Bikerdude: - questions

[Bikerdude]

Is it a custom texture or is it one available in Dark Mod 1.02?

 

its the std 1.02 texture and ive just started a basic map and the SP texture is black in that olso - Ill run an update and see if this is corrected..

[Serpentine]

Is it a custom texture or is it one available in Dark Mod 1.02?

 

It's a problem with DR, the texture/material is set correctly and DR just shows it as black which is unintended (nothing about it or similar mentioned in the changelog)

There's a bug report over here.

[Bikerdude]

There's a bug report over here.

Ah ok, so is it worth going back to 1.2.2 until the bug is fixed..?

[Serpentine]

If it annoys you and you arnt looking at using the new readable objectives stuff, I dont think you'll lose anything hopping back

[Bikerdude]

If it annoys you and you arnt looking at using the new readable objectives stuff, I dont think you'll lose anything hopping back

well I have found another object thats turned black -

 

models/darkmod/architecture/windows/round_stone01_small.lwo

 

anything other than the stock dark texture appears black, eg - windows/roundtop_diamond_pattern01_brightlit etc

 

will add this to the bug log

Edited May 28, 2010 by Bikerdude

[Bikerdude]

when to caulk -

 

 

quick question, Ive been playing with other Dm3 maps - and the ones that have a judicious amount of caulk texture on almost all hidden surfaces perform significatly better, even to the point that when the player is in the void in-game there is significantly less fps drop. I was able to fly quite a distance into the void till I could see the whole map and the fps never dropped below 10.

 

So I assume then I should make my next map as much as possible caulk every hidden surface

Edited May 29, 2010 by Bikerdude

[rich_is_bored]

The general consensus is that Doom 3 does a much better job of removing unseen faces during compile than it's predecessors. But its just an observation. Nobody has done a proper test to see how effective it really is.

 

It would be interesting if someone were to take a map that performs poorly, give it a once over with caulk, and then compare the FPS difference. Even more so, to run through both versions of the map with r_showtris in an attempt to identify the geometry that the compiler missed.

[Serpentine]

I spent quite a lot of time caulking Return to The City, since its skyportals and void- seals had a load of faces that would not be removed.

 

Generally, performance hardly changed, maybe 1-2 more fps in already high fps areas and the bottom line had no noticeable change. That said, I did start to convert parts into statics, which then need extensive caulking as they are not culled at all. Performance started getting a nice bump up from that, compile times also dived and the old visportals seemed to help a bit more for some or other reasons.

 

One other thing to observe in your doom3 experiment is that the maps are mostly using the default doom3 materials which are a lot lower res and way lighter on memory etc. AI also seems to be the largest hit when you are floating in the void (I removed all AI from training_mission and sat in the void with a noticeable increase)

[nbohr1more]

Serp ...are you saying that you are still working on Return to the City's performance? (I recall Melan mentioning that he would try some further tricks to get the FPS up...). (yeah my system is hopelessly crusty so I shouldn't bother but the map has it's playable areas... )...

 

I'm also curious about the affect of geometry on AI path-finding performance... I imagine lots of Columns, Pillars, or Trees would increase the number of places that AI would have to "look"? If so, can this be artificially narrowed (mapped) so that the AI's only see a grid or something less taxing on pathfinding? (again, a selfish infor request for us .0001% crusty 'ole CPU people... or probably just me... unless someone is running TDM on something worse than a P4 3.0ghz).

[Bikerdude]

If so, can this be artificially narrowed (mapped) so that the AI's only see a grid or something less taxing on pathfinding?

yeah I would like to know this...

[Serpentine]

Serp ...are you saying that you are still working on Return to the City's performance? (I recall Melan mentioning that he would try some further tricks to get the FPS up...). (yeah my system is hopelessly crusty so I shouldn't bother but the map has it's playable areas... )...

 

Nah, I stopped - I was working on a beta version and without moving a few buildings around I wasnt going to fit it up... it would have also taken considerable amounts of time to do it nicely

[Fidcal]

My experience is the same as Melan in that extensive caulking had little effect. This is, as said, because Doom3 does a good job of not processing hidden surfaces anyway. What you should caulk is hidden surfaces on entities. So if you make a lot of trim and brush and patchwork and convert it to func_static then change the hidden surfaces to caulk. However, I have not tested this and I suspect it would need quite a bit of surface area to make a noticeable difference. But I always try to do it anyway.

 

One way to simplify pathfinding is when you monsterclip use the furniture etc only as a guide. Don't specifically carefully wrap each piece separately. Say you have a wall. 3 feet from one end is a chair. Then a small gap and a torch on the wall. Then a couple of feet and a table, and so on up to the far corner where there is a 4 foot space. Do you really care if an AI can get into the space? What for? If you go into it he can still hit you but there is no need normally for him to go there. So one big monster clip over all that furniture along that wall. Maybe leave a wide object sticking out like the table. Then one monsterclip on that bit sticking out. I don't even other being careful any more. I drag it quick and dirty into floor and ceiling and walls and overlap clip. I am slightly careful about not wasting space though so the objects almost poke out. But the general rule is to simplify it. So when done the furniture etc doesn't exist for the AI. Just see it as a simple maze. Make sure there are no gaps he can't get through or there is no point in having those gaps. If so, move the furniture slightly to open the gap. If not possible then seal the gap with monsterclip. Keep the visible walk area for the AI simple.

 

And you already know to clip right up to the ceiling except for objects the AI can walk on. So even a chair, clip right up to the ceiling.

 

You can monsterclip worldspawn if it is not to be walked over by AI. That courtyard through the dummy gate which is just for show and nobody can actually go there. Seal the lot. Leave only the minimum area for the AI to walk on.

[nbohr1more]

Sorry to intrude again but if "notarget" cvar is invoked, this would be equivalent (performance-wise) to monsterclip-ing the whole map correct?

[Mortem Desino]

Not quite. Even if notarget is on, the AI still go through all their regular "think" cycles. (Even dead/KO AI go through a little "think" cycle so their physics don't spaz out)

 

If you want to test a map's FPS difference with AI/without AI, you'll have to use the "killMonsters" command.

[nbohr1more]

Thanks Mortem!!! :)

[Bikerdude]

Spoofing light:

 

is it possible to havea candle flame entity that dosent emit any light, so I can place a single light source in the middle of them to give the illusion that the light is coming from all of them..?

 

Also which flame entities in each class (candle, torch, fire) have the lowest tris amount - other than putting each an everyone into a test map....( now I have one use for showtris - seeing which entities are causeing my fps to fall through the floor, grrr)

Edited June 3, 2010 by Bikerdude

[Bikerdude]

other than putting each an everyone into a test map...

well i got bored and thought I would do this with all the candle entities, rather than light(candle) source entities - which seem to have a glow - which is where the hi tris count is coming from.. so in my case the simple solution is to replace all the candle light sources with candle entities...

 

but that then gives me the issue where all the candles are pulsing in-sync with each other...putting all the candle light sources in a row and the only one that dosent have a tris glow is "light_candleflame_sp" but changing to that from what I had hasn't improved the fps

Edited June 3, 2010 by Bikerdude

[Mortem Desino]

You can certainly use the particle tdm_fire_candle.prt on each candle model, and bind it all to a single light.

 

And the new TDM candle models have three different skins, so they don't all pulse in synch.

[Bikerdude]

You can certainly use the particle tdm_fire_candle.prt on each candle model, and bind it all to a single light.

 

And the new TDM candle models have three different skins, so they don't all pulse in synch.

 

whats the benefit of binding the three candles to a single light entity rather than just putting a light entity in close proximity.. and whats the best way to apply the partical, as the only way i could see to do it was via the model lit option, but the flame dosent appear once in game..

 

what are the three new models ( I think maybe Ive used tham already) but when I refer to in-sync, Im refering to the light emitted not the glowing/pulsing of the candle wax.

Edited June 3, 2010 by Bikerdude

[Mortem Desino]

I would suggest using the old candle models that don't have flickering normals (func_static --> models/darkmod/lights/extinguishable/candle_01.ase).

 

Each candle model should have a func_emitter for their candle flame.

 

There should be one light_extinguishable entity targeting every single func_emitter. When the light is doused, all the func_emitters should will turn off.

 

(Exploded view in DR)

[Bikerdude]

(Exploded view in DR)

sweet, ill have a go at that later today, shoudl allow me to have multiple candles in the same location for little cost..

[Bikerdude]

Another question:

 

If i have two rooms(leafs) and I have a large single func_static that spans both rooms (eg a vaulted ceiling). I place a Vp on the door way between the rooms, when said VP is closed does the engine render only part of the FS or the whole thing..?

Edited June 9, 2010 by Bikerdude

[Neb]

I'm pretty sure the entire thing is rendered.

[Bikerdude]

I'm pretty sure the entire thing is rendered.

 

arse, thats what I was worried about, so Ill have to break up the ceiling into two FS either side of the portal etc..

[Bikerdude]

arse, thats what I was worried about, so Ill have to break up the ceiling into two FS either side of the portal etc..

 

wupdate: it made no difference to the FPS, so I have to conclude that the engine dosent render that part of the FS thats behind the closed leaf/Vp.