TiledFlowShader

//
// Fragment shader, Tiled Directional Flow
//
// (c) 2010 frans van hoesel, university of groningen
//
//
// this shader creates animated water by transforming normalmaps
// the scaling and rotation of the normalmaps is done per tile
// and is constant per tile. Each tile can have its own parameters
// for rotation, scaling, and speed of translation
// To hide the seams between the tiles, all seams have another tile
// centered over the seam. The opacity of the tiles decreases towards the
// edge of the tiles, so the edge isn't visible at all.
// Basically, all points have four tiles (A,B,C and D), mixed together
// (although at the edges the contribution of a tile to this mix is
// reduced to zero).
// The mixing of the tiles each with different parameters gives a nice
// animated look to the water. It is no longer just sliding in one direction, but
// appears to move more like real water.

// The resulting sum of normalmaps, is used to calculate the refraction of the clouds
// in a cube map and can also be used for other nice effects. In this example the
// colormap of the material under water is distorted to fake some kind of refraction
// (for this example the water is a bit too transparent, but it shows this refraction
// better)

// A flowmap determines in the red and green channel the normalized direction of the
// flow and in the blue channel wavelength.
// The alpha channel is used for the transparency of the water. Near the edge, the
// water becomes less deep and more transparent. Also near the edge, the waves tend
// to be smaller, so the same alpha channel also scales the height of the waves.
// Currently the wavelength is in its own channel (blue), but could be premultiplied
// to the red and green channels. This makes this channel available for changing the
// speed of the waves per tile.


// Further improvements
// Besides the obvious improvements mentioned in the code (such as premultiplying
// the direction of the waves with the scale, or moving the texscale multiplication
// to the texture coordinates), one could get rid of tiling in this code and pass it
// tiled geometry. This way the whole lookup of the flowmap (which is constant over
// each tile) could be moved to the vertexshader, removing the the construction of
// the flow rotation matrix. As this is done 4 times per pixel, it might give a big
// performance boost (one does need to pass on 4 constant matrices to the fragment
// shader, which will cost you a bit of performance).
//
//////////////////////////////////////////////////////////////////////////////////
//                     This software is Creditware:
//
// you can do whatever you want with this shader except claiming rights
// you may sell it, but you cannot prevent others from selling it, giving it away
// or use it as they please.
//
// Having said that, it would be nice if you gave me some credit for it, when you
// use it.
//
//                     Frans van Hoesel, (c) 2010
//////////////////////////////////////////////////////////////////////////////////


// movie at youtube: http://www.youtube.com/watch?v=TeSuNYvXAiA?hd=1 (in Germany this is blocked by youtube)
// making of at http://www.youtube.com/watch?v=wdcvPegJ1lw&hd=1 (works even in Germany)

// Thanks to Bart Campman, Pjotr Svetachov and Martijn Kragtwijk for their help.

Shader "Custom/TiledDirectionalFlow"
{
    Properties
    {
        _MainTex ("Base (RGB)", 2D) = "white" {}
        _FlowMap ("Flow", 2D) = "red" {}
        _WaterNormalMap("Water normal", 2D) = "blue" {}
        _SkyBox("SkyBox", CUBE) = "" {}
        _FlowSpeed("Flow speed", float) = 1.0
        _FlowTileScale("Flow tile scale", float) = 35.0
        _NormalTileScale("Normal tile scale", float) = 10.0
    }
    SubShader
    {
        Tags { "RenderType"="Opaque" }
        LOD 200

        CGPROGRAM
        #pragma surface surf Lambert
        #pragma target 3.0

        sampler2D _MainTex, _FlowMap, _WaterNormalMap;
        samplerCUBE _SkyBox;
        float _FlowSpeed, _FlowTileScale, _NormalTileScale;

        struct Input
        {
            float2 uv_MainTex;
            float2 uv_FlowMap;
            float2 uv_WaterNormalMap;
            float3 viewDir;
        };

        void surf (Input IN, inout SurfaceOutput o)
        {

            // texScale determines the amount of tiles generated.
            float texScale = _FlowTileScale;
            // texScale2 determines the repeat of the water texture (the normalmap) itself
            float texScale2 = _NormalTileScale;
            float myangle;
            float transp;
            float3 myNormal;

            float2 mytexFlowCoord = IN.uv_FlowMap * texScale;
            // ff is the factor that blends the tiles.
            float2 ff =  abs(2.0*(frac(mytexFlowCoord)) - 1.0) -0.5;
            // take a third power, to make the area with more or less equal contribution
            // of more tile bigger
            ff = 0.5-4.0*ff*ff*ff;
            // ffscale is a scaling factor that compensates for the effect that
            // adding normal vectors together tends to get them closer to the average normal
            // which is a visible effect. For more or less random waves, this factor
            // compensates for it
            float2 ffscale = sqrt(ff*ff + (1-ff)*(1-ff));
            float2 Tcoord = IN.uv_WaterNormalMap  * texScale2;

            // offset makes the water move
            float2 _offset = float2(_Time.x * _FlowSpeed,0);

            // I scale the texFlowCoord and floor the value to create the tiling
            // This could have be replace by an extremely lo-res texture lookup
            // using NEAREST pixel.
            float3 flow = tex2D(_FlowMap, floor(mytexFlowCoord)/ texScale).rgb;

            // flowdir is supposed to go from -1 to 1 and the line below
            // used to be sample.xy * 2.0 - 1.0, but saves a multiply by
            // moving this factor two to the sample.b
            float2 flowdir = flow.xy -0.5;

            // sample.b is used for the inverse length of the wave
            // could be premultiplied in sample.xy, but this is easier for editing flowtexture
            flowdir *= flow.b;

            // build the rotation matrix that scales and rotates the complete tile
            float2x2 rotmat = float2x2(flowdir.x, -flowdir.y, flowdir.y ,flowdir.x);

            // this is the normal for tile A
            float2 NormalT0 = tex2D(_WaterNormalMap, mul(rotmat, Tcoord) - _offset).rg;

            // for the next tile (B) I shift by half the tile size in the x-direction
            flow = tex2D( _FlowMap, floor((mytexFlowCoord + float2(0.5,0)))/ texScale ).rgb;

            flowdir = flow.b * (flow.xy - 0.5);
            rotmat = float2x2(flowdir.x, -flowdir.y, flowdir.y ,flowdir.x);
            // and the normal for tile B...
            // multiply the offset by some number close to 1 to give it a different speed
            // The result is that after blending the water starts to animate and look
            // realistic, instead of just sliding in some direction.
            // This is also why I took the third power of ff above, so the area where the
            // water animates is as big as possible
            // adding a small arbitrary constant isn't really needed, but helps to show
            // a bit less tiling in the beginning of the program. After a few seconds, the
            // tiling cannot be seen anymore so this constant could be removed.
            // For the quick demo I leave them in. In a simulation that keeps running for
            // some time, you could just as well remove these small constant offsets
            float2 NormalT1 = tex2D(_WaterNormalMap, mul(rotmat, Tcoord) - _offset*1.06+0.62).rg ;

            // blend them together using the ff factor
            // use ff.x because this tile is shifted in the x-direction
            float2 NormalTAB = ff.x * NormalT0 + (1.0-ff.x) * NormalT1;

            // the scaling of NormalTab and NormalTCD is moved to a single scale of
            // NormalT later in the program, which is mathematically identical to
            // NormalTAB = (NormalTAB - 0.5) / ffscale.x + 0.5;

            // tile C is shifted in the y-direction
            flow = tex2D( _FlowMap, floor((mytexFlowCoord + float2(0.0,0.5)))/ texScale ).rgb;

            flowdir = flow.b * (flow.xy - 0.5);
            rotmat = float2x2(flowdir.x, -flowdir.y, flowdir.y ,flowdir.x);
            NormalT0 = tex2D(_WaterNormalMap, mul(rotmat, Tcoord) - _offset*1.33+0.27).rg;

            // tile D is shifted in both x- and y-direction
            flow = tex2D( _FlowMap, floor((mytexFlowCoord + float2(0.5,0.5)))/ texScale ).rgb;

            flowdir = flow.b * (flow.xy - 0.5);
            rotmat = float2x2(flowdir.x, -flowdir.y, flowdir.y ,flowdir.x);
            NormalT1 = tex2D(_WaterNormalMap, mul(rotmat, Tcoord) - _offset*1.24).rg ;

            float2 NormalTCD = ff.x * NormalT0 + (1.0-ff.x) * NormalT1;
            // NormalTCD = (NormalTCD - 0.5) / ffscale.x + 0.5;

            // now blend the two values together
            float2 NormalT = ff.y * NormalTAB + (1.0-ff.y) * NormalTCD;

            // this line below used to be here for scaling the result
            //NormalT = (NormalT - 0.5) / ffscale.y + 0.5;

            // below the new, direct scaling of NormalT
            NormalT = (NormalT - 0.5) / (ffscale.y * ffscale.x);
            // scaling by 0.3 is arbritrary, and could be done by just
            // changing the values in the normal map
            // without this factor, the waves look very strong
            NormalT *= 0.3;
            // to make the water more transparent
            transp = tex2D( _FlowMap, IN.uv_FlowMap ).a;
            // and scale the normals with the transparency
            NormalT *= transp*transp;

            // assume normal of plane is 0,0,1 and produce the normalized sum of adding NormalT to it
            myNormal = float3(NormalT,sqrt(1.0-NormalT.x*NormalT.x - NormalT.y*NormalT.y));

            float3 reflectDir = reflect(IN.viewDir, myNormal);
            float3 envColor = texCUBE(_SkyBox, reflectDir).rgb;

            // very ugly version of fresnel effect
            // but it gives a nice transparent water, but not too transparent
            myangle = dot(myNormal,normalize(IN.viewDir));
            myangle = 0.95-0.6*myangle*myangle;

            // blend in the color of the plane below the water

            // add in a little distortion of the colormap for the effect of a refracted
            // view of the image below the surface.
            // (this isn't really tested, just a last minute addition
            // and perhaps should be coded differently

            // the correct way, would be to use the refract routine, use the alpha channel for depth of
            // the water (and make the water disappear when depth = 0), add some watercolor to the colormap
            // depending on the depth, and use the calculated refractdir and the depth to find the right
            // pixel in the colormap.... who knows, something for the next version
            float4 base = tex2D(_MainTex, IN.uv_MainTex + myNormal.xy/texScale2*0.03*transp);

            base = float4(lerp(base.rgb,envColor,myangle*transp),1.0 );

            // note that smaller waves appear to move slower than bigger waves
            // one could use the tiles and give each tile a different speed if that
            // is what you want

            o.Albedo = base.rgb;
            o.Alpha = base.a;




        }
        ENDCG
    }
    FallBack "Diffuse"
}