yade-dev team mailing list archive

[sega <sj2001@xxxxxxxxx>]

> Let me know (sega?) if you get comparable speedups in your case.

I have next benchmarks:

STLImporterTest.py with 1000 spheres for 150k iters:

NO_BEX & VecSet 
19:45 = 1185 s, 127 iter/s
19:05 = 1145 s, 131 iter/s

BEX & VecSet
16:26 = 986 s, 152 iter/s
17:58 = 1078 s, 139 iter/s
16:19 = 979 s, 153 iter/s

NO_BEX & VecMap
15:21 = 921 s, 163 iter/s
14:58 = 898 s, 167 iter/s

BEX & VecMap 
13:53 = 833 s, 180 iter/s
13:25 = 805 s, 186 iter/s
13:34 = 814 s, 184 iter/s

BEX & VecMap & openMP OMP_NUM_THREADS=1
11:02 = 662 s, 226 iter/s
11:21 = 681 s, 220 iter/s
10:47 = 647 s, 232 iter/s (no 3D)
10:40 = 640 s, 234 iter/s (no 3D)

BEX & VecMap & openMP OMP_NUM_THREADS=2
11:24 = 684 s, 219 iter/s
11:22 = 682 s, 220 iter/s (no 3D)

BEX & VecMap & openMP OMP_NUM_THREADS=3
12:40 = 760 s, 197 iter/s

BEX & VecMap & openMP OMP_NUM_THREADS=4
11:37 = 697 s, 215 iter/s (no 3D)

Thus, the quickest option (speedup 2x) is BEX & VecMap & openMP but with 1 thread (???)... strange...
Maybe I am doing something wrong with openMP? 
The process yade-trunk-* takes no more 50-55% CPU (and 5.5% mem) with any num threads...
(I use  Core2 Duo CPU T8100 and 2GB RAM)



-- 
Best regards, Sergei D.