The Cg Tutorial

Martin Ecker writes "NVIDIA's book The Cg Tutorial: The Definitive Guide to Programmable Real-Time Graphics, published by Addison-Wesley is a book that many 3D graphics programmers have been waiting for. Finally a book is available that introduces NVIDIA's high-level shading language Cg (short for 'C for Graphics') and the concepts involved with writing shader programs for programmable graphics pipeline architectures to the interested reader." If you are such an interested reader, you'll find the rest of Ecker's review below. The Cg Tutorial: The Definitive Guide to Programmable Real-Time Graphics author Randima Fernando, Mark J. Kilgard pages 384 publisher Addison-Wesley Publishing rating 8 reviewer Martin Ecker ISBN 0321194969 summary An excellent introduction to the high-level shading language Cg (C for Graphics) and its uses in real-time 3D graphics.

The first half of the book teaches the basic language constructs of the Cg shading language and shows how to use them in concrete example shaders, whereas the second half concentrates on more advanced techniques that can be achieved on today's programmable GPUs with Cg, such as environment or bump mapping. Even these more advanced techniques are explained in a clear and easy-to-understand manner, but the authors do not neglect to present the mathematics behind the techniques in detail. Especially the more serious 3D programmer will appreciate this fact. The explanation of texture space bump mapping must be the easiest-to-understand explanation of the technique I have read to date, which alone makes it worth to have this book on my shelf. At this point it is important to note that the book does not discuss the Cg runtime which is used by applications to compile and upload shaders to the GPU. The book focuses exclusively on the Cg language itself. So if you're already familiar with Cg and want to learn how to use the Cg runtime, this book is not for you and you should rather read the freely available Cg Users Manual.

The book contains many diagrams and figures to illustrate the discussed equations and show the rendered images produced by the presented shaders. Note that most figures in the book are in black and white which sometimes leads to funny situations, such as in chapter 2.4.3 where the resulting image of a shader that renders a green triangle is shown. Since the figure is not in color the triangle that is supposed to be solid green ends up being solid gray. However, in the middle of the book there are sixteen pages with color plates that depict most of the important color images and also show some additional images of various applications, NVIDIA demos, and shaders written for Cg shader contests at www.cgshaders.org.

Accompanying the book on CD-ROM is an application framework that allows you to modify, compile, and run all the example shaders in the book without having to worry about setting up a 3D graphics API, such as OpenGL or Direct3D. The application framework uses configuration files to load meshes and textures and set up the graphics pipeline appropriately for the shaders. This way the Cg shaders can be examined and modified in isolation with the results being immediately visible in the render window of the application. Thanks to this framework application even readers that are not yet familiar with a 3D graphics API or even 3D artists interested in programmable shading on modern GPUs can begin to learn Cg and experiment with real-time shaders.

A final note for programmers using Direct3D 9: The high-level shading language included with the latest version of Direct3D, simply called HLSL for High-Level Shader Language, is syntactically equivalent to Cg. Everything written in the book about Cg equally applies to HLSL. Thus, the book is also an excellent guide for programmers that only intend to work with HLSL.

This book truly is the definitive guide for all beginners with the Cg language, and also more advanced 3D programmers will find the chapters about vertex skinning, environment mapping, bump mapping, and other advanced techniques interesting. Once you've started writing shaders in Cg you will never want to go back to writing them in low-level assembly shading languages ever again.

You can purchase The Cg Tutorial: The Definitive Guide to Programmable Real-Time Graphics from bn.com. The book's official website has additional information and ordering options besides. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.

Comments

Looks Interesting

[(54)T-Dub]

I just hope it's not too little too late. Nvidia seems to be going the way of Voodoo. Taking the same card and clocking it faster with a bigger fan. It's not bad enough that the athlon's are toasters, we have to have 2 of them in a box with enough fans to have a tornado.

nforce looks pretty cool though.

Re:Looks Interesting

[snackbar]

What are you talking about? The GeForce FX is much different than the GeForce 4, specifically in the way it processes fragments. It also has a much higher transistor count.

NVIDIA

[AbdullahHaydar]

I'm sick of NVIDIA trying to control the graphics market by controlling the language developers have to use.

...That's like someone trying to control Java

...Never mind...(Come to think of it, I can't even think of a counter-example where someone didn't try to control a market through control of the programming language.)

Re:NVIDIA

[Moonshadow]

Personally, I think that Cg is pretty cool. Sure beats writing shaders in assembly. Regardless of its status as a "marketing tool", nVidia has provided game devs with a tool that makes achieving all kinds of nifty effects a lot easier and faster than before. I'm not thrilled with the Geforce FX, but I can stick with a GF4 Ti for a while. :)

Re:NVIDIA

[(54)T-Dub]

A video card independent programing language would be nice. Though the efficiecy would probably rival java's.

I think i would prefer better eye-candy to more eye-candy.

Re:NVIDIA

[Moonshadow]

Yeah, I'd like an independent language too, but for that to happen, we'd have to have some kind of HLL standard be implemented, and we ALL know how well competing companies do with implementing standards proposed by a competitor. Usually, card makers will conform to the standards placed on them by the parent hardware (AGP, analog out, etc) and push their proprietary stuff as the end-all-be-all. There's less interest in cooperation than there is in convincing consumers that they can't live without your nifty new card.

Re:NVIDIA

[Cyberdyne]

...Never mind...(Come to think of it, I can't even think of a counter-example where someone didn't try to control a market through control of the programming language.)

BCPL springs to mind; it was developed (by one of my old supervisors!) specifically to avoid platform lockin. At the time, the university was about to acquire a second computer - but it wasn't compatible with the first. To make matters easier for the users, Martin Richards designed BCPL and an accompanying bytecode language called cintcode. Despite its age - it's an indirect ancestor of C! - it is still in use today in a few applications; apparently Ford have a custom-built setup running in BCPL on a pair of Vaxes to manage each factory outside the US. (For some reason, the US factories use a different system.) With the demise of the Vax, Ford have been supporting Martin's work in porting the whole BCPL/cintcode/Tripos (a cintcode OS) to run on Linux/x86 systems.

For that matter, I seem to recall most of the early computer languages were intended to reduce the need to be tied to a specific platform; Fortran, Pascal, C (derived from B, itself a cut-down version of BCPL), as well as the original Unix concept.

Re:NVIDIA

[br0ck]

This overview by Tom's Hardware of HLSLs says that Nvidia is pushing for Cg to be hardware independant and used by all video card vendors (see the "Which HLSL?" page). The article also explains exactly what HLSLs are and why ATI and Nvidia have created the respective languages Rendermonkey and Cg.

Re:NVIDIA

[TheRealRamone]

Ummm... As the reviewer points out, Cg is more or less equivalent to gpu-vendor-neutral HLSL for Direct3D - which belies your comment about Nvidia trying to dominate the market with a language.

However, one might say that MS and Nvidia are doing so together...

--TRR

Re:NVIDIA

[wct]

Mindless trivia: The original user components of AmigaDOS were written in BCPL, by a British company (Metacomco) contracted by Commodore. Through various revisions, they were rewritten by Commodore in C.

Excellent - Now just give me a nice DX9/HLSL Book

[SmirkingRevenge]

A copy of 3ds Max 5, and a team of artists, and I can start coding graphics stuff for fun again! Uh oh, this is /. and I said DX9, hello Troll demod!

Sounds interesting, but

[0x00000dcc]

Excuse my ignorace in this realm, but why would you want to want to learn Cg when you could extend a C/C++ library to include the various graphics that you want to use?

Seriously, I really would like someone to debunk this idea if possible, because I have picked up an interest in graphics programming and am just starting out - would like to know more. It seems like an easier / pragmatic route (due to code reusability) to go the other route ...

Re:Sounds interesting, but

[Urkki]

The shader programs are totally different anyway, they're not x86 (or whatever) assembly, they run in the graphics chip and have their own specialized assembly language suited for the stuff they do. So it can't be "just an extra library", it needs to be a new compiler and all that. The regular C-library part (which would be done through DirectX anyway) would be just for copying the shader programs to the graphics chip. But I'm not an expert, somebody correct me if I'm wrong.

Re:Sounds interesting, but

[Molt]

Cg is a very specific language which runs on the graphics card itself, and is only used for pixel and vertex shader programming. It's always used in conjunction with one (Or even more) higher level libraries.
Firstly you have the application-level library (SDL, for example), this handles the stuff like opening windows, the user interaction. This is also the bit that's often written specifically for the game.
At the next level we have the 3d library, normally OpenGL or DirectX. This handles most of the actual graphics work itself, such as displaying your polygon meshes, applying standard lighting effects, and so forth.
Finally we hit the shader level. It's here that Cg comes into it's own, with special snippets of Cg code to get the reflections on the water to look just right and ripple as the character walks through, or to make the velvet curtains actually have the distinctive sheen. Special effects work only.
It is worth noting that Direct X does have it's own way of doing shaders now, and OpenGL does have a specification for them but last time I looked no one had this implemented.
Hope this makes sense.

Re:Sounds interesting, but

[SmirkingRevenge]

C/C++ is a high level language. It's intended for writing code for the more macro level (moving objects around a scene, setting lighting sources, texturing, etc).

Cg on the other hand is used for writing what are known as procedural shaders. Shaders determine what an object/particle will look like, _procedural_ shaders can change what a polygon will look like based on any number of criteria (not the lease of which is time).

So if I'm going to texture a wall with wood, it makes sense to proceduraly generate the wood. If I'm making a flickering fire, a procedural fire texture will look a zillion times better. To do these sorts of things you need to operate VERY quickly, and very exactly, a great example of a fire shader is a perlin noise in 4-d fire from the CgShaders site: http://www.cgshaders.org/shaders/show.php?id=39

The above will blow you away.

Re:Sounds interesting, but

[AegisKnight]

The whole point of shaders is that you can now program the video card HARDWARE, which is much faster than doing everything in software.

Re:Guards, Seize Him!

[Anonymous Coward]

I smell a big fat commie rat

Yeah - chapter 17 of Das Kapital is entitled "On Creating A Card Independent Graphics Programming Language". This sort of call for standards is one of the founding tenets of Communism!

Cease and Desist!

[phraktyl]

A Definitive Guide not by O'Reilly? That's it, the gloves are off!

GCC and ANSI C standards

[gr8_phk]

I've been thinking the C standard needs native support for vector data types for some time. Sure, I have Vec3, Vec4 and M4x4 classes that I wrote, but they don't take advantage of SSE instructions and such. Intel has a compiler that supposedly works with these instruction sets, but I haven't tried it. Wider support would be available if Cg was a real standard extension to C. When is GCC going to handle Cg? This will allow all those shaders to be used in software renderers (Mesa for example) unchanged. I'm not sure Cg as defined is the correct way to extend C, but you get my point.

Re:GCC and ANSI C standards

[Hortensia Patel]

The current development version of GCC (3.3) *does* have support for automatic vectorization (i.e. using SIMD instructions where appropriate). I'm not sure whether you need to help it out by flagging decls with GCC-specific attributes, but it's definitely there.

As others have said, Cg is not an extension of C, and GCC will never and should never support it.

good book

[Horny Smurf]

I got a copy last month, and I've only read a few chapters, and skimmed some others, but it looks liek a good book.

Don't let the title foo you -- it contains high level descriptions of the algorithms as well as the mathematical concepts. They cover some advanced realtime techniques that older books don't (since the processing power wasn't there even 4 years ago), but also discuss optimizing for low-end systems.

I do recommend this book if you ahve any interest in graphic programming (whether you use Cg or not). If you use it with Coputer Graphics (3rd edition), you should have access to pretty much all graphic algorithms. (at least until TAOP volume 7: Computer Graphics is written :)

Cg and OpenGL

[dmouritsendk]

First, i dont understand why some people think its a bad idea nvidia are doing Cg. One of the goals of Cg is easier crossplatform development.. thats a noble cause if any =D

Secondly, my bet is OpenGLs shading language and Cg will eventually merge.

check: this

and notice this part:

Compatibility with DX is very important, but they're willing to weigh advantages of changes vs. cost. Cg and the proposed "OpenGL 2.0" language are similar but not identical; both look like C, and both target multiple underlying execution units. Interfaces, communication between nodes, and accessing state across nodes are different. It's very late, but not too late to contemplate merging the two languages.

Re:Other books/sources

[magic]

Why not a good GL wrapper for CG, does one exist? How about some good GL samples, period? Can anyone help here?

I released 75,000 lines of C++ code for supporting OpenGL game development on Windows and Linux as the G3D library. It is under the BSD license. The next release includes support for the shaders that are compiled by Cg-- you can grab it from the SourceForge CVS site.

G3D includes some small OpenGL demos (~200 lines), wrappers for the nasty parts of OpenGL, and wrappers for objects like textures and vertex shaders.

-m

Re:why nvidia may not be going the way of 3dfx yet

[John Hurliman]

Hoping for good ATI drivers for Linux? You must be living in a dream world. ATI hasn't written a single decent driver for ANY platform, let alone their secondary platform support. Every LAN party we have is cursed by the poor sap who read some review saying the new ATI Radeon has 3% faster performance and bought in to the worst supported cards ever made.

On a related note, does Age Of Mythology even support the ATI Radeon 9700 Pro? We messed with it for hours trying different patches, hacking the video card support files in the game, could NOT get it out of software mode. *Sigh*

Re:Guards, Seize Him!

[kisrael]

Yeah - chapter 17 of Das Kapital is entitled "On Creating A Card Independent Graphics Programming Language". This sort of call for standards is one of the founding tenets of Communism!

Remember, you can't have Communism without the "C"!

Re:Platform independence! .. uh..no nvidia.. ati..

[ShallowBlue]

Wrong! Cg and also the Cg runtime are programmed on top of the standardized APIs OpenGL and Dirct3D. So Cg shaders work on any HW that supports these standard APIs.
On top of that: Since it will probably take an other 200 years for OpenGL 2.0 to see the light of the world Cg is the only way to write high-level shader programs for OpenGL.

Cg IS NOT vendor specific

[menasius]

Sure Cg was developed and supported by NVIDIA but it works on a higer level than that. It compiles Programs down to either the DX shader language or the OpenGL ARB standards. The only vendor specific part is support for older hardware (NV_vertex_program extension and the like) but nothing is holding back someone from creating a profile to support ATI's proprietary extensions.

It is another layer and a nice one to boot. There is no performance loss running it on ATI's cards, infact the few demos I have written have run better on my friends radeon than on my Geforce3 by a long shot.

Quit trying to demonize nVidia for bringing some peace to the hectic world of writing shaders nine thousand different ways so some guy with an obscure video card doesn't complain.

-bort

There's only ONE test of graphics anything

[TerryAtWork]

And thats's what Carmack thinks of it.

If he writes Quake 4 in Cg - it's in.

Be wary...

[GarfBond]

Why should you use Cg? At this point, the only benefit one can see is if you're going to be doing crossplatform coding (DX vs. OGL). If you're going to be doing DX-only, you should stick with HLSL. Why?

Cg was developed, designed, and created by nvidia. While one of their claims is that it can be made to run on any card and is multiplatform, don't let that fool you. Cg is, at its worst, a thinly veiled attempt to convince developers to produce optimal code for nvidia cards at the expense of broad hardware support. ATI has already said that they will not be supporting Cg (in order for it to work best on ATI cards, someone needs to create profiles for it) and will instead be supporting HLSL. I doubt S3/Via or SIS have the resources to commit to 2 different projects, so I bet they're going to go with HLSL.

If you don't understand why nvidia might be looking for code that works best only on its cards (it's almost a "duh" question), look at it a different way. Look at the GFFX. In almost every instance, it's a failure. Sure, it can stick to 32-bit precision, but it runs really, really slow when you do (just look at the 3dmark03 scores recently released and john carmack's .plan comments). When it runs at 16-bit precision, it's still damn slow, almost always losing out to the Radeon 9700/9800s, but it's a little more competitive (DX9's minimum spec appears to require 24bit precision, but rumor says the jury's still out on that). It's in nvidia's best interest to make the FX appear to be fast (which it isn't), and so they're relegated to make Cg code that optimizes for nvidia cards their best interest.

Sorry I don't have links, but the beyond3d.com forums have a lot of information on this subject.

Re:Cg for NVIDIA only?

[Dr. Sp0ng]

orgive my ignorance, but is Cg made for NVIDIA only? Or is it even optimized for NVIDIA chips?

It's not optimized for anything. When you compile a Cg program (either offline or at runtime - the Cg compiler is very fast!) you specify a "profile" for it to use. Some of the currently-supported profiles are arbvp1 (which outputs code for OpenGL's ARB_vertex_program extension), vs_1_1 (DirectX 8 vertex shader), vp20 (NVIDIA's NV_vertex_program OpenGL extension), vp_2_0/vp_2_x (DirectX 9 vertex shader), vp30 (NVIDIA's NV_vertex_program2 OpenGL extension), ps_1_1/ps_1_2/ps_1_3 (DirectX 8 pixel shader), fp20 (OpenGL NV_texture_shader and NV_register_combiners extensions), arbfp1 (OpenGL's ARB_fragment_program extension - vendor-independent for older cards), ps_2_0/ps_2_x (DirectX 9 pixel shader, vendor-independent for 4th generation GPUs), and fp30 (NV_fragment_program OpenGL extension, for 4th-gen NVIDIA GPUs).

So these profiles are optimized for their target platforms, and yes, currently NVIDIA chips are better supported. However, vendors can write profiles for their chips without NVIDIA's support, so for example, ATI could write a profile for the Radeon 9800 and it would work fine. However, ATI has already written support for DX9 shaders, so the vs_2_x/ps_2_x targets would work fine for that (or vs_1_x/ps_1_x for the 8500 generation).

Don't listen to the Slashbots here - I am a professional game developer, and Cg is a godsend (and I'm even developing mostly using ATI cards). Since runtime compilation is so fast, Cg programs can be compiled when the game is played and the exact hardware being used is known. I don't imagine I have to go into more detail as to why that's a fantastic thing.

Re:Do they still make "demos"?

[CreateWindowEx]

Modern PC graphics hardware gotten to the point that there is much less emphasis on clever hackery, and more on just generating good source art. In the old demos, the general-purpose CPUs with simple bit-mapped video cards weren't powerful enough to do a lot of 3D effects, but clever programming could 'fake it out' in various ways, usually by taking advantage of some special case (such as the "2.5D" games like Wolfenstein3D). Nowadays, we have specialized 3D hardware pipelines that solve the general/correct case for most standard graphics, and it becomes more a matter of stringing together various primitive operations to get an asthetically-pleasing effect, so it's more "artistic" and less "clever" programming.

I think a console like the PS2 is a much better place for exhibiting the old-fashioned "clever hack", because its weird and exposed architecture allows for many, many more ways to do things, some of which can be very "crazy". Plus hand-coding assembly to fit in 4K or 16K for the vector units seems very similar to the old demo contests... is there a demo scene for the PS2 linux kit? There should be if there isn't...