Edit Intro: Why 'manually' mitter in the first place? This has to do with how q3map2 takes appart rectangles (our standard brush face). These need to be turned into triangles for the engine to be able to handle them. In an ideal case one rectangle would be cut up into exactly 2 triangles. Alas in the real world there are some complications, e.g. touching vertices. So it can help to manually cut up brushes... this is called mittering.

Pre Mittering (46 tris in blue frame)

Note the cut-up floor ornaments. The outer blue frame marks the area we are interested in. Light blue lines denote the axial cuts I used before mittering. As you can see I am using axial brushes, this leads to 46 tris.

Post Mittering (26 tris in red frame)

If we 'move' the light blue cuts to where the yellow cuts are in the 2nd shot, e.g. apply the (45 degree) mittering, the tris count will go down to 26. A 56% saving in tris.

Since I have 6 places in my map where I use this floor ornament, that would be 120 tris saved.

Is this worth it... in most cases probably not, if your tris counts are under 8-9k anyway. But if you are starving to reduce your tris counts, this will definately help.

Hope that clears up a few things and help the uncertain (like me just a few hours ago) to take another shot at mittering.

AEon

------------------
Do i preach to you when you'r lying stone in the gutter?
NO!
Now beat it!..

Vegard,

there is another thread in this forum, where the r_speeds where discussed. In essence they decided that the right-most r_speeds value (left of the word tris) should be less than 10k (10000). Then the map should run well on the hardware that was available when was released. Note that was several years back.

Some folks point out that these counts should actually be upped, since we just about all have better hardware now. But the consensus was again to be conservative and stick to the 10k.

It was also pointed out that if you do indeed have 10k r_speeds in a map, that this is already too high, since in the 'live' game you will have player models, and gibs, and bullet ejects and explosions. These will up the r_speeds significantly. So the suggestion was to be even more conservative and try to build maps - with *all* entities (health, armour, ammo etc.) in the map - topping out at about 7-8k leaving some 2-3k for all the player models etc. stuff for the 'live' action.

As ydnar recently pointed out though. r_speeds are not all, if you use evil multi-level shaders - a *lot* - your r_speeds may be low, but the gfx hardware may not be able to draw the scenes fast enough, so that ultimately you should keep an eye on the FPS (frags per second, cg_drawfps 1) value.

At least that is as much as I have gathered these past weeks.

posted 05-24-2003 03:46 AM          

quote:

---

Originally posted by AE-on: the FPS (frags per second

---

I had that confused with FPH, frags per hour, cause that's what really counts

posted 05-24-2003 06:10 AM           

its a shame, that even many id maps are not using this technique as it is so essential...
I always wanted to do a tut about that, but was to lazy...

Edit: That might actually be a funny thing to do for a competition. Optimize one of id's sample maps, q3dm7 has my vote, then we would need so way to run a demo in the map, that will average the number of tris seen over the time of the demo. This way one could see who optimized the map the best

That brings up another thought. Is there a value in q3map2 console output, that would show the total number of tris in a map? I was thinking of a quick lookup to see if the caulking / mittering one did actually worked as intended.

AEon

Edit: Don't see it. Might you want to draw us a picture on this? Thanx

AEon

posted 05-25-2003 04:47 AM          

this would however force you to use regular light entities (assuming yours in the middle are surfacelight ones)

you are right, I had actually mittered the above greyish floor ornament, but must have forgotten to csg merge them. This would indeed save 2 more tris.

On the texture merging, now wouldn't that be cheating

posted 05-25-2003 12:07 PM          

quote:

---

Originally posted by Ziggurat: this would however force you to use regular light entities (assuming yours in the middle are surfacelight ones)

---

If I really wanted to plan this out, I could build a new texture, build a blend channel and have the blue light shader working.

But we should hold our horses . Since there is the issue of time invested vs. gain. E.g. gain 10-20 tris maybe, by spending hours to build a new texture/shader?

But you folks are right, it would work and would save tris.

AEon

posted 05-26-2003 03:56 AM          

quote:

---

Originally posted by AE-on: The gag at this point was, that I wanted to use the existing blue light shaders, in a floor ornament texture that did not have lighting. If I really wanted to plan this out, I could build a new texture, build a blend channel and have the blue light shader working. But we should hold our horses . Since there is the issue of time invested vs. gain. E.g. gain 10-20 tris maybe, by spending hours to build a new texture/shader? But you folks are right, it would work and would save tris. AEon P.S. yes I am lazy

---

Agreed.
Time seems to be money.

posted 05-26-2003 06:32 AM          

quote:

---

Originally posted by Vegard: Is it good mapping to use mittering every its possible? Or is there some places where "a better solution can take place"?

---

Use of mitering is not recommended on structural brushes, because it overcomplicates the BSP.

http://www.mastahuk.com/index2.php?map=miter

The best way to check if mittering was the good thing to do is to run through the map with r_showtris turned on. There are cases where you can actually make things worse, mittering, because to default 'tris' generation of q3map2 is doing a better job.

In a room that had FPS problems, I actually checked and rechecked r_showtris, to reduce the number of tris, via caulking and mittering.

AEon

posted 05-26-2003 06:54 AM           

quote:

---

Originally posted by ribgib: Use of mitering is not recommended on structural brushes, because it overcomplicates the BSP. http://www.mastahuk.com/index2.php?map=miter

---

Not by much and do this only if you're using the caulk-hull method.

Only the invisible structural caulk is not mitered if using the caulk-hull. All other detail and drawn structural brushes should be mitered.

quote:

---

Originally posted by AE-on: There are cases where you can actually make things worse, mittering, because to default 'tris' generation of q3map2 is doing a better job.

---

Not really. If you're not sure or don't understand what you're doing, then certainly that's the case. But if you really understand how q3 generates tris, then you can always miter in a correct way to get the same or less tris.

------------------
Mapping Feng Shui Zen
worldfactory - LE resources

I was indeed trying to refer to the not knowing what you are doing sort of clipping.

Vangard,

fewer is better . But I have been finding out that tris is not all, as ydnar recently pointed out. I had r_speeds of 5-6K in a room but the FPS hit was still significant... as it turns out fog is evil

posted 05-26-2003 11:42 AM          

quote:

---

Originally posted by obsidian: Only the invisible structural caulk is not mitered if using the caulk-hull. All other detail and drawn structural brushes should be mitered.

---

posted 05-26-2003 06:16 PM           

Here are three of my maps for use in this thread. You'll have to find/replace the textures that are missing, or download the maps from infosprite if you want the original textures. A readme is included to help you out.
http://www.montrealserai.com/mitering-example-maps.zip

------------------

Still would rock if you could take shots of 3-4 places where clever mittering was happening. Yes, I am a junky for shots

AEon

posted 05-27-2003 06:59 AM           

quote:

---

Originally posted by krekits: I disagree to a certain extent. Could you please explain why you think so?

---

All I was trying to say was that if you're using the caulk-hull method, you are supposed to, in theory, create all the textured stuff as mitered detail brushes while the hull should be unmitered structural caulk. It is possible to not follow this rule if you know what you're doing, but if you don't then you risk overdraw, extra tris, z-fighting, etc, etc.

If you look at Q's sample map, he doesn't miter some of his corners but has caulk on the untextured sides to prevent z-fighting. Q does this because he knows what he's doing, but I wouldn't recommend it to someone who doesn't fully understand Q's method, as they are most likely to make mistakes.

posted 05-27-2003 12:10 PM          

I put up a page some time ago to demonstrate this, people interested in a technique that does not involve mitering but still keeps triangle count down might want to check it out: