Possible new T3-T5 mul_1
David Miller
davem at davemloft.net
Wed Apr 3 01:14:48 CEST 2013
From: Torbjorn Granlund <tg at gmplib.org>
Date: Wed, 03 Apr 2013 01:05:05 +0200
> I rescheduled the addmul_2 and mul_2. If I have not misunderstood this
> pipeline, we should finally reach 3.5 c/l and 3 c/l, respectively.
Attached are the output of:
tune/speed -p10000000 -s1-1000 -f1.1 -C mpn_mul_2.3
and
tune/speed -p10000000 -s1-1000 -f1.1 -C mpn_addmul_2.3
respectively.
-------------- next part --------------
overhead 6.00 cycles, precision 10000000 units of 3.51e-10 secs, CPU freq 2848.89 MHz
mpn_mul_2.3
1 n/a
2 12.7112
3 11.3335
4 10.0001
5 9.2001
6 8.9837
7 8.4287
8 8.1876
9 8.1112
10 7.9501
11 7.8183
12 7.5001
13 7.6155
14 7.5001
15 7.5567
16 7.2813
17 7.3530
18 7.3056
19 7.2106
20 7.0501
22 7.1592
24 6.9584
26 7.0385
28 6.8572
30 6.9667
33 6.9395
36 6.7917
39 6.8719
42 6.7858
46 6.7718
50 6.7201
55 6.7637
60 6.6417
66 6.6364
72 6.6042
79 6.6836
86 6.6010
94 6.5639
103 6.6408
113 6.6284
124 6.7823
136 6.7501
149 6.8054
163 6.7915
179 6.7543
196 6.6480
215 6.7070
236 6.6102
259 6.6757
284 6.5775
312 6.5609
343 6.6385
377 6.6208
414 6.5411
455 6.5979
500 6.5080
550 6.5091
605 6.5753
665 6.5685
731 6.5623
804 6.4727
884 6.4683
972 6.4629
-------------- next part --------------
overhead 6.00 cycles, precision 10000000 units of 3.51e-10 secs, CPU freq 2847.39 MHz
mpn_addmul_2.3
1 n/a
2 13.6251
3 12.4446
4 10.3751
5 10.2001
6 9.5834
7 9.2144
8 9.3751
9 8.6668
10 8.7001
11 8.6364
12 8.7501
13 8.3078
14 8.3572
15 8.2667
16 8.0626
17 8.1178
18 8.1390
19 8.0527
20 7.9251
22 8.0001
24 7.8751
26 7.8847
28 7.7501
30 7.8334
33 7.7880
36 7.6667
39 7.7437
42 7.6905
46 7.6740
50 7.6801
55 7.6728
60 7.5834
66 7.6061
72 7.5417
79 7.5950
86 7.5350
94 7.5214
103 7.5632
113 7.5487
124 7.8388
136 7.8089
149 7.7585
163 7.7301
179 7.7096
196 7.7143
215 7.6698
236 7.6780
259 7.6294
284 7.6479
312 7.6347
343 7.5831
377 7.5677
414 7.4880
455 7.5539
500 7.5840
550 7.4619
605 7.5240
665 7.5625
731 7.5541
804 7.5523
884 7.5238
972 7.5217
More information about the gmp-devel
mailing list