question about nvidia's Cg language

[Special K]

I am trying to understand why this language is supposed to be so useful. (not denying that it is, though) First of all, in non-Cg games, aren't the graphics just implemented within the rest of the source code (by #including the appropriate openGL/D3D files and then calling the included functions within the program)? How would Cg make things any different/better than this? Also when/if the graphics code is inserted within the rest of the game's code, how does the compiler know which instructions are meant for the CPU, and which the GPU? Wouldn't the compiler have to have special code to be able to translate the higher level code into each particular video cards assembly language? Does that mean every time a new vid card comes out, a new compiler must be released to use its new features?

 

[Gaunt]

I'll take a shot at this... but I haven't had any experience with NVidia's new Cg stuff.

With the more recent GPUs, and all of the future GPUs it seems, there is the ability to configure the graphics pipeline through software to produce different effects in hardware, as opposed to software. To do this, developers write code to run on the GPU. My experience with this is a little limited, as I've only done a little work with custom pixel shaders, but here's what I know... The GPU understands certain commands. It is possible write your shader routine in assembly and then send it to the GPU to be used later. To do this, your game code basically says "Hey, Shader, here's the code for this thing... use it". This can be done with DirectX calls... actually, it can be done with one function call IIRC.

The Cg language, on the other hand, is supposed to give the developers a more powerful method of writing the routines to be run on the GPU. Instead of hand-coded assembly/machine language, now it can be a compiled version of a higher level language, similar to C. The code for the GPU is still sent to the GPU in the same way.

So, the compiler for the game code doesn't do anything with the code for the GPU. It doesn't even realize it IS code. It just looks like a regular chunk of data. As for the compiler for Cg needing to change for each card... yes, it might, and NVidia has mentioned this, although the shaders are supposed to recognize the same languages, in theory, I believe. Most likely the compiler for Cg will change based on the maker of the card, and not each individual card, however.

Hopefully that helped shed some light on things a little. I may not have a ton of experience to answer the questions with, but I told you what I know, and it didn't seem like anyone else was really forthcoming with the info.