cheng/mpir
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Tuning for K10 and fixed a tuning bug.

Browse Source
This commit is contained in:
(no author) 2010-03-14 19:24:27 +00:00
parent 5bef504b53
commit c5ce40f051
2 changed files with 32 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
mpn/x86_64/k8/k10/k102/gmp-mparam.h
View File

@ -1,42 +1,29 @@
/* Generated by tuneup.c, 2010-02-18, gcc 4.4 */
/* Generated by tuneup.c, 2010-03-14, gcc 4.4 */
#define MUL_KARATSUBA_THRESHOLD 24
#define MUL_TOOM3_THRESHOLD 67
#define MUL_TOOM4_THRESHOLD 351
#define MUL_TOOM8H_THRESHOLD 418
#define MUL_TOOM8H_THRESHOLD 442
#define SQR_BASECASE_THRESHOLD 0 /* always (native) */
#define SQR_KARATSUBA_THRESHOLD 34
#define SQR_TOOM3_THRESHOLD 117
#define SQR_TOOM4_THRESHOLD 547
#define SQR_TOOM8_THRESHOLD 552
#define SQR_TOOM3_THRESHOLD 113
#define SQR_TOOM4_THRESHOLD 522
#define SQR_TOOM8_THRESHOLD 592
#define MULLOW_BASECASE_THRESHOLD 9
#define MULLOW_DC_THRESHOLD 32
#define MULLOW_DC_THRESHOLD 27
#define MULLOW_MUL_THRESHOLD 6142
#define MULHIGH_BASECASE_THRESHOLD 13
#define MULHIGH_DC_THRESHOLD 24
#define MULHIGH_MUL_THRESHOLD 5297
#define MULHIGH_DC_THRESHOLD 33
#define MULHIGH_MUL_THRESHOLD 5676
#define MULMOD_2EXPM1_THRESHOLD 18
#define MULMOD_2EXPM1_THRESHOLD 16
#define DIV_SB_PREINV_THRESHOLD 0 /* always */
#define DC_DIV_QR_THRESHOLD 60
#define DC_DIVAPPR_Q_N_THRESHOLD 156
#define INV_DIV_QR_THRESHOLD 2397
#define INV_DIVAPPR_Q_N_THRESHOLD 156
#define DC_DIV_Q_THRESHOLD 189
#define INV_DIV_Q_THRESHOLD 7071
#define DC_DIVAPPR_Q_THRESHOLD 156
#define INV_DIVAPPR_Q_THRESHOLD 9894
#define DC_BDIV_QR_THRESHOLD 58
#define DC_BDIV_Q_THRESHOLD 33
#define POWM_THRESHOLD 95
#define FAC_UI_THRESHOLD 17010
#define POWM_THRESHOLD 339
#define GCD_ACCEL_THRESHOLD 54
#define GCD_ACCEL_THRESHOLD 52
#define GCDEXT_THRESHOLD 0 /* always */
#define JACOBI_BASE_METHOD 1
@ -53,24 +40,35 @@
#define MOD_1_3_THRESHOLD 31
#define DIVREM_HENSEL_QR_1_THRESHOLD 7
#define RSH_DIVREM_HENSEL_QR_1_THRESHOLD 6
#define DIVREM_EUCLID_HENSEL_THRESHOLD 133
#define DIVREM_EUCLID_HENSEL_THRESHOLD 32
#define ROOTREM_THRESHOLD 6
#define GET_STR_DC_THRESHOLD 12
#define GET_STR_PRECOMPUTE_THRESHOLD 25
#define SET_STR_DC_THRESHOLD 410
#define SET_STR_PRECOMPUTE_THRESHOLD 1902
#define GET_STR_DC_THRESHOLD 15
#define GET_STR_PRECOMPUTE_THRESHOLD 23
#define SET_STR_DC_THRESHOLD 402
#define SET_STR_PRECOMPUTE_THRESHOLD 1083
#define MUL_FFT_TABLE { 432, 992, 2240, 2816, 7168, 28672, 114688, 327680, 0 }
#define MUL_FFT_MODF_THRESHOLD 432
#define MUL_FFT_THRESHOLD 4864
#define MUL_FFT_FULL_THRESHOLD 2368
#define SQR_FFT_TABLE { 432, 928, 2240, 2816, 7168, 28672, 114688, 327680, 0 }
#define SQR_FFT_MODF_THRESHOLD 432
#define SQR_FFT_THRESHOLD 4864
#define SQR_FFT_FULL_THRESHOLD 2368
/* Tuneup completed successfully, took 33 seconds */
#define FAC_UI_THRESHOLD 17010
#define DC_DIV_QR_THRESHOLD 58
#define DC_DIVAPPR_Q_N_THRESHOLD 138
#define INV_DIV_QR_THRESHOLD 3762
#define INV_DIVAPPR_Q_N_THRESHOLD 138
#define DC_DIV_Q_THRESHOLD 183
#define INV_DIV_Q_THRESHOLD 7071
#define DC_DIVAPPR_Q_THRESHOLD 162
#define INV_DIVAPPR_Q_THRESHOLD 15290
#define DC_BDIV_QR_THRESHOLD 58
#define DC_BDIV_Q_THRESHOLD 28
/* Tuneup completed successfully, took 41 seconds */
#define MUL_FFT_TABLE2 {{1, 3}, {205, 4}, {377, 5}, {386, 4}, {404, 5}, {831, 6}, {850, 5}, {869, 6}, {2009, 7}, {2053, 6}, {2191, 7}, {2727, 8}, {2849, 7}, {3109, 8}, {3393, 7}, {3468, 8}, {5478, 9}, {5721, 8}, {6957, 9}, {7756, 8}, {8460, 9}, {8836, 8}, {9030, 9}, {9850, 8}, {10066, 9}, {11976, 8}, {12239, 9}, {13939, 10}, {15535, 9}, {18084, 10}, {19722, 9}, {20154, 10}, {23970, 9}, {26142, 10}, {27898, 9}, {28509, 10}, {31772, 9}, {32468, 10}, {36184, 9}, {36977, 10}, {40326, 9}, {41210, 10}, {44943, 11}, {47962, 10}, {57042, 11}, {62207, 12}, {63570, 10}, {73983, 11}, {80681, 10}, {86099, 11}, {98051, 10}, {102394, 11}, {114110, 12}, {127165, 11}, {164920, 12}, {175995, 11}, {179849, 12}, {196129, 11}, {228243, 12}, {233241, 13}, {254354, 12}, {259924, 11}, {277377, 12}, {456509, 13}, {519871, 12}, {659749, 13}, {784582, 11}, {819320, 12}, {933036, 13}, {953467, 14}, {MP_SIZE_T_MAX,0}}

3
tune/tuneup.c
View File

@ -2033,7 +2033,6 @@ all (gmp_randstate_t rands)
printf("\n");
tune_powm (rands);
tune_fac_ui(rands);
printf("\n");
tune_gcd_accel (rands);
@ -2067,6 +2066,8 @@ all (gmp_randstate_t rands)
tune_fft_sqr (rands);
printf ("\n");
tune_fac_ui(rands);
/* dc_div_qr_n, dc_divappr_q, inv_div_qr, inv_divappr_q */
tune_dc_div (rands);