Re: random: Benchamrking fast_mix2

From: George Spelvin
Date: Sat Jun 14 2014 - 11:13:22 EST


Unfortunately, my test suite is not particularly lightweight.
Here are some figures (idle at full speed, and doing a parallel
kernel compile) for a 2.5 GHz Phenom.

Quite different numbers compared to the Ivy Bridge. When the processor is
bust, fast_mix takes longer, instead. And even the rolled up fast_mix2 is
(slightly) faster than the original fast_mix. The 64-bit code is basically
the same speed as the 2x unrolled 32-bit code.

Presented in the same strange order as the Ivy Bridge, even though it's
not fastest-to-slowest this time.


Rolled up, 3x:
30748 5432 49066 24126
30583 6526 50271 23788
28621 4541 52966 25610
28017 3425 54181 25733
30425 5213 54441 25878
28556 3895 48969 26499
27853 3057 50813 25431
27471 2396 54065 27685
30328 5053 55931 27500
28702 3821 50811 24877

Unrolled, 3x:
26785 4278 40386 12695
25617 3302 41349 16593
25154 2357 41166 15904
24816 1886 40571 17449
27537 4927 41716 15723
28128 6657 42085 18098
25640 3410 41453 16486
25087 2199 39964 12193
27750 5384 38960 13672
26584 4699 38280 13697

Original:
35291 10362 82640 37663
32043 7848 81887 44853
32968 8994 80527 44847
31087 7207 84894 47535
29904 5602 81426 45401
34312 9824 54889 29060
36471 12948 65112 34232
31622 8539 63531 33871
31570 7745 57946 29070
30074 6116 62829 34393

Unrolled, 2x:
24491 2411 40335 16992
27644 6721 38981 16068
25473 4347 35260 10750
24784 3372 34211 10360
24326 2689 34178 11200
26404 4896 34498 10483
24905 3517 33793 10409
24107 2150 35321 11330
23968 1903 33818 11301
26254 4671 32728 9547

64-bit:
26877 7949 52258 25946
24862 5940 55765 27634
24528 5032 56568 28142
24844 5223 55007 27671
27115 8018 55754 27873
25504 6504 54286 27471
24772 5673 52717 26609
24427 4889 51362 27109
27219 8276 51431 22619
25640 6718 52835 26329
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/