higan/extras/Shaders/NTSC-MAME.shader/mame-ntsc.fs

#version 150

// This is a port of the NTSC encode/decode shader pair in MAME and MESS, modified to use only
// one pass rather than an encode pass and a decode pass. It accurately emulates the sort of
// signal decimation one would see when viewing a composite signal, though it could benefit from a
// pre-pass to re-size the input content to more accurately reflect the actual size that would
// be incoming from a composite signal source.
//
// To encode the composite signal, I convert the RGB value to YIQ, then subsequently evaluate
// the standard NTSC composite equation. Four composite samples per RGB pixel are generated from
// the incoming linearly-interpolated texels.
//
// The decode pass implements a Fixed Impulse Response (FIR) filter designed by MAME/MESS contributor
// "austere" in matlab (if memory serves correctly) to mimic the behavior of a standard television set
// as closely as possible. The filter window is 83 composite samples wide, and there is an additional
// notch filter pass on the luminance (Y) values in order to strip the color signal from the luminance
// signal prior to processing.
//
// Yes, this code could greatly use some cleaning up.

// ported from UltraMoogleMan's "Full MAME/MESS Shader Pipe" shadertoy: https://www.shadertoy.com/view/ldf3Rf
// license: presumably MAME's license at the time, which was noncommercial

uniform sampler2D source[];
uniform vec4 sourceSize[];

in Vertex {
   vec2 vTexCoord;
};

out vec4 FragColor;

// Useful Constants
const vec4 Zero = vec4(0.0);
const vec4 Half = vec4(0.5);
const vec4 One = vec4(1.0);
const vec4 Two = vec4(2.0);
const vec3 Gray = vec3(0.3, 0.59, 0.11);
const float Pi = 3.1415926535;
const float Pi2 = 6.283185307;

// NTSC Constants
const vec4 A = vec4(0.5);
const vec4 A2 = vec4(1.0);
const vec4 B = vec4(0.5);
const float P = 1.0;
const float CCFrequency = 3.59754545;
const float NotchUpperFrequency = 5.59754545; //3.59754545 + 2.0;
const float NotchLowerFrequency = 1.59754545; //3.59754545 - 2.0;
const float YFrequency = 6.0;
const float IFrequency = 1.2;
const float QFrequency = 0.6;
const float NotchHalfWidth = 2.0;
const float ScanTime = 52.6;
const float Pi2ScanTime = 330.4955471482;// 6.283185307 * 52.6;
const float MaxC = 2.1183;
const vec4 YTransform = vec4(0.299, 0.587, 0.114, 0.0);
const vec4 ITransform = vec4(0.595716, -0.274453, -0.321263, 0.0);
const vec4 QTransform = vec4(0.211456, -0.522591, 0.311135, 0.0);
const vec3 YIQ2R = vec3(1.0, 0.956, 0.621);
const vec3 YIQ2G = vec3(1.0, -0.272, -0.647);
const vec3 YIQ2B = vec3(1.0, -1.106, 1.703);
const vec4 MinC = vec4(-1.1183);
const vec4 CRange = vec4(3.2366);
const vec4 InvCRange = vec4(1.0/3.2366);
const float Pi2Length = Pi2 / 63.0;
const vec4 NotchOffset = vec4(0.0, 1.0, 2.0, 3.0);
vec4 W = vec4(Pi2 * CCFrequency * ScanTime);

// Color Convolution Constants
const vec3 RedMatrix = vec3(1.0, 0.0, 0.0);
const vec3 GrnMatrix = vec3(0.0, 1.0, 0.0);
const vec3 BluMatrix = vec3(0.0, 0.0, 1.0);
const vec3 DCOffset = vec3(0.0, 0.0, 0.0);
const vec3 ColorScale = vec3(0.95, 0.95, 0.95);
const float Saturation = 1.4;

// Deconverge Constants
const vec3 ConvergeX = vec3(-0.4,  0.0, 0.2);
const vec3 ConvergeY = vec3( 0.0, -0.4, 0.2);
const vec3 RadialConvergeX = vec3(1.0, 1.0, 1.0);
const vec3 RadialConvergeY = vec3(1.0, 1.0, 1.0);

// Scanline/Pincushion Constants
const float PincushionAmount = 0.015;
const float CurvatureAmount = 0.015;
//const float ScanlineAmount = 0.175; <- move to parameter
const float ScanlineScale = 1.0;
const float ScanlineHeight = 1.0;
const float ScanlineBrightScale = 1.0;
const float ScanlineBrightOffset = 0.0;
const float ScanlineOffset = 0.0;
const vec3 Floor = vec3(0.05, 0.05, 0.05);

// 60Hz Bar Constants
const float SixtyHertzRate = (60.0 / 59.97 - 1.0); // Difference between NTSC and line frequency
const float SixtyHertzScale = 0.1;

vec4 CompositeSample(vec2 UV, vec2 InverseRes) {
	vec2 InverseP = vec2(P, 0.0) * InverseRes;
	
	// UVs for four linearly-interpolated samples spaced 0.25 texels apart
	vec2 C0 = UV;
	vec2 C1 = UV + InverseP * 0.25;
	vec2 C2 = UV + InverseP * 0.50;
	vec2 C3 = UV + InverseP * 0.75;
	vec4 Cx = vec4(C0.x, C1.x, C2.x, C3.x);
	vec4 Cy = vec4(C0.y, C1.y, C2.y, C3.y);

	vec4 Texel0 = texture(source[0], C0);
	vec4 Texel1 = texture(source[0], C1);
	vec4 Texel2 = texture(source[0], C2);
	vec4 Texel3 = texture(source[0], C3);
	
	float Frequency = CCFrequency;
	//Frequency = Frequency;// Uncomment for bad color sync + (sin(UV.y * 2.0 - 1.0) / CCFrequency) * 0.001;

	// Calculated the expected time of the sample.
	vec4 T = A2 * Cy * vec4(sourceSize[0].y) + B + Cx;
	vec4 W = vec4(Pi2ScanTime * Frequency);
	vec4 TW = T * W;
	vec4 Y = vec4(dot(Texel0, YTransform), dot(Texel1, YTransform), dot(Texel2, YTransform), dot(Texel3, YTransform));
	vec4 I = vec4(dot(Texel0, ITransform), dot(Texel1, ITransform), dot(Texel2, ITransform), dot(Texel3, ITransform));
	vec4 Q = vec4(dot(Texel0, QTransform), dot(Texel1, QTransform), dot(Texel2, QTransform), dot(Texel3, QTransform));

	vec4 Encoded = Y + I * cos(TW) + Q * sin(TW);
	return (Encoded - MinC) * InvCRange;
}

vec4 NTSCCodec(vec2 UV, vec2 InverseRes)
{
	vec4 YAccum = Zero;
	vec4 IAccum = Zero;
	vec4 QAccum = Zero;
	float QuadXSize = sourceSize[0].x * 4.0;
	float TimePerSample = ScanTime / QuadXSize;
	
	// Frequency cutoffs for the individual portions of the signal that we extract.
	// Y1 and Y2 are the positive and negative frequency limits of the notch filter on Y.
	// Y3 is the center of the frequency response of the Y filter.
	// I is the center of the frequency response of the I filter.
	// Q is the center of the frequency response of the Q filter.
	float Fc_y1 = NotchLowerFrequency * TimePerSample;
	float Fc_y2 = NotchUpperFrequency * TimePerSample;
	float Fc_y3 = YFrequency * TimePerSample;
	float Fc_i = IFrequency * TimePerSample;
	float Fc_q = QFrequency * TimePerSample;
	float Pi2Fc_y1 = Fc_y1 * Pi2;
	float Pi2Fc_y2 = Fc_y2 * Pi2;
	float Pi2Fc_y3 = Fc_y3 * Pi2;
	float Pi2Fc_i = Fc_i * Pi2;
	float Pi2Fc_q = Fc_q * Pi2;
	float Fc_y1_2 = Fc_y1 * 2.0;
	float Fc_y2_2 = Fc_y2 * 2.0;
	float Fc_y3_2 = Fc_y3 * 2.0;
	float Fc_i_2 = Fc_i * 2.0;
	float Fc_q_2 = Fc_q * 2.0;
	vec4 CoordY = vec4(UV.y);
	
	// 83 composite samples wide, 4 composite pixels per texel
	for(float n = -31.0; n < 32.0; n += 4.0)
	{
		vec4 n4 = n + NotchOffset;
		vec4 CoordX = UV.x + InverseRes.x * n4 * 0.25;
		vec2 TexCoord = vec2(CoordX.x, CoordY.x);
		vec4 C = CompositeSample(TexCoord, InverseRes) * CRange + MinC;
		vec4 WT = W * (CoordX  + A2 * CoordY * sourceSize[0].y + B);
		vec4 Cosine = 0.54 + 0.46 * cos(Pi2Length * n4);

		vec4 SincYIn1 = Pi2Fc_y1 * n4;
		vec4 SincYIn2 = Pi2Fc_y2 * n4;
		vec4 SincYIn3 = Pi2Fc_y3 * n4;
		vec4 SincY1 = sin(SincYIn1) / SincYIn1;
		vec4 SincY2 = sin(SincYIn2) / SincYIn2;
		vec4 SincY3 = sin(SincYIn3) / SincYIn3;
		
		// These zero-checks could be made more efficient if WebGL supported mix(vec4, vec4, bvec4)
		// Unfortunately, the universe hates us
		if(SincYIn1.x == 0.0) SincY1.x = 1.0;
		if(SincYIn1.y == 0.0) SincY1.y = 1.0;
		if(SincYIn1.z == 0.0) SincY1.z = 1.0;
		if(SincYIn1.w == 0.0) SincY1.w = 1.0;
		if(SincYIn2.x == 0.0) SincY2.x = 1.0;
		if(SincYIn2.y == 0.0) SincY2.y = 1.0;
		if(SincYIn2.z == 0.0) SincY2.z = 1.0;
		if(SincYIn2.w == 0.0) SincY2.w = 1.0;
		if(SincYIn3.x == 0.0) SincY3.x = 1.0;
		if(SincYIn3.y == 0.0) SincY3.y = 1.0;
		if(SincYIn3.z == 0.0) SincY3.z = 1.0;
		if(SincYIn3.w == 0.0) SincY3.w = 1.0;
		vec4 IdealY = (Fc_y1_2 * SincY1 - Fc_y2_2 * SincY2) + Fc_y3_2 * SincY3;
		vec4 FilterY = Cosine * IdealY;		
		
		vec4 SincIIn = Pi2Fc_i * n4;
		vec4 SincI = sin(SincIIn) / SincIIn;
		if (SincIIn.x == 0.0) SincI.x = 1.0;
		if (SincIIn.y == 0.0) SincI.y = 1.0;
		if (SincIIn.z == 0.0) SincI.z = 1.0;
		if (SincIIn.w == 0.0) SincI.w = 1.0;
		vec4 IdealI = Fc_i_2 * SincI;
		vec4 FilterI = Cosine * IdealI;
		
		vec4 SincQIn = Pi2Fc_q * n4;
		vec4 SincQ = sin(SincQIn) / SincQIn;
		if (SincQIn.x == 0.0) SincQ.x = 1.0;
		if (SincQIn.y == 0.0) SincQ.y = 1.0;
		if (SincQIn.z == 0.0) SincQ.z = 1.0;
		if (SincQIn.w == 0.0) SincQ.w = 1.0;
		vec4 IdealQ = Fc_q_2 * SincQ;
		vec4 FilterQ = Cosine * IdealQ;
		
		YAccum += C * FilterY;
		IAccum += C * cos(WT) * FilterI;
		QAccum += C * sin(WT) * FilterQ;
	}
	
	float Y = dot(YAccum, One);
	float I = dot(IAccum, One) * 2.0;
	float Q = dot(QAccum, One) * 2.0;
	
	vec3 YIQ = vec3(Y, I, Q);
	vec3 OutRGB = vec3(dot(YIQ, YIQ2R), dot(YIQ, YIQ2G), dot(YIQ, YIQ2B));
	
	return vec4(OutRGB, 1.0);
}

void main() {
   FragColor = vec4(NTSCCodec(vTexCoord, sourceSize[0].zw));
}