1#version 450
2
3// This is a port of the NTSC encode/decode shader pair in MAME and MESS, modified to use only
4// one pass rather than an encode pass and a decode pass. It accurately emulates the sort of
5// signal decimation one would see when viewing a composite signal, though it could benefit from a
6// pre-pass to re-size the input content to more accurately reflect the actual size that would
7// be incoming from a composite signal source.
8//
9// To encode the composite signal, I convert the RGB value to YIQ, then subsequently evaluate
10// the standard NTSC composite equation. Four composite samples per RGB pixel are generated from
11// the incoming linearly-interpolated texels.
12//
13// The decode pass implements a Fixed Impulse Response (FIR) filter designed by MAME/MESS contributor
14// "austere" in matlab (if memory serves correctly) to mimic the behavior of a standard television set
15// as closely as possible. The filter window is 83 composite samples wide, and there is an additional
16// notch filter pass on the luminance (Y) values in order to strip the color signal from the luminance
17// signal prior to processing.
18//
19// - UltraMoogleMan [8/2/2013]
20
21layout(push_constant) uniform Push
22{
23	vec4 SourceSize;
24	vec4 OriginalSize;
25	vec4 OutputSize;
26	uint FrameCount;
27} params;
28
29layout(std140, set = 0, binding = 0) uniform UBO
30{
31	mat4 MVP;
32} global;
33
34// Useful Constants
35const vec4 Zero = vec4(0.0);
36const vec4 Half = vec4(0.5);
37const vec4 One = vec4(1.0);
38const vec4 Two = vec4(2.0);
39const float Pi = 3.1415926535;
40const float Pi2 = 6.283185307;
41
42// NTSC Constants
43const vec4 A = vec4(0.5);
44const vec4 B = vec4(0.5);
45const float P = 1.0;
46const float CCFrequency = 3.59754545;
47const float YFrequency = 6.0;
48const float IFrequency = 1.2;
49const float QFrequency = 0.6;
50const float NotchHalfWidth = 2.0;
51const float ScanTime = 52.6;
52const float MaxC = 2.1183;
53const vec4 MinC = vec4(-1.1183);
54const vec4 CRange = vec4(3.2366);
55
56#pragma stage vertex
57layout(location = 0) in vec4 Position;
58layout(location = 1) in vec2 TexCoord;
59layout(location = 0) out vec2 vTexCoord;
60
61void main()
62{
63   gl_Position = global.MVP * Position;
64   vTexCoord = TexCoord;
65}
66
67#pragma stage fragment
68layout(location = 0) in vec2 vTexCoord;
69layout(location = 0) out vec4 FragColor;
70layout(set = 0, binding = 2) uniform sampler2D Source;
71
72vec4 CompositeSample(vec2 UV) {
73	vec2 InverseRes = params.SourceSize.zw;
74	vec2 InverseP = vec2(P, 0.0) * InverseRes;
75
76	// UVs for four linearly-interpolated samples spaced 0.25 texels apart
77	vec2 C0 = UV;
78	vec2 C1 = UV + InverseP * 0.25;
79	vec2 C2 = UV + InverseP * 0.50;
80	vec2 C3 = UV + InverseP * 0.75;
81	vec4 Cx = vec4(C0.x, C1.x, C2.x, C3.x);
82	vec4 Cy = vec4(C0.y, C1.y, C2.y, C3.y);
83
84	vec3 Texel0 = texture(Source, C0).rgb;
85	vec3 Texel1 = texture(Source, C1).rgb;
86	vec3 Texel2 = texture(Source, C2).rgb;
87	vec3 Texel3 = texture(Source, C3).rgb;
88
89	// Calculated the expected time of the sample.
90	vec4 T = A * Cy * vec4(params.SourceSize.x) * Two + B + Cx;
91
92	const vec3 YTransform = vec3(0.299, 0.587, 0.114);
93	const vec3 ITransform = vec3(0.595716, -0.274453, -0.321263);
94	const vec3 QTransform = vec3(0.211456, -0.522591, 0.311135);
95
96	float Y0 = dot(Texel0, YTransform);
97	float Y1 = dot(Texel1, YTransform);
98	float Y2 = dot(Texel2, YTransform);
99	float Y3 = dot(Texel3, YTransform);
100	vec4 Y = vec4(Y0, Y1, Y2, Y3);
101
102	float I0 = dot(Texel0, ITransform);
103	float I1 = dot(Texel1, ITransform);
104	float I2 = dot(Texel2, ITransform);
105	float I3 = dot(Texel3, ITransform);
106	vec4 I = vec4(I0, I1, I2, I3);
107
108	float Q0 = dot(Texel0, QTransform);
109	float Q1 = dot(Texel1, QTransform);
110	float Q2 = dot(Texel2, QTransform);
111	float Q3 = dot(Texel3, QTransform);
112	vec4 Q = vec4(Q0, Q1, Q2, Q3);
113
114	vec4 W = vec4(Pi2 * CCFrequency * ScanTime);
115	vec4 Encoded = Y + I * cos(T * W) + Q * sin(T * W);
116	return (Encoded - MinC) / CRange;
117}
118
119vec4 NTSCCodec(vec2 UV)
120{
121	vec2 InverseRes = params.SourceSize.zw;
122	vec4 YAccum = Zero;
123	vec4 IAccum = Zero;
124	vec4 QAccum = Zero;
125	float QuadXSize = params.SourceSize.x * 4.0;
126	float TimePerSample = ScanTime / QuadXSize;
127
128	// Frequency cutoffs for the individual portions of the signal that we extract.
129	// Y1 and Y2 are the positive and negative frequency limits of the notch filter on Y.
130	//
131	float Fc_y1 = (CCFrequency - NotchHalfWidth) * TimePerSample;
132	float Fc_y2 = (CCFrequency + NotchHalfWidth) * TimePerSample;
133	float Fc_y3 = YFrequency * TimePerSample;
134	float Fc_i = IFrequency * TimePerSample;
135	float Fc_q = QFrequency * TimePerSample;
136	float Pi2Length = Pi2 / 82.0;
137	vec4 NotchOffset = vec4(0.0, 1.0, 2.0, 3.0);
138	vec4 W = vec4(Pi2 * CCFrequency * ScanTime);
139	for(float n = -41.0; n < 42.0; n += 4.0)
140	{
141		vec4 n4 = n + NotchOffset;
142		vec4 CoordX = UV.x + InverseRes.x * n4 * 0.25;
143		vec4 CoordY = vec4(UV.y);
144		vec2 TexCoord = vec2(CoordX.r, CoordY.r);
145		vec4 C = CompositeSample(TexCoord) * CRange + MinC;
146		vec4 WT = W * (CoordX  + A * CoordY * Two * params.SourceSize.x + B);
147
148		vec4 SincYIn1 = Pi2 * Fc_y1 * n4;
149		vec4 SincYIn2 = Pi2 * Fc_y2 * n4;
150		vec4 SincYIn3 = Pi2 * Fc_y3 * n4;
151		bvec4 notEqual = notEqual(SincYIn1, Zero);
152		vec4 SincY1 = sin(SincYIn1) / SincYIn1;
153		vec4 SincY2 = sin(SincYIn2) / SincYIn2;
154		vec4 SincY3 = sin(SincYIn3) / SincYIn3;
155		if(SincYIn1.x == 0.0) SincY1.x = 1.0;
156		if(SincYIn1.y == 0.0) SincY1.y = 1.0;
157		if(SincYIn1.z == 0.0) SincY1.z = 1.0;
158		if(SincYIn1.w == 0.0) SincY1.w = 1.0;
159		if(SincYIn2.x == 0.0) SincY2.x = 1.0;
160		if(SincYIn2.y == 0.0) SincY2.y = 1.0;
161		if(SincYIn2.z == 0.0) SincY2.z = 1.0;
162		if(SincYIn2.w == 0.0) SincY2.w = 1.0;
163		if(SincYIn3.x == 0.0) SincY3.x = 1.0;
164		if(SincYIn3.y == 0.0) SincY3.y = 1.0;
165		if(SincYIn3.z == 0.0) SincY3.z = 1.0;
166		if(SincYIn3.w == 0.0) SincY3.w = 1.0;
167		//vec4 IdealY = (2.0 * Fc_y1 * SincY1 - 2.0 * Fc_y2 * SincY2) + 2.0 * Fc_y3 * SincY3;
168		vec4 IdealY = (2.0 * Fc_y1 * SincY1 - 2.0 * Fc_y2 * SincY2) + 2.0 * Fc_y3 * SincY3;
169		vec4 FilterY = (0.54 + 0.46 * cos(Pi2Length * n4)) * IdealY;
170
171		vec4 SincIIn = Pi2 * Fc_i * n4;
172		vec4 SincI = sin(SincIIn) / SincIIn;
173		if (SincIIn.x == 0.0) SincI.x = 1.0;
174		if (SincIIn.y == 0.0) SincI.y = 1.0;
175		if (SincIIn.z == 0.0) SincI.z = 1.0;
176		if (SincIIn.w == 0.0) SincI.w = 1.0;
177		vec4 IdealI = 2.0 * Fc_i * SincI;
178		vec4 FilterI = (0.54 + 0.46 * cos(Pi2Length * n4)) * IdealI;
179
180		vec4 SincQIn = Pi2 * Fc_q * n4;
181		vec4 SincQ = sin(SincQIn) / SincQIn;
182		if (SincQIn.x == 0.0) SincQ.x = 1.0;
183		if (SincQIn.y == 0.0) SincQ.y = 1.0;
184		if (SincQIn.z == 0.0) SincQ.z = 1.0;
185		if (SincQIn.w == 0.0) SincQ.w = 1.0;
186		vec4 IdealQ = 2.0 * Fc_q * SincQ;
187		vec4 FilterQ = (0.54 + 0.46 * cos(Pi2Length * n4)) * IdealQ;
188
189		YAccum = YAccum + C * FilterY;
190		IAccum = IAccum + C * cos(WT) * FilterI;
191		QAccum = QAccum + C * sin(WT) * FilterQ;
192	}
193
194	float Y = YAccum.r + YAccum.g + YAccum.b + YAccum.a;
195	float I = (IAccum.r + IAccum.g + IAccum.b + IAccum.a) * 2.0;
196	float Q = (QAccum.r + QAccum.g + QAccum.b + QAccum.a) * 2.0;
197
198	vec3 YIQ = vec3(Y, I, Q);
199
200	vec3 OutRGB = vec3(dot(YIQ, vec3(1.0, 0.956, 0.621)), dot(YIQ, vec3(1.0, -0.272, -0.647)), dot(YIQ, vec3(1.0, -1.106, 1.703)));
201
202	return vec4(OutRGB, 1.0);
203}
204
205void main()
206{
207	vec4 OutPixel = NTSCCodec(vTexCoord);
208   FragColor = vec4(OutPixel);
209}