chipd8/source/app.d

import std.stdio;
import derelict.sdl2.sdl;
import derelict.imgui.imgui;
import glad.gl.all;

import window;
import imgui;

bool checkShaderError(GLuint shader, GLuint flag, bool is_program, in char[] shader_path) nothrow {

	GLint result;

	(is_program) ? glGetProgramiv(shader, flag, &result)
		: glGetShaderiv(shader, flag, &result);

	if (result == GL_FALSE) {

		GLchar[256] log; //FIXME this is potentially fatal
		(is_program) ? glGetProgramInfoLog(shader, log.sizeof, null, log.ptr)
			: glGetShaderInfoLog(shader, log.sizeof, null, log.ptr);

		printf("[OpenGL] Error in %s: %s\n", shader_path.ptr, log.ptr);
		return false;

	}

	return true;

} // checkShaderError

struct PixelBuffer {

	struct Shader {

		private {
			GLuint shader_prog;
		}

		@disable this(this);

		void compile(const (GLchar*)* vs_source, const (GLchar*)* fs_source) {
			
			import core.stdc.stdlib : exit;

			GLuint new_vs_shader = glCreateShader(GL_VERTEX_SHADER);
			glShaderSource(new_vs_shader, 1, vs_source, null);
			glCompileShader(new_vs_shader);
			if (!checkShaderError(new_vs_shader, GL_COMPILE_STATUS, false, "vertex_shader")) {
				exit(-1);
			}

			GLuint new_fs_shader = glCreateShader(GL_FRAGMENT_SHADER);
			glShaderSource(new_fs_shader, 1, fs_source, null);
			glCompileShader(new_fs_shader);
			if (!checkShaderError(new_fs_shader, GL_COMPILE_STATUS, false, "fragment_shader")) {
				exit(-1);
			}
			
			GLuint new_shader = glCreateProgram();
			glAttachShader(new_shader, new_vs_shader);
			glAttachShader(new_shader, new_fs_shader);

			// glBindAttribLocation(new_shader, 0, "screen_size");

			glLinkProgram(new_shader);
			if (!checkShaderError(new_shader, GL_LINK_STATUS, true, "shader_program")) {
				exit(-1);
			}

			glValidateProgram(new_shader);
			if (!checkShaderError(new_shader, GL_VALIDATE_STATUS, true, "shader_program")) {
				exit(-1);
			}

			shader_prog = new_shader;
			
		} // compile

		void setUniforms(int w, int h) {
			auto attr_loc = glGetUniformLocation(shader_prog, "screen_size");
			glUniform2f(attr_loc, cast(float)w, cast(float)h);
		} // setUniforms

		void bind() {
			glUseProgram(shader_prog);
		} // bind

		void unbind() {
			glUseProgram(0);
		} // unbind

	} // Shader

	struct Texture {

		import derelict.sdl2.sdl;
		import derelict.sdl2.image;

		private {

			GLuint texture_; //OpenGL handle for texture
			GLenum input_format_, output_format_, data_type_;
			int width_, height_;

		}

		@property @nogc nothrow {

			int width() const { return width_; }
			int height() const { return height_; }
			GLuint handle() { return texture_; }

		}

		@disable this(this);

		nothrow @nogc
		this(int width, int height, GLenum input_format, GLenum output_format, GLenum unpack_alignment) {

			width_ = width;
			height_ = height;
			input_format_ = input_format;
			output_format_ = output_format;
			data_type_ = GL_UNSIGNED_BYTE;

			glGenTextures(1, &texture_);
			glBindTexture(GL_TEXTURE_2D, texture_);

			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);

			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);

			glPixelStorei(GL_UNPACK_ALIGNMENT, unpack_alignment);
			glTexImage2D(GL_TEXTURE_2D, 0, input_format_, width_, height_, 0, output_format_, GL_UNSIGNED_BYTE, cast(void*)0);

			glBindTexture(GL_TEXTURE_2D, 0);

		} //this

		nothrow @nogc
		void create(void* pixels, int width, int height, GLenum input_format = GL_RGB, GLenum output_format = GL_RGB, GLenum data_type = GL_UNSIGNED_BYTE) {

			width_ = width;
			height_ = height;
			input_format_ = input_format;
			output_format_ = output_format;
			data_type_ = data_type;

			//generate single texture, put handle in texture
			glGenTextures(1, &texture_);

			//normal 2d texture, bind to our texture handle
			glBindTexture(GL_TEXTURE_2D, texture_);

			//set texture parameters in currently bound texture, controls texture wrapping (or GL_CLAMP?)
			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);

			//linearly interpolate between pixels, MIN if texture is too small for drawing area, MAG if drawing area is smaller than texture
			glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
			glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);

			//texture type, level, format to store as, width, height, border, format loaded in
			glTexImage2D(GL_TEXTURE_2D, 0, input_format_, width_, height_, 0, output_format_, data_type_, pixels);

			//UNBIND
			glBindTexture(GL_TEXTURE_2D, 0);

		} // this

		~this() nothrow @nogc {

			if (texture_ != 0) {
				glDeleteTextures(1, &texture_);
			}

		} // ~this

		/**
		* Binds the texture handle, takes an argument for which texture unit to use.
		*/
		nothrow @nogc
		void bind(int unit) {

			assert(unit >= 0 && unit <= 31);
			glActiveTexture(GL_TEXTURE0 + unit); //since this is sequential, this works
			glBindTexture(GL_TEXTURE_2D, texture_);

		} // bind

		nothrow @nogc
		void unbind() {

			glBindTexture(GL_TEXTURE_2D, 0);

		} // unbind

		/**
		* Updates the texture in place given the new texture buffer.
		* Takes an optional offset to update only a part of the texture.
		**/
		nothrow @nogc
		void update(void[] pixels, size_t offset = 0) {

			bind(0);
			glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width_, height_, input_format_, data_type_, pixels.ptr);
			unbind();

		} // update

	} // Texture

	/**
	* Generic VertexArray structure, used to upload data of any given vertex type to the GPU.
	*/
	struct VertexArray {

		private {

			GLuint vao_;
			GLuint vbo_;
			GLenum type_; // type of vertex data, GL_TRIANGLES etc
			uint num_vertices_;

		}

		@disable this(this);

		nothrow @nogc
		void create(in float[2][] vertices, GLenum draw_type = GL_STATIC_DRAW, GLenum primitive = GL_TRIANGLES)  {

			this.num_vertices_ = cast(uint)vertices.length;
			this.type_ = primitive;

			glGenVertexArrays(1, &vao_);
			glBindVertexArray(vao_);

			glGenBuffers(1, &vbo_);
			glBindBuffer(GL_ARRAY_BUFFER, vbo_);
			glBufferData(GL_ARRAY_BUFFER, vertices.length * vertices[0].sizeof, vertices.ptr, draw_type);

			// pos
			glEnableVertexAttribArray(0);
			glVertexAttribPointer(0,
				2,
				GL_FLOAT,
				GL_FALSE,
				float.sizeof * 2,
				null
			);

			glBindVertexArray(0);

		} // this

		~this() nothrow @nogc {
			if (vao_ != 0) {
				glDeleteVertexArrays(1, &vao_);
			}
		} // ~this

		void bind() nothrow @nogc {
			glBindVertexArray(vao_);
		} // bind

		void draw() nothrow @nogc {
			glDrawArrays(type_, 0, num_vertices_);
		} // draw

		void unbind() nothrow @nogc {
			glBindVertexArray(0);
		} // unbind

	} // VertexArray

	const char* vs_shader = q{
		#version 330 core

		uniform vec2 screen_size;

		layout(location = 0) in vec2 pos;

		out vec2 frag_uv;

		void main() {
			gl_Position = vec4(pos, 0.0, 1.0);
			frag_uv = clamp(pos, vec2(0.0, 0.0), vec2(1.0, 1.0));
		}

	};

	const char* fs_shader = q{
		#version 330 core

		uniform sampler2D tex;

		in vec2 frag_uv;

		out vec4 out_col;

		void main() {
			float v = texture(tex, frag_uv.st).r * 255;
			out_col = vec4(v, v, v, 1.0);
		}

	};

	Shader shader;
	VertexArray vao;
	Texture tex;

	void create(void* pixels, int w, int h) {

		float[2][6] rect = [
			[-1.0f, -1.0f], // top left
			[1.0f, -1.0f], // top right
			[1.0f, 1.0f], // bottom right

			[-1.0f, -1.0f], // top left
			[-1.0f, 1.0f], // bottom left
			[1.0f, 1.0f], // bottom right
		];

		vao.create(rect[]);
		tex.create(pixels, w, h, GL_RED, GL_RED);
		shader.compile(&vs_shader, &fs_shader);

	} // create

	void draw(int screen_w, int screen_h) {

		vao.bind();
		shader.bind();
		shader.setUniforms(screen_w, screen_h);
		tex.bind(0);
		vao.draw();
		
	} // draw

	nothrow @nogc
	void update(void[] pixels, size_t offset = 0) {
		tex.update(pixels, offset);
	} // update

} // PixelBuffer

struct Assembler {

	/*
	Instruction Glossary:
		0x0nnn - SYS addr
		0x00E0 - CLS
		0x00EE - RET
		0x1nnn - JP addr
		0x2nnn - CALL addr
		0x3xkk - SE Vx, byte
		0x4xkk - SNE Vx, byte
		0x5xy0 - SE Vx, Vy
		0x6xkk - LD Vx, byte
		0x7xkk - ADD Vx, byte
		0x8xy0 - LD Vx, Vy
		0x8xy1 - OR Vx, Vy
		0x8xy2 - AND Vx, Vy
		0x8xy3 - XOR Vx, Vy
		0x8xy4 - ADD Vx, Vy
		0x8xy5 - SUB Vx, Vy
		0x8xy6 - SHR Vx {, Vy}
		0x8xy7 - SUBN Vx, Vy
		0x8xyE - SHL Vx {, Vy}
		0x9xy0 - SNE Vx, Vy
		0xAnnn - LD I, addr
		0xBnnn - JP V0, addr
		0xCxkk - RND Vx, byte
		0xDxyn - DRW Vx, Vy, nibble
		0xEx9E - SKP Vx
		0xExA1 - SKNP Vx
		0xFx07 - LD Vx, DT
		0xFx0A - LD Vx, K
		0xFx15 - LD DT, Vx
		0xFx18 - LD ST, Vx
		0xFx1E - ADD I, Vx
		0xFx29 - LD F, Vx
		0xFx33 - LD B, Vx
		0xFx55 - LD [I], Vx
		0xFx65 - LD Vx, [I]
	*/

	enum OpCode {

		CLS, // CLS
		RET, // RET
		CALL, // CALL addr

		ADD, // ADD Vx, Vy
			 // ADD Vx, byte
			 // ADD I, Vx

		SUB, // SUB Vx, Vy
			 // SUB Vx, byte

		SUBN, // SUBN Vx, Vy

		SE, // SE Vx, Vy
			// SE Vx, byte

		SNE, // SNE Vx, byte
		SKP, // SKP Vx
		SKNP, // SKNP Vx

		SHL, // SHL Vx {, Vy}
		SHR, // SHR Vx {, Vy}
		XOR, // XOR Vx, Vy
		AND, // AND Vx, Vy
		OR, // OR Vx, Vy

		LD, // LD Vx, DT
			// LD DT, Vx
			// LD ST, Vx
			// LD Vx, K
			// LD [I], Vx
			// LD Vx, [I]
		
		RND // RND Vx, byte
		
	}

	enum Argument {

		Vx,
		Vy,

		DT,
		ST,
		K,
		
		addr,
		val
		
	}

	struct Instruction {

	}

	import std.array;

	Appender!string dissassembly_data;

	void assemble(const char* input) {

	} // assemble

	void assemble(const char[] input) {

	} // assemble

	const (char)[] disassemble(ubyte[] instructions) {

		import std.range : chunks;

		foreach (ref ubyte[] i; instructions.chunks(2)) {

			ushort opcode = (*(cast(ushort*)(i.ptr)));

			switch (opcode & 0xF000) {
				case 0x0000:
					switch (opcode & 0x0FFF) {
						case 0x00E0: // 0x00E0 Clears the screen.

						case 0x00EE: // 0x00EE Returns from a subroutine.
						default: // 0x0NNN Calls RCA 1802 program at address NNN. Not necessary for most ROMs.
							//assert(0, "0x0NNN RCA 1802 program opcode not implemented!");
							break;
					}
					break;
				case 0x1000: // 0x1NNN Jumps to address NNN.
				case 0x2000: // 0x2NNN Calls subroutine at NNN.
				case 0x3000: // 0x3XNN Skips the next instruction if VX equals NN.
				case 0x4000: // 0x4XNN Skips the next instruction if VX doesn't equal NN.
				case 0x5000: // 0x5XYO Skips the next instruction if VX equals VY.
				case 0x6000: // 0x6XNN Sets VX to NN.
				case 0x7000: // 0x7XNN Adds NN to VX.
				case 0x8000:
					switch (opcode) {
						case 0x0000: // 0x8XY0 Sets VX to the value of VY.
						case 0x0001: // 0x8XY1 Sets VX to VX or VY.
						case 0x0002: // 0x8XY2	Sets VX to VX and VY.
						case 0x0003: // 0x8XY3 Sets VX to VX xor VY.
						case 0x0004: // 0x8XY4 Adds VY to VX. VF is set to 1 when there's a carry, and to 0 when there isn't.
						case 0x0005: // 0x8XY5 VY is subtracted from VX. VF is set to 0 when there's a borrow, and 1 when there isn't.
						case 0x0006: // 0x8XY6 Shifts VX right by one. VF is set to the value of the least significant bit of VX before the shift.
						case 0x0007: // 0x8XY7 Sets VX to VY minus VX. VF is set to 0 when there's a borrow, and 1 when there isn't.
						case 0x000E: // 0x8XYE Shifts VX left by one. VF is set to the value of the most significant bit of VX before the shift.
						default: // unhandled for some reason
							writefln("unknown opcode: 0x%x", opcode);
							break;
					}
					break;
				case 0x9000: // 0x9XYO Skips the next instruction if VX doesn't equal VY.
				case 0xA000: // 0xANNN  Sets I to the address NNN.
				case 0xB000: // 0xBNNN Jumps to the address NNN plus V0.
				case 0xC000: // 0xCXNN Sets VX to the result of a bitwise and operation on a random number and NN.

				// 0xDXYN
				// Sprites stored in memory at location in index register (I), 8bits wide.
				// Wraps around the screen. If when drawn, clears a pixel, register VF is set to 1 otherwise it is zero.
				// All drawing is XOR drawing (i.e. it toggles the screen pixels).
				// Sprites are drawn starting at position VX, VY. N is the number of 8bit rows that need to be drawn.
				// If N is greater than 1, second line continues at position VX, VY+1, and so on.
				case 0xD000:
				case 0xE000:
					switch (opcode & 0x000F) {
						case 0x000E: // 0xEX9E Skips the next instruction if the key stored in VX is pressed.
						case 0x0001: // 0xEXA1 Skips the next instruction if the key stored in VX isn't pressed.
						default: // unhandled for some reason
							writefln("unknown opcode: 0x%x", opcode);
							break;
					}
					break;
				case 0xF000:
					switch (opcode & 0x00FF) {
						case 0x0007: // 0xFX07 Sets VX to the value of the delay timer.
						case 0x000A: // 0xFX0A A key press is awaited, and then stored in VX.
						case 0x0015: // 0xFX15 Sets the delay timer to VX.
						case 0x0018: // 0xFX18 Sets the sound timer to VX.
						case 0x001E: // 0xFX1E Adds VX to I.
						case 0x0029: // 0xFX29 Sets I to the location of the sprite for the character in VX.
						// 0xFX33 Stores the Binary-coded decimal representation of VX,
						//  with the most significant of three digits at the address in I,
						//  the middle digit at I plus 1, and the least significant digit at I plus 2.
						case 0x0033:
						case 0x0055: // 0xFX55 Stores V0 to VX in memory starting at address I.
						case 0x0065: // 0xFX65 Fills V0 to VX with values from memory starting at address I.
						default: // unhandled for some reason
							writefln("unknown opcode: 0x%x", opcode);
							break;
					}
					break;
				default:
					writefln("unknown opcode: 0x%x", opcode);
			}
		}
		
		return "";

	} // dissassemble

} // Assembler

struct Chip8Status {

	// emu ptr
	Emulator* run_;
	Chip8* emu_;

	// loaded program
	const (char)* loaded_program;

	// mem editor
	// MemoryEditor mem_editor_;

	// state
	bool status_menu_ = true;
	int stack_cur_ = -1;

	void initialize(Emulator* run, Chip8* emu) {
		this.run_ = run;
		this.emu_ = emu;
	} // initialize

	alias Callback = float delegate(int idx, const char** out_text);

    static extern(C) bool doCallback(void* ptr, int idx, const (char**) out_text) {

        auto callback = *(cast(Callback*) ptr);
        callback(idx, out_text);
		return true;

    } // doCallback;

	void getStackFrame(int idx, const (char**) out_text) {

		static char[32] frame_text;

		import core.stdc.stdio : sprintf;
		sprintf(frame_text.ptr, "0x%04X", emu_.stack[idx]);

		auto p = frame_text.ptr;
		auto op = cast(char**)out_text;
		*op = p;

	} // getStackFrame

	void resetShortcut() {

		loaded_program = null;

		emu_.reset();
		
	} // resetShortcut

	void loadShortcut() {

		import std.file : read;

		loaded_program = "chip8_picture.ch8";
		auto buf = read("programs/chip8_picture.ch8");
		emu_.load(0x200, buf); // do ze load yes, will copy all the data in

	} // loadShortcut

	void saveShortcut() {

	} // saveShortcut

	void debugShortcut() {
		status_menu_ = !status_menu_;
	} // debugShortcut

	void redrawShortcut() {
		emu_.draw_flag = true;
	} // redrawShortcut

	void toggleRunShortcut() {
		emu_.run_flag = !emu_.run_flag;		
	} // toggleRunShortcut

	void stepShortcut() {
		emu_.step();
	} // stepShortcut

	void quitShortcut() {
		run_.quit();
	} // quitShortcut

	void draw() {

		if (!status_menu_) return;

		if (igBeginMainMenuBar()) {

			if (igBeginMenu("Menu")) {

				if (igMenuItem("Reset", "CTRL+R")) {
					resetShortcut();
				}

				if (igMenuItem("Load", "CTRL+L")) {
					loadShortcut();
				}

				if (igMenuItem("Debug", "CTRL+D")) {
					debugShortcut();
				}

				if (igMenuItem("Quit", "CTRL+Q")) {
					quitShortcut();
				}

				igEndMenu();
			}

			igEndMainMenuBar();
		}

		{

			import std.range : chunks;
			
			igBegin("Emulator Status");

			igBeginChild("General");
			if (!loaded_program) {
				igText("Loaded Program: none");
			} else {
				igText("Loaded Program: %s", loaded_program);
			}
			igText("Opcode: 0x%04X", emu_.cpu.opcode);
			igSameLine();
			igText("| PC: 0x%04X (%hu)", emu_.cpu.pc, emu_.cpu.pc);
			// igDragInt("##pc", cast(int*)&emu_.cpu.pc, 0.5f, 0, emu_.ram.length);

			igText("Registers (v0 - vF)");
			igColumns(4, null, false);
			igIndent();

			auto n = 0;
			foreach (ref chunk; emu_.cpu.v[].chunks(4)) {
				igText("v%0X 0x%02X ", n, chunk[0]);
				igText("v%0X 0x%02X ", n+1, chunk[1]);
				igText("v%0X 0x%02X ", n+2, chunk[2]);
				igText("v%0X 0x%02X ", n+3, chunk[3]);
				igNextColumn();
				n += 4;
			}

			igColumns(1, null, false);
			igUnindent();

			igText("Index Register: 0x%04X", emu_.cpu.i);
			igText("Delay Timer: 0x%04X", emu_.cpu.delay_timer);
			igText("Sound Timer: 0x%04X", emu_.cpu.sound_timer);

			if (igButton("Reload")) {
				resetShortcut();
				loadShortcut();
			}

			igSameLine();

			if (igButton("Reset")) {
				resetShortcut();
			}

			if (igButton("Step")) {
				emu_.step();
			}

			igSameLine();

			if (igButton((emu_.run_flag) ? "Stop" : "Run")) {
				emu_.run_flag = !emu_.run_flag;
			}

			igText("Stack");
			auto d = &getStackFrame;
			igPushItemWidth(-1);
			igListBox2("", &stack_cur_, &doCallback, cast(void*)&d, emu_.stack.length, 16);
			igPopItemWidth();

			igEndChild();
			igEnd();

		}

		// mem_editor_.draw("Emulator Memory", emu_.ram[]);

	} // draw

	void handleEvent(ref SDL_Event ev) {
		switch (ev.type) with (SDL_EventType) {
			case SDL_KEYDOWN:
				if ((ev.key.keysym.mod & KMOD_CTRL) != 0) {
					switch (ev.key.keysym.scancode) with (SDL_EventType) {
						case SDL_SCANCODE_R: resetShortcut(); break;
						case SDL_SCANCODE_L: loadShortcut(); break;
						case SDL_SCANCODE_D: debugShortcut(); break;
						case SDL_SCANCODE_G: redrawShortcut(); break;
						case SDL_SCANCODE_T: toggleRunShortcut(); break;
						case SDL_SCANCODE_S: stepShortcut(); break;
						case SDL_SCANCODE_Q: quitShortcut(); break;
						default: break;
					}
				} else {
					if (ev.key.keysym.scancode == SDL_SCANCODE_ESCAPE) {
						quitShortcut();
					}
				}
				break;
			default:
				break;
		}
	} // handleEvent

} // Chip8Status

pure
@property {
	ubyte x(ref ushort oc) {
		return (oc & 0x0F00) >> 8;
	}

	ubyte y(ref ushort oc) {
		return (oc & 0x00F0) >> 4;
	}

	ubyte n(ref ushort oc) {
		return oc & 0x000F;
	}

	ubyte nn(ref ushort oc) {
		return (oc & 0x00FF);
	}

	ushort nnn(ref ushort oc) {
		return (oc & 0x0FFF);
	}
}

struct Chip8 {

	alias OpCode = ushort;
	alias ProgramCounter = ushort;
	alias Memory = ubyte[4096];
	alias Stack = ushort[16];

	alias Register = ubyte;
	alias Registers = Register[16];
	alias IndexRegister = ushort;

	alias KeyPad = ubyte[16];

	struct CPU {

		OpCode opcode = 0; //current opcode
		ProgramCounter pc = 0x200;

		Registers v; //16 general purpose registers
		IndexRegister i;

		Register delay_timer;
		Register sound_timer;

	} // CPU

	CPU cpu;
	Memory ram;

	Stack stack;
	ubyte sp;

	KeyPad kp;

	ubyte[64*32] screen_buf;
	ubyte[3][64*32] screen_data;

	bool run_flag;
	bool draw_flag;
	bool reset_flag;

	void load(size_t offset, in void[] data) {

		assert(offset + data.length < ram.length);

		ram[offset .. offset + data.length] = cast(ubyte[])data[];

	} // load

	void reset() {
			
		reset_flag = true;

		cpu = cpu.init;
		stack = stack.init;
		ram[0x200 .. $ - 1] = 0;
		cpu.pc = 0x200;
		sp = sp.init;

		run_flag = run_flag.init;
		draw_flag = draw_flag.init;

		screen_buf = screen_buf.init;
		screen_data = screen_data.init;

	} // reset

	void step() {

		// reset if we go OOB
		if (cpu.pc >= 0xFFF) {
			reset();
			return;
		}

		cpu.opcode = ram[cpu.pc] << 8 | ram[cpu.pc + 1];
		ushort pc_target = cast(ProgramCounter)(cpu.pc + 2u);

		// writefln("opcode: 0x%X, pc: 0x%X : %d", cpu.opcode, cpu.pc, cpu.pc);

		switch (cpu.opcode & 0xF000) with (cpu) {

			case 0x0000:

				switch (cpu.opcode & 0x0FFF) {

					case 0x00E0: // 0x00E0 Clears the screen.
						screen_buf[0..$] = 0;
						draw_flag = true;
						break;

					case 0x00EE: // 0x00EE Returns from a subroutine.
						pc_target = stack[--sp];
						break;

					default: // 0x0NNN Calls RCA 1802 program at address NNN. Not necessary for most ROMs.
						//assert(0, "0x0NNN RCA 1802 program opcode not implemented!");
						break;

				}

				break;

			case 0x1000: // 0x1NNN Jumps to address NNN.
				pc_target = cpu.opcode.nnn;
				break;

			case 0x2000: // 0x2NNN Calls subroutine at NNN.

				stack[sp++] = cpu.pc;
				pc_target = cpu.opcode.nnn;
				break;

			case 0x3000: // 0x3XNN Skips the next instruction if VX equals NN.

				if (cpu.v[cpu.opcode.x] == (cpu.opcode.nn)) {
					pc_target += 2u;
				}

				break;

			case 0x4000: // 0x4XNN Skips the next instruction if VX doesn't equal NN.

				if (cpu.v[cpu.opcode.x] != (cpu.opcode.nn)) {
					pc_target += 2u;
				}

				break;

			case 0x5000: // 0x5XYO Skips the next instruction if VX equals VY.

				if (cpu.v[cpu.opcode.x] == cpu.v[cpu.opcode.y]) {
					pc_target += 2u;
				}

				break;

			case 0x6000: // 0x6XNN Sets VX to NN.
				cpu.v[cpu.opcode.x] = cpu.opcode.nn;
				break;

			case 0x7000: // 0x7XNN Adds NN to VX.
				cpu.v[cpu.opcode.x] += cpu.opcode.nn;
				break;

			case 0x8000:

				ubyte x = cpu.opcode.x;
				ubyte y = cpu.opcode.y;

				switch (cpu.opcode.n) {

					case 0x0000: // 0x8XY0 Sets VX to the value of VY.
						cpu.v[x] = cpu.v[y];
						break;

					case 0x0001: // 0x8XY1 Sets VX to VX or VY.
						cpu.v[x] = cpu.v[x] | cpu.v[y];
						break;

					case 0x0002: // 0x8XY2	Sets VX to VX and VY.
						cpu.v[x] = cpu.v[x] & cpu.v[y];
						break;

					case 0x0003: // 0x8XY3 Sets VX to VX xor VY.
						cpu.v[x] = cpu.v[x] ^ cpu.v[y];
						break;

					case 0x0004: // 0x8XY4 Adds VY to VX. VF is set to 1 when there's a carry, and to 0 when there isn't.
						ubyte vx = cpu.v[x];
						ubyte vy = cpu.v[y];
						if (cast(ushort)vx + cast(ushort)vy > 255) {
							cpu.v[0xF] = 1;
						} else { 
							cpu.v[0xF] = 0;
						}
						cpu.v[x] += cpu.v[y];
						break;

					case 0x0005: // 0x8XY5 VY is subtracted from VX. VF is set to 0 when there's a borrow, and 1 when there isn't.
						ubyte vx = cpu.v[x];
						ubyte vy = cpu.v[y];
						if (vx > vy) {
							cpu.v[0xF] = 1;
						} else {
							cpu.v[0xF] = 0;
						}
						cpu.v[x] -= cpu.v[y];
						break;

					case 0x0006: // 0x8XY6 Shifts VX right by one. VF is set to the value of the least significant bit of VX before the shift.

						ubyte vx = cpu.v[x];
						cpu.v[0xF] = vx & 0x1;
						cpu.v[x] >>= 1;
						break;

					case 0x0007: // 0x8XY7 Sets VX to VY minus VX. VF is set to 0 when there's a borrow, and 1 when there isn't.
						ubyte vx = cpu.v[x];
						ubyte vy = cpu.v[y];
						if (vy > vx) {
							cpu.v[0xF] = 1;
						} else {
							cpu.v[0xF] = 0;
						}
						cpu.v[x] = cast(Register)(vy - vx);
						break;

					case 0x000E: // 0x8XYE Shifts VX left by one. VF is set to the value of the most significant bit of VX before the shift.

						ubyte vx = cpu.v[x];
						cpu.v[0xF] = vx >> 7;
						cpu.v[x] <<= 1;
						break;

					default: // unhandled for some reason
						writefln("unknown opcode: 0x%x", cpu.opcode);
						break;

				}

				break;

			case 0x9000: // 0x9XYO Skips the next instruction if VX doesn't equal VY.

				if (cpu.v[cpu.opcode.x] != cpu.v[cpu.opcode.y]) {
					pc_target += 2u; // do skip yes
				}

				break;

			case 0xA000: // 0xANNN  Sets I to the address NNN.
				cpu.i = cpu.opcode.nnn;
				break;

			case 0xB000: // 0xBNNN Jumps to the address NNN plus V0.
				pc_target = cast(ProgramCounter)(cpu.opcode.nnn + cpu.v[0x0]);
				break;

			case 0xC000: // 0xCXNN Sets VX to the result of a bitwise and operation on a random number and NN.

				import std.random : uniform;
				ubyte x = cpu.opcode.x;
				cpu.v[x] = uniform(Register.min, Register.max) & (cpu.opcode & 0x00FF);
				break;

			// 0xDXYN
			// Sprites stored in memory at location in index register (I), 8bits wide.
			// Wraps around the screen. If when drawn, clears a pixel, register VF is set to 1 otherwise it is zero.
			// All drawing is XOR drawing (i.e. it toggles the screen pixels).
			// Sprites are drawn starting at position VX, VY. N is the number of 8bit rows that need to be drawn.
			// If N is greater than 1, second line continues at position VX, VY+1, and so on.
			case 0xD000:

				ProgramCounter spr_addr = cpu.i;
				ubyte x = cpu.opcode.x;
				ubyte y = cpu.opcode.y;
				ubyte n = cpu.opcode.n;

				foreach(int row; 0 .. n) {

					ushort pixel = ram[spr_addr + row];

					foreach (int col; 0 .. 8) {
						if ((pixel & (0x80 >> col)) != 0) {
							ubyte x_off = cast(ubyte)((x + col) % 64);
							ubyte y_off = cast(ubyte)((y + row) % 32);
							ushort offset = x_off + (y_off * 64);
							if (screen_buf[offset] == 1) {
								cpu.v[0xF] = 1;
							}
							screen_buf[offset] ^= 1;
						}
					}

				}

				draw_flag = true;

				break;

			case 0xE000:

				ubyte x = cpu.opcode.x;
				ubyte key = cpu.v[x];

				switch (cpu.opcode & 0x000F) {

					case 0x000E: // 0xEX9E Skips the next instruction if the key stored in VX is pressed.
						writefln("0xEXA1: skip instruction if VX not pressed: %x", key);
						break;

					case 0x0001: // 0xEXA1 Skips the next instruction if the key stored in VX isn't pressed.
						writefln("0xEXA1: skip instruction if VX not pressed: %x", key);
						break;

					default: //unhandled for some reason
						writefln("unknown opcode: 0x%x", cpu.opcode);
						break;

				}

				break;

			case 0xF000:

				switch (cpu.opcode & 0x00FF) {

					case 0x0007: // 0xFX07 Sets VX to the value of the delay timer.
						cpu.v[cpu.opcode.x] = cpu.delay_timer;
						break;

					case 0x000A: // 0xFX0A A key press is awaited, and then stored in VX.

						break;

					case 0x0015: // 0xFX15 Sets the delay timer to VX.
						cpu.delay_timer = cpu.v[cpu.opcode.x];
						break;

					case 0x0018: // 0xFX18 Sets the sound timer to VX.
						cpu.sound_timer = cpu.v[cpu.opcode.x];
						break;

					case 0x001E: // 0xFX1E Adds VX to I.
						if (cpu.i + cpu.v[cpu.opcode.x] > 0xFF) {
							cpu.v[0xF] = 1;
						} else {
							cpu.v[0xF] = 0;
						}
						cpu.i += cpu.v[cpu.opcode.x];
						break;

					case 0x0029: // 0xFX29 Sets I to the location of the sprite for the character in VX.

						ubyte vx = cpu.v[cpu.opcode.x];
						// ushort char_addr = 0x200 + (vx * 40); // base of char sprites + value of vx * bits per character
						ushort char_addr = vx * 0x5;
						cpu.i = char_addr;
						break;

					// 0xFX33 Stores the Binary-coded decimal representation of VX,
					//  with the most significant of three digits at the address in I,
					//  the middle digit at I plus 1, and the least significant digit at I plus 2.
					case 0x0033:

						ubyte vx = cpu.v[cpu.opcode.x];
						ram[cpu.i] = vx / 100;
						ram[cpu.i + 1] = (vx / 10) % 10;
						ram[cpu.i + 2] = (vx % 100) % 10;
						break;

					case 0x0055: // 0xFX55 Stores V0 to VX in memory starting at address I.

						IndexRegister addr = cpu.i;
						foreach (reg; cpu.v) {
							ram[addr++] = reg;
						}

						break;

					case 0x0065: // 0xFX65 Fills V0 to VX with values from memory starting at address I.

						IndexRegister addr = cpu.i;
						foreach (ref reg; cpu.v) {
							reg = ram[addr++];
						}

						break;

					default: // unhandled for some reason
						writefln("unknown opcode: 0x%x", cpu.opcode);
						break;

				}

				break;

			default:
				writefln("unknown opcode: 0x%x", cpu.opcode);

		}

		/* now update pc and timer registers. */

		cpu.pc = pc_target;

		if (cpu.delay_timer != 0u) {
			--cpu.delay_timer;
		}

		if (cpu.sound_timer != 0u) {

			if (cpu.sound_timer == 1u) {
				writefln("beep!");
			}

			--cpu.sound_timer;

		}

	} // step

	void handleEvent(ref SDL_Event ev) {

	} // handleEvent

	void tick() {

		if (run_flag) {
			step();
		}

	} // tick

} // Emulator

void loadLibs() {

	DerelictSDL2.load();
	DerelictImgui.load();

}

void initLibs() {

	SDL_Init(SDL_INIT_VIDEO);
	SDL_GL_LoadLibrary(null);

}

void setupImgui() {
	
}

void main() {

	loadLibs();
	initLibs();

	Emulator emu;
	emu.create();
	emu.run();

}

struct Emulator {

	bool running;

	// debug
	Chip8Status status;

	Window window;
	Imgui imgui;
	Chip8 emu;

	// drawing
	PixelBuffer buf;

	void create() {

		// create window
		window.createWindow(960, 768);

		// setup imgui
		imgui.initialize();
		imgui.createDeviceObjects();

		// setup debug ui
		status.initialize(&this, &emu);

		// debug data
		emu.screen_buf[0] = 1;
		emu.screen_buf[64 - 1] = 1;
		emu.screen_buf[64*32 - 64] = 1;
		emu.screen_buf[64*32 - 1] = 1;

		// set up pixel buffer to poke at
		buf.create(emu.screen_buf.ptr, 64, 32);

	}

	void handleEvents() {

		SDL_Event event;
		while (SDL_PollEvent(&event)) {

			imgui.handleEvent(event);
			status.handleEvent(event);
			emu.handleEvent(event);

			switch (event.type) with (SDL_EventType) {
				case SDL_QUIT: {
					running = false;
					break;
				}
				default: {
					break;
				}
			}

		}

	} // handleEvents

	void run() {

		running = true;

		while (running) {
			handleEvents();
			tick();
			draw();
		}

	}

	void tick() {

		emu.tick();

	}

	void draw() {

		window.renderClear(0x428bca);

		int w, h;
		window.windowSize(w, h);
		if (emu.draw_flag || emu.reset_flag) {
			if (emu.reset_flag) emu.reset_flag = false;
			buf.update(emu.screen_buf);
			emu.draw_flag = false;
		}
		buf.draw(w, h);

		imgui.newFrame(window);

		status.draw();

		imgui.endFrame();
		window.renderPresent();

	}

	void quit() {
		running = false;
	}

}