(L) [2006/06/15] [tbp] [Radius] Wayback!Lately i've been, s l o w l y, stitching & re-factoring various bits of quadrille, gpp & whatnot into something now known as radius. It doesn't cure cancer or make grass greener or hair grow back either. Yet.
It's hosted by the fine folks of Gna! & released under GPL v2;
Status: primal scream.
no packets atm.
incomplete unthreaded kd-tree compiler (but that version is somehow reliable).
a giant linux fix is pending, until then it's borked.
apart from the compiler, it's all threaded.
supports 3 compilers: gcc, icc, msvc.
OpenGL post-processing not in yet, even if the display goes through it.
no shading.
To sum it up: boring & useless but, hey, it's software libre. So... happy birthday to me & radius.
[LINK https://gna.org/projects/radius/]
[LINK http://cvs.gna.org/cvsweb/?cvsroot=radius]
(L) [2006/06/16] [Shadow007 not logged in] [Radius] Wayback!Once again, I'd like to thank the ones who share !
I don't have time yet to have a look at it (still playing with Phantom's KdTree compiler), but will soon [SMILEY Smile]
(L) [2006/06/21] [tbp] [Radius] Wayback!I've commited packet (not bundles) traversal the other day; some (read most) SSE support routines are not checked in so it's not as fast as it should be - plus it's supposed to be robust and think i've got it partially wrong, need to check further -.
As there's no fancy .obj parsing, only raw triangle loading, today i've kludged quadrille to be able to dump everything and added some code to load that into radius; so you can consider quadrille as a bloated frontend. I'll commit that code shortly + a proper binary.
Funnily enough, it seems that radius is way faster than quadrille (even considering its lack of shading); i really don't know why, i mean that's more or less the same code refactored. *scratches head*
Now that i have a way to feed varied stuff into radius while being able to directly compare it to quadrille it's about time to think about re-factoring, finalize & commit the kd compiler (read proper radix sort + clipping), even if i can safely say recent events ruined my day [SMILEY Wink]
Note about relative compiler performance wrt packet traversal: msvc 2k5 really really sucks as usual while g++ 4.2 & icc 9.1 are tied in 32bit mode; gcc is ~3-5 % faster than icc when nearing 100% coverage - that is no ray miss the bounding box - and it's the other way around when there's more miss than hits... so it's a branch prediction issue, but when icc mimics msvc (as it does on win32) there's no access to __builtin_expect() and i can't fix it. That sucks.
(L) [2006/06/21] [Phantom] [Radius] Wayback!Ruined your day eh? Sorry 'bout that. [SMILEY Wink] And Toxie is also sorry. Then again, Toxie's day was ruined also, so let's go and get drunk shall we?
_________________
--------------------------------------------------------------
Whatever
(L) [2006/06/22] [toxie] [Radius] Wayback!I'll put that on my todo-list for tomorrow evening, okay? )
_________________
what do you expect to do if you don't know what to do when you've got nothing to do?
(L) [2006/07/09] [tbp] [Radius] Wayback!Did a quick commit over to CVS before we crush those silly italians *cough* Zidane président! *cough
No time for packing a tarball with sources & bins tho.
_________________
May you live in interesting times.
[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]
(L) [2006/08/14] [tbp] [Radius] Wayback!Preliminary parallel kd-tree construction results tell me my heuristic for load balancing sucks, but for trivial cases.
Still, managed to get the buddah done in less than 5s for the first time.
_________________
May you live in interesting times.
[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]
(L) [2006/08/15] [fpsunflower] [Radius] Wayback!Cool. What was your single threaded buddah time?
Are you doing single splits in parrallel (several threads figuring out a single split plane) or doing sub-trees in parrallel?
Or both?
(L) [2006/08/15] [tbp] [Radius] Wayback!I think parallelization of splits is out of question - keyword being 'think' because i haven't tried:
a) too short to make it possible to amortize synchronization overhead.
b) locality, caching & coherency issues.
c) there's only 3 axis, i guess one could find other parts to parallelize but that would aggravate a)
I've made the thing iterative, performance wise it's about the same but it's then easy to distribute sub-trees (and it takes slightly less memory).
Single threaded buddhafication (of the data provided by toxie) takes ~0.9s to sort and ~7.4s to compile proper; it's not typical, but gives an idea.
In any case that's my first working try.
PS: all timings done on win32.
_________________
May you live in interesting times.
[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]
(L) [2006/08/16] [tbp] [Radius] Wayback!It seems coming up with a better heuristic is both trickier than first thought and eluding me. Or the other way around.
Wall, head.
Head, wall.
_________________
May you live in interesting times.
[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]
(L) [2006/08/16] [tbp] [Radius] Wayback!Made some progress wrt compilation of unbalanced trees; now throwing another cpu at the faery scene brings me near the golden ratio of 2 intead of 1.000001.
Random notes:
. have i said how entertaining it can be to try to analyze the dynamic behavior of a load balancer?
. there's nothing better than a NUMA box to spot cheesy code & race conditions.
_________________
May you live in interesting times.
[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]
back
