Hacker News Reader

Fake/real is not the most precise terminology, but I think it gets the important point across, so I’ll use it.

In a "real 3D" rasterizer, you interpolate the depth coordinate for each individual pixel. You need this value to perform two steps in the pixel pipeline that are required to get the correct look: First, you use the depth coordinate the perform depth testing (reject pixels that should be hidden). Then, you use it to perform a perspective correct texture lookup (make straight lines on the texture obey the laws of perspective).

If you throw out the depth coordinate anywhere earlier in the pipeline, as the PS1 did, you can’t do either of those things, so you get artifacts where objects flicker and warp when rotating. The rasterizer really needs to be "3D aware" right until writing out the final pixel values.

Note that both artifacts can be lessened, to some degree at least, with workarounds. For depth testing, you simply sort the polygons (which doesn’t work in every case, as triangles can overlap in cycles). And for the texture lookup problem, you subdivide the polygons (this helps because the vertex calculations are perspective correct, but it's expensive and you'd need infinite subdivision to fully solve the problem).

As an aside, the PS1 had another issue where the rasterizer didn’t support subpixel precision, which also leads to artifacts. But IMO that’s mostly unrelated to the 3D coordinate problems — 2D games need that as well to get smooth movement.

> Be as it may, I've seen the 'fake 3D' claim before and I'm starting to wonder if I'm missing something in my understanding of this technology, maybe someone can offer a third opinion?

I agree that all 3D is 'faked' in the sense that it's drawn to a 2D grid of lights eventually, I guess it just depends on where in the pipeline you make that fundamental transformation. I only meant faked in that the GPU itself is 2D, whereas the GTE is just math. So it's kinda like doing CSS transformations to make a web page all 3D-looking (a gross understatement I know, but, nevertheless). Though, semantically, I guess the '3D' part of a modern GPU is also 'just math'. Well now you've got me thinking about it harder...

Wonderful article by the way! I've read yours a few times myself.

I could see an argument being made for 'fake 3D' pretty strongly here. DF Retro has a few videos comparing various consoles, and one of the things that the PS1 and Saturn both got wrong, IIRC, was that textures would warp*, because the 3D look was being provided by affine transforms (as I understand, I may be misremembering the details), rather than perspective correct rasterization.

So, like, the vertexes are in the right spots, but the texture mapping is off. So it kinda depends on if you need texture mapping to be perspective correct to consider a given example of 3D "real" vs "fake". Dark Forces also has perspective warping that happens in it's engine when you look up/down, as an example.

* https://youtu.be/VutzIK3DqZE?t=341 DF Retro analyzes ports of Tomb Raider

Exactly, straight lines on a texture should obey perspective (think rows of bricks on a wall). I think this wasn't as widely understood back then, but notably, id software already got it right a few years earlier in the original Wolfenstein.

Great link. The artifacts are especially apparent when the walls are near the camera, for example at this timestamp: https://www.youtube.com/watch?v=VutzIK3DqZE&t=398s And note that here, the seemingly flat walls are already heavily subdivided into smaller polygons to lessen the impact of the problem. (Why they wouldn't then make use of those additional polygons to include more geometry details for free is beyond me, however.)

The way you've worded this makes it sound as though programmers of the era were somehow unaware of perspective mapping, which I can assure you is not the case at all. Affine texturing was a tradeoff of performance, nothing more.

Wolfenstein's correct texture mapping is due to it using raycasting in a 2D plane and rendering scaled vertical strips of texture, which just happens to be perspective correct because you only ever have surfaces at 90deg angles in the vertical.

As to why those extra triangles weren't used for detailing, it's likely because that would take up extra level data. Tessellation of an existing triangle into smaller triangles doesn't.

From the horses mouths (minus Abrash/Carmack) 'HandmadeCon 2016 - History of Software Texture Mapping in Games' https://www.youtube.com/watch?v=xn76r0JxqNM

Wolfenstein and Doom are 100% correct because Carmack 'cheated' by deciding to never look down/up or draw slopes :-) so whole game is drawn with 'lines of constant Z'. Or as they put it

Chris Hecker (Microsoft/Maxis/etc): that's a classic Carmack thing which is like Fuck those general problems, Im gonna solve this other problem perfectly

John Miles (ORIGIN/Miles Design/etc): Its all about not doing the math, we were still at a point in time when you won by not doing the math