Add Lanczos3 scaling (#9)

"Lanczos filtering tends to yield much better quality (sharper image) than bilinear filtering."

mkxp.json

@@ -73,6 +73,13 @@
     // "smoothScaling": false,
 
 
+    // Apply Lanczos3 interpolation when game screen
+    // is upscaled (typically higher quality than linear)
+    // (default: disabled)
+    //
+    // "lanczos3Scaling": false,
+
+
     // Sync screen redraws to the monitor refresh rate
     // (default: disabled)
     //

shader/lanczos3.frag

@@ -0,0 +1,48 @@
+// From https://raw.githubusercontent.com/Sentmoraap/doing-sdl-right/93a52a0db0ff2da5066cce12f5b9a2ac62e6f401/assets/lanczos3.frag
+// Copyright 2020 Lilian Gimenez (Sentmoraap).
+// MIT license.
+
+uniform sampler2D texture;
+uniform vec2 sourceSize;
+uniform vec2 texSizeInv;
+varying vec2 v_texCoord;
+
+float lanczos3(float x)
+{
+	x = max(abs(x), 0.00001);
+	float val = x * 3.141592654;
+	return sin(val) * sin(val / 3) / (val * val);
+}
+
+void main()
+{
+	vec2 pixel = v_texCoord * sourceSize + 0.5;
+	vec2 frac = fract(pixel);
+	vec2 onePixel = texSizeInv;
+	pixel = floor(pixel) * texSizeInv - onePixel / 2;
+
+	float lanczosX[6];
+	float sum = 0;
+	for(int x = 0; x < 6; x++)
+	{
+		lanczosX[x] = lanczos3(x - 2 - frac.x);
+		sum += lanczosX[x];
+	}
+	for(int x = 0; x < 6; x++) lanczosX[x] /= sum;
+	sum = 0;
+	float lanczosY[6];
+	for(int y = 0; y < 6; y++)
+	{
+		lanczosY[y] = lanczos3(y - 2 - frac.y);
+		sum += lanczosY[y];
+	}
+	for(int y = 0; y < 6; y++) lanczosY[y] /= sum;
+	gl_FragColor = vec4(0);
+	for(int y = -2; y <= 3; y++)
+	{
+		vec4 colour = vec4(0);
+		for(int x = -2; x <= 3; x++)
+				colour += texture2D(texture, pixel + vec2(x * onePixel.x, y * onePixel.y)).rgba * lanczosX[x + 2];
+		gl_FragColor += colour * lanczosY[y + 2];
+	}
+}

shader/meson.build

@@ -15,6 +15,7 @@ embedded_shaders = [
     'simpleAlphaUni.frag',
     'tilemap.frag',
     'flashMap.frag',
+    'lanczos3.frag',
     'minimal.vert',
     'simple.vert',
     'simpleColor.vert',

src/config.cpp

@@ -134,6 +134,7 @@ void Config::read(int argc, char *argv[]) {
         {"fullscreen", false},
         {"fixedAspectRatio", true},
         {"smoothScaling", false},
+        {"lanczos3Scaling", false},
         {"vsync", false},
         {"defScreenW", 0},
         {"defScreenH", 0},
@@ -257,6 +258,7 @@ try { exp } catch (...) {}
     SET_OPT(fullscreen, boolean);
     SET_OPT(fixedAspectRatio, boolean);
     SET_OPT(smoothScaling, boolean);
+    SET_OPT(lanczos3Scaling, boolean);
     SET_OPT(winResizable, boolean);
     SET_OPT(vsync, boolean);
     SET_STRINGOPT(windowTitle, windowTitle);

src/config.h

@@ -43,6 +43,7 @@ struct Config {
     bool fullscreen;
     bool fixedAspectRatio;
     bool smoothScaling;
+    bool lanczos3Scaling;
     bool vsync;
     
     int defScreenW;

src/display/gl/gl-meta.cpp

@@ -24,6 +24,7 @@
 #include "sharedstate.h"
 #include "glstate.h"
 #include "quad.h"
+#include "config.h"
 
 namespace GLMeta
 {
@@ -144,10 +145,22 @@ static void _blitBegin(FBO::ID fbo, const Vec2i &size)
 		FBO::bind(fbo);
 		glState.viewport.pushSet(IntRect(0, 0, size.x, size.y));
 
-		SimpleShader &shader = shState->shaders().simple;
+		if (shState->config().lanczos3Scaling)
-		shader.bind();
+		{
-		shader.applyViewportProj();
+			Lanczos3Shader &shader = shState->shaders().lanczos3;
-		shader.setTranslation(Vec2i());
+			shader.bind();
+			shader.applyViewportProj();
+			shader.setTranslation(Vec2i());
+			shader.setTexSize(Vec2i(size.x, size.y));
+		}
+		else
+		{
+			SimpleShader &shader = shState->shaders().simple;
+			shader.bind();
+			shader.applyViewportProj();
+			shader.setTranslation(Vec2i());
+			shader.setTexSize(Vec2i(size.x, size.y));
+		}
 	}
 }
 
@@ -169,8 +182,18 @@ void blitSource(TEXFBO &source)
 	}
 	else
 	{
-		SimpleShader &shader = shState->shaders().simple;
+		if (shState->config().lanczos3Scaling)
-		shader.setTexSize(Vec2i(source.width, source.height));
+		{
+			Lanczos3Shader &shader = shState->shaders().lanczos3;
+			shader.bind();
+			shader.setTexSize(Vec2i(source.width, source.height));
+		}
+		else
+		{
+			SimpleShader &shader = shState->shaders().simple;
+			shader.bind();
+			shader.setTexSize(Vec2i(source.width, source.height));
+		}
 		TEX::bind(source.tex);
 	}
 }

src/display/gl/shader.cpp

@@ -44,6 +44,7 @@
 #include "simpleAlphaUni.frag.xxd"
 #include "tilemap.frag.xxd"
 #include "flashMap.frag.xxd"
+#include "lanczos3.frag.xxd"
 #include "minimal.vert.xxd"
 #include "simple.vert.xxd"
 #include "simpleColor.vert.xxd"
@@ -746,3 +747,19 @@ void BltShader::setOpacity(float value)
 {
 	gl.Uniform1f(u_opacity, value);
 }
+
+Lanczos3Shader::Lanczos3Shader()
+{
+	INIT_SHADER(simple, lanczos3, Lanczos3Shader);
+
+	ShaderBase::init();
+
+	GET_U(texOffsetX);
+	GET_U(sourceSize);
+}
+
+void Lanczos3Shader::setTexSize(const Vec2i &value)
+{
+	ShaderBase::setTexSize(value);
+	gl.Uniform2f(u_sourceSize, (float)value.x, (float)value.y);
+}

src/display/gl/shader.h

@@ -113,7 +113,7 @@ public:
 
 	void setTexOffsetX(int value);
 
-private:
+protected:
 	GLint u_texOffsetX;
 };
 
@@ -326,6 +326,17 @@ private:
 	GLint u_source, u_destination, u_subRect, u_opacity;
 };
 
+class Lanczos3Shader : public SimpleShader
+{
+public:
+	Lanczos3Shader();
+
+	void setTexSize(const Vec2i &value);
+
+protected:
+	GLint u_sourceSize;
+};
+
 /* Global object containing all available shaders */
 struct ShaderSet
 {
@@ -347,6 +358,7 @@ struct ShaderSet
 	SimpleMatrixShader simpleMatrix;
 	BlurShader blur;
 	TilemapVXShader tilemapVX;
+	Lanczos3Shader lanczos3;
 };
 
 #endif // SHADER_H