I would like your opinion of the Nintendo Recirculating shader patent, particulary its possible implementation as a part of Hollywood.
the 'recirculating shade tree blender for a graphics system' patent covers a TEV-like apparatus (AFAICT from my limited knowledge of the original TEV). now, i did not check the figures whose description you're citing here, but i went through the whole claims section, and these are my impressions:
it does not correspond directly to any dx/gl shader model, but it's a step further from a full-fledged dx7 staged pipeline. the closest thing to the patented apparatus would possibly be nvidia's register combiners.
the 'texture enviroment unit' as they call it in the patent, otherwise known as TEV here, is a multi-steged shading pipeline (just as dx7's) but with greatly improved complexity/flexibility of the individual stage; TEV's stage comprises of a blender/combiner unit and a texture addressing/fetching unit, and the combiner takes up to
4 inputs (vs up to 3 in dx7) comprising colors from the tri rasterizer, the current output from the tex unit and the recirclulation registers from previous stages (i'd assume here one or more drawpass-wise constants too, though i don't remember the patent deliberately mentioning that). the combiner unit itself contains one multiplier and one adder, and runs at twice the rate of the tex unit; now the next is not deliberately stated by the patent, IIRC, but i'd assume this TEV could do one mul-add expression with recirculating arguments and another similar one with the current output of the tex unit, and all this would take one 'stage'. you can have as many as
16 stages like this. ..unless the patent actually means that for 16 combiner stages you can tex-fetch every other stage, which would mean that you could effectively use up to 8 textures for the max num of stages - which would be identical to the original flipper's TEV, AFAIK. i honestly don't remember the patent stating one way or the other. i guess a cube developer could really help us out here by telling how many combining stages the original TEV had. the tex unit itself can do indirection (just like the original TEV) where the output of one stage from the unit goes as addressing to the next stage of that unit (after undergoing an intermediate transform).
for what it's worth, until we get some tip from a cube dev to clear up some points, this patent may just as well be about the original flipper's TEV.
Question, if it is a part, would you expect documentation to be available for first devkits, or could the docs appear later in new dev docs? After Nintendo has come up with all the possible algorithms.
i've never worked with nintendo* so i cannot comment on this. but then again, even if i had, i guess i would not have been at libery to comment on such matters ; )
from common sense, though, whatever nintendo planned to have as a featureset in their final silicon, the same should have been conveyed to devs as early as possible, even if early devkits had very little of that exposed. for something as self-contained as a GPU, that means as soon as its design phase is done.
* i've always been in the PC segment, though at the present i'm in the amusement machines segment, which is probably the closest thing to arcades these days.
