Question

According to this nVidia CG tutorial (and my own experience) accessing uniform arrays in CG shaders with a non-constant index is either inefficient or unsupported (more often than not, unsupported it seems).

My question is; how can I circumvent this problem?

I am currently writing a GPU skinning shader in which I pass an array of bones (4×4 matrices), which I need to access using an index stored in a vertex attribute (specifically, a float4 vector who’s components are cast to ints). Obviously, because of the limitations mentioned above, this doesn’t work…perhaps I’m missing a better method of doing this?

Answer 1

That is indeed the common way of doing it, e.g. (this is HLSL but essentially the same — note the global uniform ‘boneArray’)

float4x3 CalcBoneTransform(float4 blendWeights, float4 boneIndices)
{
  // Calculate normalized fourth bone weight2
  float4 weights = float4(blendWeights.x, blendWeights.y, blendWeights.z , 1.0f - blendWeights.x - blendWeights.y - blendWeights.z);
  // Calculate bone transform
  float4x3 boneTransform;
  int4 indices = boneIndices;
  boneTransform =  weights.x * boneArray[indices.x];
  boneTransform += weights.y * boneArray[indices.y];
  boneTransform += weights.z * boneArray[indices.z];
  boneTransform += weights.w * boneArray[indices.w];
  return boneTransform;
}