Add bicubic scaling

mkxp.json

@@ -81,7 +81,7 @@
     // is upscaled
     // 0: Nearest-Neighbor
     // 1: Bilinear
-    // 2: (Reserved)
+    // 2: Bicubic
     // 3: Lanczos3
     // (default: 0)
     //

shader/bicubic.frag

@@ -0,0 +1,41 @@
+// From https://raw.githubusercontent.com/Sentmoraap/doing-sdl-right/f1a0183692abbd5d899fb432ab8dafe228a4929a/assets/bicubic.frag
+// Copyright 2020 Lilian Gimenez (Sentmoraap).
+// mkxp-z modifications Copyright 2022-2023 Splendide Imaginarius.
+// MIT license.
+
+uniform sampler2D texture;
+uniform vec2 sourceSize;
+uniform vec2 texSizeInv;
+varying vec2 v_texCoord;
+uniform vec2 bc;
+
+void main()
+{
+	vec2 pixel = v_texCoord * sourceSize + 0.5;
+	vec2 frac = fract(pixel);
+	vec2 frac2 = frac * frac;
+	vec2 frac3 = frac * frac2;
+	vec2 onePixel = texSizeInv;
+	pixel = floor(pixel) * texSizeInv - onePixel / 2.0;
+	vec4 colours[4];
+	// 16 reads, unoptimized but forks for every Mitchell-Netravali filter
+	for(int i = -1; i <= 2; i++)
+	{
+		vec4 p0 = texture2D(texture, pixel + vec2(   -onePixel.x, float(i) * onePixel.y)).rgba;
+		vec4 p1 = texture2D(texture, pixel + vec2(             0, float(i) * onePixel.y)).rgba;
+		vec4 p2 = texture2D(texture, pixel + vec2(    onePixel.x, float(i) * onePixel.y)).rgba;
+		vec4 p3 = texture2D(texture, pixel + vec2(2.0 * onePixel.x, float(i) * onePixel.y)).rgba;
+		colours[i + 1] = ((-bc.x / 6.0 - bc . y) * p0 + (- 1.5 * bc.x - bc.y + 2.0) * p1
+				+ (1.5 * bc.x + bc.y - 2.0) * p2 + (bc.x / 6.0 + bc.y) * p3) * frac3.x
+				+ ((0.5 * bc.x + 2.0 * bc.y) * p0 + (2.0 * bc.x + bc.y - 3.0) * p1
+				+ (-2.5 * bc.x - 2.0 * bc.y + 3.0) * p2 - bc.y * p3) * frac2.x
+				+ ((-0.5 * bc.x - bc.y) * p0 + (0.5 * bc.x + bc.y) * p2) * frac.x
+				+ p0 * bc.x / 6.0 + (-bc.x / 3.0 + 1.0) * p1 + p2 * bc.x / 6.0;
+	}
+	gl_FragColor = ((-bc.x / 6.0 - bc . y) * colours[0] + (- 1.5 * bc.x - bc.y + 2.0) * colours[1]
+			+ (1.5 * bc.x + bc.y - 2.0) * colours[2] + (bc.x / 6.0 + bc.y) * colours[3]) * frac3.y
+			+ ((0.5 * bc.x + 2.0 * bc.y) * colours[0] + (2.0 * bc.x + bc.y - 3.0) * colours[1]
+			+ (-2.5 * bc.x - 2.0 * bc.y + 3.0) * colours[2] - bc.y * colours[3]) * frac2.y
+			+ ((-0.5 * bc.x - bc.y) * colours[0] + (0.5 * bc.x + bc.y) * colours[2]) * frac.y
+			+ colours[0] * bc.x / 6.0 + (-bc.x / 3.0 + 1.0) * colours[1] + colours[2] * bc.x / 6.0;
+}

shader/meson.build

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

src/display/gl/gl-meta.cpp

@@ -154,6 +154,16 @@ static void _blitBegin(FBO::ID fbo, const Vec2i &size)
 
 		switch (shState->config().smoothScaling)
 		{
+		case Bicubic:
+		{
+			BicubicShader &shader = shState->shaders().bicubic;
+			shader.bind();
+			shader.applyViewportProj();
+			shader.setTranslation(Vec2i());
+			shader.setTexSize(Vec2i(size.x, size.y));
+		}
+
+			break;
 		case Lanczos3:
 		{
 			Lanczos3Shader &shader = shState->shaders().lanczos3;
@@ -234,6 +244,14 @@ void blitSource(TEXFBO &source)
 	{
 		switch (shState->config().smoothScaling)
 		{
+		case Bicubic:
+		{
+			BicubicShader &shader = shState->shaders().bicubic;
+			shader.bind();
+			shader.setTexSize(Vec2i(blitSrcWidthHires, blitSrcHeightHires));
+		}
+
+			break;
 		case Lanczos3:
 		{
 			Lanczos3Shader &shader = shState->shaders().lanczos3;

src/display/gl/shader.cpp

@@ -46,6 +46,7 @@
 #include "simpleAlphaUni.frag.xxd"
 #include "tilemap.frag.xxd"
 #include "flashMap.frag.xxd"
+#include "bicubic.frag.xxd"
 #include "lanczos3.frag.xxd"
 #include "minimal.vert.xxd"
 #include "simple.vert.xxd"
@@ -767,6 +768,20 @@ void BltShader::setOpacity(float value)
 	gl.Uniform1f(u_opacity, value);
 }
 
+BicubicShader::BicubicShader()
+{
+	INIT_SHADER(simple, bicubic, BicubicShader);
+
+	ShaderBase::init();
+
+	GET_U(texOffsetX);
+	GET_U(sourceSize);
+	GET_U(bc);
+
+	// TODO: Maybe expose this as a setting?
+	gl.Uniform2f(u_bc, 0.0, 0.5);
+}
+
 Lanczos3Shader::Lanczos3Shader()
 {
 	INIT_SHADER(simple, lanczos3, Lanczos3Shader);

src/display/gl/shader.h

@@ -341,6 +341,15 @@ protected:
 	GLint u_sourceSize;
 };
 
+class BicubicShader : public Lanczos3Shader
+{
+public:
+	BicubicShader();
+
+protected:
+	GLint u_bc;
+};
+
 /* Global object containing all available shaders */
 struct ShaderSet
 {
@@ -362,6 +371,7 @@ struct ShaderSet
 	SimpleMatrixShader simpleMatrix;
 	BlurShader blur;
 	TilemapVXShader tilemapVX;
+	BicubicShader bicubic;
 	Lanczos3Shader lanczos3;
 };
 

src/etc/etc.h

@@ -204,7 +204,7 @@ enum InterpolationMethod
 {
 	NearestNeighbor = 0,
 	Bilinear = 1,
-	// Reserving 2 for Bicubic
+	Bicubic = 2,
 	Lanczos3 = 3,
 };