Most of the tutorials and books I've been going through provide code and examples to specify object matrix data (world, view projection, world inverse transpose, etc) for individual objects by setting a constant buffer and then drawing the object, with the vertex shader referring to those values from the constant buffer.
In my case, I'm experimenting with drawing lots of the same object, instanced. Because each instance has its own world matrix and so forth, I can't just set these values in a constant buffer, which means it looks like my instance buffer needs to have all of the instance-specific matrices passed in as vertex shader parameters. In my case, that means, per instance, and in accordance with trying to translate what I saw being set in the constant buffer per object draw call (in the example code I'm referring to) to instead being set as vertex shader parameters, I have a color, a texture coordinate, a normal, a precalculated world-view-projection matrix, a world matrix for calculating lighting, and a world-inverse-transpose matrix for bring normals into world space after non-uniform scaling.
So, at this point, it means the vertex shader will have at least 3 vectors and at least 3 matrices being passed in, with each of those 3 matrices being precalculated per instance and being updated to the instance buffer every draw call. Is this ok? Or is this a red flag that I should be doing things more simply/efficiently? It seems that all those matrix calculations should be done on the GPU, but one of them involves inverting a matrix, and there doesn't seem to be a hlsl function for that.
