source: data/data_extern/programs/Cg/VTFInstancing.cg

//---------------------------------------------------------------------------
//These materials/shaders are part of the NEW InstanceManager implementation
//Written by Matias N. Goldberg ("dark_sylinc")
//---------------------------------------------------------------------------

//---------------------------------------------
//Vertex Shader Input
//---------------------------------------------
struct VS_INPUT
{
        float4 Position :       POSITION;
        float3 Normal   :       NORMAL;
        float3 Tangent  :       TANGENT;
#ifdef BONE_TWO_WEIGHTS
        float4 weights          :       BLENDWEIGHT;
#endif
        float2 uv0              :       TEXCOORD0;

        float4 m01              :       TEXCOORD1;
        float4 m23              :       TEXCOORD2;
};

#include "InstancingVertexInterpolators.cg"
#ifdef ST_DUAL_QUATERNION
#include "DualQuaternionSkinning_Shadow.cg"
#endif

//---------------------------------------------
//Main Vertex Shader
//---------------------------------------------
VS_OUTPUT main_vs( in VS_INPUT input,
                                   uniform float4x4 viewProjMatrix,
                                   
#ifdef DEPTH_SHADOWCASTER
                                   uniform sampler2D matrixTexture : register(s0)
#else
                                   uniform sampler2D matrixTexture : register(s2)
#endif

#if defined( DEPTH_SHADOWCASTER ) || defined( DEPTH_SHADOWRECEIVER )
                                ,  uniform float4 depthRange
#endif
#ifdef DEPTH_SHADOWRECEIVER
                                ,  uniform float4x4 texViewProjMatrix
#endif
                                )
{
        VS_OUTPUT output;

        float4 worldPos  = 0;
        float3 worldNorm = 0;

#ifdef ST_DUAL_QUATERNION
        float2x4 blendDQ;       
        blendDQ[0] = tex2D( matrixTexture, input.m01.xy );
        blendDQ[1] = tex2D( matrixTexture, input.m01.zw );
#ifdef BONE_TWO_WEIGHTS
        float2x4 blendDQ2;
        //Use the empty parts of m03, z and w, for the second dual quaternion
        blendDQ2[0] = tex2D( matrixTexture, input.m23.xy );
        blendDQ2[1] = tex2D( matrixTexture, input.m23.zw );
        
        //Accurate antipodality handling. For speed increase, remove the following line
        if (dot(blendDQ[0], blendDQ2[0]) < 0.0) blendDQ2 *= -1.0;
        
        //Blend the dual quaternions based on the weights
        blendDQ *= input.weights.x;
        blendDQ += input.weights.y*blendDQ2;
        //Normalize the resultant dual quaternion
        blendDQ /= length(blendDQ[0]);
#endif
        //Only dealing with one weight so normalization of the dual quaternion and weighting are unnecessary
        worldPos = float4(calculateBlendPosition(input.Position, blendDQ), 1.0);
        worldNorm = calculateBlendNormal(input.Normal.xyz, blendDQ);
#else
        float3x4 worldMatrix;
        worldMatrix[0] = tex2D( matrixTexture, input.m01.xy );
        worldMatrix[1] = tex2D( matrixTexture, input.m01.zw );
        worldMatrix[2] = tex2D( matrixTexture, input.m23.xy );

        worldPos = float4( mul( worldMatrix, input.Position ).xyz, 1.0f );
        worldNorm= mul( (float3x3)(worldMatrix), input.Normal );
#endif

        //Transform the position
        output.Position         = mul( viewProjMatrix, worldPos );

#ifdef DEPTH_SHADOWCASTER
        output.ps.unused        = float3( 0 );
        output.ps.depth         = (output.Position.z - depthRange.x + SHADOW_BIAS) * depthRange.w;
#else
        output.ps.uv0           = input.uv0;

        //Pass Normal and position for Blinn Phong lighting
        output.ps.Normal        = normalize(worldNorm);
        output.ps.vPos          = worldPos.xyz;
        
        #ifdef DEPTH_SHADOWRECEIVER
                // Calculate the position of vertex in light space to do shadows
                output.ps.lightSpacePos = mul( texViewProjMatrix, worldPos );
                // make linear
                output.ps.lightSpacePos.z = (output.ps.lightSpacePos.z - depthRange.x) * depthRange.w;
        #endif
#endif

        return output;
}