| 00:17.43 | *** part/#brlcad terrylr (i=terrylr@johann.blauedonau.com) | |
| 02:31.11 | *** join/#brlcad tegtmeye (n=tegtmeye@pool-71-248-83-186.bltmmd.east.verizon.net) | |
| 03:13.17 | Twingy | good movie |
| 03:53.31 | brlcad | worth seeing? |
| 03:53.53 | Twingy | indeed |
| 03:53.56 | brlcad | rather.. worth paying $15 to see in theater |
| 03:54.06 | brlcad | mandatory popcorn and all |
| 03:54.15 | brlcad | or just worth seeing eventually |
| 03:56.03 | Twingy | I brought two smores granolla bars and a soda |
| 03:56.13 | Twingy | with torrenting atleast |
| 03:56.16 | Twingy | *worth |
| 03:56.20 | brlcad | 33heh |
| 04:10.08 | Twingy | sean |
| 04:10.12 | Twingy | where is vgrs |
| 04:10.23 | Twingy | XXX times faster? |
| 04:10.38 | Twingy | These numbers seem to indicate that this machine is approximately 12734 times |
| 04:10.38 | Twingy | faster than the reference machine being used for comparison, a VAX 11/780 |
| 04:10.38 | Twingy | running 4.3 BSD named VGR. |
| 04:10.47 | Twingy | I just got that |
| 04:23.12 | brlcad | yes? |
| 04:23.33 | Twingy | you got 11,600 |
| 04:23.34 | brlcad | on the machine "VGR", the results for each of those tests == 1 |
| 04:23.45 | Twingy | I got 12,700 |
| 04:23.56 | brlcad | so 12700 VGRS is X times faster |
| 04:24.01 | Twingy | yes |
| 04:24.05 | Twingy | but icc is 15,000 still |
| 04:24.09 | Twingy | let me see if I can go higher |
| 04:24.49 | brlcad | right, without going deep into options for either, icc was considerably out performing |
| 04:25.16 | brlcad | e.g. didn't use the -march flag, that would have helped some |
| 04:25.28 | brlcad | or -ffast-math |
| 04:25.40 | brlcad | but then icc also hasd similar options |
| 04:26.00 | brlcad | what really steals it is that multi-pass |
| 05:21.23 | Twingy | 12,836 |
| 05:28.23 | Twingy | last test for tonight :) |
| 05:28.32 | Twingy | maybe I can sqeak out 13,000 |
| 05:32.33 | brlcad | sounds about right |
| 05:32.52 | Twingy | I should push adrt through it again |
| 05:32.55 | brlcad | i'd expect it possible to squeak 1-3k out of both with more fine tuning |
| 05:33.00 | Twingy | everytime I have used it in the past gcc was faster |
| 05:33.06 | brlcad | you should try adrt with the multi-pass |
| 05:33.11 | Twingy | I even got the pathscale guys to formally admit gcc was faster |
| 05:39.08 | brlcad | when I first tested on the pre-altix, the gap was about 50%-60% with the same options iirc |
| 05:39.42 | brlcad | so nice to see that the gap is decreasing |
| 05:41.11 | Maloeran | Ahah Justin, gcc defeated pathscale then? :) Neat |
| 05:41.26 | Twingy | there's a bug in their database as a result of my ray-tracer |
| 05:41.36 | Twingy | if I dig up old emails I have the title |
| 05:41.38 | Twingy | I think |
| 05:41.47 | Twingy | if something is faster, they report as bug |
| 05:41.50 | Maloeran | GCC is fairly horrible for Opteron optimisation, in comparison to what you can get from assembly |
| 05:42.03 | Maloeran | Eh, nice |
| 05:42.06 | Twingy | the funny thing is I was testing on opterons |
| 05:42.13 | Twingy | 2ghz quad opteron |
| 05:42.21 | Maloeran | Sure, that's what Pathscale is designed for |
| 05:42.29 | Twingy | at siggraph 2004 |
| 05:44.34 | Maloeran | I maintain that assembly is seriously under-estimated :), I seem to be the only one crazy enough to code paths optimized for archs or chips |
| 05:49.00 | Twingy | do you have a script you run to apply the mods post gcc compile? |
| 05:50.02 | Maloeran | No, it's all hand written |
| 05:50.09 | Twingy | so everytime you compile |
| 05:50.14 | Twingy | you have to open up file with editor |
| 05:50.16 | Twingy | and type it in? |
| 05:50.20 | Twingy | seems kind of laborious |
| 05:50.22 | Maloeran | Though, there are some easy changes to be made to GCC's output to increase performance on amd64 |
| 05:50.55 | Maloeran | Somewhat, it doesn't take too long when you have the C version next to it as reference |
| 05:51.17 | Twingy | seems like a little sed scripts could be useful |
| 05:51.24 | Twingy | to just go patchit.sh |
| 05:51.26 | Maloeran | It's a hand written .S file, I don't change anything |
| 05:51.26 | Twingy | everytime you compile |
| 05:51.35 | Maloeran | gcc foo.S -o foo.o |
| 05:51.39 | Twingy | so all assmembly lives in own file |
| 05:51.47 | Maloeran | A couple files, but yes |
| 05:51.53 | Twingy | I thought you were taking gcc's output |
| 05:51.56 | Twingy | everytime you compile |
| 05:52.02 | Twingy | and putting it in from scratch |
| 05:52.13 | Maloeran | Eh no, working from gcc's output wouldn't lead to the best results |
| 05:52.22 | Twingy | gotcha |
| 06:25.37 | *** join/#brlcad ibot_ (i=ibot@rikers.org) | |
| 06:25.37 | *** topic/#brlcad is http://brlcad.org/ || BRL-CAD is now Free Software! || BRL-CAD Windows port has been merged to HEAD, not yet stable yet || Archer is now on HEAD as well, kick the tires, report any bugs || there WILL be a 7.6.8 released ASAP for January after all | |
| 06:59.37 | *** join/#brlcad learner (n=brlcad@pdpc/supporter/silver/brlcad) | |
| 06:59.37 | *** mode/#brlcad [+o learner] by ChanServ | |
| 09:57.13 | *** join/#brlcad ibot (i=ibot@rikers.org) | |
| 09:57.13 | *** topic/#brlcad is http://brlcad.org/ || BRL-CAD is now Free Software! || BRL-CAD Windows port has been merged to HEAD, not yet stable yet || Archer is now on HEAD as well, kick the tires, report any bugs || there WILL be a 7.6.8 released ASAP for January after all | |
| 13:10.47 | *** join/#brlcad rogier (n=rogier@16-65-dsl.ipact.nl) | |
| 16:25.18 | *** join/#brlcad tegtmeye (n=tegtmeye@pool-71-248-83-186.bltmmd.east.verizon.net) | |
| 17:39.26 | *** join/#brlcad docelic (n=docelic@clj34-71.dial-up.arnes.si) | |
| 21:45.40 | *** join/#brlcad tegtmeye (n=tegtmeye@pool-71-248-83-186.bltmmd.east.verizon.net) | |
| 21:48.21 | *** join/#brlcad grumbel (n=grumbel@i5387ED78.versanet.de) | |
| 22:55.01 | Twingy | yep, that's a dirty job |
| 22:56.42 | Twingy | living room is oficially classified as DMZ now |