better CRT/Composite shader

This commit is contained in:
minjaesong
2025-11-26 00:39:00 +09:00
parent e3099195e4
commit acaade1062
4 changed files with 337 additions and 228 deletions

View File

@@ -2,26 +2,18 @@ if (!exec_args[1]) {
printerrln("Usage: jpdectest image.jpg")
}
filesystem.open("A", exec_args[1], "R")
const fullFilePath = _G.shell.resolvePathInput(exec_args[1])
const file = files.open(fullFilePath.full)
const fileLen = file.size
const infile = sys.malloc(file.size); file.pread(infile, fileLen, 0)
let status = com.getStatusCode(0)
let infile = undefined
if (0 != status) return status
let fileLen = filesystem.getFileLen("A")
println(`DMA reading ${fileLen} bytes from disk...`)
infile = sys.malloc(fileLen)
dma.comToRam(0, 0, infile, fileLen)
println("decoding")
//println("decoding")
// decode
const [imgw, imgh, channels, imageData] = graphics.decodeImageResample(infile, fileLen, -1, -1)
println(`dim: ${imgw}x${imgh}`)
println(`converting to displayable format...`)
//println(`dim: ${imgw}x${imgh}`)
//println(`converting to displayable format...`)
// convert colour
graphics.setGraphicsMode(0)

View File

@@ -2,26 +2,18 @@ if (!exec_args[1]) {
printerrln("Usage: jpdectesthigh image.jpg")
}
filesystem.open("A", exec_args[1], "R")
const fullFilePath = _G.shell.resolvePathInput(exec_args[1])
const file = files.open(fullFilePath.full)
const fileLen = file.size
const infile = sys.malloc(file.size); file.pread(infile, fileLen, 0)
let status = com.getStatusCode(0)
let infile = undefined
if (0 != status) return status
let fileLen = filesystem.getFileLen("A")
println(`DMA reading ${fileLen} bytes from disk...`)
infile = sys.malloc(fileLen)
dma.comToRam(0, 0, infile, fileLen)
println("decoding")
//println("decoding")
// decode
const [imgw, imgh, channels, imageData] = graphics.decodeImageResample(infile, fileLen, -1, -1)
println(`dim: ${imgw}x${imgh}`)
println(`converting to displayable format...`)
//println(`dim: ${imgw}x${imgh}`)
//println(`converting to displayable format...`)
// convert colour
graphics.setGraphicsMode(4)

View File

@@ -0,0 +1,322 @@
// ============================================================================
// CRT + NTSC Composite/S-Video Signal Simulation Shader (Enhanced Version)
// ============================================================================
// Features:
// - Runtime-switchable composite/S-Video mode (no recompilation)
// - Adjustable signal and CRT parameters via uniforms
// - Accurate NTSC color artifact simulation
// - Animated dot crawl effect
// - Trinitron phosphor mask
// - Optional bloom/glow effect
// ============================================================================
// === UNIFORMS ===
uniform float time = 0.0; // Frame count
uniform vec2 resolution = vec2(640.0, 480.0); // Virtual resolution (e.g., 640x480)
uniform sampler2D u_texture; // Input texture
uniform vec2 flip = vec2(0.0, 0.0); // UV flip control (0,1 = flip Y)
// Signal mode: 0 = S-Video, 1 = Composite, 2 = CGA Composite
// Can be changed at runtime without recompilation
uniform int signalMode = 1; // Default should be 1 for composite
// CGA-specific settings
uniform float cgaHue; // Hue adjustment for CGA (default: 0.0, range: -PI to PI)
uniform float cgaSaturation; // Saturation multiplier for CGA (default: 1.0)
// Optional adjustable parameters (set reasonable defaults if not provided)
uniform float lumaFilterWidth; // Default: 1.5
uniform float chromaIFilterWidth; // Default: 3.5
uniform float chromaQFilterWidth; // Default: 6.0
uniform float compositeFilterWidth; // Default: 1.5
uniform float phosphorIntensity; // Default: 0.25
uniform float scanlineIntensity; // Default: 0.12
in vec2 v_texCoords;
out vec4 fragColor;
// === CONSTANTS ===
const float PI = 3.14159265358979323846;
const float TAU = 6.28318530717958647692;
// NTSC color subcarrier: 3.579545 MHz
// At 640 pixels for ~52.6µs active video: cycles/pixel ≈ 0.2917
const float CC_PER_PIXEL = 0.2917;
// CGA specific: 14.318 MHz pixel clock = exactly 4× color subcarrier
// This means exactly 4 pixels per color cycle = 0.25 cycles per pixel
const float CGA_CC_PER_PIXEL = 0.25;
// Filter kernel radius (samples to each side)
const int FILTER_RADIUS = 12;
// === COLOR SPACE CONVERSION ===
// GLSL matrices are column-major
const mat3 RGB_TO_YIQ = mat3(
0.299, 0.596, 0.211, // Column 0: R coefficients for Y,I,Q
0.587, -0.274, -0.523, // Column 1: G coefficients
0.114, -0.322, 0.312 // Column 2: B coefficients
);
const mat3 YIQ_TO_RGB = mat3(
1.000, 1.000, 1.000, // Column 0: Y coefficients for R,G,B
0.956, -0.272, -1.107, // Column 1: I coefficients
0.621, -0.647, 1.704 // Column 2: Q coefficients
);
// === DEFAULT VALUES ===
// Used when uniforms aren't set (value of 0)
float getLumaFilter() {
return lumaFilterWidth > 0.0 ? lumaFilterWidth : 1.15;
}
float getChromaIFilter() {
return chromaIFilterWidth > 0.0 ? chromaIFilterWidth : 3.5;
}
float getChromaQFilter() {
return chromaQFilterWidth > 0.0 ? chromaQFilterWidth : 6.0;
}
float getCompositeFilter() {
return compositeFilterWidth > 0.0 ? compositeFilterWidth : 1.35;
}
float getPhosphorStrength() {
return phosphorIntensity > 0.0 ? phosphorIntensity : 0.25;
}
float getScanlineStrength() {
return scanlineIntensity > 0.0 ? scanlineIntensity : 0.12;
}
float getCgaSaturation() {
return cgaSaturation > 0.0 ? cgaSaturation : 1.0;
}
// === HELPER FUNCTIONS ===
float gaussianWeight(float x, float sigma) {
return exp(-0.5 * x * x / (sigma * sigma));
}
vec3 sampleTexture(vec2 uv) {
return texture(u_texture, clamp(uv, 0.0, 1.0)).rgb;
}
float calcCarrierPhase(float pixelX, float pixelY, float frameOffset) {
float phase = pixelX * TAU * CC_PER_PIXEL;
phase += pixelY * PI; // 180° per line (from 227.5 cycles/line)
phase += frameOffset;
return phase;
}
float encodeComposite(vec3 rgb, float phase) {
vec3 yiq = RGB_TO_YIQ * rgb;
return yiq.x + yiq.y * cos(phase) + yiq.z * sin(phase);
}
// === COMPOSITE SIGNAL DECODE ===
vec3 decodeComposite(vec2 uv, vec2 texelSize, float basePhase) {
float compFilter = getCompositeFilter();
float iFilter = getChromaIFilter();
float qFilter = getChromaQFilter();
float yAccum = 0.0, iAccum = 0.0, qAccum = 0.0;
float yWeight = 0.0, iWeight = 0.0, qWeight = 0.0;
for (int i = -FILTER_RADIUS; i <= FILTER_RADIUS; i++) {
float offset = float(i);
vec2 sampleUV = uv + vec2(offset * texelSize.x, 0.0);
vec3 srcRGB = sampleTexture(sampleUV);
float samplePhase = basePhase + offset * TAU * CC_PER_PIXEL;
float composite = encodeComposite(srcRGB, samplePhase);
// Low-pass for luma
float yw = gaussianWeight(offset, compFilter);
yAccum += composite * yw;
yWeight += yw;
// Demodulate and filter chroma
float iw = gaussianWeight(offset, iFilter);
float qw = gaussianWeight(offset, qFilter);
iAccum += composite * cos(samplePhase) * 2.0 * iw;
qAccum += composite * sin(samplePhase) * 2.0 * qw;
iWeight += iw;
qWeight += qw;
}
vec3 yiq = vec3(yAccum / yWeight, iAccum / iWeight, qAccum / qWeight);
return YIQ_TO_RGB * yiq;
}
// === S-VIDEO SIGNAL DECODE ===
vec3 decodeSVideo(vec2 uv, vec2 texelSize, float basePhase) {
float yFilter = getLumaFilter();
float iFilter = getChromaIFilter();
float qFilter = getChromaQFilter();
float yAccum = 0.0, iAccum = 0.0, qAccum = 0.0;
float yWeight = 0.0, iWeight = 0.0, qWeight = 0.0;
for (int i = -FILTER_RADIUS; i <= FILTER_RADIUS; i++) {
float offset = float(i);
vec2 sampleUV = uv + vec2(offset * texelSize.x, 0.0);
vec3 srcRGB = sampleTexture(sampleUV);
vec3 yiq = RGB_TO_YIQ * srcRGB;
float samplePhase = basePhase + offset * TAU * CC_PER_PIXEL;
float chromaSignal = yiq.y * cos(samplePhase) + yiq.z * sin(samplePhase);
// Luma is separate - no cross-color
float yw = gaussianWeight(offset, yFilter);
yAccum += yiq.x * yw;
yWeight += yw;
// Chroma demodulation
float iw = gaussianWeight(offset, iFilter);
float qw = gaussianWeight(offset, qFilter);
iAccum += chromaSignal * cos(samplePhase) * 2.0 * iw;
qAccum += chromaSignal * sin(samplePhase) * 2.0 * qw;
iWeight += iw;
qWeight += qw;
}
vec3 yiqOut = vec3(yAccum / yWeight, iAccum / iWeight, qAccum / qWeight);
return YIQ_TO_RGB * yiqOut;
}
// === CGA COMPOSITE DECODE ===
// CGA has exactly 4 pixels per color cycle (14.318 MHz / 3.579545 MHz = 4)
// This creates the famous artifact colors from specific bit patterns
vec3 decodeCGAComposite(vec2 uv, vec2 texelSize, float pixelX, float pixelY) {
// CGA-specific filter widths - slightly different from generic NTSC
// CGA monitors typically had less filtering, making artifacts more pronounced
float yFilter = 1.2;
float chromaFilter = 2.5;
// CGA color burst phase - this determines the base hue
// Adjusted to match the canonical CGA artifact color palette
float cgaPhaseOffset = cgaHue + PI * 0.5; // Adjust for correct color alignment
// CGA doesn't have the 227.5 cycle per line offset in the same way
// The phase is more deterministic based on pixel position
float basePhase = pixelX * TAU * CGA_CC_PER_PIXEL + cgaPhaseOffset;
// Odd lines have 180° phase shift (creates the alternating pattern)
if (mod(pixelY, 2.0) >= 1.0) {
basePhase += PI;
}
float yAccum = 0.0, iAccum = 0.0, qAccum = 0.0;
float yWeight = 0.0, chromaWeight = 0.0;
// Use smaller filter radius for sharper CGA look
const int CGA_RADIUS = 8;
for (int i = -CGA_RADIUS; i <= CGA_RADIUS; i++) {
float offset = float(i);
vec2 sampleUV = uv + vec2(offset * texelSize.x, 0.0);
// CGA outputs either black (0) or white (1) in 640x200 mode
// Get the source value (treating as monochrome for artifact generation)
vec3 srcRGB = sampleTexture(sampleUV);
float srcLuma = dot(srcRGB, vec3(0.299, 0.587, 0.114));
// For CGA artifact colors, we use the luma as the composite signal level
// In reality, CGA outputs either 0V or ~0.7V for the two states
float composite = srcLuma;
float samplePhase = basePhase + offset * TAU * CGA_CC_PER_PIXEL;
// Low-pass filter for luma
float yw = gaussianWeight(offset, yFilter);
yAccum += composite * yw;
yWeight += yw;
// Demodulate chroma
float cw = gaussianWeight(offset, chromaFilter);
iAccum += composite * cos(samplePhase) * 2.0 * cw;
qAccum += composite * sin(samplePhase) * 2.0 * cw;
chromaWeight += cw;
}
float y = yAccum / yWeight;
float i = (iAccum / chromaWeight) * getCgaSaturation();
float q = (qAccum / chromaWeight) * getCgaSaturation();
// Convert to RGB
vec3 rgb = YIQ_TO_RGB * vec3(y, i, q);
return rgb;
}
// === TRINITRON PHOSPHOR MASK ===
vec3 trinitronMask(vec2 screenPos) {
float strength = getPhosphorStrength();
float outputX = screenPos.x * 2.0; // 2x display scale
float stripe = mod(outputX, 3.0);
float bleed = 0.15;
vec3 mask;
if (stripe < 1.0) {
mask = vec3(1.0, bleed, bleed);
} else if (stripe < 2.0) {
mask = vec3(bleed, 1.0, bleed);
} else {
mask = vec3(bleed, bleed, 1.0);
}
float compensation = 1.0 / (0.333 + 0.667 * bleed);
mask *= compensation * 0.85;
return mix(vec3(1.0), mask, strength);
}
// === SCANLINE MASK ===
float scanlineMask(vec2 screenPos) {
float strength = getScanlineStrength();
float outputY = screenPos.y * 2.0; // 2x display scale
float scanline = sin(outputY * PI);
scanline = scanline * 0.5 + 0.5;
scanline = pow(scanline, 0.4);
return mix(1.0 - strength, 1.0, scanline);
}
// === MAIN ===
void main() {
vec2 uv = v_texCoords;
uv.x = mix(uv.x, 1.0 - uv.x, flip.x);
uv.y = mix(uv.y, 1.0 - uv.y, flip.y);
vec2 texelSize = 1.0 / resolution;
float pixelX = uv.x * resolution.x;
float pixelY = uv.y * resolution.y;
// Frame phase for dot crawl (4-frame cycle)
float framePhase = mod(time, 4.0) * PI * 0.5;
float basePhase = calcCarrierPhase(pixelX, pixelY, framePhase);
// Decode signal based on mode
vec3 rgb;
if (signalMode == 2) {
// CGA Composite mode - deterministic artifact colors
rgb = decodeCGAComposite(uv, texelSize, pixelX, pixelY);
} else if (signalMode == 1) {
rgb = decodeComposite(uv, texelSize, basePhase);
} else {
rgb = decodeSVideo(uv, texelSize, basePhase);
}
// CRT display effects
vec2 screenPos = vec2(pixelX, pixelY);
// rgb *= trinitronMask(screenPos);
// rgb *= scanlineMask(screenPos);
fragColor = vec4(clamp(rgb, 0.0, 1.0), 1.0);
}

View File

@@ -97,7 +97,7 @@ class VMGUI(val loaderInfo: EmulInstance, val viewportWidth: Int, val viewportHe
camera.update()
batch.projectionMatrix = camera.combined
crtShader = loadShaderInline(CRT_POST_SHADER2)
crtShader = loadShaderInline(Gdx.files.classpath("net/torvald/tsvm/shader_crt_post.frag").readString())
gpuFBO = FrameBuffer(Pixmap.Format.RGBA8888, viewportWidth, viewportHeight, false)
winFBO = FrameBuffer(Pixmap.Format.RGBA8888, viewportWidth, viewportHeight, false)
@@ -556,200 +556,3 @@ void main() {
}
"""
const val CRT_POST_SHADER2 = """
#ifdef GL_ES
precision mediump float;
#endif
in vec4 v_color;
in vec4 v_generic;
in vec2 v_texCoords;
uniform sampler2D u_texture;
uniform vec2 resolution = vec2(640.0, 480.0);
out vec4 fragColor;
uniform float time = 0.0;
const int SUBS = 6; // horizontal subsamples per pixel (oversampling)
const float PI = 3.14159265359;
// --- RGB <-> YIQ (NTSC-ish) ---
vec3 rgb2yiq(vec3 rgb) {
float y = dot(rgb, vec3(0.299, 0.587, 0.114));
float i = dot(rgb, vec3(0.596, -0.274, -0.322));
float q = dot(rgb, vec3(0.211, -0.523, 0.312));
return vec3(y, i, q);
}
vec3 yiq2rgb(vec3 yiq) {
float y = yiq.x, i = yiq.y, q = yiq.z;
vec3 r = vec3(
y + 0.956*i + 0.621*q,
y - 0.272*i - 0.647*q,
y - 1.106*i + 1.703*q
);
return clamp(r, 0.0, 1.0);
}
// --- Parameters you can tweak ---
float subcarrierCyclesPerScanline = 227.5; // NTSC approx cycles per scanline (spatial)
float chromaGain = 2.2; // strength of chroma modulation in encode
float chromaLPF_radius = 3.6; // lowpass radius in pixels for chroma (larger = more bleed)
float lumaLPF_radius = 0.7; // lowpass for luma (small = sharp)
float chromaPhaseDrift = 1.1; // extra phase offset (use time to animate dot crawl)
float subsampleSpanPixels = 2.0 / 3.0; // how wide the subsample footprint is (in pixels)
// Simple 1D gaussian weight (not normalized here)
float gaussWeight(float x, float r) {
return exp(- (x*x) / (2.0 * r * r));
}
// Sample a tiny neighborhood and get averaged Y/I/Q and also do composite modulation
// We oversample horizontally to simulate the analog sampling along the scanline.
void compositeAtUV(in vec2 uv, out float avgComposite, out float avgY, out float avgDemodI, out float avgDemodQ) {
avgComposite = 0.0;
avgY = 0.0;
avgDemodI = 0.0;
avgDemodQ = 0.0;
// pixel coordinates
float px = uv.x * resolution.x;
float py = uv.y * resolution.y;
// subcarrier spatial frequency (cycles per pixel horizontally)
// cycles per scanline / pixels per scanline = cycles per pixel (approx)
float cycles_per_pixel = subcarrierCyclesPerScanline / resolution.x;
// time-varying phase for dot-crawl
float linePhase = (py * subcarrierCyclesPerScanline) * 2.0 * PI + chromaPhaseDrift * time;
// do SUBS evenly-spaced subsamples across this pixel horizontally
float totalWeight = 0.0;
for (int s = 0; s < SUBS; ++s) {
float t = (float(s) + 0.5) / float(SUBS) - 0.5; // -0.5 .. +0.5
float sx = px + t * subsampleSpanPixels; // sample x in pixel space
vec2 sUV = vec2(sx / resolution.x, uv.y);
vec3 col = texture2D(u_texture, sUV).rgb;
vec3 yiq = rgb2yiq(col);
float Y = yiq.x;
float I = yiq.y;
float Q = yiq.z;
// subcarrier phase at this horizontal sample
float phi = linePhase + sx * cycles_per_pixel * 2.0 * PI;
// composite waveform value (analog mix)
float carrierCos = cos(phi);
float carrierSin = sin(phi);
float composite = Y + chromaGain * (I * carrierCos + Q * carrierSin);
// demodulation multipliers (we will low-pass by averaging)
float demodI = composite * carrierCos;
float demodQ = composite * carrierSin;
// weight: use gaussian centered on pixel center
float w = gaussWeight(t * subsampleSpanPixels, 0.6); // narrower weighting
totalWeight += w;
avgComposite += composite * w;
avgY += Y * w;
avgDemodI += demodI * w;
avgDemodQ += demodQ * w;
}
// normalize the local averages (this approximates a basic low-pass)
avgComposite /= totalWeight;
avgY /= totalWeight;
avgDemodI /= totalWeight;
avgDemodQ /= totalWeight;
// --- additional spatial low-pass filtering to mimic channel bandwidth ---
// We'll sample neighbours on x and average with Gaussian weights to emulate chroma/luma LPF
// Note: we do a tiny neighborhood (±2 pixels) to save performance.
float wsumC = 1.0;
float wsumY = 1.0;
float csum = avgComposite;
float idsum = avgDemodI;
float qdsum = avgDemodQ;
float ysum = avgY;
// radius in pixels for chroma or luma affects weights below; we'll use same kernel but scale contributions
int taps = 2;
for (int i = -taps; i <= taps; ++i) {
if (i == 0) continue;
float off = float(i);
// use chroma radius for chroma-related weights, luma radius for luma
float weightC = gaussWeight(off, chromaLPF_radius);
float weightY = gaussWeight(off, lumaLPF_radius);
vec2 sampleUV = vec2((px + off) / resolution.x, uv.y);
vec3 ncol = texture2D(u_texture, sampleUV).rgb;
vec3 nyiq = rgb2yiq(ncol);
float nY = nyiq.x;
float nI = nyiq.y;
float nQ = nyiq.z;
float nPhi = linePhase + (px + off) * cycles_per_pixel * 2.0 * PI;
float nComposite = nY + chromaGain * (nI * cos(nPhi) + nQ * sin(nPhi));
float ndemodI = nComposite * cos(nPhi);
float ndemodQ = nComposite * sin(nPhi);
csum += nComposite * weightC;
idsum += ndemodI * weightC;
qdsum += ndemodQ * weightC;
wsumC += weightC;
ysum += nY * weightY;
wsumY += weightY;
}
avgComposite = csum / wsumC;
avgY = ysum / wsumY;
avgDemodI = idsum / wsumC;
avgDemodQ = qdsum / wsumC;
}
// main
void mainImage(out vec4 frag_Color, in vec2 frag_Coord) {
vec2 uv = frag_Coord.xy / resolution.xy;
uv.y = 1.0 - uv.y;
// Get composite and demod values centered at uv
float compositeVal, Yval, demodI, demodQ;
compositeAtUV(uv, compositeVal, Yval, demodI, demodQ);
// Low-pass the demodulators to extract I & Q (simple normalization)
// In a real demodulator you'd low-pass filter demod*carrier; here we've already averaged spatially,
// so just apply a gain normalization and small smoothing.
// Normalize I/Q by average carrier power (~0.5)
float carrierPower = 0.5;
float I_rec = demodI / max(carrierPower, 1e-5);
float Q_rec = demodQ / max(carrierPower, 1e-5);
// Optionally reduce chroma bandwidth / smear by blurring (simulate narrow chroma filter)
// We'll lerp the recovered chroma towards zero based on chromaLPF_radius
float chromaBlurFactor = smoothstep(0.0, 5.0, chromaLPF_radius); // 0..1
I_rec *= 1.0 - 0.65 * chromaBlurFactor;
Q_rec *= 1.0 - 0.65 * chromaBlurFactor;
// For luma, we can optionally low-pass Yval slightly to simulate limited luma bandwidth
// but keep it mostly sharp
float Y_lp = Yval; // already lightly filtered above
// final reconstructed yiq
vec3 yiqRec = vec3(Y_lp, I_rec, Q_rec);
vec3 rgbRec = yiq2rgb(yiqRec);
// Optional: blend with original luminance to keep text legible (tweakable)
// vec3 orig = texture2D(u_texture, uv).rgb;
// rgbRec = mix(rgbRec, orig, 0.25);
frag_Color = vec4(rgbRec, 1.0);
}
void main() {
vec2 fragCoord = gl_FragCoord.xy;
vec4 outcol;
mainImage(outcol, fragCoord);
fragColor = outcol;
}
"""