Im using keyframed animation in my game, and to use bump mapping im storing normals and binormals for each vertice and for each keyframe( the tangent vector im calculating by crossing them), but the size of my models are huge…im thing in calculating the binormals on the fly, but i think that the normalization operation is very costly to use so many times…is it necessary to have the binormals- tangent normalized to do bump mapping? Will the normalization cube map work right?
Yes, it´s absolutely necessary to have them normalized, or else you will get wrong results.
Using a normalization-cubemap wouldn´t be a good thing, because it costs you a whole texture-unit, which you will certainly need for other stuff. Only if you got one free, you can try it and test if it´s a speed-up.
Thanx jan…another question …how can i do bump mapping without normalization cube map?
There is a paper by Cass Everitt, which explains how to normalize a vector in the register combiners. It´s not a real normalization, but it comes close to it. The results are good enough if your geometry is a bit tesselated, else it will look crappy.
Sorry, i can´t give you the code, i don´t use it. You have to google for it (“normalization in register combiners” or so).
Anyway, if you have a modell, then i think it should be possible to precalculate the normals, etc. and then simply rotate them together with the rest, when you animate it. Don´t know if that´s possible, but it might be. Then you don´t have to renormalize or recalculate the normals.
Of course you will need ONE normalization-cubemap for the light-to-vertex vector, but i´m sure you are aware of that.
Sorry dude, that´s all the advice i can give you, i don´t have much experience with animated stuff.