg { 47, 7}, { 29, 8}, { 15, 7}, { 35, 8}, \ 87*671ea119Smrg { 19, 7}, { 43, 8}, { 23, 7}, { 49, 8}, \ 88*671ea119Smrg { 27, 7}, { 55, 9}, { 15, 8}, { 31, 7}, \ 89*671ea119Smrg { 63, 8}, { 43, 9}, { 23, 8}, { 55,10}, \ 90*671ea119Smrg { 15, 9}, { 31, 8}, { 67, 9}, { 39, 8}, \ 91*671ea119Smrg { 83, 9}, { 47, 8}, { 95, 9}, { 55,10}, \ 92*671ea119Smrg { 31, 9}, { 79,10}, { 47, 9}, { 95,11}, \ 93*671ea119Smrg { 31,10}, { 63, 9}, { 135,10}, { 79, 9}, \ 94*671ea119Smrg { 159,10}, { 95, 9}, { 191,10}, { 111,11}, \ 95*671ea119Smrg { 63,10}, { 143, 9}, { 287,10}, { 159,11}, \ 96*671ea119Smrg { 95, 7}, { 1599, 8}, { 831, 9}, { 431, 8}, \ 97*671ea119Smrg { 863, 9}, { 447,10}, { 239, 9}, { 479,10}, \ 98*671ea119Smrg { 255, 9}, { 511,10}, { 287,11}, { 159,10}, \ 99*671ea119Smrg { 319, 9}, { 639,10}, { 335, 9}, { 671,11}, \ 100*671ea119Smrg { 191,10}, { 383, 9}, { 767,10}, { 399,11}, \ 101*671ea119Smrg { 223,12}, { 127,11}, { 255,10}, { 511, 9}, \ 102*671ea119Smrg { 1023,11}, { 287,10}, { 607,11}, { 319,10}, \ 103*671ea119Smrg { 671,11}, { 351,12}, { 191,11}, { 383,10}, \ 104*671ea119Smrg { 799,11}, { 415,13}, { 127,12}, { 255,11}, \ 105*671ea119Smrg { 543,10}, { 1119,11}, { 607,12}, { 319,11}, \ 106*671ea119Smrg { 671,10}, { 1343,11}, { 735,12}, { 383,11}, \ 107*671ea119Smrg { 799,10}, { 1599,11}, { 863,12}, { 447,11}, \ 108*671ea119Smrg { 959,13}, { 255,12}, { 511,11}, { 1119,12}, \ 109*671ea119Smrg { 575,11}, { 1215,12}, { 639,11}, { 1343,12}, \ 110*671ea119Smrg { 703,11}, { 1407,13}, { 383,12}, { 767,11}, \ 111*671ea119Smrg { 1599,12}, { 831,11}, { 1727,12}, { 959,14}, \ 112*671ea119Smrg { 255,13}, { 511,12}, { 1215,13}, { 639,12}, \ 113*671ea119Smrg { 1471,13}, { 767,12}, { 1727,13}, { 895,12}, \ 114*671ea119Smrg { 1919,14}, { 511,13}, { 1023,12}, { 2239,13}, \ 115*671ea119Smrg { 1151,12}, { 2431,13}, { 1279,12}, { 2623,13}, \ 116*671ea119Smrg { 1407,12}, { 2815,14}, { 767,13}, { 1535,12}, \ 117*671ea119Smrg { 3135,13}, { 1663,12}, { 3455,13}, { 1919,15}, \ 118*671ea119Smrg { 511,14}, { 1023,13}, { 2175,12}, { 4479,13}, \ 119*671ea119Smrg { 2431,14}, { 1279,13}, { 2943,12}, { 5887,14}, \ 120*671ea119Smrg { 1535,13}, { 3455,14}, { 1791,13}, { 3839,15}, \ 121*671ea119Smrg { 1023,14}, { 2047,13}, { 4479,14}, { 2303,13}, \ 122*671ea119Smrg { 4991,12}, { 9983,14}, { 2559,13}, { 5247,14}, \ 123*671ea119Smrg { 2815,13}, { 5887,15}, { 1535,14}, { 3839,16} } 124*671ea119Smrg #define MUL_FFT_TABLE3_SIZE 172 125*671ea119Smrg #define MUL_FFT_THRESHOLD 7424 126*671ea119Smrg 127*671ea119Smrg #define SQR_FFT_MODF_THRESHOLD 472 /* k = 5 */ 128*671ea119Smrg #define SQR_FFT_TABLE3 \ 129*671ea119Smrg { { 472, 5}, { 29, 6}, { 15, 5}, { 33, 6}, \ 130*671ea119Smrg { 37, 7}, { 19, 6}, { 40, 7}, { 29, 8}, \ 131*671ea119Smrg { 15, 7}, { 35, 8}, { 19, 7}, { 43, 8}, \ 132*671ea119Smrg { 23, 7}, { 49, 8}, { 27, 9}, { 15, 8}, \ 133*671ea119Smrg { 31, 7}, { 63, 8}, { 43, 9}, { 23, 8}, \ 134*671ea119Smrg { 55,10}, { 15, 9}, { 31, 8}, { 67, 9}, \ 135*671ea119Smrg { 39, 8}, { 83, 9}, { 47, 8}, { 95, 9}, \ 136*671ea119Smrg { 55,10}, { 31, 9}, { 79,10}, { 47, 9}, \ 137*671ea119Smrg { 95,11}, { 31,10}, { 63, 9}, { 135,10}, \ 138*671ea119Smrg { 79, 9}, { 159,10}, { 95,11}, { 63,10}, \ 139*671ea119Smrg { 127, 9}, { 255,10}, { 143, 9}, { 287,10}, \ 140*671ea119Smrg { 159,11}, { 95,12}, { 63,11}, { 127,10}, \ 141*671ea119Smrg { 271, 9}, { 543, 6}, { 4479, 7}, { 2431, 8}, \ 142*671ea119Smrg { 1247, 7}, { 2495, 8}, { 1279,10}, { 351,11}, \ 143*671ea119Smrg { 191,10}, { 399, 9}, { 799,10}, { 415,12}, \ 144*671ea119Smrg { 127,11}, { 255,10}, { 543,11}, { 287,10}, \ 145*671ea119Smrg { 607,11}, { 319,10}, { 639,11}, { 351,12}, \ 146*671ea119Smrg { 191,11}, { 383,10}, { 799,11}, { 415,10}, \ 147*671ea119Smrg { 831,13}, { 127,12}, { 255,11}, { 511,10}, \ 148*671ea119Smrg { 1023,11}, { 543,10}, { 1087,11}, { 607,12}, \ 149*671ea119Smrg { 319,11}, { 671,10}, { 1343,11}, { 735,12}, \ 150*671ea119Smrg { 383,11}, { 799,10}, { 1599,11}, { 863,12}, \ 151*671ea119Smrg { 447,11}, { 927,13}, { 255,12}, { 511,11}, \ 152*671ea119Smrg { 1087,12}, { 575,11}, { 1215,12}, { 639,11}, \ 153*671ea119Smrg { 1343,12}, { 703,11}, { 1471,13}, { 383,12}, \ 154*671ea119Smrg { 767,11}, { 1599,12}, { 831,11}, { 1663,12}, \ 155*671ea119Smrg { 895,11}, { 1855,14}, { 255,13}, { 511,12}, \ 156*671ea119Smrg { 1023,11}, { 2047,12}, { 1087,11}, { 2239,12}, \ 157*671ea119Smrg { 1215,13}, { 639,12}, { 1471,13}, { 767,12}, \ 158*671ea119Smrg { 1663,13}, { 895,12}, { 1983,14}, { 511,13}, \ 159*671ea119Smrg { 1023,12}, { 2239,13}, { 1151,12}, { 2495,13}, \ 160*671ea119Smrg { 1279,12}, { 2623,13}, { 1407,14}, { 767,13}, \ 161*671ea119Smrg { 1535,12}, { 3135,13}, { 1663,12}, { 3455,13}, \ 162*671ea119Smrg { 1919,15}, { 511,14}, { 1023,13}, { 2175,12}, \ 163*671ea119Smrg { 4479,13}, { 2431,14}, { 1279,13}, { 2943,12}, \ 164*671ea119Smrg { 5887,14}, { 1535,13}, { 3455,14}, { 1791,13}, \ 165*671ea119Smrg { 3839,15}, { 1023,14}, { 2047,13}, { 4479,14}, \ 166*671ea119Smrg { 2303,13}, { 4991,12}, { 9983,14}, { 2815,13}, \ 167*671ea119Smrg { 5887,15}, { 1535,14}, { 3327,13}, { 6783,14}, \ 168*671ea119Smrg { 3839,16} } 169*671ea119Smrg #define SQR_FFT_TABLE3_SIZE 157 170*671ea119Smrg #define SQR_FFT_THRESHOLD 5568 171*671ea119Smrg 172*671ea119Smrg #define MULLO_BASECASE_THRESHOLD 16 173*671ea119Smrg #define MULLO_DC_THRESHOLD 37 174*671ea119Smrg #define MULLO_MUL_N_THRESHOLD 14281 175*671ea119Smrg #define SQRLO_BASECASE_THRESHOLD 0 /* always */ 176*671ea119Smrg #define SQRLO_DC_THRESHOLD 137 177*671ea119Smrg #define SQRLO_SQR_THRESHOLD 10821 178*671ea119Smrg 179*671ea119Smrg #define DC_DIV_QR_THRESHOLD 54 180*671ea119Smrg #define DC_DIVAPPR_Q_THRESHOLD 146 181*671ea119Smrg #define DC_BDIV_QR_THRESHOLD 98 182*671ea119Smrg #define DC_BDIV_Q_THRESHOLD 218 183*671ea119Smrg 184*671ea119Smrg #define INV_MULMOD_BNM1_THRESHOLD 50 185*671ea119Smrg #define INV_NEWTON_THRESHOLD 173 186*671ea119Smrg #define INV_APPR_THRESHOLD 165 187*671ea119Smrg 188*671ea119Smrg #define BINV_NEWTON_THRESHOLD 278 189*671ea119Smrg #define REDC_1_TO_REDC_N_THRESHOLD 79 190*671ea119Smrg 191*671ea119Smrg #define MU_DIV_QR_THRESHOLD 1787 192*671ea119Smrg #define MU_DIVAPPR_Q_THRESHOLD 1787 193*671ea119Smrg #define MUPI_DIV_QR_THRESHOLD 78 194*671ea119Smrg #define MU_BDIV_QR_THRESHOLD 1589 195*671ea119Smrg #define MU_BDIV_Q_THRESHOLD 1830 196*671ea119Smrg 197*671ea119Smrg #define POWM_SEC_TABLE 1,16,126,416,932 198*671ea119Smrg 199*671ea119Smrg #define GET_STR_DC_THRESHOLD 11 200*671ea119Smrg #define GET_STR_PRECOMPUTE_THRESHOLD 17 201*671ea119Smrg #define SET_STR_DC_THRESHOLD 306 202*671ea119Smrg #define SET_STR_PRECOMPUTE_THRESHOLD 894 203*671ea119Smrg 204*671ea119Smrg #define FAC_DSC_THRESHOLD 141 205*671ea119Smrg #define FAC_ODD_THRESHOLD 34 206*671ea119Smrg 207*671ea119Smrg #define MATRIX22_STRASSEN_THRESHOLD 20 208*671ea119Smrg #define HGCD2_DIV1_METHOD 3 /* 5.97% faster than 1 */ 209*671ea119Smrg #define HGCD_THRESHOLD 73 210*671ea119Smrg #define HGCD_APPR_THRESHOLD 123 211*671ea119Smrg #define HGCD_REDUCE_THRESHOLD 3664 212*671ea119Smrg #define GCD_DC_THRESHOLD 562 213*671ea119Smrg #define GCDEXT_DC_THRESHOLD 465 214*671ea119Smrg #define JACOBI_BASE_METHOD 1 /* 31.16% faster than 3 */ 215*671ea119Smrg 216*671ea119Smrg /* Tuneup completed successfully, took 35114 seconds */ 217