First: it appears you have the concept of the "toParent" and "toRoot" matrices correct. The TransformationMatrix is primarily just a handy storage place. However, depending on how you import data, TransformationMatrix often starts out initialized as the "toParent" matrix. In that case, be sure to calculate the "toRoot" matrices before you use the TransformationMatrix for any other purpose. That is, "use it before you lose it." Once the "toRoot" and offset matrix calcs have been done, you won't need the "toParent" matrix any more.

Second: You have to decide on some maximum number of influence bones. I can't remember where I read it, but 4 influences per-vertex is usually more than enough. Your shader code (which is a very common implementation BTW) assumes a maximum of 4 influence bones.

Third: Assume you set a maximum vertex influence at 4 bones. When you set up the vertices (probably during loading or shortly thereafter), make sure each vertex has at least one influence bone, and no more than 4.

If your model has more than 4 influence bones, the best approach is to go back to the modeling program and redo the bone-weighting. At this point, I really don't think you want to get into adding code to your import routine to handle more than 4. Alternatively, when you import, you can simply ignore any bone influences greater than 4. However, depending on how the model was weighted, you won't know what you're throwing away.

In any case:

Also, when setting up the vertices, make sure the indices and weights for "unused" bones are set to 0.

For instance, assume a vertex has 2 influence bones, indexed as A and B. The vertex structure requires 4 bone indices and weights be defined.

For that vertex:

bone index 0: A

bone index 1: B

bone index 2: 0

bone index 3: 0

bone weight 0: A's weight factor

bone weight 1: B's weight factor

bone weight 2: 0

bone weight 3: 0

That should keep you busy for a little while.

If I decide to keep maximum of 6 bone influences what would be the input layout for that ?

You should write your own code. Besides, I personally wouldn't know a good way to do that. But if you use something other than 4 influences, remember to revise the shader.

Looks pretty good. You don't say what "output" is from, but looks like those vertices all have 2 bone weights which sum to 1 (which should be fine). The other weights are 0. Looks as one might expect for vertices with 2 influence bones, but can take up to 4 influence bones. I wouldn't worry about the 4th weight looking "constant." It's close enough to 0 as makes no difference.

The last value might just be the normal process of using the first 3 weights and setting the last weight to 1.0 - weight1 - weight2 - weight3.

Buckeye, it's e+ not e-

A float value of -1.07374e+008 gives me raw hex of 0xCCCCCCB6, which almost looks like the magic number for bad memory. Anyway, a weight of -107 millions sounds bad.

Hokey Smokes, Rockie! Wow. Missed that one. Good catch, unbird. However, I suspect it's immaterial.

@Saad: Where are you getting the data you posted? From what point in your code? If it's coming from the input to the shader, and you're using the shader you posted in this topic, there's only 3 weights being input to the shader (which is fine), and you only need to look at 3.

Does your vertex format support 3 weights or 4? If it supports 3 weights and you're setting 4, you may be trashing some memory somewhere.