source: GTP/trunk/App/Demos/Vis/FriendlyCulling/src/shaders/globillum.cg @ 2904

////////////////////
// SSAO + color bleeding shader
// based on shader of Alexander Kusternig

#include "../shaderenv.h"


struct fragment
{
         // normalized screen position
        float4 pos: WPOS;
        float4 texCoord: TEXCOORD0; 
        float3 view: COLOR0;
};


struct pixel2
{
        float4 ssao_col: COLOR0;
        float4 illum_col: COLOR1;
};


struct pixel
{
        float4 illum_col: COLOR0;
};


float2 myreflect(float2 pt, float2 n)
{
        // distance to plane
        float d = dot(n, pt);
        // reflect around plane
        float2 rpt = pt - d * 2.0f * n;
        return rpt;
}

struct GiStruct
{
        float3 illum;
        float2 ao;
};


/** Computes  diffuse reflections + ambient occlusion
*/
GiStruct globIllum(fragment IN,
                                 uniform sampler2D colors,
                                 uniform sampler2D positions,
                                 uniform sampler2D noiseTexture,
                                 uniform float2 samples[NUM_SAMPLES],
                                 uniform float3 currentNormal,
                                 uniform float3 currentViewDir,
                                 uniform float4 centerPosition
                                 )
{
        GiStruct gi;

        // the w coordinate from the persp. projection
        float w = centerPosition.w;

        // Check in a circular area around the current position.
        // Shoot vectors to the positions there, and check the angle to these positions.
        // Summing up these angles gives an estimation of the occlusion at the current position.

        // ao is in stored in the w component
        float3 total_color = float3(0, 0, 0);
        float total_ao = 0.0f;
        float numSamples = 0.0f;

        ////////////
        //-- the main sampling loop

        for (int i = 0; i < NUM_SAMPLES; i ++) 
        {
                float2 offset = samples[i];

#if 1
                ////////////////////
                // add random noise: reflect around random normal vector (warning: slow!)
                float2 mynoise = tex2D(noiseTexture, IN.texCoord.xy).xy;
                float2 offsetTransformed = myreflect(offset, mynoise);
#else
                float2 offsetTransformed = offset;
#endif
                // weight with projected coordinate to reach similar kernel size for near and far
                float2 texcoord = IN.texCoord.xy + offsetTransformed * AREA_SIZE * w;

                if ((texcoord.x <= 1.0f) && (texcoord.x >= 0.0f) && (texcoord.y <= 1.0f) && (texcoord.y >= 0.0f))
                        ++ numSamples;

                // use lower lod level to improve cache coherence
                float3 sample_position = tex2Dlod(positions, float4(texcoord, 0, 1)).xyz;
                float3 sample_color = tex2Dlod(colors, float4(texcoord, 0, 2)).xyz;

                float3 vector_to_sample = sample_position - centerPosition.xyz;
                const float length_to_sample = length(vector_to_sample);

                float3 direction_to_sample = vector_to_sample / length_to_sample;

                // Angle between current normal and direction to sample controls AO intensity.
                float cos_angle = max(dot(direction_to_sample, currentNormal), 0);

                // distance between current position and sample position controls AO intensity.
                const float distance_intensity = 
                        (SAMPLE_INTENSITY * DISTANCE_SCALE) / (DISTANCE_SCALE + length_to_sample * length_to_sample);

                // if normal perpenticular to view dir, only half of the samples count
                /*
                const float view_correction = 1.0f + VIEW_CORRECTION_SCALE * (1.0f - dot(currentViewDir, currentNormal));
                total_color.w -= cos_angle * distance_intensity * view_correction;
                total_color.xyz += cos_angle * distance_intensity * view_correction * sample_color * ILLUM_INTENSITY;
                */
                total_ao += cos_angle * distance_intensity;
                total_color += cos_angle * distance_intensity * sample_color * ILLUM_INTENSITY;
        }

        gi.illum = saturate(total_color);
        gi.ao = float2(max(0.0f, 1.0f - total_ao), numSamples);

        //return saturate(total_color);
        return gi;
}


/** The mrt shader for screen space ambient occlusion + indirect illumination
*/
pixel2 main(fragment IN, 
                   uniform sampler2D colors,
                   uniform sampler2D positions,
                   uniform sampler2D normals,
                   uniform sampler2D noiseTexture,
                   uniform float2 samples[NUM_SAMPLES],
                   uniform sampler2D oldSsaoTex,
                   uniform sampler2D oldIllumTex,
                   const uniform float4x4 oldModelViewProj,
                   uniform float maxDepth,
                   uniform float temporalCoherence
                   )
{
        pixel2 OUT;

        float4 norm = tex2D(normals, IN.texCoord.xy);
        
        // the ambient term
        const float amb = norm.w;

        // expand normal
        float3 normal = normalize(norm.xyz);

        /// the current view direction
        float3 viewDir;// = normalize(IN.view);

        // the current world position
        const float4 centerPosition = tex2D(positions, IN.texCoord.xy);
        
        // the current color
        const float4 currentCol = tex2Dlod(colors, float4(IN.texCoord.xy, 0, 0));
        // the current depth is stored in the w component
        const float currentDepth = currentCol.w;

        GiStruct gi = globIllum(IN, colors, positions, noiseTexture, samples, normal, viewDir, centerPosition);
        

        /////////////////
        //-- compute temporally smoothing

        float4 realPos = centerPosition * maxDepth;
        realPos.w = 1.0f;

        float4 oldPos = mul(oldModelViewProj, realPos);

        const float newDepth = oldPos.z / oldPos.w;

        float2 tex = (oldPos.xy / oldPos.w) * 0.5f + 0.5f;

        float4 oldSsao = tex2D(oldSsaoTex, tex);
        float4 oldIllum = tex2D(oldIllumTex, tex);

        const float oldDepth = oldSsao.w;
        const float depthDif = 1.0f - newDepth / oldDepth;

        float oldWeight = clamp(oldSsao.z, 0, temporalCoherence);
        float newWeight;

        const float oldNumSamples = oldSsao.y;

        if ((temporalCoherence > 0.0f) &&
                (tex.x >= 0.0f) && (tex.x < 1.0f) && 
                (tex.y >= 0.0f) && (tex.y < 1.0f) && 
                (abs(depthDif)  < 1e-3f)
                && (oldNumSamples > gi.ao.y - 1.5f) // check if something happened in the surrounding area
                )
        {
                newWeight = oldWeight + 1;

                OUT.ssao_col.xy = (gi.ao + oldSsao.xy * oldWeight) / newWeight;
                OUT.illum_col.xyz  = (gi.illum + oldIllum.xyz * oldWeight) / newWeight;
        }
        else
        {
                newWeight = 0;

                OUT.ssao_col.xy = gi.ao.xy;
                OUT.illum_col.xyz = gi.illum;
        }

        OUT.ssao_col.z = newWeight;
        OUT.ssao_col.w = currentDepth;

        return OUT;
}


pixel combine(fragment IN, 
                          uniform sampler2D colors,
                          uniform sampler2D ssaoTex,
                          uniform sampler2D illumTex
                   )
{
        pixel OUT;

        float4 col = tex2D(colors, IN.texCoord.xy);
        float ao = tex2D(ssaoTex, IN.texCoord.xy).x;
        float4 illum = tex2D(illumTex, IN.texCoord.xy);
        
        OUT.illum_col = (col + illum) * ao;
        //OUT.illum_col = ao;

        OUT.illum_col.w = col.w;

        return OUT;
}