History of Software Texture Mapping in Games — 2016 — Episode Guide

0:00Casey Muratori: Introducing Chris Hecker, John Miles, Sean Barrett and Chris Green

3:32Chris Hecker: How many people were at High School of after in 1995?

3:42CM: Continue to introduce the History of Software Texture Mapping in Games

4:22CM: Give us some background to the problem of texture mapping

4:51Chris Green: Working at subLOGIC on Flight Simulator, drawing trees on the horizon by stuffing 16-bits of binary texture into the Amiga's line-drawing hardware

7:46CM: So you wanted to get more visual detail? Did anyone know what texture mapping was?

8:12CG: Doing the 3D transforms that make texture mapping hard to slot in

9:11CM: So floating-point math came later?

9:25CG: The software was originally built for 8-bit Apple II and Commodore 64 computers

9:48CH: Bruce Artwick wrote about quaternions¹

10:14CG: Joking about "useless" quaternions

10:39CM: What year is this?

10:42CG: '88

10:44CM: So, John, in '88 you were at Origin Systems?

10:57CH: Am I the youngest person on this panel besides you?

11:20CM: When did Origin start thinking about 3D graphics?

11:30John Miles: First 3D for Wing Commander, after Chris Roberts showed up looking to publish Times of Lore

12:32CM: Could you give us some background on the Wing Commander technology?

12:39JM: Wing Commander's forward-transform mapper on the 386²

15:28CM: When were the texture mapping articles?³

15:37CH: April '95

15:39CM: So this is seven years?

15:47CH: What was SGI⁴ doing at this time?

15:52JM: It was knowing how to do all these things

15:55CG: The Flight Simulator did texture mapping on video scan-out positions

16:06CH: But towards the early 90s they got texture mapped triangles in IRIS⁵

16:22CM: Can you talk about the relationship between those communities?

16:37CH: None of the books published it⁶

17:06CM: Sean, when did you first post the text file?⁷

17:11Sean Barrett: It would've been '93 or '94

17:26CM: So in 1988 at subLOGIC, why weren't you aware of other people's work on texture mapping?

18:04CG: All information was from the library

18:28CH: Going back to older hardware with today's standard mathematical knowledge

18:46JM: We weren't any good at math, but neither were computers

19:05CH: You can avoid doing math by knowing math

19:11JM: But if you don't know the right way to do it, you don't spend time trying to convince the machine to do it

19:26CH: Knowledge would help even in a constrained environment

19:33CM: You see that sort of thing with demoscene

19:43CH: Knowledgeable efficiency anecdote: Inverse transform⁸

20:39CM: Sometimes a lack of knowledge does lead you to a solution

20:58CG: The texture mapping solution may have been rejected

21:09CM: The difficulty of defining your goal

21:20CH: We just wanted stuff on the screen

21:26CG: Ugly examples of texture mapping from Car and Driver

21:40CM: What were you trying to do in subdividing rectangles?

22:00CG: Subdividing rectangles

22:27CH: Did you affine-map it?

22:31CG: Subdividing rectangles, continued

22:58CM: Can you unpack that?

23:00CH: How perspective projection shifts the centre point

23:22CM: And this was unknown at the time?

23:34CH: It's like the quaternion thing

23:48CM: So you'd tried subdividing rectangles

23:52CG: Poor performance of subdivided rectangles, then the Apple II blitter thing

24:22CM: So then what?

24:24CG: Moving to California to work for Led Lerner on flight simulators and the game Car and Driver,⁹ with the emergence of VGA

25:21CG: EGA to VGA

26:02CM: Early '90s graphics card technology

26:42CG: Bitmap scaling and Gouraud shading on VGA

28:30CM: Self-modifying code

29:32CM: How did bitmap scaling come about?

29:40CG: Impractical bitmap scaling

30:08CG: Gouraud shading¹⁰

31:02CM: Palletised colours

31:32CG: Texture map darkening in Ultima Underworld

32:33CM: So once you'd got bitmap scaling going, you starting thinking what else is fast enough?

32:46CG: Mutation of loops

32:57CM: Getting one working example

33:12CG: Technology shipped in Car and Driver, and Ultima Underworld

35:07CM: Where does Strike Commander line up with Ultima Underworld's release?

35:18JM: Affine mapping vs perspective mapping

37:03CG: Forward-mapping in Ultima Underworld

37:52CM: Doom does not allow up-down tilting

38:04CH: John Carmack solves the specific problem perfectly

38:16JM: It's all about not doing the math

38:24CH: Looking Glass's general approach

38:36JM: Being blown away by Ultima Underworld

38:54CH: Ultima Underworld was also a tiny window

39:04CM: Potential for optimisation when you cannot tilt

39:36CH: Perspective warping

40:00CM: So you've got large polygons, and are trying to get rid of the warping?

40:41CG: I'm not exactly sure where all the warping comes from

40:56JM: I was assuming you were perspectively correct at the scanline endpoints

41:04CG: Problematic clipping in 3D

42:05CH: Did you guys have UV coordinates

42:07CG: We did, and would cut it into four polygons

42:11CH: Because Sean's original paper¹¹ didn't have UV coordinates

42:26SB: It¹² describes the UV coordinates, though

42:37CM: I just want to finish the Strike Commander part

42:41CH: Strike Commander was later, though, wasn't it?

42:45SB: Later than what?

42:46CH: You were saying 486 and Pentium which is way after…

42:51JM: Wing Commander shipped right as the 486s were coming out, '91 or so. Then Strike Commander took a long time for reasons unrelated to texture mapping

43:12CH: Texture mapping was all we talked about back then

43:26CM: How did Strike Commander's 3D technology come about?

44:33JM: Strike Commander's texture mapping technology, inspired by LucasArts' Secret Weapons of the Luftwaffe¹³ and Michael Abrash's articles in Dr Dobbs Journal¹⁴

46:49CH: Texture mapping optimisation anecdote

47:15JM: Inspiration from existence proofs¹⁵

48:58CM: Sean, when did you join Looking Glass?

49:18SB: Joining Looking Glass in '94 when System Shock was basically done

49:25CM: What happens at Looking Glass once you're there?

49:56SB: System Shock's lines of constant-z mapper

51:35CM: So walls and floors are special?

51:40SB: That was the Doom optimisation

51:41CH: Lines of constant-z on a floor, if you can't tilt, is a scanline

51:45CM: Regardless of how you tilt?

51:47SB: The lines of constant-z for walls are vertical, so tilting would change that

51:56CM: But the reason that floors and ceilings are special is because the rotation doesn't affect their plane?

52:09SB: It's an intersection of planes

52:12CM: So walls and floors were using this already?

52:16SB: Lines of constant-z mapper, continued

52:53CM: So it remains the same distance from the viewer?

52:59SB: If you had a z-buffer, it'd be the same value

53:01CH: If you take a plane and slice it through the world, there are your lines of constant-z

53:05CG: When you tilt your head while looking at a wall, those lines are still straight lines which you can just render diagonally

53:17CM: When did the concept of multiple mappers come out?

53:32SB: Underworld one

53:35CM: Could you talk about that quickly?

53:37CH: Doom had multiple mappers

53:43CG: There were special cases for scaling up and down too

53:46CM: So there were a whole bunch of different mappers for performance?

54:04CG: Or unrolling loops for tiny sections of polygons

54:13CH: The audio mixers did that too

54:17CM: Sean, can you give us a little background on the text file?¹⁶

54:49SB: Background to the Texture Mapping file,¹⁷ reading about non-realtime rendering in journals at Maryland library, Usenet and Michael Abrash's articles in Dr Dobbs Journal¹⁸

55:48CH: Did you mean "at the time" like you are now? Or compared to everyone else?

55:53SB: I am now way more scared of math

56:09CM: So you've had exposure to technical literature?

56:21SB: Unhelpful image warping book¹⁹

56:32CH: It²⁰ had a lot of filtering stuff

56:43SB: Solving texture mapping from first principles, as a less transitory record for PC Game Programmer Encyclopedia²¹^,22

58:07CM: What was the process like, coming at this from first principles?

58:23SB: Knowing the 3D transform in math

59:31SB: The basic idea in texture mapping: Interpolating between UV coordinates

1:01:11CM: So you need three well-formed points?

1:01:20CH: They need to be linearly independent

1:01:24CM: If this was a matrix, the matrix has to…

1:01:28CH: If Sean wasn't afraid of math now, he'd have said that

1:01:32CM: The matrix has to not be degenerate?

1:01:34CH: Linearly independent is the correct word for the thing

1:01:37SB: Weird insight: Assuming we have plane equations²³

1:02:41CH: I didn't know enough math to know what a plane equation was

1:02:58CM: So when did this article²⁴ make its way into the world?

1:03:05SB: '94 before Looking Glass, so before System Shock

1:03:10CM: Was that²⁵ one of the first ones published?

1:03:16CH: For perspective, the only thing

1:03:18SB: I didn't try to solve the perspective mapping problem

1:03:27CH: I thought your thing²⁶ had one plane equation over another plane equation

1:03:30SB: It²⁷ gives you the two divides

1:03:37CM: Chris, could you give us some background to your involvement?

1:03:57CH: Meeting Doug Church and everyone at CGDC, there listening to John's talk about Eye of the Beholder III

1:05:10JM: We had a sponsored session at one GDC

1:05:24CH: Writing columns for Game Developer²⁸ and reading Sean's article²⁹

1:06:32CM: So it was kind trying to solve a problem without defining the problem?

1:06:46SB: It wasn't rigorous

1:06:47CH: There was no reference implementation

1:07:17CM: The utility of reference implementations

1:07:27CH: Meeting Sean and understanding none of his article,³⁰ and being introduced to Concrete Mathematics by Ronald Graham, Donald Knuth, and Oren Patashnik³¹

1:09:25CM: Why does a fill convention come into your brain?

1:09:37CH: Improving image quality

1:10:43CM: So people knew that cracks and things were problems

1:10:54JM: It was a hard, but uninteresting problem

1:11:03SB: Doom was sub-pixel correct

1:11:11CM: Do we know why Doom was sub-pixel correct

1:11:24CH: Doom was special-case

1:11:49CM: At some point, he must have thought about sub-pixel precision

1:11:55CH: He didn't know it was called sub-pixel

1:11:58CM: How did he end up getting sub-pixel, then?

1:12:03CH: I'm assuming it wasn't actually sub-pixel

1:12:23CM: So maybe he iterated and got close, but didn't know to apply the same thought process to the triangle rasterisation?

1:12:37CH: Wolfenstein's even more special-case texture mapping, and graphics being just what you talked about

1:13:57CM: So assume you're an expert, write the article, and become that expert?

1:14:04CH: Wanting to write an article from the point of view of an expert

1:14:10JM: Attracting attention from the actual experts

1:14:19CM: So people were saying they've got these cracks and dropouts…

1:14:26CH: Bresenham being about fill conventions

1:14:42CM: So that was more archaeology?

1:14:46CH: Needing to get the fill convention right, the general texture mapper for Under a Killing Moon written by Bruce Johnson, and wanting to document³² the correct operations (floor / ceil / truncation) inspired by Concrete Mathematics³³

1:16:53CM: How did people find this information about getting their fill convention right?

1:17:32CH: Usenet

1:17:35CM: So some people just figured out there's some stuff about this?

1:17:39CH: Fill convention was well understood

1:18:18SB: Subtlety in making consistent the horizontal scan-converting floor mappers and vertical scan-converting wall mappers

1:18:40CG: People used T-junctions without caring, and we rendered concave polygons

1:19:04CM: And talk about the perspective correct part of the articles³⁴

1:19:16CH: Documenting³⁵ the math and assembly for perspective correct mapping

1:19:52CM: Was the math from Sean's article?³⁶

1:19:57CH: Figuring out the math for sub-pixel correctly mapping a triangle in the source bitmap to a triangle on the screen³⁷^,38

1:21:54CM: Thank you, everyone

Keyboard Navigation

Global Keys

Menu toggling

In-Menu and Index Controls

Quotes and References Menus and Index

Quotes, References and Credits Menus

Filter Menu

Filter and Link Menus

Credits Menu