CG shader for NES CRT emulation

// CG shader
// NES CRT simulation
// by r57shell
// thanks to feos & HardWareMan

void main_vertex
(
    float4 position : POSITION,
    out float4 oPosition : POSITION,
    uniform float4x4 modelViewProj,

    float2 tex : TEXCOORD,
    out float2 oTex : TEXCOORD
)
{
    oPosition = mul(modelViewProj, position);
    oTex = tex;
}

struct input
{
    float2 video_size;
    float2 texture_size;
    float2 output_size;
    float  frame_count;
    float  frame_direction;
    float frame_rotation;
};

const float pi = 3.142;

vec3 monitor(uniform sampler2D tex : TEXUNIT0, vec2 p, uniform input IN)
{
    float2 size = IN.texture_size;
    vec2 pos = floor(p*size);
    vec2 uv = floor(pos)/size;
    vec4 res = tex2D(tex, uv);
    vec3 yuv = mul(float3x3(
        0.2126, 0.7152, 0.0722,
        -0.09991, -0.33609, 0.436,
        0.615, -0.55861, -0.05639), res.xyz);
    float alpha = (floor(p.x*size.x*4.)/2.0)*pi;
    vec2 sincv = vec2(cos(alpha), sin(alpha));
    if (mod(pos.y + 5.,4.) < 2.)
        sincv.x = -sincv.x;
    if (mod(pos.y, 4.) >= 2.)
        sincv.y = -sincv.y;
    float mc = 1.+dot(sincv, yuv.zy)/yuv.x;

    /*vec3 rgb = vec3(
        yuv.x + 1.28033 * yuv.z,
        yuv.x - 0.21482 * yuv.y - 0.38059 * yuv.z,
        yuv.x + 2.12798 * yuv.y);*/
    return res.xyz*mc;
}

// pos (left corner, sample size)
vec4 monitor_sample(uniform sampler2D tex : TEXUNIT0, vec2 p, vec2 sample, uniform input IN)
{
    // linear interpolation was...
    // now other thing.
    // http://imgur.com/m8Z8trV
    // AT LAST IT WORKS!!!!
    // going to check in retroarch...
    float2 size = IN.texture_size;
    vec2 next = vec2(.25,1.)/size;
    vec2 f = fract(vec2(4.,1.)*size*p);
    sample *= vec2(4.,1.)*size;
    vec2 l;
    vec2 r;
    if (f.x+sample.x < 1.)
    {
        l.x = f.x+sample.x;
        r.x = 0.;
    }
    else
    {
        l.x = 1.-f.x;
        r.x = min(1.,f.x+sample.x-1.);
    }
    if (f.y+sample.y < 1.)
    {
        l.y = f.y+sample.y;
        r.y = 0.;
    }
    else
    {
        l.y = 1.-f.y;
        r.y = min(1.,f.y+sample.y-1.);
    }
    vec3 top = mix(monitor(tex, p, IN), monitor(tex, p+vec2(next.x,0.), IN), r.x/(l.x+r.x));
    vec3 bottom = mix(monitor(tex, p+vec2(0.,next.y), IN), monitor(tex, p+next, IN), r.x/(l.x+r.x));
    return vec4(mix(top,bottom, r.y/(l.y+r.y)),1.0);
}

float4 main_fragment(uniform sampler2D tex : TEXUNIT0, float2 coords : TEXCOORD0, uniform input IN) : COLOR
{
    return monitor_sample(tex, coords, 1./IN.output_size, IN);

    // difference
    //float zoom = 8;
    //vec4 sampled = monitor_sample(tex, coords, 1./zoom/IN.output_size, IN);
    //vec4 simple = vec4(monitor(tex, coords, IN),1.);
    //return vec4(length(sampled - simple)*5*sin(IN.frame_count/30.))+simple;
}