Graphics: Optimize Viewport effect rendering

Using the kitchen sink plane shader for viewport effects, even
if only a small part of them are active, incurs great performance
loss on mobile, so split the rendering into multiple optional
passes which additionally use the blending hardware for faster
mixing (lerping).
Also, don't mirror the PingPong textures if the viewport effect
covers the entire screen area anyway.

CMakeLists.txt

@@ -216,12 +216,15 @@ set(EMBEDDED_INPUT
 	shader/hue.frag
 	shader/sprite.frag
 	shader/plane.frag
+	shader/gray.frag
 	shader/bitmapBlit.frag
+	shader/flatColor.frag
 	shader/simple.frag
 	shader/simpleColor.frag
 	shader/simpleAlpha.frag
 	shader/simpleAlphaUni.frag
 	shader/flashMap.frag
+	shader/minimal.vert
 	shader/simple.vert
 	shader/simpleColor.vert
 	shader/sprite.vert

mkxp.pro

@@ -189,12 +189,15 @@ EMBED = \
 	shader/hue.frag \
 	shader/sprite.frag \
 	shader/plane.frag \
+	shader/gray.frag \
 	shader/bitmapBlit.frag \
+	shader/flatColor.frag \
 	shader/simple.frag \
 	shader/simpleColor.frag \
 	shader/simpleAlpha.frag \
 	shader/simpleAlphaUni.frag \
 	shader/flashMap.frag \
+	shader/minimal.vert \
 	shader/simple.vert \
 	shader/simpleColor.vert \
 	shader/sprite.vert \

shader/flatColor.frag

@@ -0,0 +1,7 @@
+
+uniform lowp vec4 color;
+
+void main()
+{
+	gl_FragColor = color;
+}

shader/gray.frag

@@ -0,0 +1,19 @@
+
+uniform sampler2D texture;
+uniform lowp float gray;
+
+varying vec2 v_texCoord;
+
+const vec3 lumaF = vec3(.299, .587, .114);
+
+void main()
+{
+	/* Sample source color */
+	vec4 frag = texture2D(texture, v_texCoord);
+
+	/* Apply gray */
+	float luma = dot(frag.rgb, lumaF);
+	frag.rgb = mix(frag.rgb, vec3(luma), gray);
+
+	gl_FragColor = frag;
+}

shader/minimal.vert

@@ -0,0 +1,8 @@
+
+uniform mat4 projMat;
+attribute vec2 position;
+
+void main()
+{
+	gl_Position = projMat * vec4(position, 0, 1);
+}

src/etc-internal.h

@@ -60,6 +60,11 @@ struct Vec4
 	{
 		return (x == other.x && y == other.y && z == other.z && w == other.w);
 	}
+
+	bool xyzHasEffect() const
+	{
+		return (x != 0.0 || y != 0.0 || z != 0.0);
+	}
 };
 
 struct Vec2i
@@ -129,6 +134,14 @@ struct IntRect : SDL_Rect
 		SDL_Rect r = { x, y, w, h };
 		return r;
 	}
+
+	bool encloses(const IntRect &o) const
+	{
+		return (x   <= o.x &&
+		        y   <= o.y &&
+		        x+w >= o.x+o.w &&
+		        y+h >= o.y+o.h);
+	}
 };
 
 struct StaticRect { float x, y, w, h; };

src/graphics.cpp

@@ -171,36 +171,112 @@ class ScreenScene : public Scene
 
 	void requestViewportRender(Vec4 &c, Vec4 &f, Vec4 &t)
 	{
-		pp.swapRender();
+		const IntRect &viewpRect = glState.scissorBox.get();
+		const IntRect &screenRect = geometry.rect;
 
-		/* Scissor test _does_ affect FBO blit operations,
-		 * and since we're inside the draw cycle, it will
-		 * be turned on, so turn it off temporarily */
-		glState.scissorTest.pushSet(false);
+		if (t.w != 0.0)
+		{
+			pp.swapRender();
 
-		GLMeta::blitBegin(pp.frontBuffer());
-		GLMeta::blitSource(pp.backBuffer());
-		GLMeta::blitRectangle(geometry.rect, Vec2i());
-		GLMeta::blitEnd();
+			if (!viewpRect.encloses(screenRect))
+			{
+				/* Scissor test _does_ affect FBO blit operations,
+				 * and since we're inside the draw cycle, it will
+				 * be turned on, so turn it off temporarily */
+				glState.scissorTest.pushSet(false);
+
+				GLMeta::blitBegin(pp.frontBuffer());
+				GLMeta::blitSource(pp.backBuffer());
+				GLMeta::blitRectangle(geometry.rect, Vec2i());
+				GLMeta::blitEnd();
+
+				glState.scissorTest.pop();
+			}
+
+			GrayShader &shader = shState->shaders().gray;
+			shader.bind();
+			shader.setGray(t.w);
+			shader.applyViewportProj();
+			shader.setTexSize(screenRect.size());
+
+			TEX::bind(pp.backBuffer().tex);
+
+			glState.blend.pushSet(false);
+			screenQuad.draw();
+			glState.blend.pop();
+		}
 
-		glState.scissorTest.pop();
+		bool toneEffect = t.xyzHasEffect();
+		bool colorEffect = c.xyzHasEffect();
+		bool flashEffect = f.xyzHasEffect();
 
-		PlaneShader &shader = shState->shaders().plane;
+		if (!toneEffect && !colorEffect && !flashEffect)
+			return;
+
+		FlatColorShader &shader = shState->shaders().flatColor;
 		shader.bind();
-		shader.setColor(c);
-		shader.setFlash(f);
-		shader.setTone(t);
-		shader.setOpacity(1.0);
 		shader.applyViewportProj();
-		shader.setTexSize(geometry.rect.size());
 
-		TEX::bind(pp.backBuffer().tex);
+		/* Apply tone */
+		if (toneEffect)
+		{
+			/* First split up additive / substractive components */
+			Vec4 add, sub;
+
+			if (t.x > 0)
+				add.x = t.x;
+			if (t.y > 0)
+				add.y = t.y;
+			if (t.z > 0)
+				add.z = t.z;
+
+			if (t.x < 0)
+				sub.x = -t.x;
+			if (t.y < 0)
+				sub.y = -t.y;
+			if (t.z < 0)
+				sub.z = -t.z;
+
+			/* Then apply them using hardware blending */
+			gl.BlendFuncSeparate(GL_ONE, GL_ONE, GL_ZERO, GL_ONE);
+
+			if (add.xyzHasEffect())
+			{
+				gl.BlendEquation(GL_FUNC_ADD);
+				shader.setColor(add);
+
+				screenQuad.draw();
+			}
+
+			if (sub.xyzHasEffect())
+			{
+				gl.BlendEquation(GL_FUNC_REVERSE_SUBTRACT);
+				shader.setColor(sub);
+
+				screenQuad.draw();
+			}
+		}
 
-		glState.blend.pushSet(false);
+		if (colorEffect || flashEffect)
+		{
+			gl.BlendEquation(GL_FUNC_ADD);
+			gl.BlendFuncSeparate(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA,
+			                     GL_ZERO, GL_ONE);
+		}
 
-		screenQuad.draw();
+		if (colorEffect)
+		{
+			shader.setColor(c);
+			screenQuad.draw();
+		}
+
+		if (flashEffect)
+		{
+			shader.setColor(f);
+			screenQuad.draw();
+		}
 
-		glState.blend.pop();
+		glState.blendMode.refresh();
 	}
 
 	void setBrightness(float norm)

src/shader.cpp

@@ -35,11 +35,14 @@
 #include "transSimple.frag.xxd"
 #include "bitmapBlit.frag.xxd"
 #include "plane.frag.xxd"
+#include "gray.frag.xxd"
+#include "flatColor.frag.xxd"
 #include "simple.frag.xxd"
 #include "simpleColor.frag.xxd"
 #include "simpleAlpha.frag.xxd"
 #include "simpleAlphaUni.frag.xxd"
 #include "flashMap.frag.xxd"
+#include "minimal.vert.xxd"
 #include "simple.vert.xxd"
 #include "simpleColor.vert.xxd"
 #include "sprite.vert.xxd"
@@ -269,6 +272,21 @@ void ShaderBase::setTranslation(const Vec2i &value)
 }
 
 
+FlatColorShader::FlatColorShader()
+{
+	INIT_SHADER(minimal, flatColor, FlatColorShader);
+
+	ShaderBase::init();
+
+	GET_U(color);
+}
+
+void FlatColorShader::setColor(const Vec4 &value)
+{
+	setVec4Uniform(u_color, value);
+}
+
+
 SimpleShader::SimpleShader()
 {
 	INIT_SHADER(simple, simple, SimpleShader);
@@ -480,6 +498,21 @@ void PlaneShader::setOpacity(float value)
 }
 
 
+GrayShader::GrayShader()
+{
+	INIT_SHADER(simple, gray, GrayShader);
+
+	ShaderBase::init();
+
+	GET_U(gray);
+}
+
+void GrayShader::setGray(float value)
+{
+	gl.Uniform1f(u_gray, value);
+}
+
+
 TilemapShader::TilemapShader()
 {
 	INIT_SHADER(tilemap, simple, TilemapShader);

src/shader.h

@@ -87,6 +87,17 @@ class ShaderBase : public Shader
 	GLint u_texSizeInv, u_translation;
 };
 
+class FlatColorShader : public ShaderBase
+{
+public:
+	FlatColorShader();
+
+	void setColor(const Vec4 &value);
+
+private:
+	GLint u_color;
+};
+
 class SimpleShader : public ShaderBase
 {
 public:
@@ -191,6 +202,17 @@ class PlaneShader : public ShaderBase
 	GLint u_tone, u_color, u_flash, u_opacity;
 };
 
+class GrayShader : public ShaderBase
+{
+public:
+	GrayShader();
+
+	void setGray(float value);
+
+private:
+	GLint u_gray;
+};
+
 class TilemapShader : public ShaderBase
 {
 public:
@@ -285,13 +307,15 @@ class BltShader : public ShaderBase
 /* Global object containing all available shaders */
 struct ShaderSet
 {
+	FlatColorShader flatColor;
 	SimpleShader simple;
 	SimpleColorShader simpleColor;
 	SimpleAlphaShader simpleAlpha;
 	SimpleSpriteShader simpleSprite;
 	AlphaSpriteShader alphaSprite;
 	SpriteShader sprite;
 	PlaneShader plane;
+	GrayShader gray;
 	TilemapShader tilemap;
 	FlashMapShader flashMap;
 	TransShader trans;