Merge pull request #70 from wheremyfoodat/gl_state

More efficient OpenGL state management

CMakeLists.txt

@@ -83,8 +83,8 @@ else()
     message(FATAL_ERROR "Currently unsupported CPU architecture")
 endif()
 
-set(SOURCE_FILES src/main.cpp src/emulator.cpp src/io_file.cpp src/core/CPU/cpu_dynarmic.cpp src/core/CPU/dynarmic_cycles.cpp
+set(SOURCE_FILES src/main.cpp src/emulator.cpp src/io_file.cpp src/gl_state.cpp src/core/CPU/cpu_dynarmic.cpp
-                 src/core/memory.cpp
+                 src/core/CPU/dynarmic_cycles.cpp src/core/memory.cpp
 )
 set(CRYPTO_SOURCE_FILES src/core/crypto/aes_engine.cpp)
 set(KERNEL_SOURCE_FILES src/core/kernel/kernel.cpp src/core/kernel/resource_limits.cpp
@@ -138,7 +138,7 @@ set(HEADER_FILES include/emulator.hpp include/helpers.hpp include/opengl.hpp inc
                  include/PICA/dynapica/shader_rec_emitter_x64.hpp include/PICA/pica_hash.hpp include/result/result.hpp
                  include/result/result_common.hpp include/result/result_fs.hpp include/result/result_fnd.hpp
                  include/result/result_gsp.hpp include/result/result_kernel.hpp include/result/result_os.hpp
-                 include/crypto/aes_engine.hpp include/metaprogramming.hpp include/PICA/pica_vertex.hpp
+                 include/crypto/aes_engine.hpp include/metaprogramming.hpp include/PICA/pica_vertex.hpp include/gl_state.hpp
 )
 
 set(THIRD_PARTY_SOURCE_FILES third_party/imgui/imgui.cpp

include/PICA/gpu.hpp

@@ -28,7 +28,7 @@ class GPU {
 	std::array<vec4f, 16> currentAttributes; // Vertex attributes before being passed to the shader
 
 	std::array<vec4f, 16> immediateModeAttributes; // Vertex attributes uploaded via immediate mode submission
-	std::array<PicaVertex, 3> immediateModeVertices;
+	std::array<PICA::Vertex, 3> immediateModeVertices;
 	uint immediateModeVertIndex;
 	uint immediateModeAttrIndex; // Index of the immediate mode attribute we're uploading
 
@@ -68,7 +68,7 @@ class GPU {
 	u32* cmdBuffCurr = nullptr;
 
 	Renderer renderer;
-	PicaVertex getImmediateModeVertex();
+	PICA::Vertex getImmediateModeVertex();
 
   public:
 	// 256 entries per LUT with each LUT as its own row forming a 2D image 256 * LUT_COUNT
@@ -81,7 +81,7 @@ class GPU {
 	// Set to false by the renderer when the lighting_lut is uploaded ot the GPU
 	bool lightingLUTDirty = false;
 
-	GPU(Memory& mem);
+	GPU(Memory& mem, GLStateManager& gl);
 	void initGraphicsContext() { renderer.initGraphicsContext(); }
 	void getGraphicsContext() { renderer.getGraphicsContext(); }
 	void display() { renderer.display(); }

include/PICA/pica_vertex.hpp

@@ -2,35 +2,37 @@
 #include "PICA/float_types.hpp"
 #include <array>
 
-// A representation of the output vertex as it comes out of the vertex shader, with padding and all
+namespace PICA {
-struct PicaVertex {
+	// A representation of the output vertex as it comes out of the vertex shader, with padding and all
-	using vec2f = std::array<Floats::f24, 2>;
+	struct Vertex {
-	using vec3f = std::array<Floats::f24, 3>;
+		using vec2f = std::array<Floats::f24, 2>;
-	using vec4f = std::array<Floats::f24, 4>;
+		using vec3f = std::array<Floats::f24, 3>;
+		using vec4f = std::array<Floats::f24, 4>;
 
-	union {
+		union {
-		struct {
+			struct {
-			vec4f positions;   // Vertex position
+				vec4f positions;          // Vertex position
-			vec4f quaternion;  // Quaternion specifying the normal/tangent frame (for fragment lighting)
+				vec4f quaternion;         // Quaternion specifying the normal/tangent frame (for fragment lighting)
-			vec4f colour;      // Vertex color
+				vec4f colour;             // Vertex color
-			vec2f texcoord0;   // Texcoords for texture unit 0 (Only U and V, W is stored separately for 3D textures!)
+				vec2f texcoord0;          // Texcoords for texture unit 0 (Only U and V, W is stored separately for 3D textures!)
-			vec2f texcoord1;   // Texcoords for TU 1
+				vec2f texcoord1;          // Texcoords for TU 1
-			Floats::f24 texcoord0_w;   // W component for texcoord 0 if using a 3D texture
+				Floats::f24 texcoord0_w;  // W component for texcoord 0 if using a 3D texture
-			u32 padding;       // Unused
+				u32 padding;              // Unused
 
-			vec3f view;       // View vector (for fragment lighting)
+				vec3f view;       // View vector (for fragment lighting)
-			u32 padding2;     // Unused
+				u32 padding2;     // Unused
-			vec2f texcoord2;  // Texcoords for TU 2
+				vec2f texcoord2;  // Texcoords for TU 2
-		} s;
+			} s;
 
-		// The software, non-accelerated vertex loader writes here and then reads specific components from the above struct
+			// The software, non-accelerated vertex loader writes here and then reads specific components from the above struct
-		Floats::f24 raw[0x20];
+			Floats::f24 raw[0x20];
+		};
+		Vertex() {}
 	};
-	PicaVertex() {}
+}  // namespace PICA
-};
 
 // Float is used here instead of Floats::f24 to ensure that Floats::f24 is properly sized for direct interpretations as a float by the render backend
-#define ASSERT_POS(member, pos) static_assert(offsetof(PicaVertex, s.member) == pos * sizeof(float), "PicaVertex struct is broken!");
+#define ASSERT_POS(member, pos) static_assert(offsetof(PICA::Vertex, s.member) == pos * sizeof(float), "PICA::Vertex struct is broken!");
 
 ASSERT_POS(positions, 0)
 ASSERT_POS(quaternion, 4)

include/emulator.hpp

@@ -11,7 +11,7 @@
 #include "crypto/aes_engine.hpp"
 #include "io_file.hpp"
 #include "memory.hpp"
-#include "opengl.hpp"
+#include "gl_state.hpp"
 
 enum class ROMType { None, ELF, NCSD };
 
@@ -22,6 +22,7 @@ class Emulator {
 	Kernel kernel;
 	Crypto::AESEngine aesEngine;
 
+	GLStateManager gl;
 	SDL_Window* window;
 	SDL_GLContext glContext;
 	SDL_GameController* gameController;
@@ -56,5 +57,5 @@ class Emulator {
 	bool loadNCSD(const std::filesystem::path& path);
 	bool loadELF(const std::filesystem::path& path);
 	bool loadELF(std::ifstream& file);
-	void initGraphicsContext() { gpu.initGraphicsContext(); }
+	void initGraphicsContext();
 };

include/gl_state.hpp

@@ -0,0 +1,140 @@
+#pragma once
+#include <type_traits>
+
+#include "opengl.hpp"
+
+// GL state manager object for use in the OpenGL GPU renderer and potentially other things in the future (such as a potential ImGui GUI)
+// This object is meant to help us avoid duplicate OpenGL calls (such as binding the same program twice, enabling/disabling a setting twice, etc)
+// by checking if we actually *need* a state change. This is meant to avoid expensive driver calls and minimize unneeded state changes
+// A lot of code is in the header file instead of the relevant source file to make sure stuff gets inlined even without LTO, and
+// because this header should ideally not be getting included in too many places
+// Code that does not need inlining however, like the reset() function should be in gl_state.cpp
+// This state manager may not handle every aspect of OpenGL, in which case anything not handled here should just be manipulated with raw
+// OpenGL/opengl.hpp calls However, anything that can be handled through the state manager should, or at least there should be an attempt to keep it
+// consistent with the current GL state to avoid bugs/suboptimal code.
+
+// The state manager must *also* be a trivially constructible/destructible type, to ensure that no OpenGL functions get called sneakily without us
+// knowing. This is important for when we want to eg add a Vulkan or misc backend. Would definitely not want to refactor all this. So we try to be as
+// backend-agnostic as possible
+
+struct GLStateManager {
+	bool blendEnabled;
+	bool depthEnabled;
+	bool scissorEnabled;
+
+	// Colour/depth masks
+	bool redMask, greenMask, blueMask, alphaMask;
+	bool depthMask;
+
+	GLuint boundVAO;
+	GLuint boundVBO;
+	GLuint currentProgram;
+
+	GLenum depthFunc;
+
+	void reset();
+	void resetBlend();
+	void resetColourMask();
+	void resetDepth();
+	void resetVAO();
+	void resetVBO();
+	void resetProgram();
+	void resetScissor();
+
+	void enableDepth() {
+		if (!depthEnabled) {
+			depthEnabled = true;
+			OpenGL::enableDepth();
+		}
+	}
+
+	void disableDepth() {
+		if (depthEnabled) {
+			depthEnabled = false;
+			OpenGL::disableDepth();
+		}
+	}
+
+	void enableBlend() {
+		if (!blendEnabled) {
+			blendEnabled = true;
+			OpenGL::enableBlend();
+		}
+	}
+
+	void disableBlend() {
+		if (blendEnabled) {
+			blendEnabled = false;
+			OpenGL::disableBlend();
+		}
+	}
+
+	void enableScissor() {
+		if (!scissorEnabled) {
+			scissorEnabled = true;
+			OpenGL::enableScissor();
+		}
+	}
+
+	void disableScissor() {
+		if (scissorEnabled) {
+			scissorEnabled = false;
+			OpenGL::disableScissor();
+		}
+	}
+
+	void bindVAO(GLuint handle) {
+		if (boundVAO != handle) {
+			boundVAO = handle;
+			glBindVertexArray(handle);
+		}
+	}
+
+	void bindVBO(GLuint handle) {
+		if (boundVBO != handle) {
+			boundVBO = handle;
+			glBindBuffer(GL_ARRAY_BUFFER, handle);
+		}
+	}
+
+	void useProgram(GLuint handle) {
+		if (currentProgram != handle) {
+			currentProgram = handle;
+			glUseProgram(handle);
+		}
+	}
+
+	void bindVAO(const OpenGL::VertexArray& vao) { bindVAO(vao.handle()); }
+	void bindVBO(const OpenGL::VertexBuffer& vbo) { bindVBO(vbo.handle()); }
+	void useProgram(const OpenGL::Program& program) { useProgram(program.handle()); }
+
+	void setColourMask(bool r, bool g, bool b, bool a) {
+		if (r != redMask || g != greenMask || b != blueMask || a != alphaMask) {
+			r = redMask;
+			g = greenMask;
+			b = blueMask;
+			a = alphaMask;
+
+			OpenGL::setColourMask(r, g, b, a);
+		}
+	}
+
+	void setDepthMask(bool mask) {
+		if (depthMask != mask) {
+			depthMask = mask;
+			OpenGL::setDepthMask(mask);
+		}
+	}
+
+	void setDepthFunc(GLenum func) {
+		if (depthFunc != func) {
+			depthFunc = func;
+			glDepthFunc(func);
+		}
+	}
+
+	void setDepthFunc(OpenGL::DepthFunc func) { setDepthFunc(static_cast<GLenum>(func)); }
+};
+
+static_assert(std::is_trivially_constructible<GLStateManager>(), "OpenGL State Manager class is not trivially constructible!");
+static_assert(std::is_trivially_destructible<GLStateManager>(), "OpenGL State Manager class is not trivially destructible!");

include/opengl.hpp

@@ -128,9 +128,9 @@ namespace OpenGL {
 #ifdef OPENGL_DESTRUCTORS
         ~VertexArray() { free(); }
 #endif
-        GLuint handle() { return m_handle; }
+        GLuint handle() const { return m_handle; }
-        bool exists() { return m_handle != 0; }
+        bool exists() const { return m_handle != 0; }
-        void bind() { glBindVertexArray(m_handle); }
+        void bind() const { glBindVertexArray(m_handle); }
 
         template <typename T>
         void setAttributeFloat(GLuint index, GLint size, GLsizei stride, const void* offset, bool normalized = GL_FALSE) {
@@ -299,11 +299,11 @@ namespace OpenGL {
 #ifdef OPENGL_DESTRUCTORS
         ~Texture() { free(); }
 #endif
-        GLuint handle() { return m_handle; }
+        GLuint handle() const { return m_handle; }
-        bool exists() { return m_handle != 0; }
+        bool exists() const { return m_handle != 0; }
-        void bind() { glBindTexture(m_binding, m_handle); }
+        void bind() const { glBindTexture(m_binding, m_handle); }
-        int width() { return m_width; }
+        int width() const { return m_width; }
-        int height() { return m_height; }
+        int height() const { return m_height; }
 
        void free() { glDeleteTextures(1, &m_handle); }
     };
@@ -327,10 +327,10 @@ namespace OpenGL {
 #ifdef OPENGL_DESTRUCTORS
         ~Framebuffer() { free(); }
 #endif
-        GLuint handle() { return m_handle; }
+        GLuint handle() const { return m_handle; }
-        bool exists() { return m_handle != 0; }
+        bool exists() const { return m_handle != 0; }
-        void bind(GLenum target) { glBindFramebuffer(target, m_handle); }
+        void bind(GLenum target) const { glBindFramebuffer(target, m_handle); }
-        void bind(FramebufferTypes target) { bind(static_cast<GLenum>(target)); }
+        void bind(FramebufferTypes target) const { bind(static_cast<GLenum>(target)); }
         void free() { glDeleteFramebuffers(1, &m_handle); }
 
         void createWithTexture(Texture& tex, GLenum mode = GL_FRAMEBUFFER, GLenum textureType = GL_TEXTURE_2D) {
@@ -392,8 +392,8 @@ namespace OpenGL {
             return m_handle != 0;
         }
 
-        GLuint handle() { return m_handle; }
+        GLuint handle() const { return m_handle; }
-        bool exists() { return m_handle != 0; }
+        bool exists() const { return m_handle != 0; }
     };
 
     struct Program {
@@ -421,9 +421,9 @@ namespace OpenGL {
             return m_handle != 0;
         }
 
-        GLuint handle() { return m_handle; }
+        GLuint handle() const { return m_handle; }
-        bool exists() { return m_handle != 0; }
+        bool exists() const { return m_handle != 0; }
-        void use() { glUseProgram(m_handle); }
+        void use() const { glUseProgram(m_handle); }
     };
 
     static void dispatchCompute(GLuint groupsX = 1, GLuint groupsY = 1, GLuint groupsZ = 1) {
@@ -454,9 +454,9 @@ namespace OpenGL {
 #ifdef OPENGL_DESTRUCTORS
         ~VertexBuffer() { free(); }
 #endif  
-        GLuint handle() { return m_handle; }
+        GLuint handle() const { return m_handle; }
-        bool exists() { return m_handle != 0; }
+        bool exists() const { return m_handle != 0; }
-        void bind() { glBindBuffer(GL_ARRAY_BUFFER, m_handle); }
+        void bind() const { glBindBuffer(GL_ARRAY_BUFFER, m_handle); }
         void free() { glDeleteBuffers(1, &m_handle); }
 
         // Reallocates the buffer on every call. Prefer the sub version if possible.
@@ -528,6 +528,8 @@ namespace OpenGL {
 	static void disableClipPlane(GLuint index) { glDisable(GL_CLIP_DISTANCE0 + index); }
 
     static void setDepthFunc(DepthFunc func) { glDepthFunc(static_cast<GLenum>(func)); }
+	static void setColourMask(GLboolean r, GLboolean g, GLboolean b, GLboolean a) { glColorMask(r, g, b, a); }
+	static void setDepthMask(GLboolean mask) { glDepthMask(mask); }
 
     enum Primitives {
         Triangle = GL_TRIANGLES,
@@ -667,23 +669,23 @@ namespace OpenGL {
     // We're never supporting 3D rectangles, because rectangles were never meant to be 3D in the first place
     // x, y: Coords of the top left vertex
     // width, height: Dimensions of the rectangle. Initialized to 0 if not specified.
-    template <typename T>
+	template <typename T>
-    struct Rectangle {
+	struct Rectangle {
-        T x, y, width, height;
+		T x, y, width, height;
 
-        std::pair<T, T> topLeft() { return std::make_pair(x, y); }
+		std::pair<T, T> topLeft() const { return std::make_pair(x, y); }
-        std::pair<T, T> topRight() { return std::make_pair(x + width, y); }
+		std::pair<T, T> topRight() const { return std::make_pair(x + width, y); }
-        std::pair<T, T> bottomLeft() { return std::make_pair(x, y + height); }
+		std::pair<T, T> bottomLeft() const { return std::make_pair(x, y + height); }
-        std::pair<T, T> bottomRight() { return std::make_pair(x + width, y + height); }
+		std::pair<T, T> bottomRight() const { return std::make_pair(x + width, y + height); }
 
-        Rectangle() : x(0), y(0), width(0), height(0) {}
+		Rectangle() : x(0), y(0), width(0), height(0) {}
-        Rectangle(T x, T y, T width, T height) : x(x), y(y), width(width), height(height) {}
+		Rectangle(T x, T y, T width, T height) : x(x), y(y), width(width), height(height) {}
 
-        bool isEmpty() { return width == 0 && height == 0; }
+		bool isEmpty() const { return width == 0 && height == 0; }
-        bool isLine() { return (width == 0 && height != 0) || (width != 0 && height == 0); }
+		bool isLine() const { return (width == 0 && height != 0) || (width != 0 && height == 0); }
 
-        void setEmpty() { x = y = width = height = 0; }
+		void setEmpty() { x = y = width = height = 0; }
-    };
+	};
 
     using Rect = Rectangle<GLuint>;
 

include/renderer_gl/renderer_gl.hpp

@@ -3,9 +3,9 @@
 #include <span>
 
 #include "PICA/float_types.hpp"
+#include "gl_state.hpp"
 #include "helpers.hpp"
 #include "logger.hpp"
-#include "opengl.hpp"
 #include "surface_cache.hpp"
 #include "textures.hpp"
 #include "PICA/regs.hpp"
@@ -16,13 +16,13 @@ class GPU;
 
 class Renderer {
 	GPU& gpu;
+	GLStateManager& gl;
+
 	OpenGL::Program triangleProgram;
 	OpenGL::Program displayProgram;
 
 	OpenGL::VertexArray vao;
 	OpenGL::VertexBuffer vbo;
-	GLint alphaControlLoc = -1;
-	GLint texUnitConfigLoc = -1;
 
 	// TEV configuration uniform locations
 	GLint textureEnvSourceLoc = -1;
@@ -30,8 +30,8 @@ class Renderer {
 	GLint textureEnvCombinerLoc = -1;
 	GLint textureEnvColorLoc = -1;
 	GLint textureEnvScaleLoc = -1;
-	GLint textureEnvUpdateBufferLoc = -1;
+	
-	GLint textureEnvBufferColorLoc = -1;
+	// Uniform of PICA registers
 	GLint picaRegLoc = -1;
 
 	// Depth configuration uniform locations
@@ -39,9 +39,6 @@ class Renderer {
 	GLint depthScaleLoc = -1;
 	GLint depthmapEnableLoc = -1;
 
-	u32 oldAlphaControl = 0;
-	u32 oldTexUnitConfig = 0;
-
 	float oldDepthScale = -1.0;
 	float oldDepthOffset = 0.0;
 	bool oldDepthmapEnable = false;
@@ -81,7 +78,7 @@ class Renderer {
 	void updateLightingLUT();
 
   public:
-	Renderer(GPU& gpu, const std::array<u32, regNum>& internalRegs) : gpu(gpu), regs(internalRegs) {}
+	Renderer(GPU& gpu, GLStateManager& gl, const std::array<u32, regNum>& internalRegs) : gpu(gpu), gl(gl), regs(internalRegs) {}
 
 	void reset();
 	void display();                                                                                 // Display the 3DS screen contents to the window
@@ -89,7 +86,7 @@ class Renderer {
 	void getGraphicsContext();                                                                      // Set up graphics context for rendering
 	void clearBuffer(u32 startAddress, u32 endAddress, u32 value, u32 control);                     // Clear a GPU buffer in VRAM
 	void displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32 outputSize, u32 flags);  // Perform display transfer
-	void drawVertices(PICA::PrimType primType, std::span<const PicaVertex> vertices);               // Draw the given vertices
+	void drawVertices(PICA::PrimType primType, std::span<const PICA::Vertex> vertices);             // Draw the given vertices
 
 	void setFBSize(u32 width, u32 height) {
 		fbSize.x() = width;

src/core/PICA/gpu.cpp

@@ -10,8 +10,9 @@
 
 using namespace Floats;
 
-
+// Note: For when we have multiple backends, the GL state manager can stay here and have the constructor for the Vulkan-or-whatever renderer ignore it
-GPU::GPU(Memory& mem) : mem(mem), renderer(*this, regs) {
+// Thus, our GLStateManager being here does not negatively impact renderer-agnosticness
+GPU::GPU(Memory& mem, GLStateManager& gl) : mem(mem), renderer(*this, gl, regs) {
 	vram = new u8[vramSize];
 	mem.setVRAM(vram); // Give the bus a pointer to our VRAM
 }
@@ -22,6 +23,7 @@ void GPU::reset() {
 	shaderJIT.reset();
 	std::memset(vram, 0, vramSize);
 	lightingLUT.fill(0);
+	lightingLUTDirty = true;
 
 	totalAttribCount = 0;
 	fixedAttribMask = 0;
@@ -60,7 +62,7 @@ void GPU::drawArrays(bool indexed) {
 	}
 }
 
-static std::array<PicaVertex, Renderer::vertexBufferSize> vertices;
+static std::array<PICA::Vertex, Renderer::vertexBufferSize> vertices;
 
 template <bool indexed, bool useShaderJIT>
 void GPU::drawArrays() {
@@ -248,7 +250,7 @@ void GPU::drawArrays() {
 			shaderUnit.vs.run();
 		}
 
-		PicaVertex& out = vertices[i];
+		PICA::Vertex& out = vertices[i];
 		// Map shader outputs to fixed function properties
 		const u32 totalShaderOutputs = regs[PICA::InternalRegs::ShaderOutputCount] & 7;
 		for (int i = 0; i < totalShaderOutputs; i++) {
@@ -264,8 +266,8 @@ void GPU::drawArrays() {
 	renderer.drawVertices(primType, std::span(vertices).first(vertexCount));
 }
 
-PicaVertex GPU::getImmediateModeVertex() {
+PICA::Vertex GPU::getImmediateModeVertex() {
-	PicaVertex v;
+	PICA::Vertex v;
 	const int totalAttrCount = (regs[PICA::InternalRegs::VertexShaderAttrNum] & 0xf) + 1;
 
 	// Copy immediate mode attributes to vertex shader unit

src/core/PICA/regs.cpp

@@ -188,7 +188,7 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
 
 					immediateModeAttributes[immediateModeAttrIndex++] = attr;
 					if (immediateModeAttrIndex == totalAttrCount) {
-						PicaVertex v = getImmediateModeVertex();
+						PICA::Vertex v = getImmediateModeVertex();
 						immediateModeAttrIndex = 0;
 						immediateModeVertices[immediateModeVertIndex++] = v;
 

src/core/renderer_gl/renderer_gl.cpp

@@ -34,12 +34,11 @@ const char* vertexShader = R"(
 
 	// TEV uniforms
 	uniform uint u_textureEnvColor[6];
-	uniform uint u_textureEnvBufferColor;
+	uniform uint u_picaRegs[0x200 - 0x48];
-	uniform uint u_picaRegs[0x200 - 0x47];
 
 	// Helper so that the implementation of u_pica_regs can be changed later
 	uint readPicaReg(uint reg_addr){
-		return u_picaRegs[reg_addr - 0x47];
+		return u_picaRegs[reg_addr - 0x48];
 	}
 
 	vec4 abgr8888ToVec4(uint abgr) {
@@ -96,7 +95,7 @@ const char* vertexShader = R"(
 			v_textureEnvColor[i] = abgr8888ToVec4(u_textureEnvColor[i]);
 		}
 
-		v_textureEnvBufferColor = abgr8888ToVec4(u_textureEnvBufferColor);
+		v_textureEnvBufferColor = abgr8888ToVec4(readPicaReg(0xFD));
 
 		// Parse clipping plane registers
 		// The plane registers describe a clipping plane in the form of Ax + By + Cz + D = 0 
@@ -131,15 +130,11 @@ const char* fragmentShader = R"(
 
 	out vec4 fragColour;
 
-	uniform uint u_alphaControl;
-	uniform uint u_textureConfig;
-
 	// TEV uniforms
 	uniform uint u_textureEnvSource[6];
 	uniform uint u_textureEnvOperand[6];
 	uniform uint u_textureEnvCombiner[6];
 	uniform uint u_textureEnvScale[6];
-	uniform uint u_textureEnvUpdateBuffer;
 
 	// Depth control uniforms
 	uniform float u_depthScale;
@@ -151,11 +146,11 @@ const char* fragmentShader = R"(
 	uniform sampler2D u_tex2;
 	uniform sampler1DArray u_tex_lighting_lut;
 
-	uniform uint u_picaRegs[0x200 - 0x47];
+	uniform uint u_picaRegs[0x200 - 0x48];
 
 	// Helper so that the implementation of u_pica_regs can be changed later
 	uint readPicaReg(uint reg_addr){
-		return u_picaRegs[reg_addr - 0x47];
+		return u_picaRegs[reg_addr - 0x48];
 	}
 
 	vec4 tevSources[16];
@@ -340,7 +335,7 @@ const char* fragmentShader = R"(
 
 		bool error_unimpl = false;
 
-		for (uint i = 0; i < GPUREG_LIGHTING_NUM_LIGHTS; i++){
+		for (uint i = 0; i < GPUREG_LIGHTING_NUM_LIGHTS; i++) {
 			uint light_id = bitfieldExtract(GPUREG_LIGHTING_LIGHT_PERMUTATION,int(i*3),3);
 		
 			uint GPUREG_LIGHTi_SPECULAR0 = readPicaReg(0x0140 + 0x10 * light_id);
@@ -363,8 +358,8 @@ const char* fragmentShader = R"(
 
 			vec3 half_vector = normalize(normalize(light_vector) + view);
 
-			for(int c = 0; c < 7; c++){
+			for (int c = 0; c < 7; c++) {
-				if(bitfieldExtract(GPUREG_LIGHTING_CONFIG1, 16 + c, 1) == 0){
+				if (bitfieldExtract(GPUREG_LIGHTING_CONFIG1, 16 + c, 1) == 0){
 					uint scale_id = bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_SCALE, c * 4, 3);
 					float scale = float(1u << scale_id);
 					if (scale_id >= 6u)
@@ -404,15 +399,15 @@ const char* fragmentShader = R"(
 				d[D1_LUT] = 0.0;
 				d[FR_LUT] = 0.0;
 				d[RG_LUT]= d[RB_LUT] = d[RR_LUT];
-			} else if(lookup_config == 1) {
+			} else if (lookup_config == 1) {
 				d[D0_LUT] = 0.0;
 				d[D1_LUT] = 0.0;
 				d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
-			} else if(lookup_config == 2) {
+			} else if (lookup_config == 2) {
 				d[FR_LUT] = 0.0;
 				d[SP_LUT] = 0.0;
 				d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
-			} else if(lookup_config == 3) {
+			} else if (lookup_config == 3) {
 				d[SP_LUT] = 0.0;
 				d[RG_LUT]= d[RB_LUT] = d[RR_LUT] = 1.0;
 			} else if (lookup_config == 4) {
@@ -453,20 +448,22 @@ const char* fragmentShader = R"(
 	}
 
 	void main() {
-		vec2 tex2UV = (u_textureConfig & (1u << 13)) != 0u ? v_texcoord1 : v_texcoord2;
-
 		// TODO: what do invalid sources and disabled textures read as?
 		// And what does the "previous combiner" source read initially?
 		tevSources[0] = v_colour; // Primary/vertex color
 		calcLighting(tevSources[1],tevSources[2]);
 
-		if ((u_textureConfig & 1u) != 0u) tevSources[3] = texture(u_tex0, v_texcoord0.xy);
+		uint textureConfig = readPicaReg(0x80);
-		if ((u_textureConfig & 2u) != 0u) tevSources[4] = texture(u_tex1, v_texcoord1);
+		vec2 tex2UV = (textureConfig & (1u << 13)) != 0u ? v_texcoord1 : v_texcoord2;
-		if ((u_textureConfig & 4u) != 0u) tevSources[5] = texture(u_tex2, tex2UV);
+
+		if ((textureConfig & 1u) != 0u) tevSources[3] = texture(u_tex0, v_texcoord0.xy);
+		if ((textureConfig & 2u) != 0u) tevSources[4] = texture(u_tex1, v_texcoord1);
+		if ((textureConfig & 4u) != 0u) tevSources[5] = texture(u_tex2, tex2UV);
 		tevSources[13] = vec4(0.0); // Previous buffer
 		tevSources[15] = vec4(0.0); // Previous combiner
 
 		tevNextPreviousBuffer = v_textureEnvBufferColor;
+		uint textureEnvUpdateBuffer = readPicaReg(0xE0);
 
 		for (int i = 0; i < 6; i++) {
 			tevSources[14] = v_textureEnvColor[i]; // Constant color
@@ -474,11 +471,11 @@ const char* fragmentShader = R"(
 			tevSources[13] = tevNextPreviousBuffer;
 
 			if (i < 4) {
-				if ((u_textureEnvUpdateBuffer & (0x100u << i)) != 0u) {
+				if ((textureEnvUpdateBuffer & (0x100u << i)) != 0u) {
 					tevNextPreviousBuffer.rgb = tevSources[15].rgb;
 				}
 
-				if ((u_textureEnvUpdateBuffer & (0x1000u << i)) != 0u) {
+				if ((textureEnvUpdateBuffer & (0x1000u << i)) != 0u) {
 					tevNextPreviousBuffer.a = tevSources[15].a;
 				}
 			}
@@ -503,9 +500,11 @@ const char* fragmentShader = R"(
 		// Write final fragment depth
 		gl_FragDepth = depth;
 
-		if ((u_alphaControl & 1u) != 0u) { // Check if alpha test is on
+		// Perform alpha test
-			uint func = (u_alphaControl >> 4u) & 7u;
+		uint alphaControl = readPicaReg(0x104);
-			float reference = float((u_alphaControl >> 8u) & 0xffu) / 255.0;
+		if ((alphaControl & 1u) != 0u) { // Check if alpha test is on
+			uint func = (alphaControl >> 4u) & 7u;
+			float reference = float((alphaControl >> 8u) & 0xffu) / 255.0;
 			float alpha = fragColour.a;
 
 			switch (func) {
@@ -592,21 +591,17 @@ void Renderer::reset() {
 	if (triangleProgram.exists()) {
 		const auto oldProgram = OpenGL::getProgram();
 
-		triangleProgram.use();
+		gl.useProgram(triangleProgram);
-		oldAlphaControl = 0; // Default alpha control to 0
-		oldTexUnitConfig = 0; // Default tex unit config to 0
 		
 		oldDepthScale = -1.0; // Default depth scale to -1.0, which is what games typically use
 		oldDepthOffset = 0.0; // Default depth offset to 0
 		oldDepthmapEnable = false; // Enable w buffering
 
-		glUniform1ui(alphaControlLoc, oldAlphaControl);
-		glUniform1ui(texUnitConfigLoc, oldTexUnitConfig);
 		glUniform1f(depthScaleLoc, oldDepthScale);
 		glUniform1f(depthOffsetLoc, oldDepthOffset);
 		glUniform1i(depthmapEnableLoc, oldDepthmapEnable);
 
-		glUseProgram(oldProgram); // Switch to old GL program
+		gl.useProgram(oldProgram);  // Switch to old GL program
 	}
 }
 
@@ -614,18 +609,13 @@ void Renderer::initGraphicsContext() {
 	OpenGL::Shader vert(vertexShader, OpenGL::Vertex);
 	OpenGL::Shader frag(fragmentShader, OpenGL::Fragment);
 	triangleProgram.create({ vert, frag });
-	triangleProgram.use();
+	gl.useProgram(triangleProgram);
-	
-	alphaControlLoc = OpenGL::uniformLocation(triangleProgram, "u_alphaControl");
-	texUnitConfigLoc = OpenGL::uniformLocation(triangleProgram, "u_textureConfig");
 
 	textureEnvSourceLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvSource");
 	textureEnvOperandLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvOperand");
 	textureEnvCombinerLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvCombiner");
 	textureEnvColorLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvColor");
 	textureEnvScaleLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvScale");
-	textureEnvUpdateBufferLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvUpdateBuffer");
-	textureEnvBufferColorLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvBufferColor");
 
 	depthScaleLoc = OpenGL::uniformLocation(triangleProgram, "u_depthScale");
 	depthOffsetLoc = OpenGL::uniformLocation(triangleProgram, "u_depthOffset");
@@ -642,37 +632,37 @@ void Renderer::initGraphicsContext() {
 	OpenGL::Shader fragDisplay(displayFragmentShader, OpenGL::Fragment);
 	displayProgram.create({ vertDisplay, fragDisplay });
 
-	displayProgram.use();
+	gl.useProgram(displayProgram);
 	glUniform1i(OpenGL::uniformLocation(displayProgram, "u_texture"), 0); // Init sampler object
 
-	vbo.createFixedSize(sizeof(PicaVertex) * vertexBufferSize, GL_STREAM_DRAW);
+	vbo.createFixedSize(sizeof(Vertex) * vertexBufferSize, GL_STREAM_DRAW);
-	vbo.bind();
+	gl.bindVBO(vbo);
 	vao.create();
-	vao.bind();
+	gl.bindVAO(vao);
 
 	// Position (x, y, z, w) attributes
-	vao.setAttributeFloat<float>(0, 4, sizeof(PicaVertex), offsetof(PicaVertex, s.positions));
+	vao.setAttributeFloat<float>(0, 4, sizeof(Vertex), offsetof(Vertex, s.positions));
 	vao.enableAttribute(0);
 	// Quaternion attribute
-	vao.setAttributeFloat<float>(1, 4, sizeof(PicaVertex), offsetof(PicaVertex, s.quaternion));
+	vao.setAttributeFloat<float>(1, 4, sizeof(Vertex), offsetof(Vertex, s.quaternion));
 	vao.enableAttribute(1);
 	// Colour attribute
-	vao.setAttributeFloat<float>(2, 4, sizeof(PicaVertex), offsetof(PicaVertex, s.colour));
+	vao.setAttributeFloat<float>(2, 4, sizeof(Vertex), offsetof(Vertex, s.colour));
 	vao.enableAttribute(2);
 	// UV 0 attribute
-	vao.setAttributeFloat<float>(3, 2, sizeof(PicaVertex), offsetof(PicaVertex, s.texcoord0));
+	vao.setAttributeFloat<float>(3, 2, sizeof(Vertex), offsetof(Vertex, s.texcoord0));
 	vao.enableAttribute(3);
 	// UV 1 attribute
-	vao.setAttributeFloat<float>(4, 2, sizeof(PicaVertex), offsetof(PicaVertex, s.texcoord1));
+	vao.setAttributeFloat<float>(4, 2, sizeof(Vertex), offsetof(Vertex, s.texcoord1));
 	vao.enableAttribute(4);
 	// UV 0 W-component attribute
-	vao.setAttributeFloat<float>(5, 1, sizeof(PicaVertex), offsetof(PicaVertex, s.texcoord0_w));
+	vao.setAttributeFloat<float>(5, 1, sizeof(Vertex), offsetof(Vertex, s.texcoord0_w));
 	vao.enableAttribute(5);
 	// View
-	vao.setAttributeFloat<float>(6, 3, sizeof(PicaVertex), offsetof(PicaVertex, s.view));
+	vao.setAttributeFloat<float>(6, 3, sizeof(Vertex), offsetof(Vertex, s.view));
 	vao.enableAttribute(6);
 	// UV 2 attribute
-	vao.setAttributeFloat<float>(7, 2, sizeof(PicaVertex), offsetof(PicaVertex, s.texcoord2));
+	vao.setAttributeFloat<float>(7, 2, sizeof(Vertex), offsetof(Vertex, s.texcoord2));
 	vao.enableAttribute(7);
 
 	dummyVBO.create();
@@ -725,9 +715,9 @@ void Renderer::setupBlending() {
 	};
 
 	if (!blendingEnabled) {
-		OpenGL::disableBlend();
+		gl.disableBlend();
 	} else {
-		OpenGL::enableBlend();
+		gl.enableBlend();
 
 		// Get blending equations
 		const u32 blendControl = regs[PICA::InternalRegs::BlendFunc];
@@ -783,8 +773,6 @@ void Renderer::setupTextureEnvState() {
 	glUniform1uiv(textureEnvCombinerLoc, 6, textureEnvCombinerRegs);
 	glUniform1uiv(textureEnvColorLoc, 6, textureEnvColourRegs);
 	glUniform1uiv(textureEnvScaleLoc, 6, textureEnvScaleRegs);
-	glUniform1ui(textureEnvUpdateBufferLoc, regs[PICA::InternalRegs::TexEnvUpdateBuffer]);
-	glUniform1ui(textureEnvBufferColorLoc, regs[PICA::InternalRegs::TexEnvBufferColor]);
 }
 
 void Renderer::bindTexturesToSlots() {
@@ -815,21 +803,16 @@ void Renderer::bindTexturesToSlots() {
 	glActiveTexture(GL_TEXTURE0 + 3);
 	glBindTexture(GL_TEXTURE_1D_ARRAY, lightLUTTextureArray);
 	glActiveTexture(GL_TEXTURE0);
-
-	// Update the texture unit configuration uniform if it changed
-	const u32 texUnitConfig = regs[PICA::InternalRegs::TexUnitCfg];
-	if (oldTexUnitConfig != texUnitConfig) {
-		oldTexUnitConfig = texUnitConfig;
-		glUniform1ui(texUnitConfigLoc, texUnitConfig);
-	}
 }
-void Renderer::updateLightingLUT(){
+
+void Renderer::updateLightingLUT() {
+	gpu.lightingLUTDirty = false;
 	std::array<u16, GPU::LightingLutSize> u16_lightinglut; 
 	
-	for(int i = 0; i < gpu.lightingLUT.size(); i++){
+	for (int i = 0; i < gpu.lightingLUT.size(); i++) {
 		uint64_t value =  gpu.lightingLUT[i] & ((1 << 12) - 1);
 		u16_lightinglut[i] = value * 65535 / 4095; 
-	} 
+	}
 
 	glActiveTexture(GL_TEXTURE0 + 3);
 	glBindTexture(GL_TEXTURE_1D_ARRAY, lightLUTTextureArray);
@@ -839,31 +822,19 @@ void Renderer::updateLightingLUT(){
 	glTexParameteri(GL_TEXTURE_1D_ARRAY, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
 	glTexParameteri(GL_TEXTURE_1D_ARRAY, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
 	glActiveTexture(GL_TEXTURE0);
-	gpu.lightingLUTDirty = false;
 }
 
-void Renderer::drawVertices(PICA::PrimType primType, std::span<const PicaVertex> vertices) {
+void Renderer::drawVertices(PICA::PrimType primType, std::span<const Vertex> vertices) {
 	// The fourth type is meant to be "Geometry primitive". TODO: Find out what that is
 	static constexpr std::array<OpenGL::Primitives, 4> primTypes = {
 	  OpenGL::Triangle, OpenGL::TriangleStrip, OpenGL::TriangleFan, OpenGL::Triangle
 	};
 	const auto primitiveTopology = primTypes[static_cast<usize>(primType)];
 
-    // TODO: We should implement a GL state tracker that tracks settings like scissor, blending, bound program, etc
+	gl.disableScissor();
-    // This way if we attempt to eg do multiple glEnable(GL_BLEND) calls in a row, it will say "Oh blending is already enabled"
+	gl.bindVBO(vbo);
-    // And not actually perform the very expensive driver call for it
+	gl.bindVAO(vao);
-	OpenGL::disableScissor();
+	gl.useProgram(triangleProgram);
-
-	vbo.bind();
-	vao.bind();
-	triangleProgram.use();
-
-	// Adjust alpha test if necessary
-	const u32 alphaControl = regs[PICA::InternalRegs::AlphaTestConfig];
-	if (alphaControl != oldAlphaControl) {
-		oldAlphaControl = alphaControl;
-		glUniform1ui(alphaControlLoc, alphaControl);
-	}
 
 	OpenGL::enableClipPlane(0); // Clipping plane 0 is always enabled
 	if (regs[PICA::InternalRegs::ClipEnable] & 1) {
@@ -879,7 +850,7 @@ void Renderer::drawVertices(PICA::PrimType primType, std::span<const PicaVertex>
 	const bool depthWriteEnable = getBit<12>(depthControl);
 	const int depthFunc = getBits<4, 3>(depthControl);
 	const int colourMask = getBits<8, 4>(depthControl);
-	glColorMask(colourMask & 1, colourMask & 2, colourMask & 4, colourMask & 8);
+	gl.setColourMask(colourMask & 1, colourMask & 2, colourMask & 4, colourMask & 8);
 
 	static constexpr std::array<GLenum, 8> depthModes = {
 		GL_NEVER, GL_ALWAYS, GL_EQUAL, GL_NOTEQUAL, GL_LESS, GL_LEQUAL, GL_GREATER, GL_GEQUAL
@@ -908,9 +879,9 @@ void Renderer::drawVertices(PICA::PrimType primType, std::span<const PicaVertex>
 	setupTextureEnvState();
 	bindTexturesToSlots();
 
-	// Upload PICA Registers as a single uniform. The shader needs access to the rasterizer registers (for depth, starting from index 0x47)
+	// Upload PICA Registers as a single uniform. The shader needs access to the rasterizer registers (for depth, starting from index 0x48)
 	// The texturing and the fragment lighting registers. Therefore we upload them all in one go to avoid multiple slow uniform updates
-	glUniform1uiv(picaRegLoc, 0x200 - 0x47, &regs[0x47]);
+	glUniform1uiv(picaRegLoc, 0x200 - 0x48, &regs[0x48]);
 
 	if (gpu.lightingLUTDirty) {
 		updateLightingLUT();
@@ -924,18 +895,18 @@ void Renderer::drawVertices(PICA::PrimType primType, std::span<const PicaVertex>
 	// Note: The code below must execute after we've bound the colour buffer & its framebuffer
 	// Because it attaches a depth texture to the aforementioned colour buffer
 	if (depthEnable) {
-		OpenGL::enableDepth();
+		gl.enableDepth();
-		glDepthFunc(depthModes[depthFunc]);
+		gl.setDepthMask(depthWriteEnable ? GL_TRUE : GL_FALSE);
-		glDepthMask(depthWriteEnable ? GL_TRUE : GL_FALSE);
+		gl.setDepthFunc(depthModes[depthFunc]);
 		bindDepthBuffer();
 	} else {
 		if (depthWriteEnable) {
-			OpenGL::enableDepth();
+			gl.enableDepth();
-			glDepthFunc(GL_ALWAYS);
+			gl.setDepthMask(GL_TRUE);
-			glDepthMask(GL_TRUE);
+			gl.setDepthFunc(GL_ALWAYS);
 			bindDepthBuffer();
 		} else {
-			OpenGL::disableDepth();
+			gl.disableDepth();
 		}
 	}
 
@@ -947,7 +918,7 @@ constexpr u32 topScreenBuffer = 0x1f000000;
 constexpr u32 bottomScreenBuffer = 0x1f05dc00;
 
 void Renderer::display() {
-	OpenGL::disableScissor();
+	gl.disableScissor();
 
 	glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
 	screenFramebuffer.bind(OpenGL::ReadFramebuffer);
@@ -1038,12 +1009,15 @@ void Renderer::displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32
 	tex.bind();
 	screenFramebuffer.bind(OpenGL::DrawFramebuffer);
 
-	OpenGL::disableBlend();
+	gl.disableBlend();
-	OpenGL::disableDepth();
+	gl.disableDepth();
-	OpenGL::disableScissor();
+	gl.disableScissor();
+	gl.setColourMask(true, true, true, true);
+	gl.useProgram(displayProgram);
+	gl.bindVAO(dummyVAO);
+
 	OpenGL::disableClipPlane(0);
 	OpenGL::disableClipPlane(1);
-	displayProgram.use();
 
 	// Hack: Detect whether we are writing to the top or bottom screen by checking output gap and drawing to the proper part of the output texture
 	// We consider output gap == 320 to mean bottom, and anything else to mean top
@@ -1053,6 +1027,5 @@ void Renderer::displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32
 		OpenGL::setViewport(0, 240, 400, 240); // Top screen viewport
 	}
 
-	dummyVAO.bind();
 	OpenGL::draw(OpenGL::TriangleStrip, 4); // Actually draw our 3DS screen
 }

src/emulator.cpp

@@ -1,6 +1,6 @@
 #include "emulator.hpp"
 
-Emulator::Emulator() : kernel(cpu, memory, gpu), cpu(memory, kernel), gpu(memory), memory(cpu.getTicksRef()) {
+Emulator::Emulator() : kernel(cpu, memory, gpu), cpu(memory, kernel), gpu(memory, gl), memory(cpu.getTicksRef()) {
 	if (SDL_Init(SDL_INIT_VIDEO | SDL_INIT_EVENTS) < 0) {
 		Helpers::panic("Failed to initialize SDL2");
 	}
@@ -326,3 +326,9 @@ bool Emulator::loadELF(std::ifstream& file) {
 	}
 	return true;
 }
+
+// Reset our graphics context and initialize the GPU's graphics context
+void Emulator::initGraphicsContext() {
+	gl.reset(); // TODO (For when we have multiple backends): Only do this if we are using OpenGL
+	gpu.initGraphicsContext();
+}

src/gl_state.cpp

@@ -0,0 +1,53 @@
+#include "gl_state.hpp"
+
+void GLStateManager::resetBlend() {
+	blendEnabled = false;
+	OpenGL::disableBlend();
+}
+
+void GLStateManager::resetColourMask() {
+	redMask = greenMask = blueMask = alphaMask = true;
+	OpenGL::setColourMask(redMask, greenMask, blueMask, alphaMask);
+}
+
+void GLStateManager::resetDepth() {
+	depthEnabled = false;
+	depthMask = true;
+	depthFunc = GL_LESS;
+
+	OpenGL::disableDepth();
+	OpenGL::setDepthMask(true);
+	OpenGL::setDepthFunc(OpenGL::DepthFunc::Less);
+}
+
+void GLStateManager::resetScissor() {
+	scissorEnabled = false;
+	OpenGL::disableScissor();
+	OpenGL::setScissor(0, 0, 0, 0);
+}
+
+void GLStateManager::resetVAO() {
+	boundVAO = 0;
+	glBindVertexArray(0);
+}
+
+void GLStateManager::resetVBO() {
+	boundVBO = 0;
+	glBindBuffer(GL_ARRAY_BUFFER, 0);
+}
+
+void GLStateManager::resetProgram() {
+	currentProgram = 0;
+	glUseProgram(0);
+}
+
+void GLStateManager::reset() {
+	resetBlend();
+	resetColourMask();
+	resetDepth();
+
+	resetVAO();
+	resetVBO();
+	resetProgram();
+	resetScissor();
+}