T&L speculations

T & L speculations

Since my short and simple remark started a big thrill, here is the explanation of my viewpoint.

The public sources say the following:

Rendition Verite has an on-board programmable RISC processor. The processor feeds the rasterizer with parameters. This simple CPU must be loaded with some program to perform its work.
G200 chip contains a mysterious part called WARP. (Matrox G200 leaflets)
G400 chip contains two WARPs. (Matrox G400 leaflets)
WARP is used for triangle setup. This means that the WARP is supplied with the description of a triangle (xyz, rgba and uv coordinates), and it transforms these coordinates into differential parameters used by rasterizer part. Thus it must be capable of performing division, addition and subtraction at least. (Basic operation of any graphics pipeline)
In order to process the descriptions of various kinds of triangles (with/without alpha channel, with one or two textures), the WARP must be loaded with DIFFERENT "microcode". (See the discussion on GLX driver development)

Note that the above is by mo means restricted information. You could read this in many places. Now, lets think...

The above together leads to speculation that the WARP is a programmable processor, capable of all the arithmetics operations necessary for transformation and lighting (addition, multiplication, reciprocal or division).

Neither I nor the Matrox programmers think it is practical to use the current WARPs to do T&L. It already has a lot to do, and it's better to assign these tasks to the CPU. Basically everything the computer does may be done in hardware or software. The decision is how to distribute the tasks to make the sytem fast and not too expensive.
Early Silicon Graphics workstations used fully-hardware based graphics pipeline. Later cheap models, like Indigo2 or O2, released some tasks to software running on the main CPU. The latest cheap models, like the x86-based 320, again turn towards more functions in hardware. These decisions were dictated by technology development and cost, both in terms of time and money.
I believe that the decision of what tasks should be assigned to WARPs and the main CPU depends on many factors. It's not that Matrox is so malicious that it doesn't want the users to have T&L onboard. Under current conditions the WARP is best used for what it is used - triangle setup. Maybe some future WARPs will do more or less. Just have a little trust in programmers. They usually know what they are doing, and they are trying to get the most of the hardware. And the WARP architecture brings the flexibility needed for future changes.

The goal is not to have T&L in the graphics chip. The goal is to have the fast graphics solution in a PC, and hardware T&L may or may not be the good method to achieve this.