Merge pull request #16 from rz5/master

(pal) Port @svofski's CRT shader to libretro.

pal/pal-singlepass.slang

@@ -0,0 +1,197 @@
+#version 450
+
+/*
+    pal-singlepass.slang
+
+    svo's PAL single pass shader, ported to libretro
+    ------------------------------------------------
+    "Software composite video modulation/demodulation experiments
+
+    The idea is to reproduce in GLSL shaders realistic composite-like
+    artifacting by applying PAL modulation and demodulation.
+    
+    Digital texture, passed through the model of an analog channel,
+    should suffer same effects as its analog counterpart and exhibit properties,
+    such as dot crawl and colour bleeding, that may be desirable for faithful
+    reproduction of look and feel of old computer games."
+    
+    https://github.com/svofski/CRT
+
+    Copyright (C) 2015 Viacheslav Slavinsky (svo)
+    
+*/
+
+layout(std140, set = 0, binding = 0) uniform UBO
+{
+    mat4 MVP;
+    vec4 OutputSize;
+    vec4 OriginalSize;
+    vec4 SourceSize;
+    uint FrameCount;
+} global;
+
+#pragma stage vertex
+layout(location = 0) in  vec4 Position;
+layout(location = 1) in  vec2 TexCoord;
+layout(location = 0) out vec2 vTexCoord;
+
+void main()
+{
+    gl_Position = global.MVP * Position;
+    vTexCoord   = TexCoord;
+}
+
+#pragma stage fragment
+layout(location = 0) in  vec2 vTexCoord;
+layout(location = 0) out vec4 FragColor;
+layout(set = 0, binding = 2) uniform sampler2D Source;
+
+
+/* Config options */
+
+/* FIR lowpass gain  */
+#define FIR_GAIN     1.5
+
+/* Inverse gain for luma recovery (correct for stripes) */
+#define FIR_INVGAIN  1.1
+
+/* phase noise (2.5 for crappy cable) */
+#define PHASE_NOISE  1.0
+
+
+
+
+/* Subcarrier frequency */
+#define FSC          4433618.75
+
+/* Line frequency */
+#define FLINE        15625
+
+
+
+
+#define VISIBLELINES 312
+
+#define PI           3.14159265358
+
+#define RGB_to_YIQ  mat3x3( 0.299,    0.595716,  0.211456,\
+                            0.587,   -0.274453, -0.522591,\
+                            0.114,   -0.321263,  0.311135)
+
+#define YIQ_to_RGB  mat3x3( 1.0   ,   1.0,       1.0,\
+                            0.9563,  -0.2721,   -1.1070,\
+                            0.6210,  -0.6474,    1.7046)
+
+#define RGB_to_YUV  mat3x3( 0.299,   -0.14713,   0.615,\
+                            0.587,   -0.28886,  -0.514991,\
+                            0.114,    0.436,    -0.10001)
+
+#define YUV_to_RGB  mat3x3( 1.0,      1.0,       1.0,\
+                            0.0,     -0.39465,   2.03211,\
+                            1.13983, -0.58060,   0.0)
+                            
+#define fetch(ofs,center,invx) texture(Source, vec2((ofs) * (invx) + center.x, center.y))
+
+#define FIRTAPS 20
+float FIR[FIRTAPS] = float[FIRTAPS] (
+   -0.008030271,
+    0.003107906,
+    0.016841352,
+    0.032545161,
+    0.049360136,
+    0.066256720,
+    0.082120150,
+    0.095848433,
+    0.106453014,
+    0.113151423,
+    0.115441842,
+    0.113151423,
+    0.106453014,
+    0.095848433,
+    0.082120150,
+    0.066256720,
+    0.049360136,
+    0.032545161,
+    0.016841352,
+    0.003107906
+);
+
+/* subcarrier counts per scan line = FSC/FLINE = 283.7516 */
+/* We save the reciprocal of this only to optimize it */
+float counts_per_scanline_reciprocal = 0.00352420920269701;
+
+float width_ratio;
+float height_ratio;
+float altv;
+float invx;
+
+
+/* http://byteblacksmith.com/improvements-to-the-canonical-one-liner-glsl-rand-for-opengl-es-2-0/ */
+float rand(vec2 co)
+{
+    float a  = 12.9898;
+    float b  = 78.233;
+    float c  = 43758.5453;
+    float dt = dot(co.xy, vec2(a, b));
+    float sn = mod(dt,3.14);
+
+    return fract(sin(sn) * c);
+}
+
+float modulated(vec2 xy, float sinwt, float coswt) {
+    vec3 rgb = fetch(0, xy, invx).xyz;
+    vec3 yuv = RGB_to_YUV * rgb;
+
+    return clamp(yuv.x + yuv.y * sinwt + yuv.z * coswt, 0.0, 1.0);    
+}
+
+vec2 modem_uv(vec2 xy, float ofs) {
+    float t  = (xy.x + ofs * invx) * global.SourceSize.x;
+    float wt = t * 2 * PI / width_ratio;
+
+    float sinwt = sin(wt);
+    float coswt = cos(wt + altv);
+
+    vec3 rgb = fetch(ofs, xy, invx).xyz;
+    vec3 yuv = RGB_to_YUV * rgb;
+    float signal = clamp(yuv.x + yuv.y * sinwt + yuv.z * coswt, 0.0, 1.0);
+    
+    if (PHASE_NOISE != 0)
+    {
+        /* .yy is horizontal noise, .xx looks bad, .xy is classic noise */
+        vec2 seed = vTexCoord.yy + global.FrameCount;
+
+        wt    = wt + PHASE_NOISE * (rand(seed) - 0.5);
+        sinwt = sin(wt);
+        coswt = cos(wt + altv);
+    }
+
+    return vec2(signal * sinwt, signal * coswt);
+}
+
+void main()
+{
+    vec2 xy      = vTexCoord;
+    width_ratio  = global.SourceSize.x * (counts_per_scanline_reciprocal);
+    height_ratio = global.SourceSize.y / VISIBLELINES;
+    altv         = mod(floor(xy.y * VISIBLELINES + 0.5), 2.0) * PI;
+    invx         = 0.25 * (counts_per_scanline_reciprocal); // equals 4 samples per Fsc period
+   
+    // lowpass U/V at baseband
+    vec2 filtered = vec2(0.0, 0.0);
+    for (int i = 0; i < FIRTAPS; i++) {
+        vec2 uv   = modem_uv(xy, i - FIRTAPS*0.5);
+        filtered += FIR_GAIN * uv * FIR[i];
+    }
+
+    float t  = xy.x * global.SourceSize.x;
+    float wt = t * 2 * PI / width_ratio;
+
+    float sinwt = sin(wt);
+    float coswt = cos(wt + altv);
+
+    float luma = modulated(xy, sinwt, coswt) - FIR_INVGAIN * (filtered.x * sinwt + filtered.y * coswt);
+    vec3 yuv_result = vec3(luma, filtered.x, filtered.y);
+
+    FragColor = vec4(YUV_to_RGB * yuv_result, 1.0);
+}