You really have to measure this stuff. It depends on so many things. If the models are used as prop_static: There is a partitioning done inside the set of convexes so if that partitioning is as good as the bbox one you'll get from the engine's code then it's a wash. The reason you could gain perf is that the convex/convex partitioning is done with a sphere tree and the engine's partitioning of entities is done with AABBs. So if your convexes fit to AABBs much better than they fit to spheres then you'll do fewer ray tests in the case where you split the models.
If the models are used as prop_physics: You can't really split them and save anything since all of the pieces are still moving and the cost is proportional to that. So it may not make a difference to split. > -----Original Message----- > From: [email protected] > [mailto:[email protected]] On Behalf Of > Aditya Gaddam > Sent: Tuesday, March 10, 2009 3:29 PM > To: Discussion of Half-Life Programming > Subject: Re: [hlcoders] Compiling a "complex" static prop - > Weird collision boxes > > I would think so yes. I would only be worried about the > maximum model count > (if there is such a thing). Otherwise, that makes a bit more > sense actually > :) > > - Aditya > > On Sun, Mar 8, 2009 at 9:11 PM, Minh <[email protected]> wrote: > > > I sometimes use models that are more than 20 convex pieces. > Would it be > > more efficient if I broke those models up into smaller > models? So if a > > traceline hits one of the submodels, it only has to check > that model's > > group of convex pieces. > > > > Jay Stelly wrote: > > > It's compiling each face as a separate convex because the > normals aren't > > smooth. Set a single smoothing group for each piece that > you want to be > > separate and it should work fine. > > > > > > Note that you can also configure studiomdl to output the > 76-hull model > > it's computing (which would work too) but that would be > really slow for > > physics and traces, that's why it detects and prints out an error. > > > > > > > > >> -----Original Message----- > > >> From: [email protected] > > >> [mailto:[email protected]] On Behalf Of > > >> Aditya Gaddam > > >> Sent: Saturday, March 07, 2009 7:18 AM > > >> To: Discussion of Half-Life Programming > > >> Subject: Re: [hlcoders] Compiling a "complex" static prop - > > >> Weird collision boxes > > >> > > >> Hi, > > >> > > >> Sorry it took me so long to get back to this. I tried out > > >> what people said - making edges hard. I checked the normals > > >> and they seem to pointing the right direction (they all point > > >> where the polys face. My output for studiomdl is: > > >> > > >> =============================================================== > > >> =============================================================== > > >> > > >> "C:\Program > > >> files\steam\steamapps\[email protected]\sourcesdk\bin\orangebox\ > > >> > > > bin\studiomdl.exe" > > > > > >> -game "C:\Program Files\Steam\steamapps\SourceMods\Tactics" > > >> "E:\Documents\Work\Programming\C++\Tactics\modelsrc\map\walls\ > > >> > > > floor_stairs.qc" > > > > > >> Process started >>> > > >> WARNING: AppFramework : Unable to load module p4lib.dll! > > >> qdir: > > >> "e:\documents\work\programming\c++\tactics\modelsrc\map\walls\" > > >> gamedir: "C:\Program Files\Steam\steamapps\SourceMods\Tactics\" > > >> g_path: > > >> "E:\Documents\Work\Programming\C++\Tactics\modelsrc\map\walls\ > > >> > > > floor_stairs.qc" > > > > > >> Building binary model files... > > >> Working on "floor_stairs.qc" > > >> SMD MODEL floor_stairs_ref.smd > > >> SMD MODEL floor_stairs_idle.smd > > >> SMD MODEL floor_stairs_phys.smd > > >> WARNING: Model has 2-dimensional geometry (less than 0.500 > > >> inches thick on any axis)!!! > > >> ... > > >> exactly 76 of these > > >> .... > > >> WARNING: Model has 2-dimensional geometry (less than 0.500 > > >> inches thick on any axis)!!! > > >> WARNING: COSTLY COLLISION MODEL!!!! (76 parts - 20 allowed) > > >> WARNING: Error with convex elements of floor_stairs_phys.smd, > > >> building single convex!!!! > > >> Model has 1 convex sub-parts > > >> Collision model completed. > > >> --------------------- > > >> writing C:\Program > > >> Files\Steam\steamapps\SourceMods\Tactics\models/map/walls/floo > > >> > > > r_stairs.mdl: > > > > > >> bones 964 bytes (1) > > >> animations 112 bytes (1 anims) (1 frames) [0:00] > > >> sequences 220 bytes (1 seq) > > >> ik/pose 164 bytes > > >> eyeballs 0 bytes (0 eyeballs) > > >> flexes 0 bytes (0 flexes) > > >> textures 136 bytes > > >> keyvalues 0 bytes > > >> bone transforms 0 bytes > > >> Collision model volume 4907876.50 in^3 > > >> collision 0 bytes > > >> total 1916 > > >> --------------------- > > >> writing C:\Program > > >> Files\Steam\steamapps\SourceMods\Tactics\models/map/walls/floo > > >> > > > r_stairs.vvd: > > > > > >> vertices 20400 bytes (425 vertices) > > >> tangents 6800 bytes (425 vertices) > > >> total 27264 bytes > > >> --------------------- > > >> Generating optimized mesh "C:\Program > > >> Files\Steam\steamapps\SourceMods\Tactics\models/map/walls/floo > > >> > > > r_stairs.sw.vtx": > > > > > >> body parts: 8 bytes > > >> models: 20 bytes > > >> model LODs: 12 bytes > > >> meshes: 18 bytes > > >> strip groups: 50 bytes > > >> strips: 54 bytes > > >> verts: 3825 bytes > > >> indices: 1488 bytes > > >> bone changes: 24 bytes > > >> everything: 5523 bytes > > >> --------------------- > > >> Generating optimized mesh "C:\Program > > >> Files\Steam\steamapps\SourceMods\Tactics\models/map/walls/floo > > >> > > > r_stairs.dx80.vtx": > > > > > >> body parts: 8 bytes > > >> models: 20 bytes > > >> model LODs: 12 bytes > > >> meshes: 18 bytes > > >> strip groups: 50 bytes > > >> strips: 54 bytes > > >> verts: 3825 bytes > > >> indices: 1488 bytes > > >> bone changes: 24 bytes > > >> everything: 5523 bytes > > >> --------------------- > > >> Generating optimized mesh "C:\Program > > >> Files\Steam\steamapps\SourceMods\Tactics\models/map/walls/floo > > >> > > > r_stairs.dx90.vtx": > > > > > >> body parts: 8 bytes > > >> models: 20 bytes > > >> model LODs: 12 bytes > > >> meshes: 18 bytes > > >> strip groups: 50 bytes > > >> strips: 54 bytes > > >> verts: 3825 bytes > > >> indices: 1488 bytes > > >> bone changes: 24 bytes > > >> everything: 5523 bytes > > >> > > >> Completed "floor_stairs.qc" > > >> <<< Process finished. > > >> ================ READY ================ > > >> > > >> The following two screenshots show what the collision > mesh looks like: > > >> > > >> from the top: > > >> http://screenshot.xfire.com/screenshot/large/4bd205d2698d681f8 > > >> 73d2238c2de6ee806ef4a49.jpg > > >> from the bottom: > > >> http://screenshot.xfire.com/screenshot/large/55a0bbc2167dfbb3e > > >> 771dc78b16bb7c038b2be9b.jpg > > >> > > >> (ignore the broken material) > > >> > > >> The following screenshots show normals: > > >> > > >> http://upload.pixelfaction.com/files/49b28ff8_Capture.PNG > > >> http://upload.pixelfaction.com/files/49b28fe5_Capture1.PNG > > >> > > >> > > >> Does anyone have any idea why studiomdl thinks all my parts > > >> are 2 dimensional? Does the collision model depend at all on > > >> the ref model? > > >> > > >> Thanks, > > >> Aditya > > >> > > >> > > >> On Wed, Mar 4, 2009 at 5:03 PM, Jay Stelly > > >> <[email protected]> wrote: > > >> > > >> > > >>> You should only have one $collisionmodel statement. > Studiomdl does > > >>> not support aggregating those. > > >>> > > >>> You didn't post any output. studiomdl must be reporting an > > >>> > > >> error if > > >> > > >>> it's falling back to a single convex hull. What is the > > >>> > > >> error? Post the output. > > >> > > >>> There are two likely cases: > > >>> > > >>> 1) The smoothing group suggestion. There's a simple > vertex welding > > >>> algorithm in the collision model code that requires you to share > > >>> normals at vertices to have those faces belong to a > single convex. > > >>> This is why you have to set smoothing groups. > > >>> > > >>> 2) Too many convex pieces. The default is 20 (I think 40 > > >>> > > >> in l4d on). > > >> > > >>> This is a budget, the physics system can support large > numbers but > > >>> performance may be a problem. Anyway, you can set this in > > >>> > > >> your script > > >> > > >>> with > > >>> $maxconvexpieces: > > >>> > > >>> $collisionmodel "floor_stairs_floor_phys.smd" > > >>> { > > >>> $concave > > >>> $maxconvexpieces 100 > > >>> } > > >>> > > >>> Jay > > >>> > > >>> > > >>> > > >>>> -----Original Message----- > > >>>> From: [email protected] > > >>>> [mailto:[email protected]] On > > >>>> > > >> Behalf Of Aditya > > >> > > >>>> Gaddam > > >>>> Sent: Wednesday, March 04, 2009 5:15 AM > > >>>> To: Discussion of Half-Life Programming > > >>>> Subject: [hlcoders] Compiling a "complex" static prop - Weird > > >>>> collision boxes > > >>>> > > >>>> Hi, > > >>>> > > >>>> I am currently trying to compile a "complex" static prop. I say > > >>>> "complex" > > >>>> because it is basically the little stairway prop shown in this > > >>>> screenie: > > >>>> > > >>>> http://upload.pixelfaction.com/files/49ae7dad_stairs.jpg > > >>>> > > >>>> It is however being a PITA to export. > > >>>> > > >>>> Just exporting the same merged mesh as the ref, idle and > > >>>> > > >> phys smds > > >> > > >>>> didn't work out. I get a "shrinkwrapped" (as the wiki > called it) > > >>>> collision model out of studiomdl. > > >>>> > > >>>> I then tried exporting the phys smd as several phys smds, each > > >>>> containing a piece of the model positioned from the > origin in the > > >>>> right place. So I exported a stairs_floor_phys.smd, > > >>>> stairs_toprail_phys.smd etc and then changed the qc > file to this: > > >>>> > > >>>> $modelname "map/walls/floor_stairs.mdl" > > >>>> $body "floor_stairs" "floor_stairs_ref.smd" > > >>>> $surfaceprop concrete > > >>>> $cdmaterials "models/map/walls/1/" > > >>>> $sequence idle "floor_stairs_idle.smd" loop fps 15 $scale 1.0 > > >>>> $collisionmodel "floor_stairs_floor_phys.smd" > > >>>> { > > >>>> $concave > > >>>> } > > >>>> $collisionmodel "floor_stairs_toprail_phys.smd" > > >>>> { > > >>>> $concave > > >>>> } > > >>>> $collisionmodel "floor_stairs_bottomrail_phys.smd" > > >>>> { > > >>>> $concave > > >>>> } > > >>>> $collisionmodel "floor_stairs_stairs_phys.smd" > > >>>> { > > >>>> $concave > > >>>> } > > >>>> $staticprop > > >>>> > > >>>> This didn't work out well either. Only the stairs > seemed to get a > > >>>> collision mesh and it was funky (some of the stairs worked like > > >>>> stairs, others had holes in the mesh at random places (my > > >>>> > > >> grenades > > >> > > >>>> would go through to the > > >>>> floor) and others were just covered by a flat plane type > > >>>> > > >> collision > > >> > > >>>> mesh. > > >>>> > > >>>> What is the right way to go about these kinds of meshes? Any > > >>>> pointers would be appreciated. Tutorials I have seen > > >>>> > > >> online always > > >> > > >>>> work with very simple props like a barrel etc. > > >>>> > > >>>> Thanks, > > >>>> Aditya > > >>>> > > >>>> -- > > >>>> http://www.pixelfaction.com > > >>>> AIM:ApeWithABrain > > >>>> _______________________________________________ > > >>>> To unsubscribe, edit your list preferences, or view the list > > >>>> archives, please visit: > > >>>> http://list.valvesoftware.com/mailman/listinfo/hlcoders > > >>>> > > >>>> > > >>>> > > >>> _______________________________________________ > > >>> To unsubscribe, edit your list preferences, or view the > > >>> > > >> list archives, > > >> > > >>> please visit: > > >>> http://list.valvesoftware.com/mailman/listinfo/hlcoders > > >>> > > >>> > > >>> > > >> -- > > >> http://www.pixelfaction.com > > >> AIM:ApeWithABrain > > >> _______________________________________________ > > >> To unsubscribe, edit your list preferences, or view the list > > >> archives, please visit: > > >> http://list.valvesoftware.com/mailman/listinfo/hlcoders > > >> > > >> > > >> > > > > > > _______________________________________________ > > > To unsubscribe, edit your list preferences, or view the > list archives, > > please visit: > > > http://list.valvesoftware.com/mailman/listinfo/hlcoders > > > > > > > > > > > > > _______________________________________________ > > To unsubscribe, edit your list preferences, or view the > list archives, > > please visit: > > http://list.valvesoftware.com/mailman/listinfo/hlcoders > > > > > > > -- > http://www.pixelfaction.com > AIM:ApeWithABrain > _______________________________________________ > To unsubscribe, edit your list preferences, or view the list > archives, please visit: > http://list.valvesoftware.com/mailman/listinfo/hlcoders > > _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlcoders
