source: GTP/trunk/App/Demos/Vis/FriendlyCulling/src/shaders/ssao.cg @ 3115

#include "../shaderenv.h"

////////////////////
// Screen Spaced Ambient Occlusion shader
// based on shader of Alexander Kusternig

#define USE_EYESPACE_DEPTH 1
//#extension GL_EXT_gpu_shader4 : enable




struct fragment
{
        float2 texCoord: TEXCOORD0; 
        float3 view: TEXCOORD1;
};


struct pixel
{
        float4 illum_col: COLOR0;
};


inline float occlusionPower(float radius, float dist)
{
        return 6.283185307179586476925286766559f * (1.0f - cos(asin(radius / dist)));
}


inline 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;
}


inline float3 Interpol(float2 w, float3 bl, float3 br, float3 tl, float3 tr)
{
        float3 x1 = lerp(bl, tl, w.y);
        float3 x2 = lerp(br, tr, w.y); 
        float3 v = lerp(x1, x2, w.x); 

        return v;
}


// reconstruct world space position
inline float3 ReconstructSamplePos(uniform sampler2D tex,
                                                                   float2 texcoord, 
                                                                   float3 bl, float3 br, float3 tl, float3 tr)
{
        const float eyeSpaceDepth = tex2Dlod(tex, float4(texcoord, 0, 0)).w;
        //float3 viewVec = normalize(Interpol(texcoord, bl, br, tl, tr));
        float3 viewVec = Interpol(texcoord, bl, br, tl, tr);
        float3 samplePos = -viewVec * eyeSpaceDepth;

        return samplePos;
}


inline float ComputeDifference(float2 offset,
                                                           sampler2D oldTex,
                                                           float4x4 oldModelViewProj,
                                                           sampler2D colors,
                                                           sampler2D noiseTex,
                                                           float scaleFactor,
                                                           float3 bl,
                                                           float3 br,
                                                           float3 tl,
                                                           float3 tr, 
                                                           float2 texcoord0,
                                                           float3 oldEyePos,
                                                           float3 oldbl,
                                                           float3 oldbr,
                                                           float3 oldtl,
                                                           float3 oldtr,
                                                           float eyeSpaceDepth
                                                           )
{
        const float2 mynoise = tex2Dlod(noiseTex, float4(texcoord0, 0, 0)).xy;

        const float2 offsetTransformed = myreflect(offset, mynoise);
        float2 texCoord = texcoord0 + offsetTransformed * scaleFactor;
        
        const float sampleEyeSpaceDepth = tex2Dlod(colors, float4(texCoord, 0, 0)).w;
        
        const float3 viewVec = Interpol(texCoord, bl, br, tl, tr);
        const float3 samplePos = -viewVec * sampleEyeSpaceDepth;
        const float3 translatedPos = samplePos - oldEyePos;

        // reproject into old frame and calculate projected depth
        float4 projPos = mul(oldModelViewProj, float4(translatedPos, 1.0f));
        projPos /= projPos.w;
        // fit from unit cube into 0 .. 1
        const float2 oldTexCoords = projPos.xy * 0.5f + 0.5f;
        // retrieve the sample from the last frame
        const float4 oldPixel = tex2Dlod(oldTex, float4(oldTexCoords, .0f, .0f));

        const float oldEyeSpaceDepth = oldPixel.w;
        
        // projected linear depth
        const float3 oldViewVec = Interpol(oldTexCoords, oldbl, oldbr, oldtl, oldtr);
        const float invlen = 1.0f / length(oldViewVec);
        const float projectedEyeSpaceDepth = invlen * length(translatedPos);

        float depthDif = (abs(eyeSpaceDepth - sampleEyeSpaceDepth) > 5.0f) ? 0 : 
                abs(1.0f - oldEyeSpaceDepth / projectedEyeSpaceDepth);

        return depthDif;
}


/** This shader computes the reprojection and stores 
        reprojected color / depth values as well as a boolean that 
*/
inline float4 temporalSmoothing(float4 worldPos,
                                                                float eyeSpaceDepth,
                                                                float2 ao,
                                                                float2 texcoord0,
                                                                float3 oldEyePos,
                                                                sampler2D oldTex,
                                                                float4x4 oldModelViewProj,
                                                                float temporalCoherence,
                                                                sampler2D colors,
                                                                float3 bl,
                                                                float3 br,
                                                                float3 tl,
                                                                float3 tr, 
                                                                float3 projPos,
                                                                float invW,
                                                                sampler2D noiseTex,
                                                                float2 samples[NUM_SAMPLES],
                                                                float scaleFactor,
                                                                float3 oldbl,
                                                                float3 oldbr,
                                                                float3 oldtl,
                                                                float3 oldtr,
                                                                float3 diffVec
                                                                )
{
        float4 illum_col;

        // compute position from old frame for dynamic objects + translational portion
        const float3 translatedPos = worldPos.xyz - oldEyePos + diffVec;


        /////////////////
        //-- reproject into old frame and calculate texture position of sample in old frame

        // note: the old model view matrix only holds the view orientation part
        float4 backProjPos = mul(oldModelViewProj, float4(translatedPos, 1.0f));
        backProjPos /= backProjPos.w;
        
        // fit from unit cube into 0 .. 1
        const float2 oldTexCoords = backProjPos.xy * 0.5f + 0.5f;
        // retrieve the sample from the last frame
        const float4 oldPixel = tex2Dlod(oldTex, float4(oldTexCoords, .0f, .0f));


#if USE_EYESPACE_DEPTH

        // calculate eye space position of sample in old frame
        const float oldEyeSpaceDepth = oldPixel.w;

        // vector from eye pos to old sample 
        const float3 viewVec = Interpol(oldTexCoords, oldbl, oldbr, oldtl, oldtr);
        const float invLen = 1.0f / length(viewVec);
        const float projectedEyeSpaceDepth = invLen * length(translatedPos);
        
        const float depthDif = abs(1.0f - oldEyeSpaceDepth / projectedEyeSpaceDepth);

#else
        // calculate eye space position of sample in old frame
        const float oldDepth = oldPixel.w;
        const float projectedDepth = projPos.z;
        // vector from eye pos to old sample 
        const float depthDif = abs(projectedDepth - oldDepth);
#endif

        float notValid = 0.5f;
        //float overallDepth = 0;
        const float squaredLen = diffVec.x * diffVec.x + diffVec.y * diffVec.y + diffVec.z * diffVec.z;

        if (squaredLen < 1e-8f) // object not dynamic
        {
                for (int i = 0; i < NUM_SAMPLES; ++ i) 
                {
                        float sampleDif = ComputeDifference(samples[i],
                                                                                                oldTex,
                                                                                                oldModelViewProj,
                                                                                                colors,
                                                                                                noiseTex,
                                                                                                scaleFactor,
                                                                                                bl, br, tl, tr, 
                                                                                                texcoord0,
                                                                                                oldEyePos,
                                                                                                oldbl, oldbr, oldtl, oldtr,
                                                                                                eyeSpaceDepth
                                                                                                );
                        //overallDepth += sampleDif;

                        if (sampleDif >= MIN_DEPTH_DIFF) ++ notValid;
                }
        }


        // the number of valid samples in this frame
        //const float newNumSamples = ao.y;
        //const float oldNumSamples = oldCol.y;

        const float oldWeight = clamp(oldPixel.y, .0f, temporalCoherence);
        float newWeight;

        if ((temporalCoherence > 1e-6f)
                && (oldTexCoords.x >= 0.0f) && (oldTexCoords.x < 1.0f)
                && (oldTexCoords.y >= 0.0f) && (oldTexCoords.y < 1.0f)
                && (depthDif <= MIN_DEPTH_DIFF) 
                // if visibility changed in the surrounding area we have to recompute
                //&& (oldNumSamples > 0.8f * newNumSamples)
                && (notValid < 1.0f)
                )
        {
                // increase the weight for convergence
                newWeight = oldWeight + 1.0f;
                illum_col.x = (ao.x + oldPixel.x * oldWeight) / newWeight;
                //if (notValid > 1.0f) newWeight = 2.0f;
        }
        else
        {       
                illum_col.x = ao.x;
                newWeight = .0f;
        }

        illum_col.y = newWeight;

#if USE_EYESPACE_DEPTH
        illum_col.w = eyeSpaceDepth;
#else
        illum_col.w = projectedDepth;
#endif
        //overallDepth = clamp(overallDepth*1e3, 0, 1);
        illum_col.z = invW;
        //illum_col.x = overallDepth;

        return illum_col;
}


/** The ssao shader returning the an intensity value between 0 and 1
*/
float2 ssao(fragment IN,
                        uniform sampler2D colors,
                        uniform sampler2D noiseTex,
                        uniform float2 samples[NUM_SAMPLES],
                        uniform float3 normal,
                        uniform float3 centerPosition,
                        uniform float scaleFactor,
                        uniform float3 bl,
                        uniform float3 br,
                        uniform float3 tl,
                        uniform float3 tr, 
                        uniform float3 viewDir
                        )
{
        // 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.

        float total_ao = .0f;
        float numSamples = .0f;


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

#if 1
                ////////////////////
                //-- add random noise: reflect around random normal vector (rather slow!)

                float2 mynoise = tex2Dlod(noiseTex, float4(IN.texCoord * 4.0f, 0, 0)).xy;
                const float2 offsetTransformed = myreflect(offset, mynoise);
#else
                const float2 offsetTransformed = offset;
#endif
                // weight with projected coordinate to reach similar kernel size for near and far
                const float2 texcoord = IN.texCoord.xy + offsetTransformed * scaleFactor;

                //if ((texcoord.x <= 1.0f) && (texcoord.x >= 0.0f) && (texcoord.y <= 1.0f) && (texcoord.y >= 0.0f)) ++ numSamples;
                const float3 samplePos = ReconstructSamplePos(colors, texcoord, bl, br, tl, tr);


                ////////////////
                //-- compute contribution of sample using the direction and angle

                float3 dirSample = samplePos - centerPosition;
                const float lengthToSample = max(length(dirSample), 1e-6f);

                dirSample /= lengthToSample; // normalize

                // angle between current normal and direction to sample controls AO intensity.
                float cosAngle = max(dot(dirSample, normal), .0f);
        
                // the distance_scale offset is used to avoid singularity that occurs at global illumination when 
                // the distance to a sample approaches zero
                const float aoContrib = SAMPLE_INTENSITY / (DISTANCE_SCALE + lengthToSample * lengthToSample);
                //const float aoContrib = (1.0f > lengthToSample) ? occlusionPower(9e-2f, DISTANCE_SCALE + lengthToSample): .0f;

#if 1
                // if surface normal perpenticular to view dir, approx. half of the samples will not count
                // => compensate for this (on the other hand, projected sampling area could be larger!)

                const float viewCorrection = 1.0f + VIEW_CORRECTION_SCALE * max(dot(viewDir, normal), 0.0f);
                total_ao += cosAngle * aoContrib * viewCorrection;
#else
                total_ao += cosAngle * aoContrib;
#endif
        }

        return float2(max(0.0f, 1.0f - total_ao), numSamples);
}

/** The mrt shader for screen space ambient occlusion
*/
pixel main(fragment IN, 
                   uniform sampler2D colors,
                   uniform sampler2D normals,
                   uniform sampler2D noiseTex,
                   uniform float2 samples[NUM_SAMPLES],
                   uniform sampler2D oldTex,
                   uniform float4x4 modelViewProj,
                   uniform float4x4 oldModelViewProj,
                   uniform float temporalCoherence,
                   uniform float3 bl,
                   uniform float3 br,
                   uniform float3 tl,
                   uniform float3 tr,
                   uniform float3 oldEyePos,
                   uniform float3 oldbl,
                   uniform float3 oldbr,
                   uniform float3 oldtl,
                   uniform float3 oldtr,
                   uniform sampler2D attribsTex
                   )
{
        pixel OUT;

        const float3 normal = normalize(tex2Dlod(normals, float4(IN.texCoord, 0 ,0)).xyz);

        // reconstruct position from the eye space depth
        //const float3 viewDir = normalize(IN.view);
        const float3 viewDir = IN.view;
        const float eyeSpaceDepth = tex2Dlod(colors, float4(IN.texCoord, 0, 0)).w;
        const float4 eyeSpacePos = float4(-viewDir * eyeSpaceDepth, 1.0f);


        ////////////////
        //-- calculcate the current projected posiion (also used for next frame)
        
        float4 projPos = mul(modelViewProj, eyeSpacePos);
        const float invw = 1.0f / projPos.w;
        projPos *= invw;
        float scaleFactor = SAMPLE_RADIUS * invw;
        
        float2 ao = float2(1.0f, 0);
        // note: this should be done with the stencil buffer
        if (eyeSpaceDepth < 1e10f)
        {
                ao = ssao(IN, colors, noiseTex, samples, normal, 
                          eyeSpacePos.xyz, scaleFactor, bl, br, tl, tr, normalize(viewDir));
        }

        //float3 id = tex2Dlod(attribsTex, float4(IN.texCoord, 0, 0)).xyz;
        float3 diffVec = tex2Dlod(attribsTex, float4(IN.texCoord, 0, 0)).xyz;

        /////////////////
        //-- compute temporal reprojection

        OUT.illum_col = temporalSmoothing(eyeSpacePos, eyeSpaceDepth, ao, IN.texCoord, oldEyePos,
                                              oldTex, oldModelViewProj, temporalCoherence,
                                              colors, 
                                                                          bl, br, tl, tr, 
                                                                          projPos.xyz, 
                                                                          invw, 
                                                                          noiseTex, 
                                                                          samples, 
                                                                          scaleFactor, 
                                                                          oldbl, oldbr, oldtl, oldtr,
                                                                          diffVec);

        //OUT.illum_col.xyz = id;
        //OUT.illum_col.xyz = normal * 0.5f + 0.5f;
        return OUT;
}


float Filter(float2 texCoord, 
                         uniform sampler2D ssaoTex,
                         uniform float2 filterOffs[NUM_SSAO_FILTERSAMPLES],
                         uniform float filterWeights[NUM_SSAO_FILTERSAMPLES],
                         float scale)
{
        float average = .0f;
        float w = .0f;

        for (int i = 0; i < NUM_SSAO_FILTERSAMPLES; ++ i)
        {       
                average += filterWeights[i] * tex2Dlod(ssaoTex, float4(texCoord + filterOffs[i] * scale * 2, 0, 0)).x;
                w += filterWeights[i];
        }

        average *= 1.0f / (float)w;

        return average;
}


pixel combine(fragment IN, 
                          uniform sampler2D colorsTex,
                          uniform sampler2D ssaoTex,
                          uniform float2 filterOffs[NUM_SSAO_FILTERSAMPLES],
                          uniform float filterWeights[NUM_SSAO_FILTERSAMPLES]
                          )
{
        pixel OUT;

        const float4 col = tex2Dlod(colorsTex, float4(IN.texCoord, 0, 0));
        float4 ao = tex2Dlod(ssaoTex, float4(IN.texCoord, 0, 0));

        //if ((ao.y < 10.0f) && (col.w < 1e10f))
        //      ao.x = Filter(IN.texCoord, ssaoTex, filterOffs, filterWeights, 1.0f / (1.0f + ao.y));//ao.z);

        OUT.illum_col = col * ao.x;
        //OUT.illum_col = float4(ao.y, ao.y, ao.y, col.w);
        //OUT.illum_col = float4(ao.x, ao.y, ao.z, col.w);
        //OUT.illum_col = float4(ao.x, 0, 0, col.w);
        //OUT.illum_col.xyz = float3(1.0f - ao.x, 1.0f - ao.y * 1e-2f, 1);
        //OUT.illum_col.xyz = float3(1.0f - ao.x, ao.y, 0);
        OUT.illum_col.w = col.w;

        return OUT;
}