From 06344c67e87f54f82ef4c5b98c75fdd9802b9ad1 Mon Sep 17 00:00:00 2001 From: William Hart Date: Fri, 21 Mar 2014 17:13:17 +0000 Subject: [PATCH] Added core2 tuning values. --- mpn/x86_64/core2/gmp-mparam.h | 101 ++++++++++++++++++---------------- 1 file changed, 54 insertions(+), 47 deletions(-) diff --git a/mpn/x86_64/core2/gmp-mparam.h b/mpn/x86_64/core2/gmp-mparam.h index 6f0bdc98..7fbcbf1e 100644 --- a/mpn/x86_64/core2/gmp-mparam.h +++ b/mpn/x86_64/core2/gmp-mparam.h @@ -1,22 +1,17 @@ -/* Generated by tuneup.c, 2012-10-23, gcc 4.6 */ +/* Generated by tuneup.c, 2014-03-21, gcc 4.7 */ -#define MUL_KARATSUBA_THRESHOLD 16 -#define MUL_TOOM3_THRESHOLD 81 -#define MUL_TOOM4_THRESHOLD 154 -#define MUL_TOOM8H_THRESHOLD 230 +#define MUL_KARATSUBA_THRESHOLD 17 +#define MUL_TOOM3_THRESHOLD 98 +#define MUL_TOOM4_THRESHOLD 148 +#define MUL_TOOM8H_THRESHOLD 238 #define SQR_BASECASE_THRESHOLD 0 /* always (native) */ -#define SQR_KARATSUBA_THRESHOLD 31 -#define SQR_TOOM3_THRESHOLD 91 -#define SQR_TOOM4_THRESHOLD 246 -#define SQR_TOOM8_THRESHOLD 306 +#define SQR_KARATSUBA_THRESHOLD 24 +#define SQR_TOOM3_THRESHOLD 89 +#define SQR_TOOM4_THRESHOLD 234 +#define SQR_TOOM8_THRESHOLD 321 -#define POWM_THRESHOLD 61 - -#define HGCD_THRESHOLD 31 -#define GCD_DC_THRESHOLD 1771 -#define GCDEXT_DC_THRESHOLD 1182 -#define JACOBI_BASE_METHOD 1 +#define POWM_THRESHOLD 146 #define DIVREM_1_NORM_THRESHOLD MP_SIZE_T_MAX /* never */ #define DIVREM_1_UNNORM_THRESHOLD MP_SIZE_T_MAX /* never */ @@ -27,56 +22,68 @@ #define DIVEXACT_1_THRESHOLD 0 /* always */ #define MODEXACT_1_ODD_THRESHOLD 0 /* always (native) */ #define MOD_1_1_THRESHOLD 6 -#define MOD_1_2_THRESHOLD 6 -#define MOD_1_3_THRESHOLD 17 -#define DIVREM_HENSEL_QR_1_THRESHOLD 10 +#define MOD_1_2_THRESHOLD 7 +#define MOD_1_3_THRESHOLD 13 +#define DIVREM_HENSEL_QR_1_THRESHOLD 11 #define RSH_DIVREM_HENSEL_QR_1_THRESHOLD 19 -#define DIVREM_EUCLID_HENSEL_THRESHOLD 8 +#define DIVREM_EUCLID_HENSEL_THRESHOLD 30 -#define ROOTREM_THRESHOLD 6 +#define MUL_FFT_FULL_THRESHOLD 3520 -#define GET_STR_DC_THRESHOLD 13 -#define GET_STR_PRECOMPUTE_THRESHOLD 17 -#define SET_STR_DC_THRESHOLD 6082 -#define SET_STR_PRECOMPUTE_THRESHOLD 11000 +#define SQR_FFT_FULL_THRESHOLD 2016 -#define MUL_FFT_FULL_THRESHOLD 3008 - -#define SQR_FFT_FULL_THRESHOLD 3008 - -#define MULLOW_BASECASE_THRESHOLD 17 -#define MULLOW_DC_THRESHOLD 17 +#define MULLOW_BASECASE_THRESHOLD 0 /* always */ +#define MULLOW_DC_THRESHOLD 14 #define MULLOW_MUL_THRESHOLD 2393 -#define MULHIGH_BASECASE_THRESHOLD 17 -#define MULHIGH_DC_THRESHOLD 17 +#define MULHIGH_BASECASE_THRESHOLD 9 +#define MULHIGH_DC_THRESHOLD 13 #define MULHIGH_MUL_THRESHOLD 2393 #define MULMOD_2EXPM1_THRESHOLD 12 -#define FAC_UI_THRESHOLD 1752 -#define DC_DIV_QR_THRESHOLD 100 -#define DC_DIVAPPR_Q_N_THRESHOLD 118 -#define INV_DIV_QR_THRESHOLD 465 -#define INV_DIVAPPR_Q_N_THRESHOLD 118 -#define DC_DIV_Q_THRESHOLD 144 -#define INV_DIV_Q_THRESHOLD 3837 -#define DC_DIVAPPR_Q_THRESHOLD 118 -#define INV_DIVAPPR_Q_THRESHOLD 10956 -#define DC_BDIV_QR_THRESHOLD 100 -#define DC_BDIV_Q_THRESHOLD 38 +#define SB_DIVAPPR_Q_SMALL_THRESHOLD 0 /* always */ +#define SB_DIV_QR_SMALL_THRESHOLD 0 /* always */ +#define DC_DIV_QR_THRESHOLD 50 +#define INV_DIV_QR_THRESHOLD 1589 +#define INV_DIVAPPR_Q_N_THRESHOLD 50 +#define DC_DIV_Q_THRESHOLD 65 +#define INV_DIV_Q_THRESHOLD 998 +#define DC_DIVAPPR_Q_THRESHOLD 21 +#define INV_DIVAPPR_Q_THRESHOLD 14326 +#define DC_BDIV_QR_THRESHOLD 54 +#define DC_BDIV_Q_THRESHOLD 19 + +#define ROOTREM_THRESHOLD 6 + +#define MATRIX22_STRASSEN_THRESHOLD 23 +#define HGCD_THRESHOLD 113 +#define HGCD_APPR_THRESHOLD 113 +#define HGCD_REDUCE_THRESHOLD 6852 +#define GCD_DC_THRESHOLD 460 +#define GCDEXT_DC_THRESHOLD 342 +#define JACOBI_BASE_METHOD 1 + +#define GET_STR_DC_THRESHOLD 10 +#define GET_STR_PRECOMPUTE_THRESHOLD 16 +#define SET_STR_DC_THRESHOLD 668 +#define SET_STR_PRECOMPUTE_THRESHOLD 1973 + +#define FAC_DSC_THRESHOLD 898 +#define FAC_ODD_THRESHOLD 0 /* always */ /* fft_tuning -- autogenerated by tune-fft */ #define FFT_TAB \ - { { 4, 3 }, { 3, 2 }, { 3, 2 }, { 2, 1 }, { 1, 0 } } + { { 4, 3 }, { 3, 3 }, { 2, 2 }, { 2, 1 }, { 1, 0 } } #define MULMOD_TAB \ - { 4, 3, 3, 4, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 1, 2, 2, 1, 1 } + { 4, 3, 3, 4, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1 } #define FFT_N_NUM 19 -#define FFT_MULMOD_2EXPP1_CUTOFF 256 +#define FFT_MULMOD_2EXPP1_CUTOFF 128 -/* Tuneup completed successfully, took 163 seconds */ +/* Tuneup completed successfully, took 152 seconds */ +