This is pretty cool

I don't think what you are doing is new. Maybe combining a bunch of well known techniques.

I would be worth taking further though. See if you can improve the performance. I don't see any reason why it shouldn't work on all hardware if it works on a rage 128.

I came up with something like this when I was thinking a lot about cel-shading a few months ago... it's a fairly logical progression from there. Still, he who implements gets the credit . Anyway,

GL_RENDERER: Rage 128 OpenGL Engine
drew 584.9k polys per second @ 30.1 fps (gl backface cull)
drew 579.3k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 632.5k polys per second @ 30.0 fps (gl backface cull)
drew 589.2k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 631.8k polys per second @ 30.0 fps (gl backface cull)
drew 534.5k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 577.3k polys per second @ 30.0 fps (gl backface cull)
drew 582.4k polys per second @ 30.0 fps (local backface cull + anti-alias)

GL_RENDERER: ATi Rage 128 Pro OpenGL Engine
drew 166.0k polys per second @ 30.2 fps (gl backface cull)
drew 176.1k polys per second @ 29.8 fps (local backface cull + anti-alias)
drew 195.6k polys per second @ 30.0 fps (gl backface cull)
drew 181.2k polys per second @ 29.9 fps (local backface cull + anti-alias)
drew 383.5k polys per second @ 30.3 fps (gl backface cull)
drew 231.8k polys per second @ 30.0 fps (local backface cull + anti-alias)


this is on a G4 400 b/w ati128. w_reade clearly has a beefier machine than me - rascal!

The aa looks very nice in the 'Demo' app. From 'Speed Test' it looks like there isnt much of a speed hit. When you patent it make sure to post your licensing terms!

'Demo' actually looks very cool - are you writing a level-editor editor or 3d-modeller or both. Is it purely in-house - or are you planning on making it into a product itself?

-Codemattic

hold on, memory (and ASP) tell me I've got a Rage 128 Pro, now I come to think of it. I've got a G4/400 too, so perhaps the "Rage 128" and "Rage 128 Pro" strings have got swapped somewhere? After all, you'd expect that the "Pro" one would be faster.

I like my machine a lot though - it may be a bit old now, but it stayed beefy for longer than any other machine I've had.

Quote:Originally posted by codemattic
this is on a G4 400 b/w ati128. w_reade clearly has a beefier machine than me - rascal!

That's one problem with it. Doing the backface culling and silhouette determination is a significant cost on slow CPUs.

Quote:The aa looks very nice in the 'Demo' app. From 'Speed Test' it looks like there isnt much of a speed hit. When you patent it make sure to post your licensing terms!

I'd certainly consider case-by-case licenses for small developers .

Quote:'Demo' actually looks very cool - are you writing a level-editor editor or 3d-modeller or both. Is it purely in-house - or are you planning on making it into a product itself?

It was originally going to be a generic in-house framework for editors (sorta like Kai's demo shell), but I have a few apps that use it that I'm considering releasing at some point (they're still way to buggy now to even consider releasing though).

Quote:Originally posted by w_reade
hold on, memory (and ASP) tell me I've got a Rage 128 Pro, now I come to think of it. I've got a G4/400 too, so perhaps the "Rage 128" and "Rage 128 Pro" strings have got swapped somewhere? After all, you'd expect that the "Pro" one would be faster.

Different OS?

10.2.0?

GL_RENDERER: NVIDIA GeForce2MX OpenGL Engine
drew 7578.2k polys per second @ 30.2 fps (gl backface cull)
drew 549.6k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 698.7k polys per second @ 30.0 fps (gl backface cull)
drew 566.6k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 707.4k polys per second @ 30.0 fps (gl backface cull)
drew 571.7k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 702.3k polys per second @ 30.0 fps (gl backface cull)
drew 563.2k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 710.3k polys per second @ 30.0 fps (gl backface cull)
drew 525.5k polys per second @ 30.0 fps (local backface cull + anti-alias)

Running on an 867MHz G4 with X.2.3. Definitely a speed hit with the AA, but not the 70% you said.

I am also running a G4 400, but mine has been stuffed over the years. I found the gf2mx to be a worthy companion to the CPU. Under 9 it is still a reasonalble gaming machine, unfortunately less so on X. I will do this speed test as soon as I get around to it.

- D.G

Quote:Originally posted by Bachus
GL_RENDERER: NVIDIA GeForce2MX OpenGL Engine
drew 7578.2k polys per second @ 30.2 fps (gl backface cull)
drew 549.6k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 698.7k polys per second @ 30.0 fps (gl backface cull)
drew 566.6k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 707.4k polys per second @ 30.0 fps (gl backface cull)
drew 571.7k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 702.3k polys per second @ 30.0 fps (gl backface cull)
drew 563.2k polys per second @ 30.0 fps (local backface cull + anti-alias)
drew 710.3k polys per second @ 30.0 fps (gl backface cull)
drew 525.5k polys per second @ 30.0 fps (local backface cull + anti-alias)

Running on an 867MHz G4 with X.2.3. Definitely a speed hit with the AA, but not the 70% you said.

Wow, that's weird. Shows ya how long it's been since I deved this thing. Under 10.1 the gl backface cull sets would run as fast as that first set, and the anti-aliased sets would run at about the speed that the later gl backface culls are running. I really need to look into what apple changed.

i know this thread is a little old..but i thought id throw in my 50 cents...

cant you just do glEnable(GL_POLYGON_SMOOTH); to antialias polygons?

Quote:Originally posted by ghettotek
cant you just do glEnable(GL_POLYGON_SMOOTH); to antialias polygons?

You certainly can, but you may not want to for several reasons:

1. It requires that every single pixel in the image get blended, something that is slow on older cards. and
2. It also requires that you either:
 a. Draw the entire scene back to front, foregoing any z-fail optimizations your graphic card might do, and potentially generating visible seams where two polygons meat. or
 b. draw the scene front to back, using a potentially slow source-dependant blending algorithm
3. as a result of 2, no matter what you do you have to decompose your scene into a single polygon list, and then depth-sort this list. and
4. dealing with transparency and anti-aliasing at the same time is difficult to impossible if you chose 2b.

i didnt know that. i guess it never occurred to me since im using low polygon primitives to develop, and i never noticed anything out of the ordinary relating to speed or graphics.

in cocoa you can use antialiasing in the NSOpenGLView by adding multisample attributes to the current pixel format.

AGL & CGL also let you request multisample buffers. The Rage 128 won't multisample, though, and I don't know if the Radeon/R7K cards do either.