From ad368477f336de15f8cc647ec2ac101cbef2b0b3 Mon Sep 17 00:00:00 2001
From: hunterk <hunter_kaller@yahoo.com>
Date: Fri, 12 Feb 2021 21:37:55 -0600
Subject: [PATCH] add MMPX and EPX shaders and presets

---
 scalenx/epx.slangp         |   6 +
 scalenx/mmpx.slangp        |   6 +
 scalenx/shaders/epx.slang  |  88 +++++++++++
 scalenx/shaders/mmpx.slang | 303 +++++++++++++++++++++++++++++++++++++
 4 files changed, 403 insertions(+)
 create mode 100644 scalenx/epx.slangp
 create mode 100644 scalenx/mmpx.slangp
 create mode 100644 scalenx/shaders/epx.slang
 create mode 100644 scalenx/shaders/mmpx.slang

diff --git a/scalenx/epx.slangp b/scalenx/epx.slangp
new file mode 100644
index 0000000..e746a24
--- /dev/null
+++ b/scalenx/epx.slangp
@@ -0,0 +1,6 @@
+shaders = 1
+
+shader0 = shaders/epx.slang
+filter_linear0 = false
+scale_type0 = source
+scale0 = 2.0
diff --git a/scalenx/mmpx.slangp b/scalenx/mmpx.slangp
new file mode 100644
index 0000000..ef44068
--- /dev/null
+++ b/scalenx/mmpx.slangp
@@ -0,0 +1,6 @@
+shaders = 1
+
+shader0 = shaders/mmpx.slang
+filter_linear0 = false
+scale_type0 = source
+scale0 = 2.0
diff --git a/scalenx/shaders/epx.slang b/scalenx/shaders/epx.slang
new file mode 100644
index 0000000..2affb61
--- /dev/null
+++ b/scalenx/shaders/epx.slang
@@ -0,0 +1,88 @@
+#version 450
+
+// EPX (Eric's Pixel Scaler)
+// based on the description from Wikipedia:
+// https://en.wikipedia.org/wiki/Pixel-art_scaling_algorithms#EPX/Scale2%C3%97/AdvMAME2%C3%97
+// adapted for slang by hunterk
+// license GPL, I think
+
+layout(push_constant) uniform Push
+{
+	vec4 SourceSize;
+	vec4 OriginalSize;
+	vec4 OutputSize;
+	uint FrameCount;
+} params;
+
+layout(std140, set = 0, binding = 0) uniform UBO
+{
+	mat4 MVP;
+} 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 * 1.00001;
+}
+
+#pragma stage fragment
+layout(location = 0) in vec2 vTexCoord;
+layout(location = 0) out vec4 FragColor;
+layout(set = 0, binding = 2) uniform sampler2D Source;
+
+bool same(vec3 B, vec3 A0){
+   return all(equal(B, A0));
+}
+
+bool notsame(vec3 B, vec3 A0){
+   return any(notEqual(B, A0));
+}
+
+// sample with coordinate offsets
+#define TEX(c,d) texture(Source, vTexCoord.xy + vec2(c,d) * params.SourceSize.zw).rgb
+
+void main()
+{
+// The algorithm looks at the current pixel and the 4 surrounding cardinal pixels  
+// ___|_A_|___  
+// _C_|_P_|_B_
+//    | D |  
+
+// Our current pixel, P
+   vec3 P = TEX( 0., 0.);
+
+// Input pixels
+   vec3 A = TEX( 0., 1.);
+   vec3 B = TEX( 1., 0.);
+   vec3 D = TEX( 0.,-1.);
+   vec3 C = TEX(-1., 0.);
+   
+// Output: 2x2 grid. Default to the current pixel color (Nearest magnification)
+// ___one_|_two___
+//  three | four
+   vec3 one   = P;
+   vec3 two   = P;
+   vec3 three = P;
+   vec3 four  = P;
+
+// EPX algorithm rules:   
+// IF C==A AND C!=D AND A!=B => 1=A
+// IF A==B AND A!=C AND B!=D => 2=B
+// IF D==C AND D!=B AND C!=A => 3=C
+// IF B==D AND B!=A AND D!=C => 4=D
+
+   one   = (same(C, D) && notsame(C, A) && notsame(C, B)) ? C : P;
+   two   = (same(D, B) && notsame(D, C) && notsame(D, A)) ? D : P;
+   three = (same(A, C) && notsame(A, B) && notsame(A, D)) ? A : P;
+   four  = (same(B, A) && notsame(B, D) && notsame(B, C)) ? B : P;
+
+   vec2 px = fract(vTexCoord * params.SourceSize.xy);
+// split the texels into 4 and assign one of our output pixels to each
+   FragColor.rgb = (px.x < 0.5) ? (px.y < 0.5 ? one : three) : (px.y < 0.5 ? two : four);
+   FragColor.a = 1.0;
+}
diff --git a/scalenx/shaders/mmpx.slang b/scalenx/shaders/mmpx.slang
new file mode 100644
index 0000000..9d625a6
--- /dev/null
+++ b/scalenx/shaders/mmpx.slang
@@ -0,0 +1,303 @@
+#version 450
+
+// MMPX
+// by Morgan McGuire and Mara Gagiu
+// https://casual-effects.com/research/McGuire2021PixelArt/
+// License: MIT
+// adapted for slang by hunterk
+
+layout(push_constant) uniform Push
+{
+	vec4 SourceSize;
+	vec4 OriginalSize;
+	vec4 OutputSize;
+	uint FrameCount;
+} params;
+
+layout(std140, set = 0, binding = 0) uniform UBO
+{
+	mat4 MVP;
+} 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;
+
+float luma(vec3 col){
+   return dot(col, vec3(0.2126, 0.7152, 0.0722));
+}
+
+// I tried using a hash to speed these up but it didn't really help
+bool same(vec3 B, vec3 A0){
+   return all(equal(B, A0));
+}
+
+bool notsame(vec3 B, vec3 A0){
+   return any(notEqual(B, A0));
+}
+
+bool all_eq2(vec3 B, vec3 A0, vec3 A1) {
+    return (same(B,A0) && same(B,A1));
+}
+
+bool all_eq3(vec3 B, vec3 A0, vec3 A1, vec3 A2) {
+    return (same(B,A0) && same(B,A1) && same(B,A2));
+}
+
+bool all_eq4(vec3 B, vec3 A0, vec3 A1, vec3 A2, vec3 A3) {
+    return (same(B,A0) && same(B,A1) && same(B,A2) && same(B,A3));
+}
+
+bool any_eq3(vec3 B, vec3 A0, vec3 A1, vec3 A2) {
+   return (same(B,A0) || same(B,A1) || same(B,A2));
+}
+
+bool none_eq2(vec3 B, vec3 A0, vec3 A1) {
+   return (notsame(B,A0) && notsame(B,A1));
+}
+
+bool none_eq4(vec3 B, vec3 A0, vec3 A1, vec3 A2, vec3 A3) {
+   return (notsame(B,A0) && notsame(B,A1) && notsame(B,A2) && notsame(B,A3));
+}
+
+#define src(c,d) texture(Source, vTexCoord + vec2(c,d) * params.SourceSize.zw).rgb
+
+void main()
+{
+// these do nothing, but just for consistency with the original code...
+   float srcX = 0.;
+   float srcY = 0.;
+
+// Our current pixel
+   vec3 E = src(srcX+0.,srcY+0.);
+
+// Input: A-I central 3x3 grid
+   vec3 A = src(srcX-1.,srcY-1.);
+   vec3 B = src(srcX+0.,srcY-1.);
+   vec3 C = src(srcX+1.,srcY-1.);
+
+   vec3 D = src(srcX-1.,srcY+0.);
+   vec3 F = src(srcX+1.,srcY+0.);
+
+   vec3 G = src(srcX-1.,srcY+1.);
+   vec3 H = src(srcX+0.,srcY+1.);
+   vec3 I = src(srcX+1.,srcY+1.);
+
+// Default to Nearest magnification
+   vec3 J = E;
+   vec3 K = E;
+   vec3 L = E;
+   vec3 M = E;
+// Go ahead and put this here so we can use an early return on the next
+// line to save some cycles
+   FragColor = vec4(E, 1.0);
+   
+// Skip constant 3x3 centers and just use nearest-neighbor
+// them.  This gives a good speedup on spritesheets with
+// lots of padding and full screen images with large
+// constant regions such as skies.
+// EDIT: this is a wash for me, but we'll keep it around
+ if(same(E,A) && same(E,B) && same(E,C) && same(E,D) && same(E,F) && same(E,G) && same(E,H) && same(E,I)) return;
+
+// Read additional values at the tips of the diamond pattern
+   vec3 P  = src(srcX+0.,srcY-2.);
+   vec3 Q  = src(srcX-2.,srcY+0.);
+   vec3 R  = src(srcX+2.,srcY+0.);
+   vec3 S  = src(srcX+0.,srcY+2.);
+
+// Precompute luminances
+   float Bl = luma(B);
+   float Dl = luma(D);
+   float El = luma(E);
+   float Fl = luma(F);
+   float Hl = luma(H);
+
+// Round some corners and fill in 1:1 slopes, but preserve
+// sharp right angles.
+//
+// In each expression, the left clause is from
+// EPX and the others are new. EPX
+// recognizes 1:1 single-pixel lines because it
+// applies the rounding only to the LINE, and not
+// to the background (it looks at the mirrored
+// side).  It thus fails on thick 1:1 edges
+// because it rounds *both* sides and produces an
+// aliased edge shifted by 1 dst pixel.  (This
+// also yields the mushroom-shaped arrow heads,
+// where that 1-pixel offset runs up against the
+// 2-pixel aligned end; this is an inherent
+// problem with 2X in-palette scaling.)
+//
+// The 2nd clause clauses avoid *double* diagonal
+// filling on 1:1 slopes to prevent them becoming
+// aliased again. It does this by breaking
+// symmetry ties using luminance when working with
+// thick features (it allows thin and transparent
+// features to pass always).
+//
+// The 3rd clause seeks to preserve square corners
+// by considering the center value before
+// rounding.
+//
+// The 4th clause identifies 1-pixel bumps on
+// straight lines that are darker than their
+// background, such as the tail on a pixel art
+// "4", and prevents them from being rounded. This
+// corrects for asymmetry in this case that the
+// luminance tie breaker introduced.
+
+// .------------ 1st ------------.      .----- 2nd ---------.      .------ 3rd -----.      .--------------- 4th -----------------------.
+   if (((same(D,B) && notsame(D,H) && notsame(D,F))) && ((El>=Dl) || same(E,A)) && any_eq3(E,A,C,G) && ((El<Dl) || notsame(A,D) || notsame(E,P) || notsame(E,Q))) J=D;
+   if (((same(B,F) && notsame(B,D) && notsame(B,H))) && ((El>=Bl) || same(E,C)) && any_eq3(E,A,C,I) && ((El<Bl) || notsame(C,B) || notsame(E,P) || notsame(E,R))) K=B;
+   if (((same(H,D) && notsame(H,F) && notsame(H,B))) && ((El>=Hl) || same(E,G)) && any_eq3(E,A,G,I) && ((El<Hl) || notsame(G,H) || notsame(E,S) || notsame(E,Q))) L=H;
+   if (((same(F,H) && notsame(F,B) && notsame(F,D))) && ((El>=Fl) || same(E,I)) && any_eq3(E,C,G,I) && ((El<Fl) || notsame(I,H) || notsame(E,R) || notsame(E,S))) M=F;
+
+// Clean up disconnected line intersections.
+//
+// The first clause recognizes being on the inside
+// of a diagonal corner and ensures that the "foreground"
+// has been correctly identified to avoid
+// ambiguous cases such as this:
+//
+//  o#o#
+//  oo##
+//  o#o#
+//
+// where trying to fix the center intersection of
+// either the "o" or the "#" will leave the other
+// one disconnected. This occurs, for example,
+// when a pixel-art letter "B" or "R" is next to
+// another letter on the right.
+//
+// The second clause ensures that the pattern is
+// not a notch at the edge of a checkerboard
+// dithering pattern.
+// 
+// >
+//  .--------------------- 1st ------------------------.      .--------- 2nd -----------. 
+   if ((notsame(E,F) && all_eq4(E,C,I,D,Q) && all_eq2(F,B,H)) && notsame(F,src(srcX+3.,srcY))) K=M=F;
+   if ((notsame(E,D) && all_eq4(E,A,G,F,R) && all_eq2(D,B,H)) && notsame(D,src(srcX-3.,srcY))) J=L=D;
+   if ((notsame(E,H) && all_eq4(E,G,I,B,P) && all_eq2(H,D,F)) && notsame(H,src(srcX,srcY+3.))) L=M=H;
+   if ((notsame(E,B) && all_eq4(E,A,C,H,S) && all_eq2(B,D,F)) && notsame(B,src(srcX,srcY-3.))) J=K=B;
+
+// Remove tips of bright triangles on dark
+// backgrounds. The luminance tie breaker for 1:1
+// pixel lines leaves these as sticking up squared
+// off, which makes bright triangles and diamonds
+// look bad.
+   if ((Bl<El) && all_eq4(E,G,H,I,S) && none_eq4(E,A,D,C,F)) J=K=B;
+   if ((Hl<El) && all_eq4(E,A,B,C,P) && none_eq4(E,D,G,I,F)) L=M=H;
+   if ((Fl<El) && all_eq4(E,A,D,G,Q) && none_eq4(E,B,C,I,H)) K=M=F;
+   if ((Dl<El) && all_eq4(E,C,F,I,R) && none_eq4(E,B,A,G,H)) J=L=D;
+
+//////////////////////////////////////////////////////////////////////////////////
+// Do further neighborhood peeking to identify
+// 2:1 and 1:2 slopes of constant color.
+// The first clause of each rule identifies a 2:1 slope line
+// of consistent color.
+//
+// The second clause verifies that the line is separated from
+// every adjacent pixel on one side and not part of a more
+// complex pattern. Common subexpressions from the second clause
+// are lifted to an outer test on pairs of rules.
+// 
+// The actions taken by rules are unusual in that they extend
+// a color assigned by previous rules rather than drawing from
+// the original source image.
+//
+// The comments show a diagram of the local
+// neighborhood in which letters shown with the
+// same shape and color must match each other and
+// everything else without annotation must be
+// different from the solid colored, square
+// letters.
+   if (notsame(H,B)) { // Common subexpression
+                       // Above a 2:1 slope or -2:1 slope   ◢ ◣
+                       // First:
+      if (notsame(H,A) && notsame(H,E) && notsame(H,C)) {
+                       // Second:
+                       //     P 
+                       //   Ⓐ B C .
+                       // Q D 🄴 🅵 🆁
+                       //   🅶 🅷 I
+                       //     S
+         if (all_eq3(H,G,F,R) && none_eq2(H,D,src(srcX+2.,srcY-1.))) L=M;
+                       // Third:
+                       //     P 
+                       // . A B Ⓒ
+                       // 🆀 🅳 🄴 F R
+                       //   G 🅷 🅸
+                       //     S
+         if (all_eq3(H,I,D,Q) && none_eq2(H,F,src(srcX-2.,srcY-1.))) M=L;
+      }
+                        
+                       // Below a 2:1 (◤) or -2:1 (◥) slope (reflect the above 2:1 patterns vertically)
+      if (notsame(B,I) && notsame(B,G) && notsame(B,E)) {
+                       //     P 
+                       //   🅰 🅱 C
+                       // Q D 🄴 🅵 🆁
+                       //   Ⓖ H I .
+                       //     S
+         if (all_eq3(B,A,F,R) && none_eq2(B,D,src(srcX+2.,srcY+1.))) J=K;
+                       //     P 
+                       //   A 🅱 🅲
+                       // 🆀 🅳 🄴 F R
+                       // . G H Ⓘ 
+                       //     S
+         if (all_eq3(B,C,D,Q) && none_eq2(B,F,src(srcX-2.,srcY+1.))) K=J;
+      }
+   }
+
+   if (notsame(F,D)) { // Common subexpression
+                        
+                       // Right of a -1:2 (\) or -1:2 (/) slope (reflect the left 1:2 patterns horizontally)
+      if (notsame(D,I) && notsame(D,E) && notsame(D,C)) {
+                       //     P
+                       //   🅰 B Ⓒ
+                       // Q 🅳 🄴 F R
+                       //   G 🅷 I
+                       //     🆂 .
+         if (all_eq3(D,A,H,S) && none_eq2(D,B,src(srcX+1.,srcY+2.))) J=L;
+                       //     🅿 .
+                       //   A 🅱 C
+                       // Q 🅳 🄴 F R
+                       //   🅶 H Ⓘ
+                       //     S
+         if (all_eq3(D,G,B,P) && none_eq2(D,H,src(srcX+1.,srcY-2.))) L=J;
+      }
+                        
+                       // Left of a 1:2 (/) slope or -1:2 (\) slope (transpose the above 2:1 patterns)
+                       // Pull common none_eq subexpressions out
+
+      if (notsame(F,E) && notsame(F,A) && notsame(F,G)) {
+                       //     P     
+                       //   Ⓐ B 🅲   
+                       // Q D 🄴 🅵 R 
+                       //   G 🅷 I   
+                       //   . 🆂   
+         if (all_eq3(F,C,H,S) && none_eq2(F,B,src(srcX-1.,srcY+2.))) K=M;
+                       //   . 🅿
+                       //   A 🅱 C
+                       // Q D 🄴 🅵 R
+                       //   Ⓖ H 🅸
+                       //     S
+         if (all_eq3(F,I,B,P) && none_eq2(F,H,src(srcX-1.,srcY-2.))) M=K;
+      }
+   }
+
+// Determine which of our 4 output pixels we need to use
+   vec2 a = fract(vTexCoord * params.SourceSize.xy);
+   FragColor.rgb = (a.x < 0.5) ? (a.y < 0.5 ? J : L) : (a.y < 0.5 ? K : M);
+}