ci: ensure that submodules are updated recursively

Reviewed-on: #6

.github/workflows/main.yml

@@ -0,0 +1,59 @@
+name: Nightly
+
+on:
+  push:
+    paths:
+      - "**.zig"
+    branches:
+      - main
+  schedule:
+    - cron: '0 0 * * *'
+  workflow_dispatch:
+
+jobs:
+  build:
+    strategy:
+      matrix:
+        # os: [ubuntu-latest, windows-latest, macos-latest]
+        os: [ubuntu-latest, windows-latest]
+    runs-on: ${{matrix.os}}
+    steps:
+      - uses: goto-bus-stop/setup-zig@v2
+        with:
+          version: master
+      - name: prepare-linux
+        if: runner.os == 'Linux'
+        run: |
+            sudo apt-get update
+            sudo apt-get install libsdl2-dev
+      - name: prepare-windows
+        if: runner.os == 'Windows'
+        run: |
+            vcpkg integrate install
+            vcpkg install sdl2:x64-windows
+            git config --global core.autocrlf false
+      - name: prepare-macos
+        if: runner.os == 'macOS'
+        run: |
+            brew install sdl2
+      - uses: actions/checkout@v3
+        with:
+          submodules: recursive
+      - name: build 
+        run: zig build -Doptimize=ReleaseSafe -Dcpu=baseline
+      - name: upload
+        uses: actions/upload-artifact@v3
+        with:
+          name: zba-${{matrix.os}}
+          path: zig-out/bin
+  lint:
+    runs-on: ubuntu-latest
+    steps: 
+      - uses: actions/checkout@v3
+        with:
+          submodules: recursive
+      - uses: goto-bus-stop/setup-zig@v2
+        with:
+          version: master
+      - run: zig fmt src/**/*.zig
+  

.gitignore

@@ -1,6 +1,14 @@
 /.vscode
 /bin
-/zig-cache
-/zig-out
+**/zig-cache
+**/zig-out
 /docs
-**/*.log
+**/*.log
+**/*.bin
+
+# Build on Windows
+/.build_config
+/lib/SDL2
+
+# Any Custom Scripts for Debugging purposes
+*.sh

.gitmodules

@@ -1,3 +1,18 @@
 [submodule "lib/SDL.zig"]
 	path = lib/SDL.zig
 	url = https://github.com/MasterQ32/SDL.zig
+[submodule "lib/zig-clap"]
+	path = lib/zig-clap
+	url = https://github.com/Hejsil/zig-clap
+[submodule "lib/known-folders"]
+	path = lib/known-folders
+	url = https://github.com/ziglibs/known-folders
+[submodule "lib/zig-datetime"]
+	path = lib/zig-datetime
+	url = https://github.com/frmdstryr/zig-datetime
+[submodule "lib/zig-toml"]
+	path = lib/zig-toml
+	url = https://github.com/aeronavery/zig-toml
+[submodule "lib/zba-gdbstub"]
+	path = lib/zba-gdbstub
+	url = https://git.musuka.dev/paoda/zba-gdbstub

.vscode/extensions.json

@@ -1,9 +0,0 @@
-{
-    "recommendations": [
-        "augusterame.zls-vscode",
-        "tiehuis.zig",
-        "usernamehw.errorlens",
-        "vadimcn.vscode-lldb",
-        "dan-c-underwood.arm"
-    ]
-}

README.md

@@ -0,0 +1,116 @@
+# ZBA (working title)
+
+A Game Boy Advance Emulator written in Zig ⚡!
+
+## Scope
+
+I'm hardly the first to write a Game Boy Advance Emulator nor will I be the last. This project isn't going to compete with the GOATs like [mGBA](https://github.com/mgba-emu) or [NanoBoyAdvance](https://github.com/nba-emu/NanoBoyAdvance). There aren't any interesting ideas either like in [DSHBA](https://github.com/DenSinH/DSHBA).
+
+This is a simple (read: incomplete) for-fun long-term project. I hope to get "mostly there", which to me means that I'm not missing any major hardware features and the set of possible improvements would be in memory timing or in UI/UX. With respect to that goal, here's what's outstanding:
+
+### TODO
+
+- [x] Affine Sprites
+- [ ] Windowing (see [this branch](https://git.musuka.dev/paoda/zba/src/branch/window))
+- [ ] Audio Resampler (Having issues with SDL2's)
+- [x] Immediate Mode GUI (see [this branch](https://git.musuka.dev/paoda/zba/src/branch/imgui))
+- [ ] Refactoring for easy-ish perf boosts
+
+## Usage
+
+As it currently exists, ZBA is run from the terminal. In your console of choice, type `./zba --help` to see what you can do.
+
+I typically find myself typing `./zba -b ./bin/bios.bin ./bin/test/suite.gba` to see how badly my "cool new feature" broke everything else.
+
+Need a BIOS? Why not try using the open-source [Cult-Of-GBA BIOS](https://github.com/Cult-of-GBA/BIOS) written by [fleroviux](https://github.com/fleroviux) and [DenSinH](https://github.com/DenSinH)?
+
+Finally it's worth noting that ZBA uses a TOML config file it'll store in your OS's data directory. See `example.toml` to learn about the defaults and what exactly you can mess around with.
+
+## Tests
+
+GBA Tests | [jsmolka](https://github.com/jsmolka/)
+--- | ---
+`arm.gba`,  `thumb.gba` | PASS
+`memory.gba`, `bios.gba` | PASS
+`flash64.gba`, `flash128.gba` | PASS
+`sram.gba` | PASS
+`none.gba` | PASS
+`hello.gba`, `shades.gba`, `stripes.gba` | PASS
+`nes.gba` | PASS
+
+GBARoms | [DenSinH](https://github.com/DenSinH/)
+--- | ---
+`eeprom-test`, `flash-test` | PASS
+`midikey2freq` | PASS
+`swi-tests-random` | FAIL
+
+gba_tests | [destoer](https://github.com/destoer/)
+--- | ---
+`cond_invalid.gba` | PASS
+`dma_priority.gba` | PASS
+`hello_world.gba` | PASS
+`if_ack.gba` | PASS
+`line_timing.gba` | FAIL
+`lyc_midline.gba` | FAIL
+`window_midframe.gba` | FAIL
+
+GBA Test Collection | [ladystarbreeze](https://github.com/ladystarbreeze)
+--- | ---
+`retAddr.gba` | PASS
+`helloWorld.gba` | PASS
+`helloAudio.gba` | PASS
+
+FuzzARM | [DenSinH](https://github.com/DenSinH/)
+--- | ---
+`main.gba` | PASS
+
+arm7wrestler GBA Fixed | [destoer](https://github.com/destoer)
+--- | ---
+`armwrestler-gba-fixed.gba` | PASS
+
+## Resources
+
+- [GBATEK](https://problemkaputt.de/gbatek.htm)
+- [TONC](https://coranac.com/tonc/text/toc.htm)
+- [ARM Architecture Reference Manual](https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/third-party/ddi0100e_arm_arm.pdf)
+- [ARM7TDMI Data Sheet](https://www.dca.fee.unicamp.br/cursos/EA871/references/ARM/ARM7TDMIDataSheet.pdf)
+
+## Compiling
+
+Most recently built on Zig [v0.11.0-dev.1580+a5b34a61a](https://github.com/ziglang/zig/tree/a5b34a61a)
+
+### Dependencies
+
+Dependency | Source
+--- | ---
+SDL.zig | <https://github.com/MasterQ32/SDL.zig>
+zig-clap | <https://github.com/Hejsil/zig-clap>
+known-folders | <https://github.com/ziglibs/known-folders>
+zig-toml | <https://github.com/aeronavery/zig-toml>
+zig-datetime | <https://github.com/frmdstryr/zig-datetime>
+`bitfields.zig` | [https://github.com/FlorenceOS/Florence](https://github.com/FlorenceOS/Florence/blob/aaa5a9e568/lib/util/bitfields.zig)
+`gl.zig` | <https://github.com/MasterQ32/zig-opengl>
+
+Use `git submodule update --init` from the project root to pull the git relevant git submodules
+
+Be sure to provide SDL2 using:
+
+- Linux: Your distro's package manager
+- macOS: ¯\\\_(ツ)_/¯ (try [this formula](https://formulae.brew.sh/formula/sdl2)?)
+- Windows: [`vcpkg`](https://github.com/Microsoft/vcpkg) (install `sdl2:x64-windows`)
+
+`SDL.zig` will provide a helpful compile error if the zig compiler is unable to find SDL2.
+
+Once you've got all the dependencies, execute `zig build -Doptimize=ReleaseSafe`. The executable is located at `zig-out/bin/`.
+
+## Controls
+
+Key | Button
+--- | ---
+<kbd>X</kbd> | A
+<kbd>Z</kbd> | B
+<kbd>A</kbd> | L
+<kbd>S</kbd> | R
+<kbd>Return</kbd> | Start
+<kbd>RShift</kbd> | Select
+Arrow Keys | D-Pad

build.zig

@@ -1,30 +1,55 @@
 const std = @import("std");
+const builtin = @import("builtin");
+
 const Sdk = @import("lib/SDL.zig/Sdk.zig");
+const Gdbstub = @import("lib/zba-gdbstub/build.zig");
 
 pub fn build(b: *std.build.Builder) void {
-    // Standard target options allows the person running `zig build` to choose
-    // what target to build for. Here we do not override the defaults, which
-    // means any target is allowed, and the default is native. Other options
-    // for restricting supported target set are available.
+    // Minimum Zig Version
+    const min_ver = std.SemanticVersion.parse("0.11.0-dev.1580+a5b34a61a") catch return; // https://github.com/ziglang/zig/commit/a5b34a61a
+    if (builtin.zig_version.order(min_ver).compare(.lt)) {
+        std.log.err("{s}", .{b.fmt("Zig v{} does not meet the minimum version requirement. (Zig v{})", .{ builtin.zig_version, min_ver })});
+        std.os.exit(1);
+    }
+
     const target = b.standardTargetOptions(.{});
+    const optimize = b.standardOptimizeOption(.{});
 
-    // Standard release options allow the person running `zig build` to select
-    // between Debug, ReleaseSafe, ReleaseFast, and ReleaseSmall.
-    const mode = b.standardReleaseOptions();
+    const exe = b.addExecutable(.{
+        .name = "zba",
+        .root_source_file = .{ .path = "src/main.zig" },
+        .target = target,
+        .optimize = optimize,
+    });
 
-    const exe = b.addExecutable("zba", "src/main.zig");
+    exe.setMainPkgPath("."); // Necessary so that src/main.zig can embed example.toml
+
+    // Known Folders (%APPDATA%, XDG, etc.)
+    exe.addAnonymousModule("known_folders", .{ .source_file = .{ .path = "lib/known-folders/known-folders.zig" } });
+
+    // DateTime Library
+    exe.addAnonymousModule("datetime", .{ .source_file = .{ .path = "lib/zig-datetime/src/main.zig" } });
 
     // Bitfield type from FlorenceOS: https://github.com/FlorenceOS/
-    exe.addPackage(.{ .name = "bitfield", .path = .{ .path = "lib/util/bitfield.zig" } });
+    exe.addAnonymousModule("bitfield", .{ .source_file = .{ .path = "lib/util/bitfield.zig" } });
+
+    // Argument Parsing Library
+    exe.addAnonymousModule("clap", .{ .source_file = .{ .path = "lib/zig-clap/clap.zig" } });
+
+    // TOML Library
+    exe.addAnonymousModule("toml", .{ .source_file = .{ .path = "lib/zig-toml/src/toml.zig" } });
+
+    // OpenGL 3.3 Bindings
+    exe.addAnonymousModule("gl", .{ .source_file = .{ .path = "lib/gl.zig" } });
+
+    // gdbstub
+    Gdbstub.link(exe);
 
     // Zig SDL Bindings: https://github.com/MasterQ32/SDL.zig
-    const sdk = Sdk.init(b);
+    const sdk = Sdk.init(b, null);
     sdk.link(exe, .dynamic);
+    exe.addModule("sdl2", sdk.getNativeModule());
 
-    exe.addPackage(sdk.getNativePackage("sdl2"));
-
-    exe.setTarget(target);
-    exe.setBuildMode(mode);
     exe.install();
 
     const run_cmd = exe.run();
@@ -36,9 +61,11 @@ pub fn build(b: *std.build.Builder) void {
     const run_step = b.step("run", "Run the app");
     run_step.dependOn(&run_cmd.step);
 
-    const exe_tests = b.addTest("src/main.zig");
-    exe_tests.setTarget(target);
-    exe_tests.setBuildMode(mode);
+    const exe_tests = b.addTest(.{
+        .root_source_file = .{ .path = "src/main.zig" },
+        .target = target,
+        .optimize = optimize,
+    });
 
     const test_step = b.step("test", "Run unit tests");
     test_step.dependOn(&exe_tests.step);

example.toml

@@ -0,0 +1,25 @@
+[Host]
+# Using nearest-neighbour scaling, how many times the native resolution 
+# of the game bow should the screen be?
+win_scale = 3
+# Enable VSYNC on the UI thread
+vsync = true
+# Mute ZBA
+mute = false
+
+[Guest]
+# Sync Emulation to Audio 
+audio_sync = true
+# Sync Emulation to Video
+video_sync = true
+# Force RTC support
+force_rtc = false
+# Skip BIOS
+skip_bios = false
+
+[Debug]
+# Enable detailed CPU logs
+cpu_trace = false
+# When false and builtin.mode == .Debug, ZBA will panic
+# on unknown I/O reads
+unhandled_io = true

lib/util/bitfield.zig

@@ -14,7 +14,7 @@ fn PtrCastPreserveCV(comptime T: type, comptime PtrToT: type, comptime NewT: typ
 fn BitType(comptime FieldType: type, comptime ValueType: type, comptime shamt: usize) type {
     const self_bit: FieldType = (1 << shamt);
 
-    return struct {
+    return extern struct {
         bits: Bitfield(FieldType, shamt, 1),
 
         pub fn set(self: anytype) void {
@@ -63,7 +63,7 @@ pub fn Bitfield(comptime FieldType: type, comptime shamt: usize, comptime num_bi
 
     const ValueType = std.meta.Int(.unsigned, num_bits);
 
-    return struct {
+    return extern struct {
         dummy: FieldType,
 
         fn field(self: anytype) PtrCastPreserveCV(@This(), @TypeOf(self), FieldType) {

src/Bus.zig

@@ -1,153 +0,0 @@
-const std = @import("std");
-
-const Bios = @import("bus/Bios.zig");
-const GamePak = @import("bus/GamePak.zig");
-const Io = @import("bus/io.zig").Io;
-const Ppu = @import("ppu.zig").Ppu;
-const Scheduler = @import("scheduler.zig").Scheduler;
-
-const Allocator = std.mem.Allocator;
-
-pak: GamePak,
-bios: Bios,
-ppu: Ppu,
-io: Io,
-
-pub fn init(alloc: Allocator, sched: *Scheduler, path: []const u8) !@This() {
-    return @This(){
-        .pak = try GamePak.init(alloc, path),
-        .bios = try Bios.init(alloc, "./bin/gba_bios.bin"), // TODO: don't hardcode this + bundle open-sorce Boot ROM
-        .ppu = try Ppu.init(alloc, sched),
-        .io = Io.init(),
-    };
-}
-
-pub fn deinit(self: @This()) void {
-    self.pak.deinit();
-    self.bios.deinit();
-    self.ppu.deinit();
-}
-
-pub fn read32(self: *const @This(), addr: u32) u32 {
-    return switch (addr) {
-        // General Internal Memory
-        0x0000_0000...0x0000_3FFF => self.bios.get32(@as(usize, addr)),
-        0x0200_0000...0x0203_FFFF => std.debug.panic("[Bus:32] read from 0x{X:} in IWRAM", .{addr}),
-        0x0300_0000...0x0300_7FFF => std.debug.panic("[Bus:32] read from 0x{X:} in EWRAM", .{addr}),
-        0x0400_0000...0x0400_03FE => self.read32(addr),
-
-        // Internal Display Memory
-        0x0500_0000...0x0500_03FF => self.ppu.palette.get32(@as(usize, addr - 0x0500_0000)),
-        0x0600_0000...0x0601_7FFF => self.ppu.vram.get32(@as(usize, addr - 0x0600_0000)),
-        0x0700_0000...0x0700_03FF => std.debug.panic("[Bus:32] read from 0x{X:} in OAM", .{addr}),
-
-        // External Memory (Game Pak)
-        0x0800_0000...0x09FF_FFFF => self.pak.get32(@as(usize, addr - 0x0800_0000)),
-        0x0A00_0000...0x0BFF_FFFF => self.pak.get32(@as(usize, addr - 0x0A00_0000)),
-        0x0C00_0000...0x0DFF_FFFF => self.pak.get32(@as(usize, addr - 0x0C00_0000)),
-
-        else => {
-            std.log.warn("[Bus:32] ZBA tried to read from 0x{X:}", .{addr});
-            return 0x0000_0000;
-        },
-    };
-}
-
-pub fn write32(self: *@This(), addr: u32, word: u32) void {
-    // TODO: write32 can write to GamePak Flash
-
-    switch (addr) {
-        // General Internal Memory
-        0x0200_0000...0x0203_FFFF => std.debug.panic("[Bus:32] wrote 0x{X:} to 0x{X:} in IWRAM", .{ word, addr }),
-        0x0300_0000...0x0300_7FFF => std.debug.panic("[Bus:32] wrote 0x{X:} to 0x{X:} in EWRAM", .{ word, addr }),
-        0x0400_0000...0x0400_03FE => std.debug.panic("[Bus:32] wrote 0x{X:} to 0x{X:} in I/O", .{ word, addr }),
-
-        // Internal Display Memory
-        0x0500_0000...0x0500_03FF => self.ppu.palette.set32(@as(usize, addr - 0x0500_0000), word),
-        0x0600_0000...0x0601_7FFF => self.ppu.vram.set32(@as(usize, addr - 0x0600_0000), word),
-        0x0700_0000...0x0700_03FF => std.debug.panic("[Bus:32] wrote 0x{X:} to 0x{X:} in OAM", .{ word, addr }),
-
-        else => std.log.warn("[Bus:32] ZBA tried to write 0x{X:} to 0x{X:}", .{ word, addr }),
-    }
-}
-
-pub fn read16(self: *const @This(), addr: u32) u16 {
-    return switch (addr) {
-        // General Internal Memory
-        0x0000_0000...0x0000_3FFF => self.bios.get16(@as(usize, addr)),
-        0x0200_0000...0x0203_FFFF => std.debug.panic("[Bus:16] read from 0x{X:} in IWRAM", .{addr}),
-        0x0300_0000...0x0300_7FFF => std.debug.panic("[Bus:16] read from 0x{X:} in EWRAM", .{addr}),
-        0x0400_0000...0x0400_03FE => self.io.read16(addr),
-
-        // Internal Display Memory
-        0x0500_0000...0x0500_03FF => self.ppu.palette.get16(@as(usize, addr - 0x0500_0000)),
-        0x0600_0000...0x0601_7FFF => self.ppu.vram.get16(@as(usize, addr - 0x0600_0000)),
-        0x0700_0000...0x0700_03FF => std.debug.panic("[Bus:16] read from 0x{X:} in OAM", .{addr}),
-
-        // External Memory (Game Pak)
-        0x0800_0000...0x09FF_FFFF => self.pak.get16(@as(usize, addr - 0x0800_0000)),
-        0x0A00_0000...0x0BFF_FFFF => self.pak.get16(@as(usize, addr - 0x0A00_0000)),
-        0x0C00_0000...0x0DFF_FFFF => self.pak.get16(@as(usize, addr - 0x0C00_0000)),
-
-        else => {
-            std.log.warn("[Bus:16] ZBA tried to read from 0x{X:}", .{addr});
-            return 0x0000;
-        },
-    };
-}
-
-pub fn write16(self: *@This(), addr: u32, halfword: u16) void {
-    // TODO: write16 can write to GamePak Flash
-    switch (addr) {
-        // General Internal Memory
-        0x0200_0000...0x0203_FFFF => std.debug.panic("[Bus:16] write 0x{X:} to 0x{X:} in IWRAM", .{ halfword, addr }),
-        0x0300_0000...0x0300_7FFF => std.debug.panic("[Bus:16] write 0x{X:} to 0x{X:} in EWRAM", .{ halfword, addr }),
-        0x0400_0000...0x0400_03FE => self.io.write16(addr, halfword),
-
-        // Internal Display Memory
-        0x0500_0000...0x0500_03FF => self.ppu.palette.set16(@as(usize, addr - 0x0500_0000), halfword),
-        0x0600_0000...0x0601_7FFF => self.ppu.vram.set16(@as(usize, addr - 0x0600_0000), halfword),
-        0x0700_0000...0x0700_03FF => std.debug.panic("[Bus:16] write 0x{X:} to 0x{X:} in OAM", .{ halfword, addr }),
-
-        else => std.log.warn("[Bus:16] ZBA tried to write 0x{X:} to 0x{X:}", .{ halfword, addr }),
-    }
-}
-
-pub fn read8(self: *const @This(), addr: u32) u8 {
-    return switch (addr) {
-        // General Internal Memory
-        0x0000_0000...0x0000_3FFF => self.bios.get8(@as(usize, addr)),
-        0x0200_0000...0x0203_FFFF => std.debug.panic("[Bus:8] read from 0x{X:} in IWRAM", .{addr}),
-        0x0300_0000...0x0300_7FFF => std.debug.panic("[Bus:8] read from 0x{X:} in EWRAM", .{addr}),
-        0x0400_0000...0x0400_03FE => self.io.read8(addr),
-
-        // Internal Display Memory
-        0x0500_0000...0x0500_03FF => self.ppu.palette.get8(@as(usize, addr - 0x0500_0000)),
-        0x0600_0000...0x0601_7FFF => self.ppu.vram.get8(@as(usize, addr - 0x0600_0000)),
-        0x0700_0000...0x0700_03FF => std.debug.panic("[Bus:8] read from 0x{X:} in OAM", .{addr}),
-
-        // External Memory (Game Pak)
-        0x0800_0000...0x09FF_FFFF => self.pak.get8(@as(usize, addr - 0x0800_0000)),
-        0x0A00_0000...0x0BFF_FFFF => self.pak.get8(@as(usize, addr - 0x0A00_0000)),
-        0x0C00_0000...0x0DFF_FFFF => self.pak.get8(@as(usize, addr - 0x0C00_0000)),
-        0x0E00_0000...0x0E00_FFFF => std.debug.panic("[Bus:8] read from 0x{X:} in Game Pak SRAM", .{addr}),
-
-        else => {
-            std.log.warn("[Bus:8] ZBA tried to read from 0x{X:}", .{addr});
-            return 0x00;
-        },
-    };
-}
-
-pub fn write8(_: *@This(), addr: u32, byte: u8) void {
-    switch (addr) {
-        // General Internal Memory
-        0x0200_0000...0x0203_FFFF => std.debug.panic("[Bus:8] write 0x{X:} to 0x{X:} in IWRAM", .{ byte, addr }),
-        0x0300_0000...0x0300_7FFF => std.debug.panic("[Bus:8] write 0x{X:} to 0x{X:} in EWRAM", .{ byte, addr }),
-        0x0400_0000...0x0400_03FE => std.debug.panic("[Bus:8] write 0x{X:} to 0x{X:} in I/O", .{ byte, addr }),
-
-        // External Memory (Game Pak)
-        0x0E00_0000...0x0E00_FFFF => std.debug.panic("[Bus:8] write 0x{X:} to 0x{X:} in Game Pak SRAM", .{ byte, addr }),
-        else => std.log.warn("[Bus:8] ZBA tried to write 0x{X:} to 0x{X:}", .{ byte, addr }),
-    }
-}

src/bus/Bios.zig

@@ -1,38 +0,0 @@
-const std = @import("std");
-
-const Allocator = std.mem.Allocator;
-const Self = @This();
-
-buf: []u8,
-alloc: Allocator,
-
-pub fn init(alloc: Allocator, path: []const u8) !Self {
-    const file = try std.fs.cwd().openFile(path, .{ .read = true });
-    defer file.close();
-
-    const len = try file.getEndPos();
-
-    return Self{
-        .buf = try file.readToEndAlloc(alloc, len),
-        .alloc = alloc,
-    };
-}
-
-pub fn deinit(self: Self) void {
-    self.alloc.free(self.buf);
-}
-
-pub inline fn get32(self: *const Self, idx: usize) u32 {
-    std.debug.panic("[BIOS] TODO: BIOS is not implemented", .{});
-    return (@as(u32, self.buf[idx + 3]) << 24) | (@as(u32, self.buf[idx + 2]) << 16) | (@as(u32, self.buf[idx + 1]) << 8) | (@as(u32, self.buf[idx]));
-}
-
-pub inline fn get16(self: *const Self, idx: usize) u16 {
-    std.debug.panic("[BIOS] TODO: BIOS is not implemented", .{});
-    return (@as(u16, self.buf[idx + 1]) << 8) | @as(u16, self.buf[idx]);
-}
-
-pub inline fn get8(self: *const Self, idx: usize) u8 {
-    std.debug.panic("[BIOS] TODO: BIOS is not implemented", .{});
-    return self.buf[idx];
-}

src/bus/GamePak.zig

@@ -1,35 +0,0 @@
-const std = @import("std");
-
-const Allocator = std.mem.Allocator;
-const Self = @This();
-
-buf: []u8,
-alloc: Allocator,
-
-pub fn init(alloc: Allocator, path: []const u8) !Self {
-    const file = try std.fs.cwd().openFile(path, .{ .read = true });
-    defer file.close();
-
-    const len = try file.getEndPos();
-
-    return Self{
-        .buf = try file.readToEndAlloc(alloc, len),
-        .alloc = alloc,
-    };
-}
-
-pub fn deinit(self: Self) void {
-    self.alloc.free(self.buf);
-}
-
-pub inline fn get32(self: *const Self, idx: usize) u32 {
-    return (@as(u32, self.get16(idx + 2)) << 16) | @as(u32, self.get16(idx));
-}
-
-pub inline fn get16(self: *const Self, idx: usize) u16 {
-    return (@as(u16, self.buf[idx + 1]) << 8) | @as(u16, self.buf[idx]);
-}
-
-pub inline fn get8(self: *const Self, idx: usize) u8 {
-    return self.buf[idx];
-}

src/bus/io.zig

@@ -1,81 +0,0 @@
-const std = @import("std");
-
-const Bit = @import("bitfield").Bit;
-const Bitfield = @import("bitfield").Bitfield;
-
-pub const Io = struct {
-    const Self = @This();
-
-    dispcnt: DispCnt,
-    dispstat: DispStat,
-    vcount: VCount,
-
-    pub fn init() Self {
-        return .{
-            .dispcnt = .{ .raw = 0x0000_0000 },
-            .dispstat = .{ .raw = 0x0000_0000 },
-            .vcount = .{ .raw = 0x0000_0000 },
-        };
-    }
-
-    pub fn read32(self: *const Self, addr: u32) u32 {
-        return switch (addr) {
-            0x0400_0000 => @as(u32, self.dispcnt.raw),
-            0x0400_0004 => @as(u32, self.dispstat.raw),
-            else => std.debug.panic("[I/O:32] tried to read from {X:}", .{addr}),
-        };
-    }
-
-    pub fn read16(self: *const Self, addr: u32) u16 {
-        return switch (addr) {
-            0x0400_0000 => self.dispcnt.raw,
-            0x0400_0004 => self.dispstat.raw,
-            else => std.debug.panic("[I/O:16] tried to read from {X:}", .{addr}),
-        };
-    }
-
-    pub fn write16(self: *Self, addr: u32, halfword: u16) void {
-        switch (addr) {
-            0x0400_0000 => self.dispcnt.raw = halfword,
-            0x0400_0004 => self.dispstat.raw = halfword,
-            else => std.debug.panic("[I/O:16] tried to write 0x{X:} to 0x{X:}", .{ halfword, addr }),
-        }
-    }
-
-    pub fn read8(self: *const Self, addr: u32) u8 {
-        return switch (addr) {
-            0x0400_0000 => @truncate(u8, self.dispcnt.raw),
-            0x0400_0004 => @truncate(u8, self.dispstat.raw),
-            else => std.debug.panic("[I/O:8] tried to read from {X:}", .{addr}),
-        };
-    }
-};
-
-const DispCnt = extern union {
-    bg_mode: Bitfield(u16, 0, 3),
-    frame_select: Bit(u16, 4),
-    hblank_interraw_free: Bit(u16, 5),
-    obj_mapping: Bit(u16, 6),
-    forced_blank: Bit(u16, 7),
-    bg_enable: Bitfield(u16, 8, 4),
-    obj_enable: Bit(u16, 12),
-    win_enable: Bitfield(u16, 13, 2),
-    obj_win_enable: Bit(u16, 15),
-    raw: u16,
-};
-
-const DispStat = extern union {
-    vblank: Bit(u16, 0),
-    hblank: Bit(u16, 1),
-    vcount: Bit(u16, 2),
-    vblank_irq: Bit(u16, 3),
-    hblank_irq: Bit(u16, 4),
-    vcount_irq: Bit(u16, 5),
-    vcount_setting: Bitfield(u16, 8, 7),
-    raw: u16,
-};
-
-const VCount = extern union {
-    scanline: Bitfield(u16, 0, 8),
-    raw: u16,
-};

src/config.zig

@@ -0,0 +1,86 @@
+const std = @import("std");
+const toml = @import("toml");
+
+const Allocator = std.mem.Allocator;
+
+const log = std.log.scoped(.Config);
+var state: Config = .{};
+
+const Config = struct {
+    host: Host = .{},
+    guest: Guest = .{},
+    debug: Debug = .{},
+
+    /// Settings related to the Computer the Emulator is being run on
+    const Host = struct {
+        /// Using Nearest-Neighbor, multiply the resolution of the GBA Window
+        win_scale: i64 = 3,
+        /// Enable Vsync
+        ///
+        /// Note: This does not affect whether Emulation is synced to 59Hz
+        vsync: bool = true,
+        /// Mute ZBA
+        mute: bool = false,
+    };
+
+    // Settings realted to the emulation itself
+    const Guest = struct {
+        /// Whether Emulation thread to sync to Audio Callbacks
+        audio_sync: bool = true,
+        /// Whether Emulation thread should sync to 59Hz
+        video_sync: bool = true,
+        /// Whether RTC I/O should always be enabled
+        force_rtc: bool = false,
+        /// Skip BIOS
+        skip_bios: bool = false,
+    };
+
+    /// Settings related to debugging ZBA
+    const Debug = struct {
+        /// Enable CPU Trace logs
+        cpu_trace: bool = false,
+        /// If false and ZBA is built in debug mode, ZBA will panic on unhandled I/O
+        unhandled_io: bool = true,
+    };
+};
+
+pub fn config() *const Config {
+    return &state;
+}
+
+/// Reads a config file and then loads it into the global state
+pub fn load(allocator: Allocator, file_path: []const u8) !void {
+    var config_file = try std.fs.cwd().openFile(file_path, .{});
+    defer config_file.close();
+
+    log.info("loaded from {s}", .{file_path});
+
+    const contents = try config_file.readToEndAlloc(allocator, try config_file.getEndPos());
+    defer allocator.free(contents);
+
+    var parser = try toml.parseFile(allocator, file_path);
+    defer parser.deinit();
+
+    const table = try parser.parse();
+    defer table.deinit();
+
+    // TODO: Report unknown config options
+
+    if (table.keys.get("Host")) |host| {
+        if (host.Table.keys.get("win_scale")) |scale| state.host.win_scale = scale.Integer;
+        if (host.Table.keys.get("vsync")) |vsync| state.host.vsync = vsync.Boolean;
+        if (host.Table.keys.get("mute")) |mute| state.host.mute = mute.Boolean;
+    }
+
+    if (table.keys.get("Guest")) |guest| {
+        if (guest.Table.keys.get("audio_sync")) |sync| state.guest.audio_sync = sync.Boolean;
+        if (guest.Table.keys.get("video_sync")) |sync| state.guest.video_sync = sync.Boolean;
+        if (guest.Table.keys.get("force_rtc")) |forced| state.guest.force_rtc = forced.Boolean;
+        if (guest.Table.keys.get("skip_bios")) |skip| state.guest.skip_bios = skip.Boolean;
+    }
+
+    if (table.keys.get("Debug")) |debug| {
+        if (debug.Table.keys.get("cpu_trace")) |trace| state.debug.cpu_trace = trace.Boolean;
+        if (debug.Table.keys.get("unhandled_io")) |unhandled| state.debug.unhandled_io = unhandled.Boolean;
+    }
+}

src/core/Bus.zig

@@ -0,0 +1,500 @@
+const std = @import("std");
+
+const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
+const Bios = @import("bus/Bios.zig");
+const Ewram = @import("bus/Ewram.zig");
+const GamePak = @import("bus/GamePak.zig");
+const Io = @import("bus/io.zig").Io;
+const Iwram = @import("bus/Iwram.zig");
+const Ppu = @import("ppu.zig").Ppu;
+const Apu = @import("apu.zig").Apu;
+const DmaTuple = @import("bus/dma.zig").DmaTuple;
+const TimerTuple = @import("bus/timer.zig").TimerTuple;
+const Scheduler = @import("scheduler.zig").Scheduler;
+const FilePaths = @import("../util.zig").FilePaths;
+
+const io = @import("bus/io.zig");
+const Allocator = std.mem.Allocator;
+const log = std.log.scoped(.Bus);
+
+const createDmaTuple = @import("bus/dma.zig").create;
+const createTimerTuple = @import("bus/timer.zig").create;
+const rotr = @import("../util.zig").rotr;
+
+const timings: [2][0x10]u8 = [_][0x10]u8{
+    // BIOS, Unused, EWRAM, IWRAM, I/0, PALRAM, VRAM, OAM, ROM0, ROM0, ROM1, ROM1, ROM2, ROM2, SRAM, Unused
+    [_]u8{ 1, 1, 3, 1, 1, 1, 1, 1, 5, 5, 5, 5, 5, 5, 5, 5 }, // 8-bit & 16-bit
+    [_]u8{ 1, 1, 6, 1, 1, 2, 2, 1, 8, 8, 8, 8, 8, 8, 8, 8 }, // 32-bit
+};
+
+pub const fetch_timings: [2][0x10]u8 = [_][0x10]u8{
+    // BIOS, Unused, EWRAM, IWRAM, I/0, PALRAM, VRAM, OAM, ROM0, ROM0, ROM1, ROM1, ROM2, ROM2, SRAM, Unused
+    [_]u8{ 1, 1, 3, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 5, 5 }, // 8-bit & 16-bit
+    [_]u8{ 1, 1, 6, 1, 1, 2, 2, 1, 4, 4, 4, 4, 4, 4, 8, 8 }, // 32-bit
+};
+
+// Fastmem Related
+const page_size = 1 * 0x400; // 1KiB
+const address_space_size = 0x1000_0000;
+const table_len = address_space_size / page_size;
+
+const Self = @This();
+
+pak: GamePak,
+bios: Bios,
+ppu: Ppu,
+apu: Apu,
+dma: DmaTuple,
+tim: TimerTuple,
+iwram: Iwram,
+ewram: Ewram,
+io: Io,
+
+cpu: *Arm7tdmi,
+sched: *Scheduler,
+
+read_table: *const [table_len]?*const anyopaque,
+write_tables: [2]*const [table_len]?*anyopaque,
+allocator: Allocator,
+
+pub fn init(self: *Self, allocator: Allocator, sched: *Scheduler, cpu: *Arm7tdmi, paths: FilePaths) !void {
+    const tables = try allocator.alloc(?*anyopaque, 3 * table_len); // Allocate all tables
+
+    const read_table = tables[0..table_len];
+    const write_tables = .{ tables[table_len .. 2 * table_len], tables[2 * table_len .. 3 * table_len] };
+
+    self.* = .{
+        .pak = try GamePak.init(allocator, cpu, paths.rom, paths.save),
+        .bios = try Bios.init(allocator, paths.bios),
+        .ppu = try Ppu.init(allocator, sched),
+        .apu = Apu.init(sched),
+        .iwram = try Iwram.init(allocator),
+        .ewram = try Ewram.init(allocator),
+        .dma = createDmaTuple(),
+        .tim = createTimerTuple(sched),
+        .io = Io.init(),
+        .cpu = cpu,
+        .sched = sched,
+
+        .read_table = read_table,
+        .write_tables = write_tables,
+        .allocator = allocator,
+    };
+
+    self.fillReadTable(read_table);
+
+    // Internal Display Memory behaves differently on 8-bit reads
+    self.fillWriteTable(u32, write_tables[0]);
+    self.fillWriteTable(u8, write_tables[1]);
+}
+
+pub fn deinit(self: *Self) void {
+    self.iwram.deinit();
+    self.ewram.deinit();
+    self.pak.deinit();
+    self.bios.deinit();
+    self.ppu.deinit();
+
+    // This is so I can deallocate the original `allocator.alloc`. I have to re-make the type
+    // since I'm not keeping it around, This is very jank and bad though
+    // FIXME: please figure out another way
+    self.allocator.free(@ptrCast([*]const ?*anyopaque, self.read_table[0..])[0 .. 3 * table_len]);
+    self.* = undefined;
+}
+
+fn fillReadTable(self: *Self, table: *[table_len]?*const anyopaque) void {
+    const vramMirror = @import("ppu/Vram.zig").mirror;
+
+    for (table, 0..) |*ptr, i| {
+        const addr = @intCast(u32, page_size * i);
+
+        ptr.* = switch (addr) {
+            // General Internal Memory
+            0x0000_0000...0x0000_3FFF => null, // BIOS has it's own checks
+            0x0200_0000...0x02FF_FFFF => &self.ewram.buf[addr & 0x3FFFF],
+            0x0300_0000...0x03FF_FFFF => &self.iwram.buf[addr & 0x7FFF],
+            0x0400_0000...0x0400_03FF => null, // I/O
+
+            // Internal Display Memory
+            0x0500_0000...0x05FF_FFFF => &self.ppu.palette.buf[addr & 0x3FF],
+            0x0600_0000...0x06FF_FFFF => &self.ppu.vram.buf[vramMirror(addr)],
+            0x0700_0000...0x07FF_FFFF => &self.ppu.oam.buf[addr & 0x3FF],
+
+            // External Memory (Game Pak)
+            0x0800_0000...0x0DFF_FFFF => self.fillReadTableExternal(addr),
+            0x0E00_0000...0x0FFF_FFFF => null, // SRAM
+            else => null,
+        };
+    }
+}
+
+fn fillWriteTable(self: *Self, comptime T: type, table: *[table_len]?*const anyopaque) void {
+    comptime std.debug.assert(T == u32 or T == u16 or T == u8);
+    const vramMirror = @import("ppu/Vram.zig").mirror;
+
+    for (table, 0..) |*ptr, i| {
+        const addr = @intCast(u32, page_size * i);
+
+        ptr.* = switch (addr) {
+            // General Internal Memory
+            0x0000_0000...0x0000_3FFF => null, // BIOS has it's own checks
+            0x0200_0000...0x02FF_FFFF => &self.ewram.buf[addr & 0x3FFFF],
+            0x0300_0000...0x03FF_FFFF => &self.iwram.buf[addr & 0x7FFF],
+            0x0400_0000...0x0400_03FF => null, // I/O
+
+            // Internal Display Memory
+            0x0500_0000...0x05FF_FFFF => if (T != u8) &self.ppu.palette.buf[addr & 0x3FF] else null,
+            0x0600_0000...0x06FF_FFFF => if (T != u8) &self.ppu.vram.buf[vramMirror(addr)] else null,
+            0x0700_0000...0x07FF_FFFF => if (T != u8) &self.ppu.oam.buf[addr & 0x3FF] else null,
+
+            // External Memory (Game Pak)
+            0x0800_0000...0x0DFF_FFFF => null, // ROM
+            0x0E00_0000...0x0FFF_FFFF => null, // SRAM
+            else => null,
+        };
+    }
+}
+
+fn fillReadTableExternal(self: *Self, addr: u32) ?*anyopaque {
+    // see `GamePak.zig` for more information about what conditions need to be true
+    // so that a simple pointer dereference isn't possible
+
+    std.debug.assert(addr & @as(u32, page_size - 1) == 0); // addr is guaranteed to be page-aligned
+
+    const start_addr = addr;
+    const end_addr = start_addr + page_size;
+
+    {
+        const data = start_addr <= 0x0800_00C4 and 0x0800_00C4 < end_addr; // GPIO Data
+        const direction = start_addr <= 0x0800_00C6 and 0x0800_00C6 < end_addr; // GPIO Direction
+        const control = start_addr <= 0x0800_00C8 and 0x0800_00C8 < end_addr; // GPIO Control
+
+        const has_gpio = data or direction or control;
+        const gpio_kind = self.pak.gpio.device.kind;
+
+        // There is a GPIO Device, and the current page contains at least one memory-mapped GPIO register
+        if (gpio_kind != .None and has_gpio) return null;
+    }
+
+    if (self.pak.backup.kind == .Eeprom) {
+        if (self.pak.buf.len > 0x100_000) {
+            // We are using a "large" EEPROM which means that if the below check is true
+            // this page has an address that's reserved for the EEPROM and therefore must
+            // be handled in slowmem
+            if (addr & 0x1FF_FFFF > 0x1FF_FEFF) return null;
+        } else {
+            // We are using a "small" EEPROM which means that if the below check is true
+            // (that is, we're in the 0xD address page) then we must handle at least one
+            // address in this page in slowmem
+            if (@truncate(u4, addr >> 24) == 0xD) return null;
+        }
+    }
+
+    // Finally, the GamePak has some unique behaviour for reads past the end of the ROM,
+    // so those will be handled by slowmem as well
+    const masked_addr = addr & 0x1FF_FFFF;
+    if (masked_addr >= self.pak.buf.len) return null;
+
+    return &self.pak.buf[masked_addr];
+}
+
+fn readIo(self: *const Self, comptime T: type, address: u32) T {
+    return io.read(self, T, address) orelse self.openBus(T, address);
+}
+
+fn openBus(self: *const Self, comptime T: type, address: u32) T {
+    @setCold(true);
+    const r15 = self.cpu.r[15];
+
+    const word = blk: {
+        // If Arm, get the most recently fetched instruction (PC + 8)
+        //
+        // FIXME: This is most likely a faulty assumption.
+        // I think what *actually* happens is that the Bus has a latch for the most
+        // recently fetched piece of data, which is then returned during Open Bus (also DMA open bus?)
+        // I can "get away" with this because it's very statistically likely that the most recently latched value is
+        // the most recently fetched instruction by the pipeline
+        if (!self.cpu.cpsr.t.read()) break :blk self.cpu.pipe.stage[1].?;
+
+        const page = @truncate(u8, r15 >> 24);
+
+        // PC + 2 = stage[0]
+        // PC + 4 = stage[1]
+        // PC + 6 = Need a Debug Read for this?
+
+        switch (page) {
+            // EWRAM, PALRAM, VRAM, and Game ROM (16-bit)
+            0x02, 0x05, 0x06, 0x08...0x0D => {
+                const halfword: u32 = @truncate(u16, self.cpu.pipe.stage[1].?);
+                break :blk halfword << 16 | halfword;
+            },
+
+            // BIOS or OAM (32-bit)
+            0x00, 0x07 => {
+                // Aligned: (PC + 6) | (PC + 4)
+                // Unaligned: (PC + 4) | (PC + 2)
+                const aligned = address & 3 == 0b00;
+
+                // TODO: What to do on PC + 6?
+                const high: u32 = if (aligned) self.dbgRead(u16, r15 + 4) else @truncate(u16, self.cpu.pipe.stage[1].?);
+                const low: u32 = @truncate(u16, self.cpu.pipe.stage[@boolToInt(aligned)].?);
+
+                break :blk high << 16 | low;
+            },
+
+            // IWRAM (16-bit but special)
+            0x03 => {
+                // Aligned: (PC + 2) | (PC + 4)
+                // Unaligned: (PC + 4) | (PC + 2)
+                const aligned = address & 3 == 0b00;
+
+                const high: u32 = @truncate(u16, self.cpu.pipe.stage[1 - @boolToInt(aligned)].?);
+                const low: u32 = @truncate(u16, self.cpu.pipe.stage[@boolToInt(aligned)].?);
+
+                break :blk high << 16 | low;
+            },
+            else => {
+                log.err("THUMB open bus read from 0x{X:0>2} page @0x{X:0>8}", .{ page, address });
+                @panic("invariant most-likely broken");
+            },
+        }
+    };
+
+    return @truncate(T, word);
+}
+
+pub fn read(self: *Self, comptime T: type, unaligned_address: u32) T {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // whether or not we do this in slowmem or fastmem, we should advance the scheduler
+    self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, unaligned_address >> 24)];
+
+    // We're doing some serious out-of-bounds open-bus reads
+    if (page >= table_len) return self.openBus(T, unaligned_address);
+
+    if (self.read_table[page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]const T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        return ptr[forceAlign(T, offset) / @sizeOf(T)];
+    }
+
+    return self.slowRead(T, unaligned_address);
+}
+
+pub fn dbgRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // We're doing some serious out-of-bounds open-bus reads
+    if (page >= table_len) return self.openBus(T, unaligned_address);
+
+    if (self.read_table[page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]const T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        return ptr[forceAlign(T, offset) / @sizeOf(T)];
+    }
+
+    return self.dbgSlowRead(T, unaligned_address);
+}
+
+fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T {
+    @setCold(true);
+
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    return switch (page) {
+        // General Internal Memory
+        0x00 => blk: {
+            if (address < Bios.size)
+                break :blk self.bios.read(T, self.cpu.r[15], unaligned_address);
+
+            break :blk self.openBus(T, address);
+        },
+        0x02 => unreachable, // completely handled by fastmeme
+        0x03 => unreachable, // completely handled by fastmeme
+        0x04 => self.readIo(T, address),
+
+        // Internal Display Memory
+        0x05 => unreachable, // completely handled by fastmeme
+        0x06 => unreachable, // completely handled by fastmeme
+        0x07 => unreachable, // completely handled by fastmeme
+
+        // External Memory (Game Pak)
+        0x08...0x0D => self.pak.read(T, address),
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
+        else => self.openBus(T, address),
+    };
+}
+
+fn dbgSlowRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    return switch (page) {
+        // General Internal Memory
+        0x00 => blk: {
+            if (address < Bios.size)
+                break :blk self.bios.dbgRead(T, self.cpu.r[15], unaligned_address);
+
+            break :blk self.openBus(T, address);
+        },
+        0x02 => unreachable, // handled by fastmem
+        0x03 => unreachable, // handled by fastmem
+        0x04 => self.readIo(T, address),
+
+        // Internal Display Memory
+        0x05 => unreachable, // handled by fastmem
+        0x06 => unreachable, // handled by fastmem
+        0x07 => unreachable, // handled by fastmem
+
+        // External Memory (Game Pak)
+        0x08...0x0D => self.pak.dbgRead(T, address),
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
+        else => self.openBus(T, address),
+    };
+}
+
+fn readBackup(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const value = self.pak.backup.read(unaligned_address);
+
+    const multiplier = switch (T) {
+        u32 => 0x01010101,
+        u16 => 0x0101,
+        u8 => 1,
+        else => @compileError("Backup: Unsupported read width"),
+    };
+
+    return @as(T, value) * multiplier;
+}
+
+pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // whether or not we do this in slowmem or fastmem, we should advance the scheduler
+    self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, unaligned_address >> 24)];
+
+    // We're doing some serious out-of-bounds open-bus writes, they do nothing though
+    if (page >= table_len) return;
+
+    if (self.write_tables[@boolToInt(T == u8)][page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        ptr[forceAlign(T, offset) / @sizeOf(T)] = value;
+    } else {
+        // we can return early if this is an 8-bit OAM write
+        if (T == u8 and @truncate(u8, unaligned_address >> 24) == 0x07) return;
+
+        self.slowWrite(T, unaligned_address, value);
+    }
+}
+
+/// Mostly Identical to `Bus.write`, slowmeme is handled by `Bus.dbgSlowWrite`
+pub fn dbgWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // We're doing some serious out-of-bounds open-bus writes, they do nothing though
+    if (page >= table_len) return;
+
+    if (self.write_tables[@boolToInt(T == u8)][page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        ptr[forceAlign(T, offset) / @sizeOf(T)] = value;
+    } else {
+        // we can return early if this is an 8-bit OAM write
+        if (T == u8 and @truncate(u8, unaligned_address >> 24) == 0x07) return;
+
+        self.dbgSlowWrite(T, unaligned_address, value);
+    }
+}
+
+fn slowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    @setCold(true);
+
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    switch (page) {
+        // General Internal Memory
+        0x00 => self.bios.write(T, address, value),
+        0x02 => unreachable, // completely handled by fastmem
+        0x03 => unreachable, // completely handled by fastmem
+        0x04 => io.write(self, T, address, value),
+
+        // Internal Display Memory
+        0x05 => self.ppu.palette.write(T, address, value),
+        0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, address, value),
+        0x07 => unreachable, // completely handled by fastmem
+
+        // External Memory (Game Pak)
+        0x08...0x0D => self.pak.write(T, self.dma[3].word_count, address, value),
+        0x0E...0x0F => self.pak.backup.write(unaligned_address, @truncate(u8, rotr(T, value, 8 * rotateBy(T, unaligned_address)))),
+        else => {},
+    }
+}
+
+fn dbgSlowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    @setCold(true);
+
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    switch (page) {
+        // General Internal Memory
+        0x00 => self.bios.write(T, address, value),
+        0x02 => unreachable, // completely handled by fastmem
+        0x03 => unreachable, // completely handled by fastmem
+        0x04 => return, // FIXME: Let debug writes mess with I/O
+
+        // Internal Display Memory
+        0x05 => self.ppu.palette.write(T, address, value),
+        0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, address, value),
+        0x07 => unreachable, // completely handled by fastmem
+
+        // External Memory (Game Pak)
+        0x08...0x0D => return, // FIXME: Debug Write to Backup/GPIO w/out messing with state
+        0x0E...0x0F => return, // FIXME: Debug Write to Backup w/out messing with state
+        else => {},
+    }
+}
+
+inline fn rotateBy(comptime T: type, address: u32) u32 {
+    return switch (T) {
+        u32 => address & 3,
+        u16 => address & 1,
+        u8 => 0,
+        else => @compileError("Unsupported write width"),
+    };
+}
+
+pub inline fn forceAlign(comptime T: type, address: u32) u32 {
+    return switch (T) {
+        u32 => address & ~@as(u32, 3),
+        u16 => address & ~@as(u32, 1),
+        u8 => address,
+        else => @compileError("Bus: Invalid read/write type"),
+    };
+}

src/core/apu.zig

@@ -0,0 +1,582 @@
+const std = @import("std");
+const SDL = @import("sdl2");
+const io = @import("bus/io.zig");
+const util = @import("../util.zig");
+
+const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
+const Scheduler = @import("scheduler.zig").Scheduler;
+const ToneSweep = @import("apu/ToneSweep.zig");
+const Tone = @import("apu/Tone.zig");
+const Wave = @import("apu/Wave.zig");
+const Noise = @import("apu/Noise.zig");
+
+const SoundFifo = std.fifo.LinearFifo(u8, .{ .Static = 0x20 });
+
+const getHalf = util.getHalf;
+const setHalf = util.setHalf;
+const intToBytes = util.intToBytes;
+
+const log = std.log.scoped(.APU);
+
+pub const host_rate = @import("../platform.zig").sample_rate;
+pub const host_format = @import("../platform.zig").sample_format;
+
+pub fn read(comptime T: type, apu: *const Apu, addr: u32) ?T {
+    const byte_addr = @truncate(u8, addr);
+
+    return switch (T) {
+        u32 => switch (byte_addr) {
+            0x60 => @as(T, apu.ch1.sound1CntH()) << 16 | apu.ch1.sound1CntL(),
+            0x64 => apu.ch1.sound1CntX(),
+            0x68 => apu.ch2.sound2CntL(),
+            0x6C => apu.ch2.sound2CntH(),
+            0x70 => @as(T, apu.ch3.sound3CntH()) << 16 | apu.ch3.sound3CntL(),
+            0x74 => apu.ch3.sound3CntX(),
+            0x78 => apu.ch4.sound4CntL(),
+            0x7C => apu.ch4.sound4CntH(),
+            0x80 => @as(T, apu.dma_cnt.raw) << 16 | apu.psg_cnt.raw, // SOUNDCNT_H, SOUNDCNT_L
+            0x84 => apu.soundCntX(),
+            0x88 => apu.bias.raw, // SOUNDBIAS, high is unused
+            0x8C => null,
+            0x90, 0x94, 0x98, 0x9C => apu.ch3.wave_dev.read(T, apu.ch3.select, addr),
+            0xA0 => null, // FIFO_A
+            0xA4 => null, // FIFO_B
+            else => util.io.read.err(T, log, "unaligned {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        u16 => switch (byte_addr) {
+            0x60 => apu.ch1.sound1CntL(),
+            0x62 => apu.ch1.sound1CntH(),
+            0x64 => apu.ch1.sound1CntX(),
+            0x66 => 0x0000, // suite.gba expects 0x0000, not 0xDEAD
+            0x68 => apu.ch2.sound2CntL(),
+            0x6A => 0x0000,
+            0x6C => apu.ch2.sound2CntH(),
+            0x6E => 0x0000,
+            0x70 => apu.ch3.sound3CntL(),
+            0x72 => apu.ch3.sound3CntH(),
+            0x74 => apu.ch3.sound3CntX(),
+            0x76 => 0x0000,
+            0x78 => apu.ch4.sound4CntL(),
+            0x7A => 0x0000,
+            0x7C => apu.ch4.sound4CntH(),
+            0x7E => 0x0000,
+            0x80 => apu.soundCntL(),
+            0x82 => apu.soundCntH(),
+            0x84 => apu.soundCntX(),
+            0x86 => 0x0000,
+            0x88 => apu.bias.raw, // SOUNDBIAS
+            0x8A => 0x0000,
+            0x8C, 0x8E => null,
+            0x90, 0x92, 0x94, 0x96, 0x98, 0x9A, 0x9C, 0x9E => apu.ch3.wave_dev.read(T, apu.ch3.select, addr),
+            0xA0, 0xA2 => null, // FIFO_A
+            0xA4, 0xA6 => null, // FIFO_B
+            else => util.io.read.err(T, log, "unaligned {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        u8 => switch (byte_addr) {
+            0x60, 0x61 => @truncate(T, @as(u16, apu.ch1.sound1CntL()) >> getHalf(byte_addr)),
+            0x62, 0x63 => @truncate(T, apu.ch1.sound1CntH() >> getHalf(byte_addr)),
+            0x64, 0x65 => @truncate(T, apu.ch1.sound1CntX() >> getHalf(byte_addr)),
+            0x66, 0x67 => 0x00, // assuming behaviour is identical to that of 16-bit reads
+            0x68, 0x69 => @truncate(T, apu.ch2.sound2CntL() >> getHalf(byte_addr)),
+            0x6A, 0x6B => 0x00,
+            0x6C, 0x6D => @truncate(T, apu.ch2.sound2CntH() >> getHalf(byte_addr)),
+            0x6E, 0x6F => 0x00,
+            0x70, 0x71 => @truncate(T, @as(u16, apu.ch3.sound3CntL()) >> getHalf(byte_addr)), // SOUND3CNT_L
+            0x72, 0x73 => @truncate(T, apu.ch3.sound3CntH() >> getHalf(byte_addr)),
+            0x74, 0x75 => @truncate(T, apu.ch3.sound3CntX() >> getHalf(byte_addr)), // SOUND3CNT_L
+            0x76, 0x77 => 0x00,
+            0x78, 0x79 => @truncate(T, apu.ch4.sound4CntL() >> getHalf(byte_addr)),
+            0x7A, 0x7B => 0x00,
+            0x7C, 0x7D => @truncate(T, apu.ch4.sound4CntH() >> getHalf(byte_addr)),
+            0x7E, 0x7F => 0x00,
+            0x80, 0x81 => @truncate(T, apu.soundCntL() >> getHalf(byte_addr)), // SOUNDCNT_L
+            0x82, 0x83 => @truncate(T, apu.soundCntH() >> getHalf(byte_addr)), // SOUNDCNT_H
+            0x84, 0x85 => @truncate(T, @as(u16, apu.soundCntX()) >> getHalf(byte_addr)),
+            0x86, 0x87 => 0x00,
+            0x88, 0x89 => @truncate(T, apu.bias.raw >> getHalf(byte_addr)), // SOUNDBIAS
+            0x8A, 0x8B => 0x00,
+            0x8C...0x8F => null,
+            0x90...0x9F => apu.ch3.wave_dev.read(T, apu.ch3.select, addr),
+            0xA0, 0xA1, 0xA2, 0xA3 => null, // FIFO_A
+            0xA4, 0xA5, 0xA6, 0xA7 => null, // FIFO_B
+            else => util.io.read.err(T, log, "unexpected {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        else => @compileError("APU: Unsupported read width"),
+    };
+}
+
+pub fn write(comptime T: type, apu: *Apu, addr: u32, value: T) void {
+    const byte_addr = @truncate(u8, addr);
+
+    if (byte_addr <= 0x81 and !apu.cnt.apu_enable.read()) return;
+
+    switch (T) {
+        u32 => {
+            // 0x80 and 0x81 handled in setSoundCnt
+            if (byte_addr < 0x80 and !apu.cnt.apu_enable.read()) return;
+
+            switch (byte_addr) {
+                0x60 => apu.ch1.setSound1Cnt(value),
+                0x64 => apu.ch1.setSound1CntX(&apu.fs, @truncate(u16, value)),
+
+                0x68 => apu.ch2.setSound2CntL(@truncate(u16, value)),
+                0x6C => apu.ch2.setSound2CntH(&apu.fs, @truncate(u16, value)),
+
+                0x70 => apu.ch3.setSound3Cnt(value),
+                0x74 => apu.ch3.setSound3CntX(&apu.fs, @truncate(u16, value)),
+
+                0x78 => apu.ch4.setSound4CntL(@truncate(u16, value)),
+                0x7C => apu.ch4.setSound4CntH(&apu.fs, @truncate(u16, value)),
+
+                0x80 => apu.setSoundCnt(value),
+                0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
+                0x88 => apu.bias.raw = @truncate(u16, value),
+                0x8C => {},
+
+                0x90, 0x94, 0x98, 0x9C => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0xA0 => apu.chA.push(value), // FIFO_A
+                0xA4 => apu.chB.push(value), // FIFO_B
+                else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, addr }),
+            }
+        },
+        u16 => {
+            if (byte_addr <= 0x81 and !apu.cnt.apu_enable.read()) return;
+
+            switch (byte_addr) {
+                0x60 => apu.ch1.setSound1CntL(@truncate(u8, value)), // SOUND1CNT_L
+                0x62 => apu.ch1.setSound1CntH(value),
+                0x64 => apu.ch1.setSound1CntX(&apu.fs, value),
+                0x66 => {},
+
+                0x68 => apu.ch2.setSound2CntL(value),
+                0x6A => {},
+                0x6C => apu.ch2.setSound2CntH(&apu.fs, value),
+                0x6E => {},
+
+                0x70 => apu.ch3.setSound3CntL(@truncate(u8, value)),
+                0x72 => apu.ch3.setSound3CntH(value),
+                0x74 => apu.ch3.setSound3CntX(&apu.fs, value),
+                0x76 => {},
+
+                0x78 => apu.ch4.setSound4CntL(value),
+                0x7A => {},
+                0x7C => apu.ch4.setSound4CntH(&apu.fs, value),
+                0x7E => {},
+
+                0x80 => apu.setSoundCntL(value),
+                0x82 => apu.setSoundCntH(value),
+                0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
+                0x86 => {},
+                0x88 => apu.bias.raw = value, // SOUNDBIAS
+                0x8A, 0x8C, 0x8E => {},
+
+                0x90, 0x92, 0x94, 0x96, 0x98, 0x9A, 0x9C, 0x9E => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0xA0, 0xA2 => log.err("Tried to write 0x{X:0>4}{} to FIFO_A", .{ value, T }),
+                0xA4, 0xA6 => log.err("Tried to write 0x{X:0>4}{} to FIFO_B", .{ value, T }),
+                else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, addr }),
+            }
+        },
+        u8 => {
+            if (byte_addr <= 0x81 and !apu.cnt.apu_enable.read()) return;
+
+            switch (byte_addr) {
+                0x60 => apu.ch1.setSound1CntL(value),
+                0x61 => {},
+                0x62 => apu.ch1.setNr11(value),
+                0x63 => apu.ch1.setNr12(value),
+                0x64 => apu.ch1.setNr13(value),
+                0x65 => apu.ch1.setNr14(&apu.fs, value),
+                0x66, 0x67 => {},
+
+                0x68 => apu.ch2.setNr21(value),
+                0x69 => apu.ch2.setNr22(value),
+                0x6A, 0x6B => {},
+                0x6C => apu.ch2.setNr23(value),
+                0x6D => apu.ch2.setNr24(&apu.fs, value),
+                0x6E, 0x6F => {},
+
+                0x70 => apu.ch3.setSound3CntL(value), // NR30
+                0x71 => {},
+                0x72 => apu.ch3.setNr31(value),
+                0x73 => apu.ch3.vol.raw = value, // NR32
+                0x74 => apu.ch3.setNr33(value),
+                0x75 => apu.ch3.setNr34(&apu.fs, value),
+                0x76, 0x77 => {},
+
+                0x78 => apu.ch4.setNr41(value),
+                0x79 => apu.ch4.setNr42(value),
+                0x7A, 0x7B => {},
+                0x7C => apu.ch4.poly.raw = value, // NR 43
+                0x7D => apu.ch4.setNr44(&apu.fs, value),
+                0x7E, 0x7F => {},
+
+                0x80, 0x81 => apu.setSoundCntL(setHalf(u16, apu.psg_cnt.raw, byte_addr, value)),
+                0x82, 0x83 => apu.setSoundCntH(setHalf(u16, apu.dma_cnt.raw, byte_addr, value)),
+                0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
+                0x85 => {},
+                0x86, 0x87 => {},
+                0x88, 0x89 => apu.bias.raw = setHalf(u16, apu.bias.raw, byte_addr, value), // SOUNDBIAS
+                0x8A...0x8F => {},
+
+                0x90...0x9F => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0xA0...0xA3 => log.err("Tried to write 0x{X:0>2}{} to FIFO_A", .{ value, T }),
+                0xA4...0xA7 => log.err("Tried to write 0x{X:0>2}{} to FIFO_B", .{ value, T }),
+                else => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, addr }),
+            }
+        },
+        else => @compileError("APU: Unsupported write width"),
+    }
+}
+
+pub const Apu = struct {
+    const Self = @This();
+
+    ch1: ToneSweep,
+    ch2: Tone,
+    ch3: Wave,
+    ch4: Noise,
+    chA: DmaSound(.A),
+    chB: DmaSound(.B),
+
+    bias: io.SoundBias,
+    /// NR50, NR51
+    psg_cnt: io.ChannelVolumeControl,
+    dma_cnt: io.DmaSoundControl,
+    cnt: io.SoundControl,
+
+    sampling_cycle: u2,
+
+    stream: *SDL.SDL_AudioStream,
+    sched: *Scheduler,
+
+    fs: FrameSequencer,
+    capacitor: f32,
+
+    is_buffer_full: bool,
+
+    pub const Tick = enum { Length, Envelope, Sweep };
+
+    pub fn init(sched: *Scheduler) Self {
+        const apu: Self = .{
+            .ch1 = ToneSweep.init(sched),
+            .ch2 = Tone.init(sched),
+            .ch3 = Wave.init(sched),
+            .ch4 = Noise.init(sched),
+            .chA = DmaSound(.A).init(),
+            .chB = DmaSound(.B).init(),
+
+            .psg_cnt = .{ .raw = 0 },
+            .dma_cnt = .{ .raw = 0 },
+            .cnt = .{ .raw = 0 },
+            .bias = .{ .raw = 0x0200 },
+
+            .sampling_cycle = 0b00,
+            .stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, 1 << 15, host_format, 2, host_rate).?,
+            .sched = sched,
+
+            .capacitor = 0,
+            .fs = FrameSequencer.init(),
+            .is_buffer_full = false,
+        };
+
+        sched.push(.SampleAudio, apu.interval());
+        sched.push(.{ .ApuChannel = 0 }, @import("apu/signal/Square.zig").interval);
+        sched.push(.{ .ApuChannel = 1 }, @import("apu/signal/Square.zig").interval);
+        sched.push(.{ .ApuChannel = 2 }, @import("apu/signal/Wave.zig").interval);
+        sched.push(.{ .ApuChannel = 3 }, @import("apu/signal/Lfsr.zig").interval);
+        sched.push(.FrameSequencer, FrameSequencer.interval);
+
+        return apu;
+    }
+
+    fn reset(self: *Self) void {
+        // All PSG Registers between 0x0400_0060..0x0400_0081 are zeroed
+        // 0x0400_0082 and 0x0400_0088 retain their values
+        self.ch1.reset();
+        self.ch2.reset();
+        self.ch3.reset();
+        self.ch4.reset();
+
+        // GBATEK says 4000060h..4000081h I take this to mean inclusive
+        self.psg_cnt.raw = 0x0000;
+    }
+
+    /// SOUNDCNT
+    fn setSoundCnt(self: *Self, value: u32) void {
+        if (self.cnt.apu_enable.read()) self.setSoundCntL(@truncate(u16, value));
+        self.setSoundCntH(@truncate(u16, value >> 16));
+    }
+
+    /// SOUNDCNT_L
+    pub fn soundCntL(self: *const Self) u16 {
+        return self.psg_cnt.raw & 0xFF77;
+    }
+
+    /// SOUNDCNT_L
+    pub fn setSoundCntL(self: *Self, value: u16) void {
+        self.psg_cnt.raw = value;
+    }
+
+    /// SOUNDCNT_H
+    pub fn setSoundCntH(self: *Self, value: u16) void {
+        const new: io.DmaSoundControl = .{ .raw = value };
+
+        // Reinitializing instead of resetting is fine because
+        // the FIFOs I'm using are stack allocated and 0x20 bytes big
+        if (new.chA_reset.read()) self.chA.fifo = SoundFifo.init();
+        if (new.chB_reset.read()) self.chB.fifo = SoundFifo.init();
+
+        self.dma_cnt = new;
+    }
+
+    /// SOUNDCNT_H
+    pub fn soundCntH(self: *const Self) u16 {
+        return self.dma_cnt.raw & 0x770F;
+    }
+
+    /// NR52
+    pub fn setSoundCntX(self: *Self, value: bool) void {
+        self.cnt.apu_enable.write(value);
+
+        if (value) {
+            self.fs.step = 0; // Reset Frame Sequencer
+
+            // Reset Square Wave Offsets
+            self.ch1.square.reset();
+            self.ch2.square.reset();
+
+            // Reset Wave
+            self.ch3.wave_dev.reset();
+
+            // Rest Noise
+            self.ch4.lfsr.reset();
+        } else {
+            self.reset();
+        }
+    }
+
+    /// NR52
+    pub fn soundCntX(self: *const Self) u8 {
+        const apu_enable: u8 = @boolToInt(self.cnt.apu_enable.read());
+
+        const ch1_enable: u8 = @boolToInt(self.ch1.enabled);
+        const ch2_enable: u8 = @boolToInt(self.ch2.enabled);
+        const ch3_enable: u8 = @boolToInt(self.ch3.enabled);
+        const ch4_enable: u8 = @boolToInt(self.ch4.enabled);
+
+        return apu_enable << 7 | ch4_enable << 3 | ch3_enable << 2 | ch2_enable << 1 | ch1_enable;
+    }
+
+    pub fn sampleAudio(self: *Self, late: u64) void {
+        self.sched.push(.SampleAudio, self.interval() -| late);
+
+        // Whether the APU is busy or not is determined  by the main loop in emu.zig
+        // This should only ever be true (because this side of the emu is single threaded)
+        // When audio sync is disaabled
+        if (self.is_buffer_full) return;
+
+        var left: i16 = 0;
+        var right: i16 = 0;
+
+        // SOUNDCNT_L Channel Enable flags
+        const ch_left: u4 = self.psg_cnt.ch_left.read();
+        const ch_right: u4 = self.psg_cnt.ch_right.read();
+
+        // Determine SOUNDCNT_H volume modifications
+        const gba_vol: u4 = switch (self.dma_cnt.ch_vol.read()) {
+            0b00 => 2,
+            0b01 => 1,
+            else => 0,
+        };
+
+        // Add all PSG channels together
+        left += if (ch_left & 1 == 1) @as(i16, self.ch1.sample) else 0;
+        left += if (ch_left >> 1 & 1 == 1) @as(i16, self.ch2.sample) else 0;
+        left += if (ch_left >> 2 & 1 == 1) @as(i16, self.ch3.sample) else 0;
+        left += if (ch_left >> 3 == 1) @as(i16, self.ch4.sample) else 0;
+
+        right += if (ch_right & 1 == 1) @as(i16, self.ch1.sample) else 0;
+        right += if (ch_right >> 1 & 1 == 1) @as(i16, self.ch2.sample) else 0;
+        right += if (ch_right >> 2 & 1 == 1) @as(i16, self.ch3.sample) else 0;
+        right += if (ch_right >> 3 == 1) @as(i16, self.ch4.sample) else 0;
+
+        // Multiply by master channel volume
+        left *= 1 + @as(i16, self.psg_cnt.left_vol.read());
+        right *= 1 + @as(i16, self.psg_cnt.right_vol.read());
+
+        // Apply GBA volume modifications to PSG Channels
+        left >>= gba_vol;
+        right >>= gba_vol;
+
+        const chA_sample = self.chA.amplitude() << if (self.dma_cnt.chA_vol.read()) @as(u4, 2) else 1;
+        const chB_sample = self.chB.amplitude() << if (self.dma_cnt.chB_vol.read()) @as(u4, 2) else 1;
+
+        left += if (self.dma_cnt.chA_left.read()) chA_sample else 0;
+        left += if (self.dma_cnt.chB_left.read()) chB_sample else 0;
+
+        right += if (self.dma_cnt.chA_right.read()) chA_sample else 0;
+        right += if (self.dma_cnt.chB_right.read()) chB_sample else 0;
+
+        // Add SOUNDBIAS
+        // FIXME: SOUNDBIAS is 10-bit but The waveform is centered around 0 if I treat it as 11-bit
+        const bias = @as(i16, self.bias.level.read()) << 2;
+        left += bias;
+        right += bias;
+
+        const clamped_left = std.math.clamp(@bitCast(u16, left), std.math.minInt(u11), std.math.maxInt(u11));
+        const clamped_right = std.math.clamp(@bitCast(u16, right), std.math.minInt(u11), std.math.maxInt(u11));
+
+        // Extend to 16-bit signed audio samples
+        const ext_left = (clamped_left << 5) | (clamped_left >> 6);
+        const ext_right = (clamped_right << 5) | (clamped_right >> 6);
+
+        if (self.sampling_cycle != self.bias.sampling_cycle.read()) self.replaceSDLResampler();
+
+        _ = SDL.SDL_AudioStreamPut(self.stream, &[2]u16{ ext_left, ext_right }, 2 * @sizeOf(u16));
+    }
+
+    fn replaceSDLResampler(self: *Self) void {
+        @setCold(true);
+        const sample_rate = Self.sampleRate(self.bias.sampling_cycle.read());
+        log.info("Sample Rate changed from {}Hz to {}Hz", .{ Self.sampleRate(self.sampling_cycle), sample_rate });
+
+        // Sampling Cycle (Sample Rate) changed, Craete a new SDL Audio Resampler
+        // FIXME: Replace SDL's Audio Resampler with either a custom or more reliable one
+        const old_stream = self.stream;
+        defer SDL.SDL_FreeAudioStream(old_stream);
+
+        self.sampling_cycle = self.bias.sampling_cycle.read();
+        self.stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, @intCast(c_int, sample_rate), host_format, 2, host_rate).?;
+    }
+
+    fn interval(self: *const Self) u64 {
+        return (1 << 24) / Self.sampleRate(self.bias.sampling_cycle.read());
+    }
+
+    fn sampleRate(cycle: u2) u64 {
+        return @as(u64, 1) << (15 + @as(u6, cycle));
+    }
+
+    pub fn onSequencerTick(self: *Self, late: u64) void {
+        self.fs.tick();
+
+        switch (self.fs.step) {
+            7 => self.tick(.Envelope), // Clock Envelope
+            0, 4 => self.tick(.Length), // Clock Length
+            2, 6 => {
+                // Clock Length and Sweep
+                self.tick(.Length);
+                self.tick(.Sweep);
+            },
+            1, 3, 5 => {},
+        }
+
+        self.sched.push(.FrameSequencer, ((1 << 24) / 512) -| late);
+    }
+
+    fn tick(self: *Self, comptime kind: Tick) void {
+        self.ch1.tick(kind);
+
+        switch (kind) {
+            .Length => {
+                self.ch2.tick(kind);
+                self.ch3.tick(kind);
+                self.ch4.tick(kind);
+            },
+            .Envelope => {
+                self.ch2.tick(kind);
+                self.ch4.tick(kind);
+            },
+            .Sweep => {}, // Already handled above (only for Ch1)
+        }
+    }
+
+    pub fn onDmaAudioSampleRequest(self: *Self, cpu: *Arm7tdmi, tim_id: u3) void {
+        if (!self.cnt.apu_enable.read()) return;
+
+        if (@boolToInt(self.dma_cnt.chA_timer.read()) == tim_id) {
+            if (!self.chA.enabled) return;
+
+            self.chA.updateSample();
+            if (self.chA.len() <= 15) cpu.bus.dma[1].requestAudio(0x0400_00A0);
+        }
+
+        if (@boolToInt(self.dma_cnt.chB_timer.read()) == tim_id) {
+            if (!self.chB.enabled) return;
+
+            self.chB.updateSample();
+            if (self.chB.len() <= 15) cpu.bus.dma[2].requestAudio(0x0400_00A4);
+        }
+    }
+};
+
+pub fn DmaSound(comptime kind: DmaSoundKind) type {
+    return struct {
+        const Self = @This();
+
+        fifo: SoundFifo,
+        kind: DmaSoundKind,
+        sample: i8,
+        enabled: bool,
+
+        fn init() Self {
+            return .{
+                .fifo = SoundFifo.init(),
+                .kind = kind,
+                .sample = 0,
+                .enabled = false,
+            };
+        }
+
+        pub fn push(self: *Self, value: u32) void {
+            if (!self.enabled) self.enable();
+
+            self.fifo.write(&intToBytes(u32, value)) catch |e| log.err("{} Error: {}", .{ kind, e });
+        }
+
+        fn enable(self: *Self) void {
+            @setCold(true);
+            self.enabled = true;
+        }
+
+        pub fn len(self: *const Self) usize {
+            return self.fifo.readableLength();
+        }
+
+        pub fn updateSample(self: *Self) void {
+            if (self.fifo.readItem()) |sample| self.sample = @bitCast(i8, sample);
+        }
+
+        pub fn amplitude(self: *const Self) i16 {
+            return @as(i16, self.sample);
+        }
+    };
+}
+
+const DmaSoundKind = enum {
+    A,
+    B,
+};
+
+pub const FrameSequencer = struct {
+    const Self = @This();
+    pub const interval = (1 << 24) / 512;
+
+    step: u3,
+
+    pub fn init() Self {
+        return .{ .step = 0 };
+    }
+
+    pub fn tick(self: *Self) void {
+        self.step +%= 1;
+    }
+
+    pub fn isLengthNext(self: *const Self) bool {
+        return (self.step +% 1) & 1 == 0; // Steps, 0, 2, 4, and 6 clock length
+    }
+
+    pub fn isEnvelopeNext(self: *const Self) bool {
+        return (self.step +% 1) == 7;
+    }
+};

src/core/apu/Noise.zig

@@ -0,0 +1,145 @@
+const io = @import("../bus/io.zig");
+const util = @import("../../util.zig");
+
+const Scheduler = @import("../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../apu.zig").FrameSequencer;
+const Tick = @import("../apu.zig").Apu.Tick;
+const Envelope = @import("device/Envelope.zig");
+const Length = @import("device/Length.zig");
+const Lfsr = @import("signal/Lfsr.zig");
+
+const Self = @This();
+
+/// Write-only
+/// NR41
+len: u6,
+/// NR42
+envelope: io.Envelope,
+/// NR43
+poly: io.PolyCounter,
+/// NR44
+cnt: io.NoiseControl,
+
+/// Length Functionarlity
+len_dev: Length,
+
+/// Envelope Functionality
+env_dev: Envelope,
+
+// Linear Feedback Shift Register
+lfsr: Lfsr,
+
+enabled: bool,
+sample: i8,
+
+pub fn init(sched: *Scheduler) Self {
+    return .{
+        .len = 0,
+        .envelope = .{ .raw = 0 },
+        .poly = .{ .raw = 0 },
+        .cnt = .{ .raw = 0 },
+        .enabled = false,
+
+        .len_dev = Length.create(),
+        .env_dev = Envelope.create(),
+        .lfsr = Lfsr.create(sched),
+
+        .sample = 0,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.len = 0; // NR41
+    self.envelope.raw = 0; // NR42
+    self.poly.raw = 0; // NR43
+    self.cnt.raw = 0; // NR44
+
+    self.len_dev.reset();
+    self.env_dev.reset();
+
+    self.sample = 0;
+    self.enabled = false;
+}
+
+pub fn tick(self: *Self, comptime kind: Tick) void {
+    switch (kind) {
+        .Length => self.len_dev.tick(self.cnt.length_enable.read(), &self.enabled),
+        .Envelope => self.env_dev.tick(self.envelope),
+        .Sweep => @compileError("Channel 4 does not implement Sweep"),
+    }
+}
+
+/// NR41, NR42
+pub fn sound4CntL(self: *const Self) u16 {
+    return @as(u16, self.envelope.raw) << 8;
+}
+
+/// NR41, NR42
+pub fn setSound4CntL(self: *Self, value: u16) void {
+    self.setNr41(@truncate(u8, value));
+    self.setNr42(@truncate(u8, value >> 8));
+}
+
+/// NR41
+pub fn setNr41(self: *Self, len: u8) void {
+    self.len = @truncate(u6, len);
+    self.len_dev.timer = @as(u7, 64) - @truncate(u6, len);
+}
+
+/// NR42
+pub fn setNr42(self: *Self, value: u8) void {
+    self.envelope.raw = value;
+    if (!self.isDacEnabled()) self.enabled = false;
+}
+
+/// NR43, NR44
+pub fn sound4CntH(self: *const Self) u16 {
+    return @as(u16, self.poly.raw & 0x40) << 8 | self.cnt.raw;
+}
+
+/// NR43, NR44
+pub fn setSound4CntH(self: *Self, fs: *const FrameSequencer, value: u16) void {
+    self.poly.raw = @truncate(u8, value);
+    self.setNr44(fs, @truncate(u8, value >> 8));
+}
+
+/// NR44
+pub fn setNr44(self: *Self, fs: *const FrameSequencer, byte: u8) void {
+    var new: io.NoiseControl = .{ .raw = byte };
+
+    if (new.trigger.read()) {
+        self.enabled = true;
+
+        if (self.len_dev.timer == 0) {
+            self.len_dev.timer =
+                if (!fs.isLengthNext() and new.length_enable.read()) 63 else 64;
+        }
+
+        // Update The Frequency Timer
+        self.lfsr.reload(self.poly);
+        self.lfsr.shift = 0x7FFF;
+
+        // Update Envelope and Volume
+        self.env_dev.timer = self.envelope.period.read();
+        if (fs.isEnvelopeNext() and self.env_dev.timer != 0b111) self.env_dev.timer += 1;
+
+        self.env_dev.vol = self.envelope.init_vol.read();
+
+        self.enabled = self.isDacEnabled();
+    }
+
+    util.audio.length.ch4.update(self, fs, new);
+    self.cnt = new;
+}
+
+pub fn onNoiseEvent(self: *Self, late: u64) void {
+    self.lfsr.onLfsrTimerExpire(self.poly, late);
+
+    self.sample = 0;
+    if (!self.isDacEnabled()) return;
+    self.sample = if (self.enabled) self.lfsr.sample() * @as(i8, self.env_dev.vol) else 0;
+}
+
+fn isDacEnabled(self: *const Self) bool {
+    return self.envelope.raw & 0xF8 != 0x00;
+}

src/core/apu/Tone.zig

@@ -0,0 +1,141 @@
+const io = @import("../bus/io.zig");
+const util = @import("../../util.zig");
+
+const Scheduler = @import("../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../apu.zig").FrameSequencer;
+const Tick = @import("../apu.zig").Apu.Tick;
+const Length = @import("device/Length.zig");
+const Envelope = @import("device/Envelope.zig");
+const Square = @import("signal/Square.zig");
+
+const Self = @This();
+
+/// NR21
+duty: io.Duty,
+/// NR22
+envelope: io.Envelope,
+/// NR23, NR24
+freq: io.Frequency,
+
+/// Length Functionarlity
+len_dev: Length,
+/// Envelope Functionality
+env_dev: Envelope,
+/// FrequencyTimer Functionality
+square: Square,
+
+enabled: bool,
+sample: i8,
+
+pub fn init(sched: *Scheduler) Self {
+    return .{
+        .duty = .{ .raw = 0 },
+        .envelope = .{ .raw = 0 },
+        .freq = .{ .raw = 0 },
+        .enabled = false,
+
+        .square = Square.init(sched),
+        .len_dev = Length.create(),
+        .env_dev = Envelope.create(),
+
+        .sample = 0,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.duty.raw = 0; // NR21
+    self.envelope.raw = 0; // NR22
+    self.freq.raw = 0; // NR32, NR24
+
+    self.len_dev.reset();
+    self.env_dev.reset();
+
+    self.sample = 0;
+    self.enabled = false;
+}
+
+pub fn tick(self: *Self, comptime kind: Tick) void {
+    switch (kind) {
+        .Length => self.len_dev.tick(self.freq.length_enable.read(), &self.enabled),
+        .Envelope => self.env_dev.tick(self.envelope),
+        .Sweep => @compileError("Channel 2 does not implement Sweep"),
+    }
+}
+
+pub fn onToneEvent(self: *Self, late: u64) void {
+    self.square.onSquareTimerExpire(Self, self.freq, late);
+
+    self.sample = 0;
+    if (!self.isDacEnabled()) return;
+    self.sample = if (self.enabled) self.square.sample(self.duty) * @as(i8, self.env_dev.vol) else 0;
+}
+
+/// NR21, NR22
+pub fn sound2CntL(self: *const Self) u16 {
+    return @as(u16, self.envelope.raw) << 8 | (self.duty.raw & 0xC0);
+}
+
+/// NR21, NR22
+pub fn setSound2CntL(self: *Self, value: u16) void {
+    self.setNr21(@truncate(u8, value));
+    self.setNr22(@truncate(u8, value >> 8));
+}
+
+/// NR21
+pub fn setNr21(self: *Self, value: u8) void {
+    self.duty.raw = value;
+    self.len_dev.timer = @as(u7, 64) - @truncate(u6, value);
+}
+
+/// NR22
+pub fn setNr22(self: *Self, value: u8) void {
+    self.envelope.raw = value;
+    if (!self.isDacEnabled()) self.enabled = false;
+}
+
+/// NR23, NR24
+pub fn sound2CntH(self: *const Self) u16 {
+    return self.freq.raw & 0x4000;
+}
+
+/// NR23, NR24
+pub fn setSound2CntH(self: *Self, fs: *const FrameSequencer, value: u16) void {
+    self.setNr23(@truncate(u8, value));
+    self.setNr24(fs, @truncate(u8, value >> 8));
+}
+
+/// NR23
+pub fn setNr23(self: *Self, byte: u8) void {
+    self.freq.raw = (self.freq.raw & 0xFF00) | byte;
+}
+
+/// NR24
+pub fn setNr24(self: *Self, fs: *const FrameSequencer, byte: u8) void {
+    var new: io.Frequency = .{ .raw = (@as(u16, byte) << 8) | (self.freq.raw & 0xFF) };
+
+    if (new.trigger.read()) {
+        self.enabled = true;
+
+        if (self.len_dev.timer == 0) {
+            self.len_dev.timer =
+                if (!fs.isLengthNext() and new.length_enable.read()) 63 else 64;
+        }
+
+        self.square.reload(Self, self.freq.frequency.read());
+
+        // Reload Envelope period and timer
+        self.env_dev.timer = self.envelope.period.read();
+        if (fs.isEnvelopeNext() and self.env_dev.timer != 0b111) self.env_dev.timer += 1;
+
+        self.env_dev.vol = self.envelope.init_vol.read();
+
+        self.enabled = self.isDacEnabled();
+    }
+
+    util.audio.length.update(Self, self, fs, new);
+    self.freq = new;
+}
+
+fn isDacEnabled(self: *const Self) bool {
+    return self.envelope.raw & 0xF8 != 0;
+}

src/core/apu/ToneSweep.zig

@@ -0,0 +1,185 @@
+const io = @import("../bus/io.zig");
+const util = @import("../../util.zig");
+
+const Scheduler = @import("../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../apu.zig").FrameSequencer;
+const Length = @import("device/Length.zig");
+const Envelope = @import("device/Envelope.zig");
+const Sweep = @import("device/Sweep.zig");
+const Square = @import("signal/Square.zig");
+
+const Tick = @import("../apu.zig").Apu.Tick;
+
+const Self = @This();
+
+/// NR10
+sweep: io.Sweep,
+/// NR11
+duty: io.Duty,
+/// NR12
+envelope: io.Envelope,
+/// NR13, NR14
+freq: io.Frequency,
+
+/// Length Functionality
+len_dev: Length,
+/// Sweep Functionality
+sweep_dev: Sweep,
+/// Envelope Functionality
+env_dev: Envelope,
+/// Frequency Timer Functionality
+square: Square,
+enabled: bool,
+
+sample: i8,
+
+pub fn init(sched: *Scheduler) Self {
+    return .{
+        .sweep = .{ .raw = 0 },
+        .duty = .{ .raw = 0 },
+        .envelope = .{ .raw = 0 },
+        .freq = .{ .raw = 0 },
+        .sample = 0,
+        .enabled = false,
+
+        .square = Square.init(sched),
+        .len_dev = Length.create(),
+        .sweep_dev = Sweep.create(),
+        .env_dev = Envelope.create(),
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.sweep.raw = 0; // NR10
+    self.duty.raw = 0; // NR11
+    self.envelope.raw = 0; // NR12
+    self.freq.raw = 0; // NR13, NR14
+
+    self.len_dev.reset();
+    self.sweep_dev.reset();
+    self.env_dev.reset();
+
+    self.sample = 0;
+    self.enabled = false;
+}
+
+pub fn tick(self: *Self, comptime kind: Tick) void {
+    switch (kind) {
+        .Length => self.len_dev.tick(self.freq.length_enable.read(), &self.enabled),
+        .Envelope => self.env_dev.tick(self.envelope),
+        .Sweep => self.sweep_dev.tick(self),
+    }
+}
+
+pub fn onToneSweepEvent(self: *Self, late: u64) void {
+    self.square.onSquareTimerExpire(Self, self.freq, late);
+
+    self.sample = 0;
+    if (!self.isDacEnabled()) return;
+    self.sample = if (self.enabled) self.square.sample(self.duty) * @as(i8, self.env_dev.vol) else 0;
+}
+
+/// NR10, NR11, NR12
+pub fn setSound1Cnt(self: *Self, value: u32) void {
+    self.setSound1CntL(@truncate(u8, value));
+    self.setSound1CntH(@truncate(u16, value >> 16));
+}
+
+/// NR10
+pub fn sound1CntL(self: *const Self) u8 {
+    return self.sweep.raw & 0x7F;
+}
+
+/// NR10
+pub fn setSound1CntL(self: *Self, value: u8) void {
+    const new = io.Sweep{ .raw = value };
+
+    if (!new.direction.read()) {
+        // If at least one (1) sweep calculation has been made with
+        // the negate bit set (since last trigger), disable the channel
+
+        if (self.sweep_dev.calc_performed) self.enabled = false;
+    }
+
+    self.sweep.raw = value;
+}
+
+/// NR11, NR12
+pub fn sound1CntH(self: *const Self) u16 {
+    return @as(u16, self.envelope.raw) << 8 | (self.duty.raw & 0xC0);
+}
+
+/// NR11, NR12
+pub fn setSound1CntH(self: *Self, value: u16) void {
+    self.setNr11(@truncate(u8, value));
+    self.setNr12(@truncate(u8, value >> 8));
+}
+
+/// NR11
+pub fn setNr11(self: *Self, value: u8) void {
+    self.duty.raw = value;
+    self.len_dev.timer = @as(u7, 64) - @truncate(u6, value);
+}
+
+/// NR12
+pub fn setNr12(self: *Self, value: u8) void {
+    self.envelope.raw = value;
+    if (!self.isDacEnabled()) self.enabled = false;
+}
+
+/// NR13, NR14
+pub fn sound1CntX(self: *const Self) u16 {
+    return self.freq.raw & 0x4000;
+}
+
+/// NR13, NR14
+pub fn setSound1CntX(self: *Self, fs: *const FrameSequencer, value: u16) void {
+    self.setNr13(@truncate(u8, value));
+    self.setNr14(fs, @truncate(u8, value >> 8));
+}
+
+/// NR13
+pub fn setNr13(self: *Self, byte: u8) void {
+    self.freq.raw = (self.freq.raw & 0xFF00) | byte;
+}
+
+/// NR14
+pub fn setNr14(self: *Self, fs: *const FrameSequencer, byte: u8) void {
+    var new: io.Frequency = .{ .raw = (@as(u16, byte) << 8) | (self.freq.raw & 0xFF) };
+
+    if (new.trigger.read()) {
+        self.enabled = true;
+
+        if (self.len_dev.timer == 0) {
+            self.len_dev.timer =
+                if (!fs.isLengthNext() and new.length_enable.read()) 63 else 64;
+        }
+
+        self.square.reload(Self, self.freq.frequency.read());
+
+        // Reload Envelope period and timer
+        self.env_dev.timer = self.envelope.period.read();
+        if (fs.isEnvelopeNext() and self.env_dev.timer != 0b111) self.env_dev.timer += 1;
+
+        self.env_dev.vol = self.envelope.init_vol.read();
+
+        // Sweep Trigger Behaviour
+        const sw_period = self.sweep.period.read();
+        const sw_shift = self.sweep.shift.read();
+
+        self.sweep_dev.calc_performed = false;
+        self.sweep_dev.shadow = self.freq.frequency.read();
+        self.sweep_dev.timer = if (sw_period == 0) 8 else sw_period;
+        self.sweep_dev.enabled = sw_period != 0 or sw_shift != 0;
+        if (sw_shift != 0) _ = self.sweep_dev.calculate(self.sweep, &self.enabled);
+
+        self.enabled = self.isDacEnabled();
+    }
+
+    util.audio.length.update(Self, self, fs, new);
+    self.freq = new;
+}
+
+fn isDacEnabled(self: *const Self) bool {
+    return self.envelope.raw & 0xF8 != 0;
+}

src/core/apu/Wave.zig

@@ -0,0 +1,145 @@
+const io = @import("../bus/io.zig");
+const util = @import("../../util.zig");
+
+const Scheduler = @import("../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../apu.zig").FrameSequencer;
+const Tick = @import("../apu.zig").Apu.Tick;
+
+const Length = @import("device/Length.zig");
+const Wave = @import("signal/Wave.zig");
+
+const Self = @This();
+
+/// Write-only
+/// NR30
+select: io.WaveSelect,
+/// NR31
+length: u8,
+/// NR32
+vol: io.WaveVolume,
+/// NR33, NR34
+freq: io.Frequency,
+
+/// Length Functionarlity
+len_dev: Length,
+wave_dev: Wave,
+
+enabled: bool,
+sample: i8,
+
+pub fn init(sched: *Scheduler) Self {
+    return .{
+        .select = .{ .raw = 0 },
+        .vol = .{ .raw = 0 },
+        .freq = .{ .raw = 0 },
+        .length = 0,
+
+        .len_dev = Length.create(),
+        .wave_dev = Wave.init(sched),
+        .enabled = false,
+        .sample = 0,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.select.raw = 0; // NR30
+    self.length = 0; // NR31
+    self.vol.raw = 0; // NR32
+    self.freq.raw = 0; // NR33, NR34
+
+    self.len_dev.reset();
+    self.wave_dev.reset();
+
+    self.sample = 0;
+    self.enabled = false;
+}
+
+pub fn tick(self: *Self, comptime kind: Tick) void {
+    switch (kind) {
+        .Length => self.len_dev.tick(self.freq.length_enable.read(), &self.enabled),
+        .Envelope => @compileError("Channel 3 does not implement Envelope"),
+        .Sweep => @compileError("Channel 3 does not implement Sweep"),
+    }
+}
+
+/// NR30, NR31, NR32
+pub fn setSound3Cnt(self: *Self, value: u32) void {
+    self.setSound3CntL(@truncate(u8, value));
+    self.setSound3CntH(@truncate(u16, value >> 16));
+}
+
+/// NR30
+pub fn setSound3CntL(self: *Self, value: u8) void {
+    self.select.raw = value;
+    if (!self.select.enabled.read()) self.enabled = false;
+}
+
+/// NR30
+pub fn sound3CntL(self: *const Self) u8 {
+    return self.select.raw & 0xE0;
+}
+
+/// NR31, NR32
+pub fn sound3CntH(self: *const Self) u16 {
+    return @as(u16, self.length & 0xE0) << 8;
+}
+
+/// NR31, NR32
+pub fn setSound3CntH(self: *Self, value: u16) void {
+    self.setNr31(@truncate(u8, value));
+    self.vol.raw = (@truncate(u8, value >> 8));
+}
+
+/// NR31
+pub fn setNr31(self: *Self, len: u8) void {
+    self.length = len;
+    self.len_dev.timer = 256 - @as(u9, len);
+}
+
+/// NR33, NR34
+pub fn setSound3CntX(self: *Self, fs: *const FrameSequencer, value: u16) void {
+    self.setNr33(@truncate(u8, value));
+    self.setNr34(fs, @truncate(u8, value >> 8));
+}
+
+/// NR33, NR34
+pub fn sound3CntX(self: *const Self) u16 {
+    return self.freq.raw & 0x4000;
+}
+
+/// NR33
+pub fn setNr33(self: *Self, byte: u8) void {
+    self.freq.raw = (self.freq.raw & 0xFF00) | byte;
+}
+
+/// NR34
+pub fn setNr34(self: *Self, fs: *const FrameSequencer, byte: u8) void {
+    var new: io.Frequency = .{ .raw = (@as(u16, byte) << 8) | (self.freq.raw & 0xFF) };
+
+    if (new.trigger.read()) {
+        self.enabled = true;
+
+        if (self.len_dev.timer == 0) {
+            self.len_dev.timer =
+                if (!fs.isLengthNext() and new.length_enable.read()) 255 else 256;
+        }
+
+        // Update The Frequency Timer
+        self.wave_dev.reload(self.freq.frequency.read());
+        self.wave_dev.offset = 0;
+
+        self.enabled = self.select.enabled.read();
+    }
+
+    util.audio.length.update(Self, self, fs, new);
+    self.freq = new;
+}
+
+pub fn onWaveEvent(self: *Self, late: u64) void {
+    self.wave_dev.onWaveTimerExpire(self.freq, self.select, late);
+
+    self.sample = 0;
+    if (!self.select.enabled.read()) return;
+    // Convert unsigned 4-bit wave sample to signed 8-bit sample
+    self.sample = (2 * @as(i8, self.wave_dev.sample(self.select)) - 15) >> self.wave_dev.shift(self.vol);
+}

src/core/apu/device/Envelope.zig

@@ -0,0 +1,33 @@
+const io = @import("../../bus/io.zig");
+
+const Self = @This();
+
+/// Period Timer
+timer: u3,
+/// Current Volume
+vol: u4,
+
+pub fn create() Self {
+    return .{ .timer = 0, .vol = 0 };
+}
+
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.vol = 0;
+}
+
+pub fn tick(self: *Self, nrx2: io.Envelope) void {
+    if (nrx2.period.read() != 0) {
+        if (self.timer != 0) self.timer -= 1;
+
+        if (self.timer == 0) {
+            self.timer = nrx2.period.read();
+
+            if (nrx2.direction.read()) {
+                if (self.vol < 0xF) self.vol += 1;
+            } else {
+                if (self.vol > 0x0) self.vol -= 1;
+            }
+        }
+    }
+}

src/core/apu/device/Length.zig

@@ -0,0 +1,22 @@
+const Self = @This();
+
+timer: u9,
+
+pub fn create() Self {
+    return .{ .timer = 0 };
+}
+
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+}
+
+pub fn tick(self: *Self, enabled: bool, ch_enable: *bool) void {
+    if (enabled) {
+        if (self.timer == 0) return;
+        self.timer -= 1;
+
+        // By returning early if timer == 0, this is only
+        // true if timer == 0 because of the decrement we just did
+        if (self.timer == 0) ch_enable.* = false;
+    }
+}

src/core/apu/device/Sweep.zig

@@ -0,0 +1,61 @@
+const io = @import("../../bus/io.zig");
+const ToneSweep = @import("../ToneSweep.zig");
+
+const Self = @This();
+
+timer: u8,
+enabled: bool,
+shadow: u11,
+
+calc_performed: bool,
+
+pub fn create() Self {
+    return .{
+        .timer = 0,
+        .enabled = false,
+        .shadow = 0,
+        .calc_performed = false,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.enabled = false;
+    self.shadow = 0;
+    self.calc_performed = false;
+}
+
+pub fn tick(self: *Self, ch1: *ToneSweep) void {
+    if (self.timer != 0) self.timer -= 1;
+
+    if (self.timer == 0) {
+        const period = ch1.sweep.period.read();
+        self.timer = if (period == 0) 8 else period;
+
+        if (self.enabled and period != 0) {
+            const new_freq = self.calculate(ch1.sweep, &ch1.enabled);
+
+            if (new_freq <= 0x7FF and ch1.sweep.shift.read() != 0) {
+                ch1.freq.frequency.write(@truncate(u11, new_freq));
+                self.shadow = @truncate(u11, new_freq);
+
+                _ = self.calculate(ch1.sweep, &ch1.enabled);
+            }
+        }
+    }
+}
+
+/// Calculates the Sweep Frequency
+pub fn calculate(self: *Self, sweep: io.Sweep, ch_enable: *bool) u12 {
+    const shadow = @as(u12, self.shadow);
+    const shadow_shifted = shadow >> sweep.shift.read();
+    const decrease = sweep.direction.read();
+
+    const freq = if (decrease) blk: {
+        self.calc_performed = true;
+        break :blk shadow - shadow_shifted;
+    } else shadow + shadow_shifted;
+    if (freq > 0x7FF) ch_enable.* = false;
+
+    return freq;
+}

src/core/apu/signal/Lfsr.zig

@@ -0,0 +1,62 @@
+//! Linear Feedback Shift Register
+const io = @import("../../bus/io.zig");
+
+const Scheduler = @import("../../scheduler.zig").Scheduler;
+
+const Self = @This();
+pub const interval: u64 = (1 << 24) / (1 << 22);
+
+shift: u15,
+timer: u16,
+
+sched: *Scheduler,
+
+pub fn create(sched: *Scheduler) Self {
+    return .{
+        .shift = 0,
+        .timer = 0,
+        .sched = sched,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.shift = 0;
+    self.timer = 0;
+}
+
+pub fn sample(self: *const Self) i8 {
+    return if ((~self.shift & 1) == 1) 1 else -1;
+}
+
+/// Reload LFSR Timer
+pub fn reload(self: *Self, poly: io.PolyCounter) void {
+    self.sched.removeScheduledEvent(.{ .ApuChannel = 3 });
+
+    const div = Self.divisor(poly.div_ratio.read());
+    const timer = div << poly.shift.read();
+    self.sched.push(.{ .ApuChannel = 3 }, @as(u64, timer) * interval);
+}
+
+/// Scheduler Event Handler for LFSR Timer Expire
+/// FIXME: This gets called a lot, slowing down the scheduler
+pub fn onLfsrTimerExpire(self: *Self, poly: io.PolyCounter, late: u64) void {
+    // Obscure: "Using a noise channel clock shift of 14 or 15
+    // results in the LFSR receiving no clocks."
+    if (poly.shift.read() >= 14) return;
+
+    const div = Self.divisor(poly.div_ratio.read());
+    const timer = div << poly.shift.read();
+
+    const tmp = (self.shift & 1) ^ ((self.shift & 2) >> 1);
+    self.shift = (self.shift >> 1) | (tmp << 14);
+
+    if (poly.width.read())
+        self.shift = (self.shift & ~@as(u15, 0x40)) | tmp << 6;
+
+    self.sched.push(.{ .ApuChannel = 3 }, @as(u64, timer) * interval -| late);
+}
+
+fn divisor(code: u3) u16 {
+    if (code == 0) return 8;
+    return @as(u16, code) << 4;
+}

src/core/apu/signal/Square.zig

@@ -0,0 +1,62 @@
+const std = @import("std");
+const io = @import("../../bus/io.zig");
+
+const Scheduler = @import("../../scheduler.zig").Scheduler;
+const ToneSweep = @import("../ToneSweep.zig");
+const Tone = @import("../Tone.zig");
+
+const Self = @This();
+pub const interval: u64 = (1 << 24) / (1 << 22);
+
+pos: u3,
+sched: *Scheduler,
+timer: u16,
+
+pub fn init(sched: *Scheduler) Self {
+    return .{
+        .timer = 0,
+        .pos = 0,
+        .sched = sched,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.pos = 0;
+}
+
+/// Scheduler Event Handler for Square Synth Timer Expire
+pub fn onSquareTimerExpire(self: *Self, comptime T: type, nrx34: io.Frequency, late: u64) void {
+    comptime std.debug.assert(T == ToneSweep or T == Tone);
+    self.pos +%= 1;
+
+    self.timer = (@as(u16, 2048) - nrx34.frequency.read()) * 4;
+    self.sched.push(.{ .ApuChannel = if (T == ToneSweep) 0 else 1 }, @as(u64, self.timer) * interval -| late);
+}
+
+/// Reload Square Wave Timer
+pub fn reload(self: *Self, comptime T: type, value: u11) void {
+    comptime std.debug.assert(T == ToneSweep or T == Tone);
+    const channel = if (T == ToneSweep) 0 else 1;
+
+    self.sched.removeScheduledEvent(.{ .ApuChannel = channel });
+
+    const tmp = (@as(u16, 2048) - value) * 4; // What Freq Timer should be assuming no weird behaviour
+    self.timer = (tmp & ~@as(u16, 0x3)) | self.timer & 0x3; // Keep the last two bits from the old timer;
+
+    self.sched.push(.{ .ApuChannel = channel }, @as(u64, self.timer) * interval);
+}
+
+pub fn sample(self: *const Self, nrx1: io.Duty) i8 {
+    const pattern = nrx1.pattern.read();
+
+    const i = self.pos ^ 7; // index of 0 should get highest bit
+    const result = switch (pattern) {
+        0b00 => @as(u8, 0b00000001) >> i, // 12.5%
+        0b01 => @as(u8, 0b00000011) >> i, // 25%
+        0b10 => @as(u8, 0b00001111) >> i, // 50%
+        0b11 => @as(u8, 0b11111100) >> i, // 75%
+    };
+
+    return if (result & 1 == 1) 1 else -1;
+}

src/core/apu/signal/Wave.zig

@@ -0,0 +1,84 @@
+const std = @import("std");
+const io = @import("../../bus/io.zig");
+
+const Scheduler = @import("../../scheduler.zig").Scheduler;
+
+const buf_len = 0x20;
+pub const interval: u64 = (1 << 24) / (1 << 22);
+const Self = @This();
+
+buf: [buf_len]u8,
+timer: u16,
+offset: u12,
+
+sched: *Scheduler,
+
+pub fn read(self: *const Self, comptime T: type, nr30: io.WaveSelect, addr: u32) T {
+    // TODO: Handle reads when Channel 3 is disabled
+    const base = if (!nr30.bank.read()) @as(u32, 0x10) else 0; // Read from the Opposite Bank in Use
+
+    const i = base + addr - 0x0400_0090;
+    return std.mem.readIntSliceLittle(T, self.buf[i..][0..@sizeOf(T)]);
+}
+
+pub fn write(self: *Self, comptime T: type, nr30: io.WaveSelect, addr: u32, value: T) void {
+    // TODO: Handle writes when Channel 3 is disabled
+    const base = if (!nr30.bank.read()) @as(u32, 0x10) else 0; // Write to the Opposite Bank in Use
+
+    const i = base + addr - 0x0400_0090;
+    std.mem.writeIntSliceLittle(T, self.buf[i..][0..@sizeOf(T)], value);
+}
+
+pub fn init(sched: *Scheduler) Self {
+    return .{
+        .buf = [_]u8{0x00} ** buf_len,
+        .timer = 0,
+        .offset = 0,
+        .sched = sched,
+    };
+}
+
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.offset = 0;
+
+    // sample buffer isn't reset because it's outside of the range of what NR52{7}'s effects
+}
+
+/// Reload internal Wave Timer
+pub fn reload(self: *Self, value: u11) void {
+    self.sched.removeScheduledEvent(.{ .ApuChannel = 2 });
+
+    self.timer = (@as(u16, 2048) - value) * 2;
+    self.sched.push(.{ .ApuChannel = 2 }, @as(u64, self.timer) * interval);
+}
+
+/// Scheduler Event Handler
+pub fn onWaveTimerExpire(self: *Self, nrx34: io.Frequency, nr30: io.WaveSelect, late: u64) void {
+    if (nr30.dimension.read()) {
+        self.offset = (self.offset + 1) % 0x40; // 0x20 bytes (both banks), which contain 2 samples each
+    } else {
+        self.offset = (self.offset + 1) % 0x20; // 0x10 bytes, which contain 2 samples each
+    }
+
+    self.timer = (@as(u16, 2048) - nrx34.frequency.read()) * 2;
+    self.sched.push(.{ .ApuChannel = 2 }, @as(u64, self.timer) * interval -| late);
+}
+
+/// Generate Sample from Wave Synth
+pub fn sample(self: *const Self, nr30: io.WaveSelect) u4 {
+    const base = if (nr30.bank.read()) @as(u32, 0x10) else 0;
+
+    const value = self.buf[base + self.offset / 2];
+    return if (self.offset & 1 == 0) @truncate(u4, value >> 4) else @truncate(u4, value);
+}
+
+/// TODO: Write comment
+pub fn shift(_: *const Self, nr32: io.WaveVolume) u2 {
+    return switch (nr32.kind.read()) {
+        0b00 => 3, // Mute / Zero
+        0b01 => 0, // 100% Volume
+        0b10 => 1, // 50% Volume
+        0b11 => 2, // 25% Volume
+    };
+}

src/core/bus/Bios.zig

@@ -0,0 +1,83 @@
+const std = @import("std");
+
+const Allocator = std.mem.Allocator;
+const log = std.log.scoped(.Bios);
+
+const rotr = @import("../../util.zig").rotr;
+const forceAlign = @import("../Bus.zig").forceAlign;
+
+/// Size of the BIOS in bytes
+pub const size = 0x4000;
+const Self = @This();
+
+buf: ?[]u8,
+allocator: Allocator,
+
+addr_latch: u32 = 0,
+
+// https://github.com/ITotalJustice/notorious_beeg/issues/106
+pub fn read(self: *Self, comptime T: type, r15: u32, address: u32) T {
+    if (r15 < Self.size) {
+        const addr = forceAlign(T, address);
+
+        self.addr_latch = addr;
+        return self._read(T, addr);
+    }
+
+    log.warn("Open Bus! Read from 0x{X:0>8}, but PC was 0x{X:0>8}", .{ address, r15 });
+    const value = self._read(u32, self.addr_latch);
+
+    return @truncate(T, rotr(u32, value, 8 * rotateBy(T, address)));
+}
+
+fn rotateBy(comptime T: type, address: u32) u32 {
+    return switch (T) {
+        u8 => address & 3,
+        u16 => address & 2,
+        u32 => 0,
+        else => @compileError("bios: unsupported read width"),
+    };
+}
+
+pub fn dbgRead(self: *const Self, comptime T: type, r15: u32, address: u32) T {
+    if (r15 < Self.size) return self._read(T, forceAlign(T, address));
+
+    const value = self._read(u32, self.addr_latch);
+    return @truncate(T, rotr(u32, value, 8 * rotateBy(T, address)));
+}
+
+/// Read without the GBA safety checks
+fn _read(self: *const Self, comptime T: type, addr: u32) T {
+    const buf = self.buf orelse std.debug.panic("[BIOS] ZBA tried to read {} from 0x{X:0>8} but not BIOS was present", .{ T, addr });
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.readIntSliceLittle(T, buf[addr..][0..@sizeOf(T)]),
+        else => @compileError("BIOS: Unsupported read width"),
+    };
+}
+
+pub fn write(_: *Self, comptime T: type, addr: u32, value: T) void {
+    @setCold(true);
+    log.debug("Tried to write {} 0x{X:} to 0x{X:0>8} ", .{ T, value, addr });
+}
+
+pub fn init(allocator: Allocator, maybe_path: ?[]const u8) !Self {
+    if (maybe_path == null) return .{ .buf = null, .allocator = allocator };
+    const path = maybe_path.?;
+
+    const buf = try allocator.alloc(u8, Self.size);
+    errdefer allocator.free(buf);
+
+    const file = try std.fs.cwd().openFile(path, .{});
+    defer file.close();
+
+    const file_len = try file.readAll(buf);
+    if (file_len != Self.size) log.err("Expected BIOS to be {}B, was {}B", .{ Self.size, file_len });
+
+    return Self{ .buf = buf, .allocator = allocator };
+}
+
+pub fn deinit(self: *Self) void {
+    if (self.buf) |buf| self.allocator.free(buf);
+    self.* = undefined;
+}

src/core/bus/Ewram.zig

@@ -0,0 +1,41 @@
+const std = @import("std");
+
+const Allocator = std.mem.Allocator;
+const ewram_size = 0x40000;
+const Self = @This();
+
+buf: []u8,
+allocator: Allocator,
+
+pub fn read(self: *const Self, comptime T: type, address: usize) T {
+    const addr = address & 0x3FFFF;
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.readIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)]),
+        else => @compileError("EWRAM: Unsupported read width"),
+    };
+}
+
+pub fn write(self: *const Self, comptime T: type, address: usize, value: T) void {
+    const addr = address & 0x3FFFF;
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.writeIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)], value),
+        else => @compileError("EWRAM: Unsupported write width"),
+    };
+}
+
+pub fn init(allocator: Allocator) !Self {
+    const buf = try allocator.alloc(u8, ewram_size);
+    std.mem.set(u8, buf, 0);
+
+    return Self{
+        .buf = buf,
+        .allocator = allocator,
+    };
+}
+
+pub fn deinit(self: *Self) void {
+    self.allocator.free(self.buf);
+    self.* = undefined;
+}

src/core/bus/GamePak.zig

@@ -0,0 +1,256 @@
+const std = @import("std");
+const config = @import("../../config.zig");
+
+const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
+const Backup = @import("backup.zig").Backup;
+const Gpio = @import("gpio.zig").Gpio;
+const Allocator = std.mem.Allocator;
+
+const log = std.log.scoped(.GamePak);
+
+const Self = @This();
+
+title: [12]u8,
+buf: []u8,
+allocator: Allocator,
+backup: Backup,
+gpio: *Gpio,
+
+pub fn read(self: *Self, comptime T: type, address: u32) T {
+    const addr = address & 0x1FF_FFFF;
+
+    if (self.backup.kind == .Eeprom) {
+        if (self.buf.len > 0x100_0000) { // Large
+            // Addresses 0x1FF_FF00 to 0x1FF_FFFF are reserved from EEPROM accesses if
+            // * Backup type is EEPROM
+            // * Large ROM (Size is greater than 16MB)
+            if (addr > 0x1FF_FEFF)
+                return self.backup.eeprom.read();
+        } else {
+            // Addresses 0x0D00_0000 to 0x0DFF_FFFF are reserved for EEPROM accesses if
+            // * Backup type is EEPROM
+            // * Small ROM (less than 16MB)
+            if (@truncate(u8, address >> 24) == 0x0D)
+                return self.backup.eeprom.read();
+        }
+    }
+
+    if (self.gpio.cnt == 1) {
+        // GPIO Can be read from
+        // We assume that this will only be true when a ROM actually does want something from GPIO
+
+        switch (T) {
+            u32 => switch (address) {
+                // TODO: Do I even need to implement these?
+                0x0800_00C4 => std.debug.panic("Handle 32-bit GPIO Data/Direction Reads", .{}),
+                0x0800_00C6 => std.debug.panic("Handle 32-bit GPIO Direction/Control Reads", .{}),
+                0x0800_00C8 => std.debug.panic("Handle 32-bit GPIO Control Reads", .{}),
+                else => {},
+            },
+            u16 => switch (address) {
+                // FIXME: What do 16-bit GPIO Reads look like?
+                0x0800_00C4 => return self.gpio.read(.Data),
+                0x0800_00C6 => return self.gpio.read(.Direction),
+                0x0800_00C8 => return self.gpio.read(.Control),
+                else => {},
+            },
+            u8 => switch (address) {
+                0x0800_00C4 => return self.gpio.read(.Data),
+                0x0800_00C6 => return self.gpio.read(.Direction),
+                0x0800_00C8 => return self.gpio.read(.Control),
+                else => {},
+            },
+            else => @compileError("GamePak[GPIO]: Unsupported read width"),
+        }
+    }
+
+    return switch (T) {
+        u32 => (@as(T, self.get(addr + 3)) << 24) | (@as(T, self.get(addr + 2)) << 16) | (@as(T, self.get(addr + 1)) << 8) | (@as(T, self.get(addr))),
+        u16 => (@as(T, self.get(addr + 1)) << 8) | @as(T, self.get(addr)),
+        u8 => self.get(addr),
+        else => @compileError("GamePak: Unsupported read width"),
+    };
+}
+
+inline fn get(self: *const Self, i: u32) u8 {
+    @setRuntimeSafety(false);
+    if (i < self.buf.len) return self.buf[i];
+
+    const lhs = i >> 1 & 0xFFFF;
+    return @truncate(u8, lhs >> 8 * @truncate(u5, i & 1));
+}
+
+pub fn dbgRead(self: *const Self, comptime T: type, address: u32) T {
+    const addr = address & 0x1FF_FFFF;
+
+    if (self.backup.kind == .Eeprom) {
+        if (self.buf.len > 0x100_0000) { // Large
+            // Addresses 0x1FF_FF00 to 0x1FF_FFFF are reserved from EEPROM accesses if
+            // * Backup type is EEPROM
+            // * Large ROM (Size is greater than 16MB)
+            if (addr > 0x1FF_FEFF)
+                return self.backup.eeprom.dbgRead();
+        } else {
+            // Addresses 0x0D00_0000 to 0x0DFF_FFFF are reserved for EEPROM accesses if
+            // * Backup type is EEPROM
+            // * Small ROM (less than 16MB)
+            if (@truncate(u8, address >> 24) == 0x0D)
+                return self.backup.eeprom.dbgRead();
+        }
+    }
+
+    if (self.gpio.cnt == 1) {
+        // GPIO Can be read from
+        // We assume that this will only be true when a ROM actually does want something from GPIO
+
+        switch (T) {
+            u32 => switch (address) {
+                // FIXME: Do I even need to implement these?
+                0x0800_00C4 => std.debug.panic("Handle 32-bit GPIO Data/Direction Reads", .{}),
+                0x0800_00C6 => std.debug.panic("Handle 32-bit GPIO Direction/Control Reads", .{}),
+                0x0800_00C8 => std.debug.panic("Handle 32-bit GPIO Control Reads", .{}),
+                else => {},
+            },
+            u16 => switch (address) {
+                0x0800_00C4 => return self.gpio.read(.Data),
+                0x0800_00C6 => return self.gpio.read(.Direction),
+                0x0800_00C8 => return self.gpio.read(.Control),
+                else => {},
+            },
+            u8 => switch (address) {
+                0x0800_00C4 => return self.gpio.read(.Data),
+                0x0800_00C6 => return self.gpio.read(.Direction),
+                0x0800_00C8 => return self.gpio.read(.Control),
+                else => {},
+            },
+            else => @compileError("GamePak[GPIO]: Unsupported read width"),
+        }
+    }
+
+    return switch (T) {
+        u32 => (@as(T, self.get(addr + 3)) << 24) | (@as(T, self.get(addr + 2)) << 16) | (@as(T, self.get(addr + 1)) << 8) | (@as(T, self.get(addr))),
+        u16 => (@as(T, self.get(addr + 1)) << 8) | @as(T, self.get(addr)),
+        u8 => self.get(addr),
+        else => @compileError("GamePak: Unsupported read width"),
+    };
+}
+
+pub fn write(self: *Self, comptime T: type, word_count: u16, address: u32, value: T) void {
+    const addr = address & 0x1FF_FFFF;
+
+    if (self.backup.kind == .Eeprom) {
+        const bit = @truncate(u1, value);
+
+        if (self.buf.len > 0x100_0000) { // Large
+            // Addresses 0x1FF_FF00 to 0x1FF_FFFF are reserved from EEPROM accesses if
+            // * Backup type is EEPROM
+            // * Large ROM (Size is greater than 16MB)
+            if (addr > 0x1FF_FEFF)
+                return self.backup.eeprom.write(word_count, &self.backup.buf, bit);
+        } else {
+            // Addresses 0x0D00_0000 to 0x0DFF_FFFF are reserved for EEPROM accesses if
+            // * Backup type is EEPROM
+            // * Small ROM (less than 16MB)
+            if (@truncate(u8, address >> 24) == 0x0D)
+                return self.backup.eeprom.write(word_count, &self.backup.buf, bit);
+        }
+    }
+
+    switch (T) {
+        u32 => switch (address) {
+            0x0800_00C4 => {
+                self.gpio.write(.Data, @truncate(u4, value));
+                self.gpio.write(.Direction, @truncate(u4, value >> 16));
+            },
+            0x0800_00C6 => {
+                self.gpio.write(.Direction, @truncate(u4, value));
+                self.gpio.write(.Control, @truncate(u1, value >> 16));
+            },
+            else => log.err("Wrote {} 0x{X:0>8} to 0x{X:0>8}, Unhandled", .{ T, value, address }),
+        },
+        u16 => switch (address) {
+            0x0800_00C4 => self.gpio.write(.Data, @truncate(u4, value)),
+            0x0800_00C6 => self.gpio.write(.Direction, @truncate(u4, value)),
+            0x0800_00C8 => self.gpio.write(.Control, @truncate(u1, value)),
+            else => log.err("Wrote {} 0x{X:0>4} to 0x{X:0>8}, Unhandled", .{ T, value, address }),
+        },
+        u8 => log.debug("Wrote {} 0x{X:0>2} to 0x{X:0>8}, Ignored.", .{ T, value, address }),
+        else => @compileError("GamePak: Unsupported write width"),
+    }
+}
+
+pub fn init(allocator: Allocator, cpu: *Arm7tdmi, rom_path: []const u8, save_path: ?[]const u8) !Self {
+    const file = try std.fs.cwd().openFile(rom_path, .{});
+    defer file.close();
+
+    const file_buf = try file.readToEndAlloc(allocator, try file.getEndPos());
+    const title = file_buf[0xA0..0xAC].*;
+    const kind = Backup.guess(file_buf);
+    const device = if (config.config().guest.force_rtc) .Rtc else guessDevice(file_buf);
+
+    logHeader(file_buf, &title);
+
+    return .{
+        .buf = file_buf,
+        .allocator = allocator,
+        .title = title,
+        .backup = try Backup.init(allocator, kind, title, save_path),
+        .gpio = try Gpio.init(allocator, cpu, device),
+    };
+}
+
+pub fn deinit(self: *Self) void {
+    self.backup.deinit();
+    self.gpio.deinit(self.allocator);
+    self.allocator.destroy(self.gpio);
+    self.allocator.free(self.buf);
+    self.* = undefined;
+}
+
+/// Searches the ROM to see if it can determine whether the ROM it's searching uses
+/// any GPIO device, like a RTC for example.
+fn guessDevice(buf: []const u8) Gpio.Device.Kind {
+    // Try to Guess if ROM uses RTC
+    const needle = "RTC_V"; // I was told SIIRTC_V, though Pokemen Firered (USA) is a false negative
+
+    // TODO: Use new for loop syntax?
+    var i: usize = 0;
+    while ((i + needle.len) < buf.len) : (i += 1) {
+        if (std.mem.eql(u8, needle, buf[i..(i + needle.len)])) return .Rtc;
+    }
+
+    // TODO: Detect other GPIO devices
+
+    return .None;
+}
+
+fn logHeader(buf: []const u8, title: *const [12]u8) void {
+    const code = buf[0xAC..0xB0];
+    const maker = buf[0xB0..0xB2];
+    const version = buf[0xBC];
+
+    log.info("Title: {s}", .{title});
+    if (version != 0) log.info("Version: {}", .{version});
+    log.info("Game Code: {s}", .{code});
+    log.info("Maker Code: {s}", .{maker});
+}
+
+test "OOB Access" {
+    const title = .{ 'H', 'E', 'L', 'L', 'O', ' ', 'W', 'O', 'R', 'L', 'D', '!' };
+    const alloc = std.testing.allocator;
+    const pak = Self{
+        .buf = &.{},
+        .alloc = alloc,
+        .title = title,
+        .backup = try Backup.init(alloc, .None, title, null),
+    };
+
+    std.debug.assert(pak.get(0) == 0x00); // 0x0000
+    std.debug.assert(pak.get(1) == 0x00);
+
+    std.debug.assert(pak.get(2) == 0x01); // 0x0001
+    std.debug.assert(pak.get(3) == 0x00);
+
+    std.debug.assert(pak.get(4) == 0x02); // 0x0002
+    std.debug.assert(pak.get(5) == 0x00);
+}

src/core/bus/Iwram.zig

@@ -0,0 +1,41 @@
+const std = @import("std");
+
+const Allocator = std.mem.Allocator;
+const iwram_size = 0x8000;
+const Self = @This();
+
+buf: []u8,
+allocator: Allocator,
+
+pub fn read(self: *const Self, comptime T: type, address: usize) T {
+    const addr = address & 0x7FFF;
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.readIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)]),
+        else => @compileError("IWRAM: Unsupported read width"),
+    };
+}
+
+pub fn write(self: *const Self, comptime T: type, address: usize, value: T) void {
+    const addr = address & 0x7FFF;
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.writeIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)], value),
+        else => @compileError("IWRAM: Unsupported write width"),
+    };
+}
+
+pub fn init(allocator: Allocator) !Self {
+    const buf = try allocator.alloc(u8, iwram_size);
+    std.mem.set(u8, buf, 0);
+
+    return Self{
+        .buf = buf,
+        .allocator = allocator,
+    };
+}
+
+pub fn deinit(self: *Self) void {
+    self.allocator.free(self.buf);
+    self.* = undefined;
+}

src/core/bus/backup.zig

@@ -0,0 +1,219 @@
+const std = @import("std");
+const Allocator = std.mem.Allocator;
+const log = std.log.scoped(.Backup);
+
+const Eeprom = @import("backup/eeprom.zig").Eeprom;
+const Flash = @import("backup/Flash.zig");
+
+const escape = @import("../../util.zig").escape;
+
+const Needle = struct { str: []const u8, kind: Backup.Kind };
+const backup_kinds = [6]Needle{
+    .{ .str = "EEPROM_V", .kind = .Eeprom },
+    .{ .str = "SRAM_V", .kind = .Sram },
+    .{ .str = "SRAM_F_V", .kind = .Sram },
+    .{ .str = "FLASH_V", .kind = .Flash },
+    .{ .str = "FLASH512_V", .kind = .Flash },
+    .{ .str = "FLASH1M_V", .kind = .Flash1M },
+};
+
+const SaveError = error{Unsupported};
+
+pub const Backup = struct {
+    const Self = @This();
+
+    buf: []u8,
+    allocator: Allocator,
+    kind: Kind,
+
+    title: [12]u8,
+    save_path: ?[]const u8,
+
+    flash: Flash,
+    eeprom: Eeprom,
+
+    const Kind = enum {
+        Eeprom,
+        Sram,
+        Flash,
+        Flash1M,
+        None,
+    };
+
+    pub fn read(self: *const Self, address: usize) u8 {
+        const addr = address & 0xFFFF;
+
+        switch (self.kind) {
+            .Flash => {
+                switch (addr) {
+                    0x0000 => if (self.flash.id_mode) return 0x32, // Panasonic manufacturer ID
+                    0x0001 => if (self.flash.id_mode) return 0x1B, // Panasonic device ID
+                    else => {},
+                }
+
+                return self.flash.read(self.buf, addr);
+            },
+            .Flash1M => {
+                switch (addr) {
+                    0x0000 => if (self.flash.id_mode) return 0x62, // Sanyo manufacturer ID
+                    0x0001 => if (self.flash.id_mode) return 0x13, // Sanyo device ID
+                    else => {},
+                }
+
+                return self.flash.read(self.buf, addr);
+            },
+            .Sram => return self.buf[addr & 0x7FFF], // 32K SRAM chip is mirrored
+            .None, .Eeprom => return 0xFF,
+        }
+    }
+
+    pub fn write(self: *Self, address: usize, byte: u8) void {
+        const addr = address & 0xFFFF;
+
+        switch (self.kind) {
+            .Flash, .Flash1M => {
+                if (self.flash.prep_write) return self.flash.write(self.buf, addr, byte);
+                if (self.flash.shouldEraseSector(addr, byte)) return self.flash.erase(self.buf, addr);
+
+                switch (addr) {
+                    0x0000 => if (self.kind == .Flash1M and self.flash.set_bank) {
+                        self.flash.bank = @truncate(u1, byte);
+                    },
+                    0x5555 => {
+                        if (self.flash.state == .Command) {
+                            self.flash.handleCommand(self.buf, byte);
+                        } else if (byte == 0xAA and self.flash.state == .Ready) {
+                            self.flash.state = .Set;
+                        } else if (byte == 0xF0) {
+                            self.flash.state = .Ready;
+                        }
+                    },
+                    0x2AAA => if (byte == 0x55 and self.flash.state == .Set) {
+                        self.flash.state = .Command;
+                    },
+                    else => {},
+                }
+            },
+            .Sram => self.buf[addr & 0x7FFF] = byte,
+            .None, .Eeprom => {},
+        }
+    }
+
+    pub fn init(allocator: Allocator, kind: Kind, title: [12]u8, path: ?[]const u8) !Self {
+        log.info("Kind: {}", .{kind});
+
+        const buf_size: usize = switch (kind) {
+            .Sram => 0x8000, // 32K
+            .Flash => 0x10000, // 64K
+            .Flash1M => 0x20000, // 128K
+            .None, .Eeprom => 0, // EEPROM is handled upon first Read Request to it
+        };
+
+        const buf = try allocator.alloc(u8, buf_size);
+        std.mem.set(u8, buf, 0xFF);
+
+        var backup = Self{
+            .buf = buf,
+            .allocator = allocator,
+            .kind = kind,
+            .title = title,
+            .save_path = path,
+            .flash = Flash.create(),
+            .eeprom = Eeprom.create(allocator),
+        };
+
+        if (backup.save_path) |p| backup.readSave(allocator, p) catch |e| log.err("Failed to load save: {}", .{e});
+        return backup;
+    }
+
+    pub fn deinit(self: *Self) void {
+        if (self.save_path) |path| self.writeSave(self.allocator, path) catch |e| log.err("Failed to write save: {}", .{e});
+        self.allocator.free(self.buf);
+        self.* = undefined;
+    }
+
+    /// Guesses the Backup Kind of a GBA ROM
+    pub fn guess(rom: []const u8) Kind {
+        for (backup_kinds) |needle| {
+            const needle_len = needle.str.len;
+
+            // TODO: Use new for loop syntax?
+            var i: usize = 0;
+            while ((i + needle_len) < rom.len) : (i += 1) {
+                if (std.mem.eql(u8, needle.str, rom[i..][0..needle_len])) return needle.kind;
+            }
+        }
+
+        return .None;
+    }
+
+    fn readSave(self: *Self, allocator: Allocator, path: []const u8) !void {
+        const file_path = try self.savePath(allocator, path);
+        defer allocator.free(file_path);
+
+        const expected = "untitled.sav";
+        if (std.mem.eql(u8, file_path[file_path.len - expected.len .. file_path.len], expected)) {
+            return log.err("ROM header lacks title, no save loaded", .{});
+        }
+
+        const file: std.fs.File = try std.fs.openFileAbsolute(file_path, .{});
+        const file_buf = try file.readToEndAlloc(allocator, try file.getEndPos());
+        defer allocator.free(file_buf);
+
+        switch (self.kind) {
+            .Sram, .Flash, .Flash1M => {
+                if (self.buf.len == file_buf.len) {
+                    std.mem.copy(u8, self.buf, file_buf);
+                    return log.info("Loaded Save from {s}", .{file_path});
+                }
+
+                log.err("{s} is {} bytes, but we expected {} bytes", .{ file_path, file_buf.len, self.buf.len });
+            },
+            .Eeprom => {
+                if (file_buf.len == 0x200 or file_buf.len == 0x2000) {
+                    self.eeprom.kind = if (file_buf.len == 0x200) .Small else .Large;
+
+                    self.buf = try allocator.alloc(u8, file_buf.len);
+                    std.mem.copy(u8, self.buf, file_buf);
+                    return log.info("Loaded Save from {s}", .{file_path});
+                }
+
+                log.err("EEPROM can either be 0x200 bytes or 0x2000 byes, but {s} was {X:} bytes", .{
+                    file_path,
+                    file_buf.len,
+                });
+            },
+            .None => return SaveError.Unsupported,
+        }
+    }
+
+    fn savePath(self: *const Self, allocator: Allocator, path: []const u8) ![]const u8 {
+        const filename = try self.saveName(allocator);
+        defer allocator.free(filename);
+
+        return try std.fs.path.join(allocator, &[_][]const u8{ path, filename });
+    }
+
+    fn saveName(self: *const Self, allocator: Allocator) ![]const u8 {
+        const title_str = std.mem.sliceTo(&escape(self.title), 0);
+        const name = if (title_str.len != 0) title_str else "untitled";
+
+        return try std.mem.concat(allocator, u8, &[_][]const u8{ name, ".sav" });
+    }
+
+    fn writeSave(self: Self, allocator: Allocator, path: []const u8) !void {
+        const file_path = try self.savePath(allocator, path);
+        defer allocator.free(file_path);
+
+        switch (self.kind) {
+            .Sram, .Flash, .Flash1M, .Eeprom => {
+                const file = try std.fs.createFileAbsolute(file_path, .{});
+                defer file.close();
+
+                try file.writeAll(self.buf);
+                log.info("Wrote Save to {s}", .{file_path});
+            },
+            else => return SaveError.Unsupported,
+        }
+    }
+};

src/core/bus/backup/Flash.zig

@@ -0,0 +1,72 @@
+const std = @import("std");
+
+const Self = @This();
+
+state: State,
+
+id_mode: bool,
+set_bank: bool,
+prep_erase: bool,
+prep_write: bool,
+
+bank: u1,
+
+const State = enum {
+    Ready,
+    Set,
+    Command,
+};
+
+pub fn read(self: *const Self, buf: []u8, idx: usize) u8 {
+    return buf[self.address() + idx];
+}
+
+pub fn write(self: *Self, buf: []u8, idx: usize, byte: u8) void {
+    buf[self.address() + idx] = byte;
+    self.prep_write = false;
+}
+
+pub fn create() Self {
+    return .{
+        .state = .Ready,
+        .id_mode = false,
+        .set_bank = false,
+        .prep_erase = false,
+        .prep_write = false,
+        .bank = 0,
+    };
+}
+
+pub fn handleCommand(self: *Self, buf: []u8, byte: u8) void {
+    switch (byte) {
+        0x90 => self.id_mode = true,
+        0xF0 => self.id_mode = false,
+        0xB0 => self.set_bank = true,
+        0x80 => self.prep_erase = true,
+        0x10 => {
+            std.mem.set(u8, buf, 0xFF);
+            self.prep_erase = false;
+        },
+        0xA0 => self.prep_write = true,
+        else => std.debug.panic("Unhandled Flash Command: 0x{X:0>2}", .{byte}),
+    }
+
+    self.state = .Ready;
+}
+
+pub fn shouldEraseSector(self: *const Self, addr: usize, byte: u8) bool {
+    return self.state == .Command and self.prep_erase and byte == 0x30 and addr & 0xFFF == 0x000;
+}
+
+pub fn erase(self: *Self, buf: []u8, sector: usize) void {
+    const start = self.address() + (sector & 0xF000);
+
+    std.mem.set(u8, buf[start..][0..0x1000], 0xFF);
+    self.prep_erase = false;
+    self.state = .Ready;
+}
+
+/// Base Address
+inline fn address(self: *const Self) usize {
+    return if (self.bank == 1) 0x10000 else @as(usize, 0);
+}

src/core/bus/backup/eeprom.zig

@@ -0,0 +1,269 @@
+const std = @import("std");
+
+const Allocator = std.mem.Allocator;
+
+const log = std.log.scoped(.Eeprom);
+
+pub const Eeprom = struct {
+    const Self = @This();
+
+    addr: u14,
+
+    kind: Kind,
+    state: State,
+    writer: Writer,
+    reader: Reader,
+
+    allocator: Allocator,
+
+    const Kind = enum {
+        Unknown,
+        Small, // 512B
+        Large, // 8KB
+    };
+
+    const State = enum {
+        Ready,
+        Read,
+        Write,
+        WriteTransfer,
+        RequestEnd,
+    };
+
+    pub fn read(self: *Self) u1 {
+        return self.reader.read();
+    }
+
+    pub fn dbgRead(self: *const Self) u1 {
+        return self.reader.dbgRead();
+    }
+
+    pub fn write(self: *Self, word_count: u16, buf: *[]u8, bit: u1) void {
+        if (self.guessKind(word_count)) |found| {
+            log.info("EEPROM Kind: {}", .{found});
+            self.kind = found;
+
+            // buf.len will not equal zero when a save file was found and loaded.
+            // Right now, we assume that the save file is of the correct size which
+            // isn't necessarily true, since we can't trust anything a user can influence
+            // TODO: use ?[]u8 instead of a 0-sized slice?
+            if (buf.len == 0) {
+                const len: usize = switch (found) {
+                    .Small => 0x200,
+                    .Large => 0x2000,
+                    else => unreachable,
+                };
+
+                buf.* = self.allocator.alloc(u8, len) catch |e| {
+                    log.err("Failed to resize EEPROM buf to {} bytes", .{len});
+                    std.debug.panic("EEPROM entered irrecoverable state {}", .{e});
+                };
+                std.mem.set(u8, buf.*, 0xFF);
+            }
+        }
+
+        if (self.state == .RequestEnd) {
+            // if (bit != 0) log.debug("EEPROM Request did not end in 0u1. TODO: is this ok?", .{});
+            self.state = .Ready;
+            return;
+        }
+
+        switch (self.state) {
+            .Ready => self.writer.requestWrite(bit),
+            .Read, .Write => self.writer.addressWrite(self.kind, bit),
+            .WriteTransfer => self.writer.dataWrite(bit),
+            .RequestEnd => unreachable, // We return early just above this block
+        }
+
+        self.tick(buf.*);
+    }
+
+    pub fn create(allocator: Allocator) Self {
+        return .{
+            .kind = .Unknown,
+            .state = .Ready,
+            .writer = Writer.create(),
+            .reader = Reader.create(),
+            .addr = 0,
+            .allocator = allocator,
+        };
+    }
+
+    fn guessKind(self: *const Self, word_count: u16) ?Kind {
+        if (self.kind != .Unknown or self.state != .Read) return null;
+
+        return switch (word_count) {
+            17 => .Large,
+            9 => .Small,
+            else => blk: {
+                log.err("Unexpected length of DMA3 Transfer upon initial EEPROM read: {}", .{word_count});
+                break :blk null;
+            },
+        };
+    }
+
+    fn tick(self: *Self, buf: []u8) void {
+        switch (self.state) {
+            .Ready => {
+                if (self.writer.len() == 2) {
+                    const req = @intCast(u2, self.writer.finish());
+                    switch (req) {
+                        0b11 => self.state = .Read,
+                        0b10 => self.state = .Write,
+                        else => log.err("Unknown EEPROM Request 0b{b:0>2}", .{req}),
+                    }
+                }
+            },
+            .Read => {
+                switch (self.kind) {
+                    .Large => {
+                        if (self.writer.len() == 14) {
+                            const addr = @intCast(u10, self.writer.finish());
+                            const value = std.mem.readIntSliceLittle(u64, buf[@as(u13, addr) * 8 ..][0..8]);
+
+                            self.reader.configure(value);
+                            self.state = .RequestEnd;
+                        }
+                    },
+                    .Small => {
+                        if (self.writer.len() == 6) {
+                            // FIXME: Duplicated code from above
+                            const addr = @intCast(u6, self.writer.finish());
+                            const value = std.mem.readIntSliceLittle(u64, buf[@as(u13, addr) * 8 ..][0..8]);
+
+                            self.reader.configure(value);
+                            self.state = .RequestEnd;
+                        }
+                    },
+                    else => log.err("Unable to calculate EEPROM read address. EEPROM size UNKNOWN", .{}),
+                }
+            },
+            .Write => {
+                switch (self.kind) {
+                    .Large => {
+                        if (self.writer.len() == 14) {
+                            self.addr = @intCast(u10, self.writer.finish());
+                            self.state = .WriteTransfer;
+                        }
+                    },
+                    .Small => {
+                        if (self.writer.len() == 6) {
+                            self.addr = @intCast(u6, self.writer.finish());
+                            self.state = .WriteTransfer;
+                        }
+                    },
+                    else => log.err("Unable to calculate EEPROM write address. EEPROM size UNKNOWN", .{}),
+                }
+            },
+            .WriteTransfer => {
+                if (self.writer.len() == 64) {
+                    std.mem.writeIntSliceLittle(u64, buf[self.addr * 8 ..][0..8], self.writer.finish());
+                    self.state = .RequestEnd;
+                }
+            },
+            .RequestEnd => unreachable, // We return early in write() if state is .RequestEnd
+        }
+    }
+};
+
+const Reader = struct {
+    const Self = @This();
+
+    data: u64,
+    i: u8,
+    enabled: bool,
+
+    fn create() Self {
+        return .{
+            .data = 0,
+            .i = 0,
+            .enabled = false,
+        };
+    }
+
+    fn read(self: *Self) u1 {
+        if (!self.enabled) return 1;
+
+        const bit = if (self.i < 4) blk: {
+            break :blk 0;
+        } else blk: {
+            const idx = @intCast(u6, 63 - (self.i - 4));
+            break :blk @truncate(u1, self.data >> idx);
+        };
+
+        self.i = (self.i + 1) % (64 + 4);
+        if (self.i == 0) self.enabled = false;
+
+        return bit;
+    }
+
+    fn dbgRead(self: *const Self) u1 {
+        if (!self.enabled) return 1;
+
+        const bit = if (self.i < 4) blk: {
+            break :blk 0;
+        } else blk: {
+            const idx = @intCast(u6, 63 - (self.i - 4));
+            break :blk @truncate(u1, self.data >> idx);
+        };
+
+        return bit;
+    }
+
+    fn configure(self: *Self, value: u64) void {
+        self.data = value;
+        self.i = 0;
+        self.enabled = true;
+    }
+};
+
+const Writer = struct {
+    const Self = @This();
+
+    data: u64,
+    i: u8,
+
+    fn create() Self {
+        return .{ .data = 0, .i = 0 };
+    }
+
+    fn requestWrite(self: *Self, bit: u1) void {
+        const idx = @intCast(u1, 1 - self.i);
+        self.data = (self.data & ~(@as(u64, 1) << idx)) | (@as(u64, bit) << idx);
+        self.i += 1;
+    }
+
+    fn addressWrite(self: *Self, kind: Eeprom.Kind, bit: u1) void {
+        if (kind == .Unknown) return;
+
+        const size: u4 = switch (kind) {
+            .Large => 13,
+            .Small => 5,
+            .Unknown => unreachable,
+        };
+
+        const idx = @intCast(u4, size - self.i);
+        self.data = (self.data & ~(@as(u64, 1) << idx)) | (@as(u64, bit) << idx);
+        self.i += 1;
+    }
+
+    fn dataWrite(self: *Self, bit: u1) void {
+        const idx = @intCast(u6, 63 - self.i);
+        self.data = (self.data & ~(@as(u64, 1) << idx)) | (@as(u64, bit) << idx);
+        self.i += 1;
+    }
+
+    fn len(self: *const Self) u8 {
+        return self.i;
+    }
+
+    fn finish(self: *Self) u64 {
+        defer self.reset();
+        return self.data;
+    }
+
+    fn reset(self: *Self) void {
+        self.i = 0;
+        self.data = 0;
+    }
+};

src/core/bus/dma.zig

@@ -0,0 +1,357 @@
+const std = @import("std");
+const util = @import("../../util.zig");
+
+const DmaControl = @import("io.zig").DmaControl;
+const Bus = @import("../Bus.zig");
+const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
+
+pub const DmaTuple = struct { DmaController(0), DmaController(1), DmaController(2), DmaController(3) };
+const log = std.log.scoped(.DmaTransfer);
+
+const getHalf = util.getHalf;
+const setHalf = util.setHalf;
+const setQuart = util.setQuart;
+
+const rotr = @import("../../util.zig").rotr;
+
+pub fn create() DmaTuple {
+    return .{ DmaController(0).init(), DmaController(1).init(), DmaController(2).init(), DmaController(3).init() };
+}
+
+pub fn read(comptime T: type, dma: *const DmaTuple, addr: u32) ?T {
+    const byte_addr = @truncate(u8, addr);
+
+    return switch (T) {
+        u32 => switch (byte_addr) {
+            0xB0, 0xB4 => null, // DMA0SAD, DMA0DAD,
+            0xB8 => @as(T, dma.*[0].dmacntH()) << 16, // DMA0CNT_L is write-only
+            0xBC, 0xC0 => null, // DMA1SAD, DMA1DAD
+            0xC4 => @as(T, dma.*[1].dmacntH()) << 16, // DMA1CNT_L is write-only
+            0xC8, 0xCC => null, // DMA2SAD, DMA2DAD
+            0xD0 => @as(T, dma.*[2].dmacntH()) << 16, // DMA2CNT_L is write-only
+            0xD4, 0xD8 => null, // DMA3SAD, DMA3DAD
+            0xDC => @as(T, dma.*[3].dmacntH()) << 16, // DMA3CNT_L is write-only
+            else => util.io.read.err(T, log, "unaligned {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        u16 => switch (byte_addr) {
+            0xB0, 0xB2, 0xB4, 0xB6 => null, // DMA0SAD, DMA0DAD
+            0xB8 => 0x0000, // DMA0CNT_L, suite.gba expects 0x0000 instead of 0xDEAD
+            0xBA => dma.*[0].dmacntH(),
+
+            0xBC, 0xBE, 0xC0, 0xC2 => null, // DMA1SAD, DMA1DAD
+            0xC4 => 0x0000, // DMA1CNT_L
+            0xC6 => dma.*[1].dmacntH(),
+
+            0xC8, 0xCA, 0xCC, 0xCE => null, // DMA2SAD, DMA2DAD
+            0xD0 => 0x0000, // DMA2CNT_L
+            0xD2 => dma.*[2].dmacntH(),
+
+            0xD4, 0xD6, 0xD8, 0xDA => null, // DMA3SAD, DMA3DAD
+            0xDC => 0x0000, // DMA3CNT_L
+            0xDE => dma.*[3].dmacntH(),
+            else => util.io.read.err(T, log, "unaligned {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        u8 => switch (byte_addr) {
+            0xB0...0xB7 => null, // DMA0SAD, DMA0DAD
+            0xB8, 0xB9 => 0x00, // DMA0CNT_L
+            0xBA, 0xBB => @truncate(T, dma.*[0].dmacntH() >> getHalf(byte_addr)),
+
+            0xBC...0xC3 => null, // DMA1SAD, DMA1DAD
+            0xC4, 0xC5 => 0x00, // DMA1CNT_L
+            0xC6, 0xC7 => @truncate(T, dma.*[1].dmacntH() >> getHalf(byte_addr)),
+
+            0xC8...0xCF => null, // DMA2SAD, DMA2DAD
+            0xD0, 0xD1 => 0x00, // DMA2CNT_L
+            0xD2, 0xD3 => @truncate(T, dma.*[2].dmacntH() >> getHalf(byte_addr)),
+
+            0xD4...0xDB => null, // DMA3SAD, DMA3DAD
+            0xDC, 0xDD => 0x00, // DMA3CNT_L
+            0xDE, 0xDF => @truncate(T, dma.*[3].dmacntH() >> getHalf(byte_addr)),
+            else => util.io.read.err(T, log, "unexpected {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        else => @compileError("DMA: Unsupported read width"),
+    };
+}
+
+pub fn write(comptime T: type, dma: *DmaTuple, addr: u32, value: T) void {
+    const byte_addr = @truncate(u8, addr);
+
+    switch (T) {
+        u32 => switch (byte_addr) {
+            0xB0 => dma.*[0].setDmasad(value),
+            0xB4 => dma.*[0].setDmadad(value),
+            0xB8 => dma.*[0].setDmacnt(value),
+
+            0xBC => dma.*[1].setDmasad(value),
+            0xC0 => dma.*[1].setDmadad(value),
+            0xC4 => dma.*[1].setDmacnt(value),
+
+            0xC8 => dma.*[2].setDmasad(value),
+            0xCC => dma.*[2].setDmadad(value),
+            0xD0 => dma.*[2].setDmacnt(value),
+
+            0xD4 => dma.*[3].setDmasad(value),
+            0xD8 => dma.*[3].setDmadad(value),
+            0xDC => dma.*[3].setDmacnt(value),
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, addr }),
+        },
+        u16 => switch (byte_addr) {
+            0xB0, 0xB2 => dma.*[0].setDmasad(setHalf(u32, dma.*[0].sad, byte_addr, value)),
+            0xB4, 0xB6 => dma.*[0].setDmadad(setHalf(u32, dma.*[0].dad, byte_addr, value)),
+            0xB8 => dma.*[0].setDmacntL(value),
+            0xBA => dma.*[0].setDmacntH(value),
+
+            0xBC, 0xBE => dma.*[1].setDmasad(setHalf(u32, dma.*[1].sad, byte_addr, value)),
+            0xC0, 0xC2 => dma.*[1].setDmadad(setHalf(u32, dma.*[1].dad, byte_addr, value)),
+            0xC4 => dma.*[1].setDmacntL(value),
+            0xC6 => dma.*[1].setDmacntH(value),
+
+            0xC8, 0xCA => dma.*[2].setDmasad(setHalf(u32, dma.*[2].sad, byte_addr, value)),
+            0xCC, 0xCE => dma.*[2].setDmadad(setHalf(u32, dma.*[2].dad, byte_addr, value)),
+            0xD0 => dma.*[2].setDmacntL(value),
+            0xD2 => dma.*[2].setDmacntH(value),
+
+            0xD4, 0xD6 => dma.*[3].setDmasad(setHalf(u32, dma.*[3].sad, byte_addr, value)),
+            0xD8, 0xDA => dma.*[3].setDmadad(setHalf(u32, dma.*[3].dad, byte_addr, value)),
+            0xDC => dma.*[3].setDmacntL(value),
+            0xDE => dma.*[3].setDmacntH(value),
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, addr }),
+        },
+        u8 => switch (byte_addr) {
+            0xB0, 0xB1, 0xB2, 0xB3 => dma.*[0].setDmasad(setQuart(dma.*[0].sad, byte_addr, value)),
+            0xB4, 0xB5, 0xB6, 0xB7 => dma.*[0].setDmadad(setQuart(dma.*[0].dad, byte_addr, value)),
+            0xB8, 0xB9 => dma.*[0].setDmacntL(setHalf(u16, dma.*[0].word_count, byte_addr, value)),
+            0xBA, 0xBB => dma.*[0].setDmacntH(setHalf(u16, dma.*[0].cnt.raw, byte_addr, value)),
+
+            0xBC, 0xBD, 0xBE, 0xBF => dma.*[1].setDmasad(setQuart(dma.*[1].sad, byte_addr, value)),
+            0xC0, 0xC1, 0xC2, 0xC3 => dma.*[1].setDmadad(setQuart(dma.*[1].dad, byte_addr, value)),
+            0xC4, 0xC5 => dma.*[1].setDmacntL(setHalf(u16, dma.*[1].word_count, byte_addr, value)),
+            0xC6, 0xC7 => dma.*[1].setDmacntH(setHalf(u16, dma.*[1].cnt.raw, byte_addr, value)),
+
+            0xC8, 0xC9, 0xCA, 0xCB => dma.*[2].setDmasad(setQuart(dma.*[2].sad, byte_addr, value)),
+            0xCC, 0xCD, 0xCE, 0xCF => dma.*[2].setDmadad(setQuart(dma.*[2].dad, byte_addr, value)),
+            0xD0, 0xD1 => dma.*[2].setDmacntL(setHalf(u16, dma.*[2].word_count, byte_addr, value)),
+            0xD2, 0xD3 => dma.*[2].setDmacntH(setHalf(u16, dma.*[2].cnt.raw, byte_addr, value)),
+
+            0xD4, 0xD5, 0xD6, 0xD7 => dma.*[3].setDmasad(setQuart(dma.*[3].sad, byte_addr, value)),
+            0xD8, 0xD9, 0xDA, 0xDB => dma.*[3].setDmadad(setQuart(dma.*[3].dad, byte_addr, value)),
+            0xDC, 0xDD => dma.*[3].setDmacntL(setHalf(u16, dma.*[3].word_count, byte_addr, value)),
+            0xDE, 0xDF => dma.*[3].setDmacntH(setHalf(u16, dma.*[3].cnt.raw, byte_addr, value)),
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, addr }),
+        },
+        else => @compileError("DMA: Unsupported write width"),
+    }
+}
+
+/// Function that creates a DMAController. Determines unique DMA Controller behaiour at compile-time
+fn DmaController(comptime id: u2) type {
+    return struct {
+        const Self = @This();
+
+        const sad_mask: u32 = if (id == 0) 0x07FF_FFFF else 0x0FFF_FFFF;
+        const dad_mask: u32 = if (id != 3) 0x07FF_FFFF else 0x0FFF_FFFF;
+        const WordCount = if (id == 3) u16 else u14;
+
+        /// Write-only. The first address in a DMA transfer. (DMASAD)
+        /// Note: use writeSrc instead of manipulating src_addr directly
+        sad: u32,
+        /// Write-only. The final address in a DMA transffer. (DMADAD)
+        /// Note: Use writeDst instead of manipulatig dst_addr directly
+        dad: u32,
+        /// Write-only. The Word Count for the DMA Transfer (DMACNT_L)
+        word_count: WordCount,
+        /// Read / Write. DMACNT_H
+        /// Note: Use writeControl instead of manipulating cnt directly.
+        cnt: DmaControl,
+
+        /// Internal. The last successfully read value
+        data_latch: u32,
+        /// Internal. Currrent Source Address
+        sad_latch: u32,
+        /// Internal. Current Destination Address
+        dad_latch: u32,
+        /// Internal. Word Count
+        _word_count: WordCount,
+
+        /// Some DMA Transfers are enabled during Hblank / VBlank and / or
+        /// have delays. Thefore bit 15 of DMACNT isn't actually something
+        /// we can use to control when we do or do not execute a step in a DMA Transfer
+        in_progress: bool,
+
+        pub fn init() Self {
+            return .{
+                .sad = 0,
+                .dad = 0,
+                .word_count = 0,
+                .cnt = .{ .raw = 0x000 },
+
+                // Internals
+                .sad_latch = 0,
+                .dad_latch = 0,
+                .data_latch = 0,
+
+                ._word_count = 0,
+                .in_progress = false,
+            };
+        }
+
+        pub fn setDmasad(self: *Self, addr: u32) void {
+            self.sad = addr & sad_mask;
+        }
+
+        pub fn setDmadad(self: *Self, addr: u32) void {
+            self.dad = addr & dad_mask;
+        }
+
+        pub fn setDmacntL(self: *Self, halfword: u16) void {
+            self.word_count = @truncate(@TypeOf(self.word_count), halfword);
+        }
+
+        pub fn dmacntH(self: *const Self) u16 {
+            return self.cnt.raw & if (id == 3) 0xFFE0 else 0xF7E0;
+        }
+
+        pub fn setDmacntH(self: *Self, halfword: u16) void {
+            const new = DmaControl{ .raw = halfword };
+
+            if (!self.cnt.enabled.read() and new.enabled.read()) {
+                // Reload Internals on Rising Edge.
+                self.sad_latch = self.sad;
+                self.dad_latch = self.dad;
+                self._word_count = if (self.word_count == 0) std.math.maxInt(WordCount) else self.word_count;
+
+                // Only a Start Timing of 00 has a DMA Transfer immediately begin
+                self.in_progress = new.start_timing.read() == 0b00;
+            }
+
+            self.cnt.raw = halfword;
+        }
+
+        pub fn setDmacnt(self: *Self, word: u32) void {
+            self.setDmacntL(@truncate(u16, word));
+            self.setDmacntH(@truncate(u16, word >> 16));
+        }
+
+        pub fn step(self: *Self, cpu: *Arm7tdmi) void {
+            const is_fifo = (id == 1 or id == 2) and self.cnt.start_timing.read() == 0b11;
+            const sad_adj = @intToEnum(Adjustment, self.cnt.sad_adj.read());
+            const dad_adj = if (is_fifo) .Fixed else @intToEnum(Adjustment, self.cnt.dad_adj.read());
+
+            const transfer_type = is_fifo or self.cnt.transfer_type.read();
+            const offset: u32 = if (transfer_type) @sizeOf(u32) else @sizeOf(u16);
+
+            const mask = if (transfer_type) ~@as(u32, 3) else ~@as(u32, 1);
+            const sad_addr = self.sad_latch & mask;
+            const dad_addr = self.dad_latch & mask;
+
+            if (transfer_type) {
+                if (sad_addr >= 0x0200_0000) self.data_latch = cpu.bus.read(u32, sad_addr);
+                cpu.bus.write(u32, dad_addr, self.data_latch);
+            } else {
+                if (sad_addr >= 0x0200_0000) {
+                    const value: u32 = cpu.bus.read(u16, sad_addr);
+                    self.data_latch = value << 16 | value;
+                }
+
+                cpu.bus.write(u16, dad_addr, @truncate(u16, rotr(u32, self.data_latch, 8 * (dad_addr & 3))));
+            }
+
+            switch (@truncate(u8, sad_addr >> 24)) {
+                // according to fleroviux, DMAs with a source address in ROM misbehave
+                // the resultant behaviour is that the source address will increment despite what DMAXCNT says
+                0x08...0x0D => self.sad_latch +%= offset, // obscure behaviour
+                else => switch (sad_adj) {
+                    .Increment => self.sad_latch +%= offset,
+                    .Decrement => self.sad_latch -%= offset,
+                    .IncrementReload => log.err("{} is a prohibited adjustment on SAD", .{sad_adj}),
+                    .Fixed => {},
+                },
+            }
+
+            switch (dad_adj) {
+                .Increment, .IncrementReload => self.dad_latch +%= offset,
+                .Decrement => self.dad_latch -%= offset,
+                .Fixed => {},
+            }
+
+            self._word_count -= 1;
+
+            if (self._word_count == 0) {
+                if (self.cnt.irq.read()) {
+                    switch (id) {
+                        0 => cpu.bus.io.irq.dma0.set(),
+                        1 => cpu.bus.io.irq.dma1.set(),
+                        2 => cpu.bus.io.irq.dma2.set(),
+                        3 => cpu.bus.io.irq.dma3.set(),
+                    }
+
+                    cpu.handleInterrupt();
+                }
+
+                // If we're not repeating, Fire the IRQs and disable the DMA
+                if (!self.cnt.repeat.read()) self.cnt.enabled.unset();
+
+                // We want to disable our internal enabled flag regardless of repeat
+                // because we only want to step A DMA that repeats during it's specific
+                // timing window
+                self.in_progress = false;
+            }
+        }
+
+        fn poll(self: *Self, comptime kind: DmaKind) void {
+            if (self.in_progress) return; // If there's an ongoing DMA Transfer, exit early
+
+            // No ongoing DMA Transfer, We want to check if we should repeat an existing one
+            // Determined by the repeat bit and whether the DMA is in the right start_timing
+            switch (kind) {
+                .VBlank => self.in_progress = self.cnt.enabled.read() and self.cnt.start_timing.read() == 0b01,
+                .HBlank => self.in_progress = self.cnt.enabled.read() and self.cnt.start_timing.read() == 0b10,
+                .Immediate, .Special => {},
+            }
+
+            // If we determined that the repeat bit is set (and now the Hblank / Vblank DMA is now in progress)
+            // Reload internal word count latch
+            // Reload internal DAD latch if we are in IncrementRelaod
+            if (self.in_progress) {
+                self._word_count = if (self.word_count == 0) std.math.maxInt(@TypeOf(self._word_count)) else self.word_count;
+                if (@intToEnum(Adjustment, self.cnt.dad_adj.read()) == .IncrementReload) self.dad_latch = self.dad;
+            }
+        }
+
+        pub fn requestAudio(self: *Self, _: u32) void {
+            comptime std.debug.assert(id == 1 or id == 2);
+            if (self.in_progress) return; // APU must wait their turn
+
+            // DMA May not be configured for handling DMAs
+            if (self.cnt.start_timing.read() != 0b11) return;
+
+            // We Assume the Repeat Bit is Set
+            // We Assume that DAD is set to 0x0400_00A0 or 0x0400_00A4 (fifo_addr)
+            // We Assume DMACNT_L is set to 4
+
+            // FIXME: Safe to just assume whatever DAD is set to is the FIFO Address?
+            // self.dad_latch = fifo_addr;
+            self.cnt.repeat.set();
+            self._word_count = 4;
+            self.in_progress = true;
+        }
+    };
+}
+
+pub fn onBlanking(bus: *Bus, comptime kind: DmaKind) void {
+    inline for (0..4) |i| bus.dma[i].poll(kind);
+}
+
+const Adjustment = enum(u2) {
+    Increment = 0,
+    Decrement = 1,
+    Fixed = 2,
+    IncrementReload = 3,
+};
+
+const DmaKind = enum(u2) {
+    Immediate = 0,
+    HBlank,
+    VBlank,
+    Special,
+};

src/core/bus/gpio.zig

@@ -0,0 +1,456 @@
+const std = @import("std");
+const Bit = @import("bitfield").Bit;
+const DateTime = @import("datetime").datetime.Datetime;
+
+const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
+const Allocator = std.mem.Allocator;
+
+/// GPIO Register Implementation
+pub const Gpio = struct {
+    const Self = @This();
+    const log = std.log.scoped(.Gpio);
+
+    data: u4,
+    direction: u4,
+    cnt: u1,
+
+    device: Device,
+
+    const Register = enum { Data, Direction, Control };
+
+    pub const Device = struct {
+        ptr: ?*anyopaque,
+        kind: Kind, // TODO: Make comptime known?
+
+        pub const Kind = enum { Rtc, None };
+
+        fn step(self: *Device, value: u4) u4 {
+            return switch (self.kind) {
+                .Rtc => blk: {
+                    const clock = @ptrCast(*Clock, @alignCast(@alignOf(*Clock), self.ptr.?));
+                    break :blk clock.step(Clock.Data{ .raw = value });
+                },
+                .None => value,
+            };
+        }
+
+        fn init(kind: Kind, ptr: ?*anyopaque) Device {
+            return .{ .kind = kind, .ptr = ptr };
+        }
+    };
+
+    pub fn write(self: *Self, comptime reg: Register, value: if (reg == .Control) u1 else u4) void {
+        switch (reg) {
+            .Data => {
+                const masked_value = value & self.direction;
+
+                // The value which is actually stored in the GPIO register
+                // might be modified by the device implementing the GPIO interface e.g. RTC reads
+                self.data = self.device.step(masked_value);
+            },
+            .Direction => self.direction = value,
+            .Control => self.cnt = value,
+        }
+    }
+
+    pub fn read(self: *const Self, comptime reg: Register) if (reg == .Control) u1 else u4 {
+        if (self.cnt == 0) return 0;
+
+        return switch (reg) {
+            .Data => self.data & ~self.direction,
+            .Direction => self.direction,
+            .Control => self.cnt,
+        };
+    }
+
+    pub fn init(allocator: Allocator, cpu: *Arm7tdmi, kind: Device.Kind) !*Self {
+        log.info("Device: {}", .{kind});
+
+        const self = try allocator.create(Self);
+        errdefer allocator.destroy(self);
+
+        self.* = .{
+            .data = 0b0000,
+            .direction = 0b1111, // TODO: What is GPIO Direction set to by default?
+            .cnt = 0b0,
+
+            .device = switch (kind) {
+                .Rtc => blk: {
+                    const clock = try allocator.create(Clock);
+                    clock.init(cpu, self);
+
+                    break :blk Device{ .kind = kind, .ptr = clock };
+                },
+                .None => Device{ .kind = kind, .ptr = null },
+            },
+        };
+
+        return self;
+    }
+
+    pub fn deinit(self: *Self, allocator: Allocator) void {
+        switch (self.device.kind) {
+            .Rtc => allocator.destroy(@ptrCast(*Clock, @alignCast(@alignOf(*Clock), self.device.ptr.?))),
+            .None => {},
+        }
+
+        self.* = undefined;
+    }
+};
+
+/// GBA Real Time Clock
+pub const Clock = struct {
+    const Self = @This();
+    const log = std.log.scoped(.Rtc);
+
+    writer: Writer,
+    reader: Reader,
+    state: State,
+    cnt: Control,
+
+    year: u8,
+    month: u5,
+    day: u6,
+    weekday: u3,
+    hour: u6,
+    minute: u7,
+    second: u7,
+
+    cpu: *Arm7tdmi,
+    gpio: *const Gpio,
+
+    const Register = enum {
+        Control,
+        DateTime,
+        Time,
+    };
+
+    const State = union(enum) {
+        Idle,
+        Command,
+        Write: Register,
+        Read: Register,
+    };
+
+    const Reader = struct {
+        i: u4,
+        count: u8,
+
+        /// Reads a bit from RTC registers. Which bit it reads is dependent on
+        ///
+        /// 1. The RTC State Machine, whitch tells us which register we're accessing
+        /// 2. A `count`, which keeps track of which byte is currently being read
+        /// 3. An index, which keeps track of which bit of the byte determined by `count` is being read
+        fn read(self: *Reader, clock: *const Clock, register: Register) u1 {
+            const idx = @intCast(u3, self.i);
+            defer self.i += 1;
+
+            // FIXME: What do I do about the unused bits?
+            return switch (register) {
+                .Control => @truncate(u1, switch (self.count) {
+                    0 => clock.cnt.raw >> idx,
+                    else => std.debug.panic("Tried to read from byte #{} of {} (hint: there's only 1 byte)", .{ self.count, register }),
+                }),
+                .DateTime => @truncate(u1, switch (self.count) {
+                    // Date
+                    0 => clock.year >> idx,
+                    1 => @as(u8, clock.month) >> idx,
+                    2 => @as(u8, clock.day) >> idx,
+                    3 => @as(u8, clock.weekday) >> idx,
+
+                    // Time
+                    4 => @as(u8, clock.hour) >> idx,
+                    5 => @as(u8, clock.minute) >> idx,
+                    6 => @as(u8, clock.second) >> idx,
+                    else => std.debug.panic("Tried to read from byte #{} of {} (hint: there's only 7 bytes)", .{ self.count, register }),
+                }),
+                .Time => @truncate(u1, switch (self.count) {
+                    0 => @as(u8, clock.hour) >> idx,
+                    1 => @as(u8, clock.minute) >> idx,
+                    2 => @as(u8, clock.second) >> idx,
+                    else => std.debug.panic("Tried to read from byte #{} of {} (hint: there's only 3 bytes)", .{ self.count, register }),
+                }),
+            };
+        }
+
+        /// Is true when a Reader has read a u8's worth of bits
+        fn finished(self: *const Reader) bool {
+            return self.i >= 8;
+        }
+
+        /// Resets the index used to shift bits out of RTC registers
+        /// and `count`, which is used to keep track of which byte we're reading
+        /// is incremeneted
+        fn lap(self: *Reader) void {
+            self.i = 0;
+            self.count += 1;
+        }
+
+        /// Resets the state of a `Reader` in preparation for a future
+        /// read command
+        fn reset(self: *Reader) void {
+            self.i = 0;
+            self.count = 0;
+        }
+    };
+
+    const Writer = struct {
+        buf: u8,
+        i: u4,
+
+        /// The Number of bytes written since last reset
+        count: u8,
+
+        /// Append a bit to the internal bit buffer (aka an integer)
+        fn push(self: *Writer, value: u1) void {
+            const idx = @intCast(u3, self.i);
+            self.buf = (self.buf & ~(@as(u8, 1) << idx)) | @as(u8, value) << idx;
+            self.i += 1;
+        }
+
+        /// Takes the contents of the internal buffer and writes it to an RTC register
+        /// Where it writes to is dependent on:
+        ///
+        /// 1. The RTC State Machine, whitch tells us which register we're accessing
+        /// 2. A `count`, which keeps track of which byte is currently being read
+        fn write(self: *const Writer, clock: *Clock, register: Register) void {
+            // FIXME: What do do about unused bits?
+            switch (register) {
+                .Control => switch (self.count) {
+                    0 => clock.cnt.raw = (clock.cnt.raw & 0x80) | (self.buf & 0x7F), // Bit 7 read-only
+                    else => std.debug.panic("Tried to write to byte #{} of {} (hint: there's only 1 byte)", .{ self.count, register }),
+                },
+                .DateTime, .Time => log.debug("Ignoring {} write", .{register}),
+            }
+        }
+
+        /// Is true when 8 bits have been shifted into the internal buffer
+        fn finished(self: *const Writer) bool {
+            return self.i >= 8;
+        }
+
+        /// Resets the internal buffer
+        /// resets the index used to shift bits into the internal buffer
+        /// increments `count` (which keeps track of byte offsets) by one
+        fn lap(self: *Writer) void {
+            self.buf = 0;
+            self.i = 0;
+            self.count += 1;
+        }
+
+        /// Resets `Writer` to a clean state in preparation for a future write command
+        fn reset(self: *Writer) void {
+            self.buf = 0;
+            self.i = 0;
+            self.count = 0;
+        }
+    };
+
+    const Data = extern union {
+        sck: Bit(u8, 0),
+        sio: Bit(u8, 1),
+        cs: Bit(u8, 2),
+        raw: u8,
+    };
+
+    const Control = extern union {
+        /// Unknown, value should be preserved though
+        unk: Bit(u8, 1),
+        /// Per-minute IRQ
+        /// If set, fire a Gamepak IRQ every 30s,
+        irq: Bit(u8, 3),
+        /// 12/24 Hour Bit
+        /// If set, 12h mode
+        /// If cleared, 24h mode
+        mode: Bit(u8, 6),
+        /// Read-Only, bit cleared on read
+        /// If is set, means that there has been a failure / time has been lost
+        off: Bit(u8, 7),
+        raw: u8,
+    };
+
+    fn init(ptr: *Self, cpu: *Arm7tdmi, gpio: *const Gpio) void {
+        ptr.* = .{
+            .writer = .{ .buf = 0, .i = 0, .count = 0 },
+            .reader = .{ .i = 0, .count = 0 },
+            .state = .Idle,
+            .cnt = .{ .raw = 0 },
+            .year = 0x01,
+            .month = 0x6,
+            .day = 0x13,
+            .weekday = 0x3,
+            .hour = 0x23,
+            .minute = 0x59,
+            .second = 0x59,
+            .cpu = cpu,
+            .gpio = gpio, // Can't use Arm7tdmi ptr b/c not initialized yet
+        };
+
+        cpu.sched.push(.RealTimeClock, 1 << 24); // Every Second
+    }
+
+    pub fn onClockUpdate(self: *Self, late: u64) void {
+        self.cpu.sched.push(.RealTimeClock, (1 << 24) -| late); // Reschedule
+
+        const now = DateTime.now();
+        self.year = bcd(@intCast(u8, now.date.year - 2000));
+        self.month = @truncate(u5, bcd(now.date.month));
+        self.day = @truncate(u6, bcd(now.date.day));
+        self.weekday = @truncate(u3, bcd((now.date.weekday() + 1) % 7)); // API is Monday = 0, Sunday = 6. We want Sunday = 0, Saturday = 6
+        self.hour = @truncate(u6, bcd(now.time.hour));
+        self.minute = @truncate(u7, bcd(now.time.minute));
+        self.second = @truncate(u7, bcd(now.time.second));
+    }
+
+    fn step(self: *Self, value: Data) u4 {
+        const cache: Data = .{ .raw = self.gpio.data };
+
+        return switch (self.state) {
+            .Idle => blk: {
+                // FIXME: Maybe check incoming value to see if SCK is also high?
+                if (cache.sck.read()) {
+                    if (!cache.cs.read() and value.cs.read()) {
+                        log.debug("Entering Command Mode", .{});
+                        self.state = .Command;
+                    }
+                }
+
+                break :blk @truncate(u4, value.raw);
+            },
+            .Command => blk: {
+                if (!value.cs.read()) log.err("Expected CS to be set during {}, however CS was cleared", .{self.state});
+
+                // If SCK rises, sample SIO
+                if (!cache.sck.read() and value.sck.read()) {
+                    self.writer.push(@boolToInt(value.sio.read()));
+
+                    if (self.writer.finished()) {
+                        self.state = self.processCommand(self.writer.buf);
+                        self.writer.reset();
+
+                        log.debug("Switching to {}", .{self.state});
+                    }
+                }
+
+                break :blk @truncate(u4, value.raw);
+            },
+            .Write => |register| blk: {
+                if (!value.cs.read()) log.err("Expected CS to be set during {}, however CS was cleared", .{self.state});
+
+                // If SCK rises, sample SIO
+                if (!cache.sck.read() and value.sck.read()) {
+                    self.writer.push(@boolToInt(value.sio.read()));
+
+                    const register_width: u32 = switch (register) {
+                        .Control => 1,
+                        .DateTime => 7,
+                        .Time => 3,
+                    };
+
+                    if (self.writer.finished()) {
+                        self.writer.write(self, register); // write inner buffer to RTC register
+                        self.writer.lap();
+
+                        if (self.writer.count == register_width) {
+                            self.writer.reset();
+                            self.state = .Idle;
+                        }
+                    }
+                }
+
+                break :blk @truncate(u4, value.raw);
+            },
+            .Read => |register| blk: {
+                if (!value.cs.read()) log.err("Expected CS to be set during {}, however CS was cleared", .{self.state});
+                var ret = value;
+
+                // if SCK rises, sample SIO
+                if (!cache.sck.read() and value.sck.read()) {
+                    ret.sio.write(self.reader.read(self, register) == 0b1);
+
+                    const register_width: u32 = switch (register) {
+                        .Control => 1,
+                        .DateTime => 7,
+                        .Time => 3,
+                    };
+
+                    if (self.reader.finished()) {
+                        self.reader.lap();
+
+                        if (self.reader.count == register_width) {
+                            self.reader.reset();
+                            self.state = .Idle;
+                        }
+                    }
+                }
+
+                break :blk @truncate(u4, ret.raw);
+            },
+        };
+    }
+
+    fn reset(self: *Self) void {
+        // mGBA and NBA only zero the control register. We will do the same
+        log.debug("Reset (control register was zeroed)", .{});
+
+        self.cnt.raw = 0;
+    }
+
+    fn irq(self: *Self) void {
+        // TODO: Confirm that this is the right behaviour
+        log.debug("Force GamePak IRQ", .{});
+
+        self.cpu.bus.io.irq.game_pak.set();
+        self.cpu.handleInterrupt();
+    }
+
+    fn processCommand(self: *Self, raw_command: u8) State {
+        const command = blk: {
+            // If High Nybble is 0x6, no need to switch the endianness
+            if (raw_command >> 4 & 0xF == 0x6) break :blk raw_command;
+
+            // Turns out reversing the order of bits isn't trivial at all
+            // https://stackoverflow.com/questions/2602823/in-c-c-whats-the-simplest-way-to-reverse-the-order-of-bits-in-a-byte
+            var ret = raw_command;
+            ret = (ret & 0xF0) >> 4 | (ret & 0x0F) << 4;
+            ret = (ret & 0xCC) >> 2 | (ret & 0x33) << 2;
+            ret = (ret & 0xAA) >> 1 | (ret & 0x55) << 1;
+
+            break :blk ret;
+        };
+        log.debug("Handling Command 0x{X:0>2} [0b{b:0>8}]", .{ command, command });
+
+        const is_write = command & 1 == 0;
+        const rtc_register = @truncate(u3, command >> 1 & 0x7);
+
+        if (is_write) {
+            return switch (rtc_register) {
+                0 => blk: {
+                    self.reset();
+                    break :blk .Idle;
+                },
+                1 => .{ .Write = .Control },
+                2 => .{ .Write = .DateTime },
+                3 => .{ .Write = .Time },
+                6 => blk: {
+                    self.irq();
+                    break :blk .Idle;
+                },
+                4, 5, 7 => .Idle,
+            };
+        } else {
+            return switch (rtc_register) {
+                1 => .{ .Read = .Control },
+                2 => .{ .Read = .DateTime },
+                3 => .{ .Read = .Time },
+                0, 4, 5, 6, 7 => .Idle, // Do Nothing
+            };
+        }
+    }
+};
+
+/// Converts an 8-bit unsigned integer to its BCD representation.
+/// Note: Algorithm only works for values between 0 and 99 inclusive.
+fn bcd(value: u8) u8 {
+    return ((value / 10) << 4) + (value % 10);
+}

src/core/bus/io.zig

@@ -0,0 +1,675 @@
+const std = @import("std");
+const timer = @import("timer.zig");
+const dma = @import("dma.zig");
+const apu = @import("../apu.zig");
+const ppu = @import("../ppu.zig");
+const util = @import("../../util.zig");
+
+const Bit = @import("bitfield").Bit;
+const Bitfield = @import("bitfield").Bitfield;
+const Bus = @import("../Bus.zig");
+
+const getHalf = util.getHalf;
+const setHalf = util.setHalf;
+
+const log = std.log.scoped(.@"I/O");
+
+pub const Io = struct {
+    const Self = @This();
+
+    /// Read / Write
+    ime: bool,
+    ie: InterruptEnable,
+    irq: InterruptRequest,
+    postflg: PostFlag,
+    waitcnt: WaitControl,
+    haltcnt: HaltControl,
+    keyinput: AtomicKeyInput,
+
+    pub fn init() Self {
+        return .{
+            .ime = false,
+            .ie = .{ .raw = 0x0000 },
+            .irq = .{ .raw = 0x0000 },
+            .keyinput = AtomicKeyInput.init(.{ .raw = 0x03FF }),
+            .waitcnt = .{ .raw = 0x0000_0000 }, // Bit 15 == 0 for GBA
+            .postflg = .FirstBoot,
+            .haltcnt = .Execute,
+        };
+    }
+
+    fn setIrqs(self: *Io, word: u32) void {
+        self.ie.raw = @truncate(u16, word);
+        self.irq.raw &= ~@truncate(u16, word >> 16);
+    }
+};
+
+pub fn read(bus: *const Bus, comptime T: type, address: u32) ?T {
+    return switch (T) {
+        u32 => switch (address) {
+            // Display
+            0x0400_0000...0x0400_0054 => ppu.read(T, &bus.ppu, address),
+
+            // Sound
+            0x0400_0060...0x0400_00A4 => apu.read(T, &bus.apu, address),
+
+            // DMA Transfers
+            0x0400_00B0...0x0400_00DC => dma.read(T, &bus.dma, address),
+
+            // Timers
+            0x0400_0100...0x0400_010C => timer.read(T, &bus.tim, address),
+
+            // Serial Communication 1
+            0x0400_0128 => util.io.read.todo(log, "Read {} from SIOCNT and SIOMLT_SEND", .{T}),
+
+            // Keypad Input
+            0x0400_0130 => util.io.read.todo(log, "Read {} from KEYINPUT", .{T}),
+
+            // Serial Communication 2
+            0x0400_0150 => util.io.read.todo(log, "Read {} from JOY_RECV", .{T}),
+
+            // Interrupts
+            0x0400_0200 => @as(u32, bus.io.irq.raw) << 16 | bus.io.ie.raw,
+            0x0400_0204 => bus.io.waitcnt.raw,
+            0x0400_0208 => @boolToInt(bus.io.ime),
+            0x0400_0300 => @enumToInt(bus.io.postflg),
+            else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, address }),
+        },
+        u16 => switch (address) {
+            // Display
+            0x0400_0000...0x0400_0054 => ppu.read(T, &bus.ppu, address),
+
+            // Sound
+            0x0400_0060...0x0400_00A6 => apu.read(T, &bus.apu, address),
+
+            // DMA Transfers
+            0x0400_00B0...0x0400_00DE => dma.read(T, &bus.dma, address),
+
+            // Timers
+            0x0400_0100...0x0400_010E => timer.read(T, &bus.tim, address),
+
+            // Serial Communication 1
+            0x0400_0128 => util.io.read.todo(log, "Read {} from SIOCNT", .{T}),
+
+            // Keypad Input
+            0x0400_0130 => bus.io.keyinput.load(.Monotonic).raw,
+
+            // Serial Communication 2
+            0x0400_0134 => util.io.read.todo(log, "Read {} from RCNT", .{T}),
+            0x0400_0136 => 0x0000,
+            0x0400_0142 => 0x0000,
+            0x0400_015A => 0x0000,
+
+            // Interrupts
+            0x0400_0200 => bus.io.ie.raw,
+            0x0400_0202 => bus.io.irq.raw,
+            0x0400_0204 => bus.io.waitcnt.raw,
+            0x0400_0206 => 0x0000,
+            0x0400_0208 => @boolToInt(bus.io.ime),
+            0x0400_020A => 0x0000,
+            0x0400_0300 => @enumToInt(bus.io.postflg),
+            0x0400_0302 => 0x0000,
+            else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, address }),
+        },
+        u8 => return switch (address) {
+            // Display
+            0x0400_0000...0x0400_0055 => ppu.read(T, &bus.ppu, address),
+
+            // Sound
+            0x0400_0060...0x0400_00A7 => apu.read(T, &bus.apu, address),
+
+            // DMA Transfers
+            0x0400_00B0...0x0400_00DF => dma.read(T, &bus.dma, address),
+
+            // Timers
+            0x0400_0100...0x0400_010F => timer.read(T, &bus.tim, address),
+
+            // Serial Communication 1
+            0x0400_0128 => util.io.read.todo(log, "Read {} from SIOCNT_L", .{T}),
+
+            // Keypad Input
+            0x0400_0130 => util.io.read.todo(log, "read {} from KEYINPUT_L", .{T}),
+
+            // Serial Communication 2
+            0x0400_0135 => util.io.read.todo(log, "Read {} from RCNT_H", .{T}),
+            0x0400_0136, 0x0400_0137 => 0x00,
+            0x0400_0142, 0x0400_0143 => 0x00,
+            0x0400_015A, 0x0400_015B => 0x00,
+
+            // Interrupts
+            0x0400_0200, 0x0400_0201 => @truncate(T, bus.io.ie.raw >> getHalf(@truncate(u8, address))),
+            0x0400_0202, 0x0400_0203 => @truncate(T, bus.io.irq.raw >> getHalf(@truncate(u8, address))),
+            0x0400_0204, 0x0400_0205 => @truncate(T, bus.io.waitcnt.raw >> getHalf(@truncate(u8, address))),
+            0x0400_0206, 0x0400_0207 => 0x00,
+            0x0400_0208, 0x0400_0209 => @truncate(T, @as(u16, @boolToInt(bus.io.ime)) >> getHalf(@truncate(u8, address))),
+            0x0400_020A, 0x0400_020B => 0x00,
+            0x0400_0300 => @enumToInt(bus.io.postflg),
+            0x0400_0301 => null,
+            0x0400_0302, 0x0400_0303 => 0x00,
+            else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, address }),
+        },
+        else => @compileError("I/O: Unsupported read width"),
+    };
+}
+
+pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
+    return switch (T) {
+        u32 => switch (address) {
+            // Display
+            0x0400_0000...0x0400_0054 => ppu.write(T, &bus.ppu, address, value),
+            0x0400_0058...0x0400_005C => {}, // Unused
+
+            // Sound
+            0x0400_0060...0x0400_00A4 => apu.write(T, &bus.apu, address, value),
+            0x0400_00A8, 0x0400_00AC => {}, // Unused
+
+            // DMA Transfers
+            0x0400_00B0...0x0400_00DC => dma.write(T, &bus.dma, address, value),
+            0x0400_00E0...0x0400_00FC => {}, // Unused
+
+            // Timers
+            0x0400_0100...0x0400_010C => timer.write(T, &bus.tim, address, value),
+            0x0400_0110...0x0400_011C => {}, // Unused
+
+            // Serial Communication 1
+            0x0400_0120 => log.debug("Wrote 0x{X:0>8} to SIODATA32/(SIOMULTI0 and SIOMULTI1)", .{value}),
+            0x0400_0124 => log.debug("Wrote 0x{X:0>8} to SIOMULTI2 and SIOMULTI3", .{value}),
+            0x0400_0128 => log.debug("Wrote 0x{X:0>8} to SIOCNT and SIOMLT_SEND/SIODATA8", .{value}),
+            0x0400_012C => {}, // Unused
+
+            // Keypad Input
+            0x0400_0130 => log.debug("Wrote 0x{X:0>8} to KEYINPUT and KEYCNT", .{value}),
+            0x0400_0134 => log.debug("Wrote 0x{X:0>8} to RCNT and IR", .{value}),
+            0x0400_0138, 0x0400_013C => {}, // Unused
+
+            // Serial Communication 2
+            0x0400_0140 => log.debug("Wrote 0x{X:0>8} to JOYCNT", .{value}),
+            0x0400_0150 => log.debug("Wrote 0x{X:0>8} to JOY_RECV", .{value}),
+            0x0400_0154 => log.debug("Wrote 0x{X:0>8} to JOY_TRANS", .{value}),
+            0x0400_0158 => log.debug("Wrote 0x{X:0>8} to JOYSTAT (?)", .{value}),
+            0x0400_0144...0x0400_014C, 0x0400_015C => {}, // Unused
+            0x0400_0160...0x0400_01FC => {},
+
+            // Interrupts
+            0x0400_0200 => bus.io.setIrqs(value),
+            0x0400_0204 => bus.io.waitcnt.set(@truncate(u16, value)),
+            0x0400_0208 => bus.io.ime = value & 1 == 1,
+            0x0400_0300 => {
+                bus.io.postflg = @intToEnum(PostFlag, value & 1);
+                bus.io.haltcnt = if (value >> 15 & 1 == 0) .Halt else @panic("TODO: Implement STOP");
+            },
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, address }),
+        },
+        u16 => switch (address) {
+            // Display
+            0x0400_0000...0x0400_0054 => ppu.write(T, &bus.ppu, address, value),
+            0x0400_0056 => {}, // Not used
+
+            // Sound
+            0x0400_0060...0x0400_00A6 => apu.write(T, &bus.apu, address, value),
+
+            // Dma Transfers
+            0x0400_00B0...0x0400_00DE => dma.write(T, &bus.dma, address, value),
+
+            // Timers
+            0x0400_0100...0x0400_010E => timer.write(T, &bus.tim, address, value),
+            0x0400_0114 => {},
+            0x0400_0110 => {}, // Not Used,
+
+            // Serial Communication 1
+            0x0400_0120 => log.debug("Wrote 0x{X:0>4} to SIOMULTI0", .{value}),
+            0x0400_0122 => log.debug("Wrote 0x{X:0>4} to SIOMULTI1", .{value}),
+            0x0400_0124 => log.debug("Wrote 0x{X:0>4} to SIOMULTI2", .{value}),
+            0x0400_0126 => log.debug("Wrote 0x{X:0>4} to SIOMULTI3", .{value}),
+            0x0400_0128 => log.debug("Wrote 0x{X:0>4} to SIOCNT", .{value}),
+            0x0400_012A => log.debug("Wrote 0x{X:0>4} to SIOMLT_SEND", .{value}),
+
+            // Keypad Input
+            0x0400_0130 => log.debug("Wrote 0x{X:0>4} to KEYINPUT. Ignored", .{value}),
+            0x0400_0132 => log.debug("Wrote 0x{X:0>4} to KEYCNT", .{value}),
+
+            // Serial Communication 2
+            0x0400_0134 => log.debug("Wrote 0x{X:0>4} to RCNT", .{value}),
+            0x0400_0140 => log.debug("Wrote 0x{X:0>4} to JOYCNT", .{value}),
+            0x0400_0158 => log.debug("Wrote 0x{X:0>4} to JOYSTAT", .{value}),
+            0x0400_0142, 0x0400_015A => {}, // Not Used
+
+            // Interrupts
+            0x0400_0200 => bus.io.ie.raw = value,
+            0x0400_0202 => bus.io.irq.raw &= ~value,
+            0x0400_0204 => bus.io.waitcnt.set(value),
+            0x0400_0206 => {},
+            0x0400_0208 => bus.io.ime = value & 1 == 1,
+            0x0400_020A => {},
+            0x0400_0300 => {
+                bus.io.postflg = @intToEnum(PostFlag, value & 1);
+                bus.io.haltcnt = if (value >> 15 & 1 == 0) .Halt else @panic("TODO: Implement STOP");
+            },
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, address }),
+        },
+        u8 => switch (address) {
+            // Display
+            0x0400_0000...0x0400_0055 => ppu.write(T, &bus.ppu, address, value),
+
+            // Sound
+            0x0400_0060...0x0400_00A7 => apu.write(T, &bus.apu, address, value),
+
+            // Dma Transfers
+            0x0400_00B0...0x0400_00DF => dma.write(T, &bus.dma, address, value),
+
+            // Timers
+            0x0400_0100...0x0400_010F => timer.write(T, &bus.tim, address, value),
+
+            // Serial Communication 1
+            0x0400_0120 => log.debug("Wrote 0x{X:0>2} to SIODATA32_L_L", .{value}),
+            0x0400_0128 => log.debug("Wrote 0x{X:0>2} to SIOCNT_L", .{value}),
+
+            // Serial Communication 2
+            0x0400_0135 => log.debug("Wrote 0x{X:0>2} to RCNT_H", .{value}),
+            0x0400_0140 => log.debug("Wrote 0x{X:0>2} to JOYCNT_L", .{value}),
+
+            // Interrupts
+            0x0400_0200, 0x0400_0201 => bus.io.ie.raw = setHalf(u16, bus.io.ie.raw, @truncate(u8, address), value),
+            0x0400_0202 => bus.io.irq.raw &= ~@as(u16, value),
+            0x0400_0203 => bus.io.irq.raw &= ~@as(u16, value) << 8, // TODO: Is this good?
+            0x0400_0204, 0x0400_0205 => bus.io.waitcnt.set(setHalf(u16, @truncate(u16, bus.io.waitcnt.raw), @truncate(u8, address), value)),
+            0x0400_0206, 0x0400_0207 => {},
+            0x0400_0208 => bus.io.ime = value & 1 == 1,
+            0x0400_0209 => {},
+            0x0400_020A, 0x0400_020B => {},
+
+            0x0400_0300 => bus.io.postflg = @intToEnum(PostFlag, value & 1),
+            0x0400_0301 => bus.io.haltcnt = if (value >> 7 & 1 == 0) .Halt else std.debug.panic("TODO: Implement STOP", .{}),
+
+            0x0400_0410 => log.debug("Wrote 0x{X:0>2} to the common yet undocumented 0x{X:0>8}", .{ value, address }),
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, address }),
+        },
+        else => @compileError("I/O: Unsupported write width"),
+    };
+}
+
+/// Read / Write
+pub const PostFlag = enum(u1) {
+    FirstBoot = 0,
+    FurtherBoots = 1,
+};
+
+/// Write Only
+pub const HaltControl = enum {
+    Halt,
+    Stop,
+    Execute,
+};
+
+/// Read / Write
+pub const DisplayControl = extern union {
+    bg_mode: Bitfield(u16, 0, 3),
+    frame_select: Bit(u16, 4),
+    hblank_interval_free: Bit(u16, 5),
+    obj_mapping: Bit(u16, 6),
+    forced_blank: Bit(u16, 7),
+    bg_enable: Bitfield(u16, 8, 4),
+    obj_enable: Bit(u16, 12),
+    win_enable: Bitfield(u16, 13, 2),
+    obj_win_enable: Bit(u16, 15),
+    raw: u16,
+};
+
+/// Read / Write
+pub const DisplayStatus = extern union {
+    /// read-only
+    vblank: Bit(u16, 0),
+    /// read-only
+    hblank: Bit(u16, 1),
+    // read-only
+    coincidence: Bit(u16, 2),
+    vblank_irq: Bit(u16, 3),
+    hblank_irq: Bit(u16, 4),
+    vcount_irq: Bit(u16, 5),
+    vcount_trigger: Bitfield(u16, 8, 8),
+    raw: u16,
+
+    pub fn set(self: *DisplayStatus, value: u16) void {
+        const mask: u16 = 0x00C7; // set bits are read-only
+        self.raw = (self.raw & mask) | (value & ~mask);
+    }
+};
+
+/// Read Only
+pub const VCount = extern union {
+    scanline: Bitfield(u16, 0, 8),
+    raw: u16,
+};
+
+/// Read / Write
+const InterruptEnable = extern union {
+    vblank: Bit(u16, 0),
+    hblank: Bit(u16, 1),
+    coincidence: Bit(u16, 2),
+    tm0_overflow: Bit(u16, 3),
+    tm1_overflow: Bit(u16, 4),
+    tm2_overflow: Bit(u16, 5),
+    tm3_overflow: Bit(u16, 6),
+    serial: Bit(u16, 7),
+    dma0: Bit(u16, 8),
+    dma1: Bit(u16, 9),
+    dma2: Bit(u16, 10),
+    dma3: Bit(u16, 11),
+    keypad: Bit(u16, 12),
+    game_pak: Bit(u16, 13),
+    raw: u16,
+};
+
+/// Read Only
+/// 0 = Pressed, 1 = Released
+const KeyInput = extern union {
+    a: Bit(u16, 0),
+    b: Bit(u16, 1),
+    select: Bit(u16, 2),
+    start: Bit(u16, 3),
+    right: Bit(u16, 4),
+    left: Bit(u16, 5),
+    up: Bit(u16, 6),
+    down: Bit(u16, 7),
+    shoulder_r: Bit(u16, 8),
+    shoulder_l: Bit(u16, 9),
+    raw: u16,
+};
+
+const AtomicKeyInput = struct {
+    const Self = @This();
+    const Ordering = std.atomic.Ordering;
+
+    inner: KeyInput,
+
+    pub fn init(value: KeyInput) Self {
+        return .{ .inner = value };
+    }
+
+    pub inline fn load(self: *const Self, comptime ordering: Ordering) KeyInput {
+        return .{ .raw = switch (ordering) {
+            .AcqRel, .Release => @compileError("not supported for atomic loads"),
+            else => @atomicLoad(u16, &self.inner.raw, ordering),
+        } };
+    }
+
+    pub inline fn store(self: *Self, value: u16, comptime ordering: Ordering) void {
+        switch (ordering) {
+            .AcqRel, .Acquire => @compileError("not supported for atomic stores"),
+            else => @atomicStore(u16, &self.inner.raw, value, ordering),
+        }
+    }
+};
+
+// Read / Write
+pub const BackgroundControl = extern union {
+    priority: Bitfield(u16, 0, 2),
+    char_base: Bitfield(u16, 2, 2),
+    mosaic_enable: Bit(u16, 6),
+    colour_mode: Bit(u16, 7),
+    screen_base: Bitfield(u16, 8, 5),
+    display_overflow: Bit(u16, 13),
+    size: Bitfield(u16, 14, 2),
+    raw: u16,
+};
+
+/// Write Only
+pub const BackgroundOffset = extern union {
+    offset: Bitfield(u16, 0, 9),
+    raw: u16,
+};
+
+/// Read / Write
+pub const BldCnt = extern union {
+    /// BLDCNT{0} is BG0 A
+    /// BLDCNT{4} is OBJ A
+    /// BLDCNT{5} is BD  A
+    layer_a: Bitfield(u16, 0, 6),
+    mode: Bitfield(u16, 6, 2),
+
+    /// BLDCNT{8} is BG0 B
+    /// BLDCNT{12} is OBJ B
+    /// BLDCNT{13} is BD  B
+    layer_b: Bitfield(u16, 8, 6),
+    raw: u16,
+};
+
+/// Read-only?
+/// Alpha Blending Coefficients
+pub const BldAlpha = extern union {
+    eva: Bitfield(u16, 0, 5),
+    evb: Bitfield(u16, 8, 5),
+    raw: u16,
+};
+
+/// Write-only?
+/// Brightness COefficients
+pub const BldY = extern union {
+    evy: Bitfield(u16, 0, 5),
+    raw: u16,
+};
+
+const u8WriteKind = enum { Hi, Lo };
+
+/// Write-only
+pub const WinH = extern union {
+    x2: Bitfield(u16, 0, 8),
+    x1: Bitfield(u16, 8, 8),
+    raw: u16,
+};
+
+/// Write-only
+pub const WinV = extern union {
+    const Self = @This();
+
+    y2: Bitfield(u16, 0, 8),
+    y1: Bitfield(u16, 8, 8),
+    raw: u16,
+};
+
+pub const WinIn = extern union {
+    w0_bg: Bitfield(u16, 0, 4),
+    w0_obj: Bit(u16, 4),
+    w0_bld: Bit(u16, 5),
+    w1_bg: Bitfield(u16, 8, 4),
+    w1_obj: Bit(u16, 12),
+    w1_bld: Bit(u16, 13),
+    raw: u16,
+};
+
+pub const WinOut = extern union {
+    out_bg: Bitfield(u16, 0, 4),
+    out_obj: Bit(u16, 4),
+    out_bld: Bit(u16, 5),
+    obj_bg: Bitfield(u16, 8, 4),
+    obj_obj: Bit(u16, 12),
+    obj_bld: Bit(u16, 13),
+    raw: u16,
+};
+
+/// Read / Write
+const InterruptRequest = extern union {
+    vblank: Bit(u16, 0),
+    hblank: Bit(u16, 1),
+    coincidence: Bit(u16, 2),
+    tim0: Bit(u16, 3),
+    tim1: Bit(u16, 4),
+    tim2: Bit(u16, 5),
+    tim3: Bit(u16, 6),
+    serial: Bit(u16, 7),
+    dma0: Bit(u16, 8),
+    dma1: Bit(u16, 9),
+    dma2: Bit(u16, 10),
+    dma3: Bit(u16, 11),
+    keypad: Bit(u16, 12),
+    game_pak: Bit(u16, 13),
+    raw: u16,
+};
+
+/// Read / Write
+pub const DmaControl = extern union {
+    dad_adj: Bitfield(u16, 5, 2),
+    sad_adj: Bitfield(u16, 7, 2),
+    repeat: Bit(u16, 9),
+    transfer_type: Bit(u16, 10),
+    pak_drq: Bit(u16, 11),
+    start_timing: Bitfield(u16, 12, 2),
+    irq: Bit(u16, 14),
+    enabled: Bit(u16, 15),
+    raw: u16,
+};
+
+/// Read / Write
+pub const TimerControl = extern union {
+    frequency: Bitfield(u16, 0, 2),
+    cascade: Bit(u16, 2),
+    irq: Bit(u16, 6),
+    enabled: Bit(u16, 7),
+    raw: u16,
+};
+
+/// Read / Write
+/// NR10
+pub const Sweep = extern union {
+    shift: Bitfield(u8, 0, 3),
+    direction: Bit(u8, 3),
+    period: Bitfield(u8, 4, 3),
+    raw: u8,
+};
+
+/// Read / Write
+/// This represents the Duty / Len
+/// NRx1
+pub const Duty = extern union {
+    /// Write-only
+    /// Only used when bit 6 is set
+    length: Bitfield(u16, 0, 6),
+    pattern: Bitfield(u16, 6, 2),
+    raw: u8,
+};
+
+/// Read / Write
+/// NRx2
+pub const Envelope = extern union {
+    period: Bitfield(u8, 0, 3),
+    direction: Bit(u8, 3),
+    init_vol: Bitfield(u8, 4, 4),
+    raw: u8,
+};
+
+/// Read / Write
+/// NRx3, NRx4
+pub const Frequency = extern union {
+    /// Write-only
+    frequency: Bitfield(u16, 0, 11),
+    length_enable: Bit(u16, 14),
+    /// Write-only
+    trigger: Bit(u16, 15),
+
+    raw: u16,
+};
+
+/// Read / Write
+/// NR30
+pub const WaveSelect = extern union {
+    dimension: Bit(u8, 5),
+    bank: Bit(u8, 6),
+    enabled: Bit(u8, 7),
+    raw: u8,
+};
+
+/// Read / Write
+/// NR32
+pub const WaveVolume = extern union {
+    kind: Bitfield(u8, 5, 2),
+    force: Bit(u8, 7),
+    raw: u8,
+};
+
+/// Read / Write
+/// NR43
+pub const PolyCounter = extern union {
+    div_ratio: Bitfield(u8, 0, 3),
+    width: Bit(u8, 3),
+    shift: Bitfield(u8, 4, 4),
+    raw: u8,
+};
+
+/// Read / Write
+/// NR44
+pub const NoiseControl = extern union {
+    length_enable: Bit(u8, 6),
+    trigger: Bit(u8, 7),
+    raw: u8,
+};
+
+/// Read / Write
+pub const ChannelVolumeControl = extern union {
+    right_vol: Bitfield(u16, 0, 3),
+    left_vol: Bitfield(u16, 4, 3),
+    ch_right: Bitfield(u16, 8, 4),
+    ch_left: Bitfield(u16, 12, 4),
+    raw: u16,
+};
+
+/// Read / Write
+pub const DmaSoundControl = extern union {
+    ch_vol: Bitfield(u16, 0, 2),
+    chA_vol: Bit(u16, 2),
+    chB_vol: Bit(u16, 3),
+
+    chA_right: Bit(u16, 8),
+    chA_left: Bit(u16, 9),
+    chA_timer: Bit(u16, 10),
+    /// Write only?
+    chA_reset: Bit(u16, 11),
+
+    chB_right: Bit(u16, 12),
+    chB_left: Bit(u16, 13),
+    chB_timer: Bit(u16, 14),
+    /// Write only?
+    chB_reset: Bit(u16, 15),
+    raw: u16,
+};
+
+/// Read / Write
+pub const SoundControl = extern union {
+    /// Read-only
+    ch1_enable: Bit(u8, 0),
+    /// Read-only
+    ch2_enable: Bit(u8, 1),
+    /// Read-only
+    ch3_enable: Bit(u8, 2),
+    /// Read-only
+    ch4_enable: Bit(u8, 3),
+    apu_enable: Bit(u8, 7),
+    raw: u8,
+};
+
+/// Read / Write
+pub const SoundBias = extern union {
+    level: Bitfield(u16, 1, 9),
+    sampling_cycle: Bitfield(u16, 14, 2),
+    raw: u16,
+};
+
+/// Read / Write
+pub const WaitControl = extern union {
+    sram_cnt: Bitfield(u16, 0, 2),
+    s0_first: Bitfield(u16, 2, 2),
+    s0_second: Bit(u16, 4),
+    s1_first: Bitfield(u16, 5, 2),
+    s1_second: Bit(u16, 7),
+    s2_first: Bitfield(u16, 8, 2),
+    s2_second: Bit(u16, 10),
+    phi_out: Bitfield(u16, 11, 2),
+
+    prefetch_enable: Bit(u16, 14),
+    pak_kind: Bit(u16, 15),
+    raw: u16,
+
+    pub fn set(self: *WaitControl, value: u16) void {
+        const mask: u16 = 0x8000; // set bits are read-only
+        self.raw = (self.raw & mask) | (value & ~mask);
+    }
+};

src/core/bus/timer.zig

@@ -0,0 +1,243 @@
+const std = @import("std");
+const util = @import("../../util.zig");
+
+const TimerControl = @import("io.zig").TimerControl;
+const Scheduler = @import("../scheduler.zig").Scheduler;
+const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
+
+pub const TimerTuple = struct { Timer(0), Timer(1), Timer(2), Timer(3) };
+const log = std.log.scoped(.Timer);
+
+const getHalf = util.getHalf;
+const setHalf = util.setHalf;
+
+pub fn create(sched: *Scheduler) TimerTuple {
+    return .{ Timer(0).init(sched), Timer(1).init(sched), Timer(2).init(sched), Timer(3).init(sched) };
+}
+
+pub fn read(comptime T: type, tim: *const TimerTuple, addr: u32) ?T {
+    const nybble_addr = @truncate(u4, addr);
+
+    return switch (T) {
+        u32 => switch (nybble_addr) {
+            0x0 => @as(T, tim.*[0].cnt.raw) << 16 | tim.*[0].timcntL(),
+            0x4 => @as(T, tim.*[1].cnt.raw) << 16 | tim.*[1].timcntL(),
+            0x8 => @as(T, tim.*[2].cnt.raw) << 16 | tim.*[2].timcntL(),
+            0xC => @as(T, tim.*[3].cnt.raw) << 16 | tim.*[3].timcntL(),
+            else => util.io.read.err(T, log, "unaligned {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        u16 => switch (nybble_addr) {
+            0x0 => tim.*[0].timcntL(),
+            0x2 => tim.*[0].cnt.raw,
+
+            0x4 => tim.*[1].timcntL(),
+            0x6 => tim.*[1].cnt.raw,
+
+            0x8 => tim.*[2].timcntL(),
+            0xA => tim.*[2].cnt.raw,
+
+            0xC => tim.*[3].timcntL(),
+            0xE => tim.*[3].cnt.raw,
+            else => util.io.read.err(T, log, "unaligned {} read from 0x{X:0>8}", .{ T, addr }),
+        },
+        u8 => switch (nybble_addr) {
+            0x0, 0x1 => @truncate(T, tim.*[0].timcntL() >> getHalf(nybble_addr)),
+            0x2, 0x3 => @truncate(T, tim.*[0].cnt.raw >> getHalf(nybble_addr)),
+
+            0x4, 0x5 => @truncate(T, tim.*[1].timcntL() >> getHalf(nybble_addr)),
+            0x6, 0x7 => @truncate(T, tim.*[1].cnt.raw >> getHalf(nybble_addr)),
+
+            0x8, 0x9 => @truncate(T, tim.*[2].timcntL() >> getHalf(nybble_addr)),
+            0xA, 0xB => @truncate(T, tim.*[2].cnt.raw >> getHalf(nybble_addr)),
+
+            0xC, 0xD => @truncate(T, tim.*[3].timcntL() >> getHalf(nybble_addr)),
+            0xE, 0xF => @truncate(T, tim.*[3].cnt.raw >> getHalf(nybble_addr)),
+        },
+        else => @compileError("TIM: Unsupported read width"),
+    };
+}
+
+pub fn write(comptime T: type, tim: *TimerTuple, addr: u32, value: T) void {
+    const nybble_addr = @truncate(u4, addr);
+
+    return switch (T) {
+        u32 => switch (nybble_addr) {
+            0x0 => tim.*[0].setTimcnt(value),
+            0x4 => tim.*[1].setTimcnt(value),
+            0x8 => tim.*[2].setTimcnt(value),
+            0xC => tim.*[3].setTimcnt(value),
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, addr }),
+        },
+        u16 => switch (nybble_addr) {
+            0x0 => tim.*[0].setTimcntL(value),
+            0x2 => tim.*[0].setTimcntH(value),
+
+            0x4 => tim.*[1].setTimcntL(value),
+            0x6 => tim.*[1].setTimcntH(value),
+
+            0x8 => tim.*[2].setTimcntL(value),
+            0xA => tim.*[2].setTimcntH(value),
+
+            0xC => tim.*[3].setTimcntL(value),
+            0xE => tim.*[3].setTimcntH(value),
+            else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, addr }),
+        },
+        u8 => switch (nybble_addr) {
+            0x0, 0x1 => tim.*[0].setTimcntL(setHalf(u16, tim.*[0]._reload, nybble_addr, value)),
+            0x2, 0x3 => tim.*[0].setTimcntH(setHalf(u16, tim.*[0].cnt.raw, nybble_addr, value)),
+
+            0x4, 0x5 => tim.*[1].setTimcntL(setHalf(u16, tim.*[1]._reload, nybble_addr, value)),
+            0x6, 0x7 => tim.*[1].setTimcntH(setHalf(u16, tim.*[1].cnt.raw, nybble_addr, value)),
+
+            0x8, 0x9 => tim.*[2].setTimcntL(setHalf(u16, tim.*[2]._reload, nybble_addr, value)),
+            0xA, 0xB => tim.*[2].setTimcntH(setHalf(u16, tim.*[2].cnt.raw, nybble_addr, value)),
+
+            0xC, 0xD => tim.*[3].setTimcntL(setHalf(u16, tim.*[3]._reload, nybble_addr, value)),
+            0xE, 0xF => tim.*[3].setTimcntH(setHalf(u16, tim.*[3].cnt.raw, nybble_addr, value)),
+        },
+        else => @compileError("TIM: Unsupported write width"),
+    };
+}
+
+fn Timer(comptime id: u2) type {
+    return struct {
+        const Self = @This();
+
+        /// Read Only, Internal. Please use self.timcntL()
+        _counter: u16,
+
+        /// Write Only, Internal. Please use self.setTimcntL()
+        _reload: u16,
+
+        /// Write Only, Internal. Please use self.setTimcntH()
+        cnt: TimerControl,
+
+        /// Internal.
+        sched: *Scheduler,
+
+        /// Internal
+        _start_timestamp: u64,
+
+        pub fn init(sched: *Scheduler) Self {
+            return .{
+                ._reload = 0,
+                ._counter = 0,
+                .cnt = .{ .raw = 0x0000 },
+                .sched = sched,
+                ._start_timestamp = 0,
+            };
+        }
+
+        /// TIMCNT_L Getter
+        pub fn timcntL(self: *const Self) u16 {
+            if (self.cnt.cascade.read() or !self.cnt.enabled.read()) return self._counter;
+
+            return self._counter +% @truncate(u16, (self.sched.now() - self._start_timestamp) / self.frequency());
+        }
+
+        /// TIMCNT_L Setter
+        pub fn setTimcntL(self: *Self, halfword: u16) void {
+            self._reload = halfword;
+        }
+
+        /// TIMCNT_L & TIMCNT_H
+        pub fn setTimcnt(self: *Self, word: u32) void {
+            self.setTimcntL(@truncate(u16, word));
+            self.setTimcntH(@truncate(u16, word >> 16));
+        }
+
+        /// TIMCNT_H
+        pub fn setTimcntH(self: *Self, halfword: u16) void {
+            const new = TimerControl{ .raw = halfword };
+
+            if (self.cnt.enabled.read()) {
+                // timer was already enabled
+
+                // If enabled falling edge or cascade falling edge, timer is paused
+                if (!new.enabled.read() or (!self.cnt.cascade.read() and new.cascade.read())) {
+                    self.sched.removeScheduledEvent(.{ .TimerOverflow = id });
+
+                    // Counter should hold the value it stopped at meaning we have to calculate it now
+                    self._counter +%= @truncate(u16, (self.sched.now() - self._start_timestamp) / self.frequency());
+                }
+
+                // the timer has always been enabled, but the cascade bit which was blocking the timer has been unset
+                if (new.enabled.read() and (self.cnt.cascade.read() and !new.cascade.read())) {
+                    // we want to reschedule the timer event, however we won't reload the counter.
+                    // the invariant here is that self._counter holds the already calculated paused value
+
+                    self.rescheduleTimerExpire(0);
+                }
+            } else {
+                // the timer was previously disabeld
+
+                if (new.enabled.read()) {
+                    // timer should start counting (with a reloaded counter value)
+                    self._counter = self._reload;
+
+                    // if cascade happens to be set, the timer doesn't actually do anything though
+                    if (!new.cascade.read()) self.rescheduleTimerExpire(0);
+                }
+            }
+
+            self.cnt.raw = halfword;
+        }
+
+        pub fn onTimerExpire(self: *Self, cpu: *Arm7tdmi, late: u64) void {
+            // Fire IRQ if enabled
+            const io = &cpu.bus.io;
+
+            if (self.cnt.irq.read()) {
+                switch (id) {
+                    0 => io.irq.tim0.set(),
+                    1 => io.irq.tim1.set(),
+                    2 => io.irq.tim2.set(),
+                    3 => io.irq.tim3.set(),
+                }
+
+                cpu.handleInterrupt();
+            }
+
+            // DMA Sound Things
+            if (id == 0 or id == 1) {
+                cpu.bus.apu.onDmaAudioSampleRequest(cpu, id);
+            }
+
+            // Perform Cascade Behaviour
+            switch (id) {
+                inline 0, 1, 2 => |idx| {
+                    const next = idx + 1;
+
+                    if (cpu.bus.tim[next].cnt.cascade.read()) {
+                        cpu.bus.tim[next]._counter +%= 1;
+                        if (cpu.bus.tim[next]._counter == 0) cpu.bus.tim[next].onTimerExpire(cpu, late);
+                    }
+                },
+                3 => {}, // THere is no timer for TIM3 to cascade to
+            }
+
+            // Reschedule Timer if we're not cascading
+            // TIM0 cascade value is N/A
+            if (id == 0 or !self.cnt.cascade.read()) {
+                self._counter = self._reload;
+                self.rescheduleTimerExpire(late);
+            }
+        }
+
+        fn rescheduleTimerExpire(self: *Self, late: u64) void {
+            const when = (@as(u64, 0x10000) - self._counter) * self.frequency();
+
+            self._start_timestamp = self.sched.now();
+            self.sched.push(.{ .TimerOverflow = id }, when -| late);
+        }
+
+        fn frequency(self: *const Self) u16 {
+            return switch (self.cnt.frequency.read()) {
+                0 => 1,
+                1 => 64,
+                2 => 256,
+                3 => 1024,
+            };
+        }
+    };
+}

src/core/cpu.zig

@@ -0,0 +1,673 @@
+const std = @import("std");
+
+const Bus = @import("Bus.zig");
+const Bit = @import("bitfield").Bit;
+const Bitfield = @import("bitfield").Bitfield;
+const Scheduler = @import("scheduler.zig").Scheduler;
+const Logger = @import("../util.zig").Logger;
+
+const File = std.fs.File;
+const log = std.log.scoped(.Arm7Tdmi);
+
+// ARM Instructions
+pub const arm = struct {
+    pub const InstrFn = *const fn (*Arm7tdmi, *Bus, u32) void;
+    const lut: [0x1000]InstrFn = populate();
+
+    const processing = @import("cpu/arm/data_processing.zig").dataProcessing;
+    const psrTransfer = @import("cpu/arm/psr_transfer.zig").psrTransfer;
+    const transfer = @import("cpu/arm/single_data_transfer.zig").singleDataTransfer;
+    const halfSignedTransfer = @import("cpu/arm/half_signed_data_transfer.zig").halfAndSignedDataTransfer;
+    const blockTransfer = @import("cpu/arm/block_data_transfer.zig").blockDataTransfer;
+    const branch = @import("cpu/arm/branch.zig").branch;
+    const branchExchange = @import("cpu/arm/branch.zig").branchAndExchange;
+    const swi = @import("cpu/arm/software_interrupt.zig").armSoftwareInterrupt;
+    const swap = @import("cpu/arm/single_data_swap.zig").singleDataSwap;
+
+    const multiply = @import("cpu/arm/multiply.zig").multiply;
+    const multiplyLong = @import("cpu/arm/multiply.zig").multiplyLong;
+
+    /// Determine index into ARM InstrFn LUT
+    fn idx(opcode: u32) u12 {
+        return @truncate(u12, opcode >> 20 & 0xFF) << 4 | @truncate(u12, opcode >> 4 & 0xF);
+    }
+
+    // Undefined ARM Instruction handler
+    fn und(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+        const id = idx(opcode);
+        cpu.panic("[CPU/Decode] ID: 0x{X:0>3} 0x{X:0>8} is an illegal opcode", .{ id, opcode });
+    }
+
+    fn populate() [0x1000]InstrFn {
+        comptime {
+            @setEvalBranchQuota(0xE000);
+            var table = [_]InstrFn{und} ** 0x1000;
+
+            for (&table, 0..) |*handler, i| {
+                handler.* = switch (@as(u2, i >> 10)) {
+                    0b00 => if (i == 0x121) blk: {
+                        break :blk branchExchange;
+                    } else if (i & 0xFCF == 0x009) blk: {
+                        const A = i >> 5 & 1 == 1;
+                        const S = i >> 4 & 1 == 1;
+                        break :blk multiply(A, S);
+                    } else if (i & 0xFBF == 0x109) blk: {
+                        const B = i >> 6 & 1 == 1;
+                        break :blk swap(B);
+                    } else if (i & 0xF8F == 0x089) blk: {
+                        const U = i >> 6 & 1 == 1;
+                        const A = i >> 5 & 1 == 1;
+                        const S = i >> 4 & 1 == 1;
+                        break :blk multiplyLong(U, A, S);
+                    } else if (i & 0xE49 == 0x009 or i & 0xE49 == 0x049) blk: {
+                        const P = i >> 8 & 1 == 1;
+                        const U = i >> 7 & 1 == 1;
+                        const I = i >> 6 & 1 == 1;
+                        const W = i >> 5 & 1 == 1;
+                        const L = i >> 4 & 1 == 1;
+                        break :blk halfSignedTransfer(P, U, I, W, L);
+                    } else if (i & 0xD90 == 0x100) blk: {
+                        const I = i >> 9 & 1 == 1;
+                        const R = i >> 6 & 1 == 1;
+                        const kind = i >> 4 & 0x3;
+                        break :blk psrTransfer(I, R, kind);
+                    } else blk: {
+                        const I = i >> 9 & 1 == 1;
+                        const S = i >> 4 & 1 == 1;
+                        const instrKind = i >> 5 & 0xF;
+                        break :blk processing(I, S, instrKind);
+                    },
+                    0b01 => if (i >> 9 & 1 == 1 and i & 1 == 1) und else blk: {
+                        const I = i >> 9 & 1 == 1;
+                        const P = i >> 8 & 1 == 1;
+                        const U = i >> 7 & 1 == 1;
+                        const B = i >> 6 & 1 == 1;
+                        const W = i >> 5 & 1 == 1;
+                        const L = i >> 4 & 1 == 1;
+                        break :blk transfer(I, P, U, B, W, L);
+                    },
+                    else => switch (@as(u2, i >> 9 & 0x3)) {
+                        // MSB is guaranteed to be 1
+                        0b00 => blk: {
+                            const P = i >> 8 & 1 == 1;
+                            const U = i >> 7 & 1 == 1;
+                            const S = i >> 6 & 1 == 1;
+                            const W = i >> 5 & 1 == 1;
+                            const L = i >> 4 & 1 == 1;
+                            break :blk blockTransfer(P, U, S, W, L);
+                        },
+                        0b01 => blk: {
+                            const L = i >> 8 & 1 == 1;
+                            break :blk branch(L);
+                        },
+                        0b10 => und, // COP Data Transfer
+                        0b11 => if (i >> 8 & 1 == 1) swi() else und, // COP Data Operation + Register Transfer
+                    },
+                };
+            }
+
+            return table;
+        }
+    }
+};
+
+// THUMB Instructions
+pub const thumb = struct {
+    pub const InstrFn = *const fn (*Arm7tdmi, *Bus, u16) void;
+    const lut: [0x400]InstrFn = populate();
+
+    const processing = @import("cpu/thumb/data_processing.zig");
+    const alu = @import("cpu/thumb/alu.zig").fmt4;
+    const transfer = @import("cpu/thumb/data_transfer.zig");
+    const block_transfer = @import("cpu/thumb/block_data_transfer.zig");
+    const swi = @import("cpu/thumb/software_interrupt.zig").fmt17;
+    const branch = @import("cpu/thumb/branch.zig");
+
+    /// Determine index into THUMB InstrFn LUT
+    fn idx(opcode: u16) u10 {
+        return @truncate(u10, opcode >> 6);
+    }
+
+    /// Undefined THUMB Instruction Handler
+    fn und(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+        const id = idx(opcode);
+        cpu.panic("[CPU/Decode] ID: 0b{b:0>10} 0x{X:0>2} is an illegal opcode", .{ id, opcode });
+    }
+
+    fn populate() [0x400]InstrFn {
+        comptime {
+            @setEvalBranchQuota(5025); // This is exact
+            var table = [_]InstrFn{und} ** 0x400;
+
+            for (&table, 0..) |*handler, i| {
+                handler.* = switch (@as(u3, i >> 7 & 0x7)) {
+                    0b000 => if (i >> 5 & 0x3 == 0b11) blk: {
+                        const I = i >> 4 & 1 == 1;
+                        const is_sub = i >> 3 & 1 == 1;
+                        const rn = i & 0x7;
+                        break :blk processing.fmt2(I, is_sub, rn);
+                    } else blk: {
+                        const op = i >> 5 & 0x3;
+                        const offset = i & 0x1F;
+                        break :blk processing.fmt1(op, offset);
+                    },
+                    0b001 => blk: {
+                        const op = i >> 5 & 0x3;
+                        const rd = i >> 2 & 0x7;
+                        break :blk processing.fmt3(op, rd);
+                    },
+                    0b010 => switch (@as(u2, i >> 5 & 0x3)) {
+                        0b00 => if (i >> 4 & 1 == 1) blk: {
+                            const op = i >> 2 & 0x3;
+                            const h1 = i >> 1 & 1;
+                            const h2 = i & 1;
+                            break :blk processing.fmt5(op, h1, h2);
+                        } else blk: {
+                            const op = i & 0xF;
+                            break :blk alu(op);
+                        },
+                        0b01 => blk: {
+                            const rd = i >> 2 & 0x7;
+                            break :blk transfer.fmt6(rd);
+                        },
+                        else => blk: {
+                            const op = i >> 4 & 0x3;
+                            const T = i >> 3 & 1 == 1;
+                            break :blk transfer.fmt78(op, T);
+                        },
+                    },
+                    0b011 => blk: {
+                        const B = i >> 6 & 1 == 1;
+                        const L = i >> 5 & 1 == 1;
+                        const offset = i & 0x1F;
+                        break :blk transfer.fmt9(B, L, offset);
+                    },
+                    else => switch (@as(u3, i >> 6 & 0x7)) {
+                        // MSB is guaranteed to be 1
+                        0b000 => blk: {
+                            const L = i >> 5 & 1 == 1;
+                            const offset = i & 0x1F;
+                            break :blk transfer.fmt10(L, offset);
+                        },
+                        0b001 => blk: {
+                            const L = i >> 5 & 1 == 1;
+                            const rd = i >> 2 & 0x7;
+                            break :blk transfer.fmt11(L, rd);
+                        },
+                        0b010 => blk: {
+                            const isSP = i >> 5 & 1 == 1;
+                            const rd = i >> 2 & 0x7;
+                            break :blk processing.fmt12(isSP, rd);
+                        },
+                        0b011 => if (i >> 4 & 1 == 1) blk: {
+                            const L = i >> 5 & 1 == 1;
+                            const R = i >> 2 & 1 == 1;
+                            break :blk block_transfer.fmt14(L, R);
+                        } else blk: {
+                            const S = i >> 1 & 1 == 1;
+                            break :blk processing.fmt13(S);
+                        },
+                        0b100 => blk: {
+                            const L = i >> 5 & 1 == 1;
+                            const rb = i >> 2 & 0x7;
+
+                            break :blk block_transfer.fmt15(L, rb);
+                        },
+                        0b101 => if (i >> 2 & 0xF == 0b1111) blk: {
+                            break :blk thumb.swi();
+                        } else blk: {
+                            const cond = i >> 2 & 0xF;
+                            break :blk branch.fmt16(cond);
+                        },
+                        0b110 => branch.fmt18(),
+                        0b111 => blk: {
+                            const is_low = i >> 5 & 1 == 1;
+                            break :blk branch.fmt19(is_low);
+                        },
+                    },
+                };
+            }
+
+            return table;
+        }
+    }
+};
+
+pub const Arm7tdmi = struct {
+    const Self = @This();
+
+    r: [16]u32,
+    pipe: Pipeline,
+    sched: *Scheduler,
+    bus: *Bus,
+    cpsr: PSR,
+    spsr: PSR,
+
+    bank: Bank,
+
+    logger: ?Logger,
+
+    /// Bank of Registers from other CPU Modes
+    const Bank = struct {
+        /// Storage for r13_<mode>, r14_<mode>
+        /// e.g. [r13, r14, r13_svc, r14_svc]
+        r: [2 * 6]u32,
+
+        /// Storage  for R8_fiq -> R12_fiq and their normal counterparts
+        /// e.g [r[0 + 8], fiq_r[0 + 8], r[1 + 8], fiq_r[1 + 8]...]
+        fiq: [2 * 5]u32,
+
+        spsr: [5]PSR,
+
+        const Kind = enum(u1) {
+            R13 = 0,
+            R14,
+        };
+
+        pub fn create() Bank {
+            return .{
+                .r = [_]u32{0x00} ** 12,
+                .fiq = [_]u32{0x00} ** 10,
+                .spsr = [_]PSR{.{ .raw = 0x0000_0000 }} ** 5,
+            };
+        }
+
+        inline fn regIdx(mode: Mode, kind: Kind) usize {
+            const idx: usize = switch (mode) {
+                .User, .System => 0,
+                .Supervisor => 1,
+                .Abort => 2,
+                .Undefined => 3,
+                .Irq => 4,
+                .Fiq => 5,
+            };
+
+            return (idx * 2) + if (kind == .R14) @as(usize, 1) else 0;
+        }
+
+        inline fn spsrIdx(mode: Mode) usize {
+            return switch (mode) {
+                .Supervisor => 0,
+                .Abort => 1,
+                .Undefined => 2,
+                .Irq => 3,
+                .Fiq => 4,
+                else => std.debug.panic("[CPU/Mode] {} does not have a SPSR Register", .{mode}),
+            };
+        }
+
+        inline fn fiqIdx(i: usize, mode: Mode) usize {
+            return (i * 2) + if (mode == .Fiq) @as(usize, 1) else 0;
+        }
+    };
+
+    pub fn init(sched: *Scheduler, bus: *Bus, log_file: ?std.fs.File) Self {
+        return Self{
+            .r = [_]u32{0x00} ** 16,
+            .pipe = Pipeline.init(),
+            .sched = sched,
+            .bus = bus,
+            .cpsr = .{ .raw = 0x0000_001F },
+            .spsr = .{ .raw = 0x0000_0000 },
+            .bank = Bank.create(),
+            .logger = if (log_file) |file| Logger.init(file) else null,
+        };
+    }
+
+    pub inline fn hasSPSR(self: *const Self) bool {
+        const mode = getModeChecked(self, self.cpsr.mode.read());
+        return switch (mode) {
+            .System, .User => false,
+            else => true,
+        };
+    }
+
+    pub inline fn isPrivileged(self: *const Self) bool {
+        const mode = getModeChecked(self, self.cpsr.mode.read());
+        return switch (mode) {
+            .User => false,
+            else => true,
+        };
+    }
+
+    pub inline fn isHalted(self: *const Self) bool {
+        return self.bus.io.haltcnt == .Halt;
+    }
+
+    pub fn setCpsr(self: *Self, value: u32) void {
+        if (value & 0x1F != self.cpsr.raw & 0x1F) self.changeModeFromIdx(@truncate(u5, value & 0x1F));
+        self.cpsr.raw = value;
+    }
+
+    fn changeModeFromIdx(self: *Self, next: u5) void {
+        self.changeMode(getModeChecked(self, next));
+    }
+
+    pub fn setUserModeRegister(self: *Self, idx: usize, value: u32) void {
+        const current = getModeChecked(self, self.cpsr.mode.read());
+
+        switch (idx) {
+            8...12 => {
+                if (current == .Fiq) {
+                    self.bank.fiq[Bank.fiqIdx(idx - 8, .User)] = value;
+                } else self.r[idx] = value;
+            },
+            13, 14 => switch (current) {
+                .User, .System => self.r[idx] = value,
+                else => {
+                    const kind = std.meta.intToEnum(Bank.Kind, idx - 13) catch unreachable;
+                    self.bank.r[Bank.regIdx(.User, kind)] = value;
+                },
+            },
+            else => self.r[idx] = value, // R0 -> R7  and R15
+        }
+    }
+
+    pub fn getUserModeRegister(self: *Self, idx: usize) u32 {
+        const current = getModeChecked(self, self.cpsr.mode.read());
+
+        return switch (idx) {
+            8...12 => if (current == .Fiq) self.bank.fiq[Bank.fiqIdx(idx - 8, .User)] else self.r[idx],
+            13, 14 => switch (current) {
+                .User, .System => self.r[idx],
+                else => blk: {
+                    const kind = std.meta.intToEnum(Bank.Kind, idx - 13) catch unreachable;
+                    break :blk self.bank.r[Bank.regIdx(.User, kind)];
+                },
+            },
+            else => self.r[idx], // R0 -> R7  and R15
+        };
+    }
+
+    pub fn changeMode(self: *Self, next: Mode) void {
+        const now = getModeChecked(self, self.cpsr.mode.read());
+
+        // Bank R8 -> r12
+        for (0..5) |i| {
+            self.bank.fiq[Bank.fiqIdx(i, now)] = self.r[8 + i];
+        }
+
+        // Bank r13, r14, SPSR
+        switch (now) {
+            .User, .System => {
+                self.bank.r[Bank.regIdx(now, .R13)] = self.r[13];
+                self.bank.r[Bank.regIdx(now, .R14)] = self.r[14];
+            },
+            else => {
+                self.bank.r[Bank.regIdx(now, .R13)] = self.r[13];
+                self.bank.r[Bank.regIdx(now, .R14)] = self.r[14];
+                self.bank.spsr[Bank.spsrIdx(now)] = self.spsr;
+            },
+        }
+
+        // Grab R8 -> R12
+        for (0..5) |i| {
+            self.r[8 + i] = self.bank.fiq[Bank.fiqIdx(i, next)];
+        }
+
+        // Grab r13, r14, SPSR
+        switch (next) {
+            .User, .System => {
+                self.r[13] = self.bank.r[Bank.regIdx(next, .R13)];
+                self.r[14] = self.bank.r[Bank.regIdx(next, .R14)];
+            },
+            else => {
+                self.r[13] = self.bank.r[Bank.regIdx(next, .R13)];
+                self.r[14] = self.bank.r[Bank.regIdx(next, .R14)];
+                self.spsr = self.bank.spsr[Bank.spsrIdx(next)];
+            },
+        }
+
+        self.cpsr.mode.write(@enumToInt(next));
+    }
+
+    /// Advances state so that the BIOS is skipped
+    ///
+    /// Note: This accesses the CPU's bus ptr so it only may be called
+    /// once the Bus has been properly initialized
+    ///
+    /// TODO: Make above notice impossible to do in code
+    pub fn fastBoot(self: *Self) void {
+        self.r = std.mem.zeroes([16]u32);
+
+        // self.r[0] = 0x08000000;
+        // self.r[1] = 0x000000EA;
+        self.r[13] = 0x0300_7F00;
+        self.r[15] = 0x0800_0000;
+
+        self.bank.r[Bank.regIdx(.Irq, .R13)] = 0x0300_7FA0;
+        self.bank.r[Bank.regIdx(.Supervisor, .R13)] = 0x0300_7FE0;
+
+        // self.cpsr.raw = 0x6000001F;
+        self.cpsr.raw = 0x0000_001F;
+
+        self.bus.bios.addr_latch = 0x0000_00DC + 8;
+    }
+
+    pub fn step(self: *Self) void {
+        defer {
+            if (!self.pipe.flushed) self.r[15] += if (self.cpsr.t.read()) 2 else @as(u32, 4);
+            self.pipe.flushed = false;
+        }
+
+        if (self.cpsr.t.read()) {
+            const opcode = @truncate(u16, self.pipe.step(self, u16) orelse return);
+            if (self.logger) |*trace| trace.mgbaLog(self, opcode);
+
+            thumb.lut[thumb.idx(opcode)](self, self.bus, opcode);
+        } else {
+            const opcode = self.pipe.step(self, u32) orelse return;
+            if (self.logger) |*trace| trace.mgbaLog(self, opcode);
+
+            if (checkCond(self.cpsr, @truncate(u4, opcode >> 28))) {
+                arm.lut[arm.idx(opcode)](self, self.bus, opcode);
+            }
+        }
+    }
+
+    pub fn stepDmaTransfer(self: *Self) bool {
+        inline for (0..4) |i| {
+            if (self.bus.dma[i].in_progress) {
+                self.bus.dma[i].step(self);
+                return true;
+            }
+        }
+
+        return false;
+    }
+
+    pub fn handleInterrupt(self: *Self) void {
+        const should_handle = self.bus.io.ie.raw & self.bus.io.irq.raw;
+
+        // Return if IME is disabled, CPSR I is set or there is nothing to handle
+        if (!self.bus.io.ime or self.cpsr.i.read() or should_handle == 0) return;
+
+        // If Pipeline isn't full, we have a bug
+        std.debug.assert(self.pipe.isFull());
+
+        // log.debug("Handling Interrupt!", .{});
+        self.bus.io.haltcnt = .Execute;
+
+        // FIXME: This seems weird, but retAddr.gba suggests I need to make these changes
+        const ret_addr = self.r[15] - if (self.cpsr.t.read()) 0 else @as(u32, 4);
+        const new_spsr = self.cpsr.raw;
+
+        self.changeMode(.Irq);
+        self.cpsr.t.write(false);
+        self.cpsr.i.write(true);
+
+        self.r[14] = ret_addr;
+        self.spsr.raw = new_spsr;
+        self.r[15] = 0x0000_0018;
+        self.pipe.reload(self);
+    }
+
+    inline fn fetch(self: *Self, comptime T: type, address: u32) T {
+        comptime std.debug.assert(T == u32 or T == u16); // Opcode may be 32-bit (ARM) or 16-bit (THUMB)
+
+        // Bus.read will advance the scheduler. There are different timings for CPU fetches,
+        // so we want to undo what Bus.read will apply. We can do this by caching the current tick
+        // This is very dumb.
+        //
+        // FIXME: Please rework this
+        const tick_cache = self.sched.tick;
+        defer self.sched.tick = tick_cache + Bus.fetch_timings[@boolToInt(T == u32)][@truncate(u4, address >> 24)];
+
+        return self.bus.read(T, address);
+    }
+
+    pub fn panic(self: *const Self, comptime format: []const u8, args: anytype) noreturn {
+        var i: usize = 0;
+        while (i < 16) : (i += 4) {
+            const i_1 = i + 1;
+            const i_2 = i + 2;
+            const i_3 = i + 3;
+            std.debug.print("R{}: 0x{X:0>8}\tR{}: 0x{X:0>8}\tR{}: 0x{X:0>8}\tR{}: 0x{X:0>8}\n", .{ i, self.r[i], i_1, self.r[i_1], i_2, self.r[i_2], i_3, self.r[i_3] });
+        }
+        std.debug.print("cpsr: 0x{X:0>8} ", .{self.cpsr.raw});
+        self.cpsr.toString();
+
+        std.debug.print("spsr: 0x{X:0>8} ", .{self.spsr.raw});
+        self.spsr.toString();
+
+        std.debug.print("pipeline: {??X:0>8}\n", .{self.pipe.stage});
+
+        if (self.cpsr.t.read()) {
+            const opcode = self.bus.dbgRead(u16, self.r[15] - 4);
+            const id = thumb.idx(opcode);
+            std.debug.print("opcode: ID: 0x{b:0>10} 0x{X:0>4}\n", .{ id, opcode });
+        } else {
+            const opcode = self.bus.dbgRead(u32, self.r[15] - 4);
+            const id = arm.idx(opcode);
+            std.debug.print("opcode: ID: 0x{X:0>3} 0x{X:0>8}\n", .{ id, opcode });
+        }
+
+        std.debug.print("tick: {}\n\n", .{self.sched.tick});
+
+        std.debug.panic(format, args);
+    }
+};
+
+const condition_lut = [_]u16{
+    0xF0F0, // EQ - Equal
+    0x0F0F, // NE - Not Equal
+    0xCCCC, // CS - Unsigned higher or same
+    0x3333, // CC - Unsigned lower
+    0xFF00, // MI - Negative
+    0x00FF, // PL - Positive or Zero
+    0xAAAA, // VS - Overflow
+    0x5555, // VC - No Overflow
+    0x0C0C, // HI - unsigned hierh
+    0xF3F3, // LS - unsigned lower or same
+    0xAA55, // GE - greater or equal
+    0x55AA, // LT - less than
+    0x0A05, // GT - greater than
+    0xF5FA, // LE - less than or equal
+    0xFFFF, // AL - always
+    0x0000, // NV - never
+};
+
+pub inline fn checkCond(cpsr: PSR, cond: u4) bool {
+    const flags = @truncate(u4, cpsr.raw >> 28);
+
+    return condition_lut[cond] & (@as(u16, 1) << flags) != 0;
+}
+
+const Pipeline = struct {
+    const Self = @This();
+    stage: [2]?u32,
+    flushed: bool,
+
+    fn init() Self {
+        return .{
+            .stage = [_]?u32{null} ** 2,
+            .flushed = false,
+        };
+    }
+
+    pub fn isFull(self: *const Self) bool {
+        return self.stage[0] != null and self.stage[1] != null;
+    }
+
+    pub fn step(self: *Self, cpu: *Arm7tdmi, comptime T: type) ?u32 {
+        comptime std.debug.assert(T == u32 or T == u16);
+
+        const opcode = self.stage[0];
+        self.stage[0] = self.stage[1];
+        self.stage[1] = cpu.fetch(T, cpu.r[15]);
+
+        return opcode;
+    }
+
+    pub fn reload(self: *Self, cpu: *Arm7tdmi) void {
+        if (cpu.cpsr.t.read()) {
+            self.stage[0] = cpu.fetch(u16, cpu.r[15]);
+            self.stage[1] = cpu.fetch(u16, cpu.r[15] + 2);
+            cpu.r[15] += 4;
+        } else {
+            self.stage[0] = cpu.fetch(u32, cpu.r[15]);
+            self.stage[1] = cpu.fetch(u32, cpu.r[15] + 4);
+            cpu.r[15] += 8;
+        }
+
+        self.flushed = true;
+    }
+};
+
+pub const PSR = extern union {
+    mode: Bitfield(u32, 0, 5),
+    t: Bit(u32, 5),
+    f: Bit(u32, 6),
+    i: Bit(u32, 7),
+    v: Bit(u32, 28),
+    c: Bit(u32, 29),
+    z: Bit(u32, 30),
+    n: Bit(u32, 31),
+    raw: u32,
+
+    fn toString(self: PSR) void {
+        std.debug.print("[", .{});
+
+        if (self.n.read()) std.debug.print("N", .{}) else std.debug.print("-", .{});
+        if (self.z.read()) std.debug.print("Z", .{}) else std.debug.print("-", .{});
+        if (self.c.read()) std.debug.print("C", .{}) else std.debug.print("-", .{});
+        if (self.v.read()) std.debug.print("V", .{}) else std.debug.print("-", .{});
+        if (self.i.read()) std.debug.print("I", .{}) else std.debug.print("-", .{});
+        if (self.f.read()) std.debug.print("F", .{}) else std.debug.print("-", .{});
+        if (self.t.read()) std.debug.print("T", .{}) else std.debug.print("-", .{});
+        std.debug.print("|", .{});
+        if (getMode(self.mode.read())) |m| std.debug.print("{s}", .{m.toString()}) else std.debug.print("---", .{});
+
+        std.debug.print("]\n", .{});
+    }
+};
+
+const Mode = enum(u5) {
+    User = 0b10000,
+    Fiq = 0b10001,
+    Irq = 0b10010,
+    Supervisor = 0b10011,
+    Abort = 0b10111,
+    Undefined = 0b11011,
+    System = 0b11111,
+
+    fn toString(self: Mode) []const u8 {
+        return switch (self) {
+            .User => "usr",
+            .Fiq => "fiq",
+            .Irq => "irq",
+            .Supervisor => "svc",
+            .Abort => "abt",
+            .Undefined => "und",
+            .System => "sys",
+        };
+    }
+};
+
+fn getMode(bits: u5) ?Mode {
+    return std.meta.intToEnum(Mode, bits) catch null;
+}
+
+fn getModeChecked(cpu: *const Arm7tdmi, bits: u5) Mode {
+    return getMode(bits) orelse cpu.panic("[CPU/CPSR] 0b{b:0>5} is an invalid CPU mode", .{bits});
+}

src/core/cpu/arm/block_data_transfer.zig

@@ -0,0 +1,111 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+pub fn blockDataTransfer(comptime P: bool, comptime U: bool, comptime S: bool, comptime W: bool, comptime L: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
+            const rn = @truncate(u4, opcode >> 16 & 0xF);
+            const rlist = opcode & 0xFFFF;
+            const r15 = rlist >> 15 & 1 == 1;
+
+            var count: u32 = 0;
+            var i: u5 = 0;
+            var first: u4 = 0;
+            var write_to_base = true;
+
+            while (i < 16) : (i += 1) {
+                const r = @truncate(u4, 15 - i);
+                if (rlist >> r & 1 == 1) {
+                    first = r;
+                    count += 1;
+                }
+            }
+
+            var start = cpu.r[rn];
+            if (U) {
+                start += if (P) 4 else 0;
+            } else {
+                start = start - (4 * count) + if (!P) 4 else 0;
+            }
+
+            var end = cpu.r[rn];
+            if (U) {
+                end = end + (4 * count) - if (!P) 4 else 0;
+            } else {
+                end -= if (P) 4 else 0;
+            }
+
+            var new_base = cpu.r[rn];
+            if (U) {
+                new_base += 4 * count;
+            } else {
+                new_base -= 4 * count;
+            }
+
+            var address = start;
+
+            if (rlist == 0) {
+                var und_addr = cpu.r[rn];
+                if (U) {
+                    und_addr += if (P) 4 else 0;
+                } else {
+                    und_addr -= 0x40 - if (!P) 4 else 0;
+                }
+
+                if (L) {
+                    cpu.r[15] = bus.read(u32, und_addr);
+                    cpu.pipe.reload(cpu);
+                } else {
+                    bus.write(u32, und_addr, cpu.r[15] + 4);
+                }
+
+                cpu.r[rn] = if (U) cpu.r[rn] + 0x40 else cpu.r[rn] - 0x40;
+                return;
+            }
+
+            i = first;
+            while (i < 16) : (i += 1) {
+                if (rlist >> i & 1 == 1) {
+                    transfer(cpu, bus, r15, i, address);
+                    address += 4;
+
+                    if (W and !L and write_to_base) {
+                        cpu.r[rn] = new_base;
+                        write_to_base = false;
+                    }
+                }
+            }
+
+            if (W and L and rlist >> rn & 1 == 0) cpu.r[rn] = new_base;
+        }
+
+        fn transfer(cpu: *Arm7tdmi, bus: *Bus, r15_present: bool, i: u5, address: u32) void {
+            if (L) {
+                if (S and !r15_present) {
+                    // Always Transfer User mode Registers
+                    cpu.setUserModeRegister(i, bus.read(u32, address));
+                } else {
+                    const value = bus.read(u32, address);
+
+                    cpu.r[i] = value;
+                    if (i == 0xF) {
+                        cpu.r[i] &= ~@as(u32, 3); // Align r15
+                        cpu.pipe.reload(cpu);
+
+                        if (S) cpu.setCpsr(cpu.spsr.raw);
+                    }
+                }
+            } else {
+                if (S) {
+                    // Always Transfer User mode Registers
+                    // This happens regardless if r15 is in the list
+                    const value = cpu.getUserModeRegister(i);
+                    bus.write(u32, address, value + if (i == 0xF) 4 else @as(u32, 0)); // PC is already 8 ahead to make 12
+                } else {
+                    bus.write(u32, address, cpu.r[i] + if (i == 0xF) 4 else @as(u32, 0));
+                }
+            }
+        }
+    }.inner;
+}

src/core/cpu/arm/branch.zig

@@ -0,0 +1,26 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+const sext = @import("../../../util.zig").sext;
+
+pub fn branch(comptime L: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+            if (L) cpu.r[14] = cpu.r[15] - 4;
+
+            cpu.r[15] +%= sext(u32, u24, opcode) << 2;
+            cpu.pipe.reload(cpu);
+        }
+    }.inner;
+}
+
+pub fn branchAndExchange(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+    const rn = opcode & 0xF;
+
+    const thumb = cpu.r[rn] & 1 == 1;
+    cpu.r[15] = cpu.r[rn] & if (thumb) ~@as(u32, 1) else ~@as(u32, 3);
+
+    cpu.cpsr.t.write(thumb);
+    cpu.pipe.reload(cpu);
+}

src/core/cpu/arm/data_processing.zig

@@ -0,0 +1,185 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+const exec = @import("../barrel_shifter.zig").exec;
+const ror = @import("../barrel_shifter.zig").ror;
+
+pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime kind: u4) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+            const rd = @truncate(u4, opcode >> 12 & 0xF);
+            const rn = opcode >> 16 & 0xF;
+            const old_carry = @boolToInt(cpu.cpsr.c.read());
+
+            // If certain conditions are met, PC is 12 ahead instead of 8
+            // TODO: Why these conditions?
+            if (!I and opcode >> 4 & 1 == 1) cpu.r[15] += 4;
+            const op1 = cpu.r[rn];
+
+            const amount = @truncate(u8, (opcode >> 8 & 0xF) << 1);
+            const op2 = if (I) ror(S, &cpu.cpsr, opcode & 0xFF, amount) else exec(S, cpu, opcode);
+
+            // Undo special condition from above
+            if (!I and opcode >> 4 & 1 == 1) cpu.r[15] -= 4;
+
+            var result: u32 = undefined;
+            var overflow: u1 = undefined;
+
+            // Perform Data Processing Logic
+            switch (kind) {
+                0x0 => result = op1 & op2, // AND
+                0x1 => result = op1 ^ op2, // EOR
+                0x2 => result = op1 -% op2, // SUB
+                0x3 => result = op2 -% op1, // RSB
+                0x4 => result = add(&overflow, op1, op2), // ADD
+                0x5 => result = adc(&overflow, op1, op2, old_carry), // ADC
+                0x6 => result = sbc(op1, op2, old_carry), // SBC
+                0x7 => result = sbc(op2, op1, old_carry), // RSC
+                0x8 => {
+                    // TST
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
+
+                    result = op1 & op2;
+                },
+                0x9 => {
+                    // TEQ
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
+
+                    result = op1 ^ op2;
+                },
+                0xA => {
+                    // CMP
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
+
+                    result = op1 -% op2;
+                },
+                0xB => {
+                    // CMN
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
+
+                    const tmp = @addWithOverflow(op1, op2);
+                    result = tmp[0];
+                    overflow = tmp[1];
+                },
+                0xC => result = op1 | op2, // ORR
+                0xD => result = op2, // MOV
+                0xE => result = op1 & ~op2, // BIC
+                0xF => result = ~op2, // MVN
+            }
+
+            // Write to Destination Register
+            switch (kind) {
+                0x8, 0x9, 0xA, 0xB => {}, // Test Operations
+                else => {
+                    cpu.r[rd] = result;
+                    if (rd == 0xF) {
+                        if (S) cpu.setCpsr(cpu.spsr.raw);
+                        cpu.pipe.reload(cpu);
+                    }
+                },
+            }
+
+            // Write Flags
+            switch (kind) {
+                0x0, 0x1, 0xC, 0xD, 0xE, 0xF => if (S and rd != 0xF) {
+                    // Logic Operation Flags
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    // C set by Barrel Shifter, V is unaffected
+
+                },
+                0x2, 0x3 => if (S and rd != 0xF) {
+                    // SUB, RSB Flags
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+
+                    if (kind == 0x2) {
+                        // SUB specific
+                        cpu.cpsr.c.write(op2 <= op1);
+                        cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                    } else {
+                        // RSB Specific
+                        cpu.cpsr.c.write(op1 <= op2);
+                        cpu.cpsr.v.write(((op2 ^ result) & (~op1 ^ result)) >> 31 & 1 == 1);
+                    }
+                },
+                0x4, 0x5 => if (S and rd != 0xF) {
+                    // ADD, ADC Flags
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    cpu.cpsr.c.write(overflow == 0b1);
+                    cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
+                },
+                0x6, 0x7 => if (S and rd != 0xF) {
+                    // SBC, RSC Flags
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+
+                    if (kind == 0x6) {
+                        // SBC specific
+                        const subtrahend = @as(u64, op2) -% old_carry +% 1;
+                        cpu.cpsr.c.write(subtrahend <= op1);
+                        cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                    } else {
+                        // RSC Specific
+                        const subtrahend = @as(u64, op1) -% old_carry +% 1;
+                        cpu.cpsr.c.write(subtrahend <= op2);
+                        cpu.cpsr.v.write(((op2 ^ result) & (~op1 ^ result)) >> 31 & 1 == 1);
+                    }
+                },
+                0x8, 0x9, 0xA, 0xB => {
+                    // Test Operation Flags
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+
+                    if (kind == 0xA) {
+                        // CMP specific
+                        cpu.cpsr.c.write(op2 <= op1);
+                        cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                    } else if (kind == 0xB) {
+                        // CMN specific
+                        cpu.cpsr.c.write(overflow == 0b1);
+                        cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
+                    } else {
+                        // TST, TEQ specific
+                        // Barrel Shifter should always calc CPSR C in TST
+                        if (!S) _ = exec(true, cpu, opcode);
+                    }
+                },
+            }
+        }
+    }.inner;
+}
+
+pub fn sbc(left: u32, right: u32, old_carry: u1) u32 {
+    // TODO: Make your own version (thanks peach.bot)
+    const subtrahend = @as(u64, right) -% old_carry +% 1;
+    const ret = @truncate(u32, left -% subtrahend);
+
+    return ret;
+}
+
+pub fn add(overflow: *u1, left: u32, right: u32) u32 {
+    const ret = @addWithOverflow(left, right);
+    overflow.* = ret[1];
+
+    return ret[0];
+}
+
+pub fn adc(overflow: *u1, left: u32, right: u32, old_carry: u1) u32 {
+    const tmp = @addWithOverflow(left, right);
+    const ret = @addWithOverflow(tmp[0], old_carry);
+    overflow.* = tmp[1] | ret[1];
+
+    return ret[0];
+}
+
+fn undefinedTestBehaviour(cpu: *Arm7tdmi) void {
+    @setCold(true);
+    cpu.setCpsr(cpu.spsr.raw);
+}

src/core/cpu/arm/half_signed_data_transfer.zig

@@ -0,0 +1,53 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+const sext = @import("../../../util.zig").sext;
+const rotr = @import("../../../util.zig").rotr;
+
+pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I: bool, comptime W: bool, comptime L: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
+            const rn = opcode >> 16 & 0xF;
+            const rd = opcode >> 12 & 0xF;
+            const rm = opcode & 0xF;
+            const imm_offset_high = opcode >> 8 & 0xF;
+
+            const base = cpu.r[rn] + if (!L and rn == 0xF) 4 else @as(u32, 0);
+            const offset = if (I) imm_offset_high << 4 | rm else cpu.r[rm];
+
+            const modified_base = if (U) base +% offset else base -% offset;
+            var address = if (P) modified_base else base;
+
+            var result: u32 = undefined;
+            if (L) {
+                switch (@truncate(u2, opcode >> 5)) {
+                    0b01 => {
+                        // LDRH
+                        const value = bus.read(u16, address);
+                        result = rotr(u32, value, 8 * (address & 1));
+                    },
+                    0b10 => {
+                        // LDRSB
+                        result = sext(u32, u8, bus.read(u8, address));
+                    },
+                    0b11 => {
+                        // LDRSH
+                        const value = bus.read(u16, address);
+                        result = if (address & 1 == 1) sext(u32, u8, @truncate(u8, value >> 8)) else sext(u32, u16, value);
+                    },
+                    0b00 => unreachable, // SWP
+                }
+            } else {
+                if (opcode >> 5 & 0x01 == 0x01) {
+                    // STRH
+                    bus.write(u16, address, @truncate(u16, cpu.r[rd]));
+                } else unreachable; // SWP
+            }
+
+            address = modified_base;
+            if (W and P or !P) cpu.r[rn] = address;
+            if (L) cpu.r[rd] = result; // // This emulates the LDR rd == rn behaviour
+        }
+    }.inner;
+}

src/core/cpu/arm/multiply.zig

@@ -0,0 +1,57 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+pub fn multiply(comptime A: bool, comptime S: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+            const rd = opcode >> 16 & 0xF;
+            const rn = opcode >> 12 & 0xF;
+            const rs = opcode >> 8 & 0xF;
+            const rm = opcode & 0xF;
+
+            const temp: u64 = @as(u64, cpu.r[rm]) * @as(u64, cpu.r[rs]) + if (A) cpu.r[rn] else 0;
+            const result = @truncate(u32, temp);
+            cpu.r[rd] = result;
+
+            if (S) {
+                cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                cpu.cpsr.z.write(result == 0);
+                // V is unaffected, C is *actually* undefined in ARMv4
+            }
+        }
+    }.inner;
+}
+
+pub fn multiplyLong(comptime U: bool, comptime A: bool, comptime S: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+            const rd_hi = opcode >> 16 & 0xF;
+            const rd_lo = opcode >> 12 & 0xF;
+            const rs = opcode >> 8 & 0xF;
+            const rm = opcode & 0xF;
+
+            if (U) {
+                // Signed (WHY IS IT U THEN?)
+                var result: i64 = @as(i64, @bitCast(i32, cpu.r[rm])) * @as(i64, @bitCast(i32, cpu.r[rs]));
+                if (A) result +%= @bitCast(i64, @as(u64, cpu.r[rd_hi]) << 32 | @as(u64, cpu.r[rd_lo]));
+
+                cpu.r[rd_hi] = @bitCast(u32, @truncate(i32, result >> 32));
+                cpu.r[rd_lo] = @bitCast(u32, @truncate(i32, result));
+            } else {
+                // Unsigned
+                var result: u64 = @as(u64, cpu.r[rm]) * @as(u64, cpu.r[rs]);
+                if (A) result +%= @as(u64, cpu.r[rd_hi]) << 32 | @as(u64, cpu.r[rd_lo]);
+
+                cpu.r[rd_hi] = @truncate(u32, result >> 32);
+                cpu.r[rd_lo] = @truncate(u32, result);
+            }
+
+            if (S) {
+                cpu.cpsr.z.write(cpu.r[rd_hi] == 0 and cpu.r[rd_lo] == 0);
+                cpu.cpsr.n.write(cpu.r[rd_hi] >> 31 & 1 == 1);
+                // C and V are set to meaningless values
+            }
+        }
+    }.inner;
+}

src/core/cpu/arm/psr_transfer.zig

@@ -0,0 +1,59 @@
+const std = @import("std");
+
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+const PSR = @import("../../cpu.zig").PSR;
+
+const log = std.log.scoped(.PsrTransfer);
+
+const rotr = @import("../../../util.zig").rotr;
+
+pub fn psrTransfer(comptime I: bool, comptime R: bool, comptime kind: u2) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
+            switch (kind) {
+                0b00 => {
+                    // MRS
+                    const rd = opcode >> 12 & 0xF;
+
+                    if (R and !cpu.hasSPSR()) log.err("Tried to read SPSR from User/System Mode", .{});
+                    cpu.r[rd] = if (R) cpu.spsr.raw else cpu.cpsr.raw;
+                },
+                0b10 => {
+                    // MSR
+                    const field_mask = @truncate(u4, opcode >> 16 & 0xF);
+                    const rm_idx = opcode & 0xF;
+                    const right = if (I) rotr(u32, opcode & 0xFF, (opcode >> 8 & 0xF) * 2) else cpu.r[rm_idx];
+
+                    if (R and !cpu.hasSPSR()) log.err("Tried to write to SPSR in User/System Mode", .{});
+
+                    if (R) {
+                        // arm.gba seems to expect the SPSR to do somethign in SYS mode,
+                        // so we just assume that despite writing to the SPSR in USR or SYS mode
+                        // being UNPREDICTABLE, it just magically has a working SPSR somehow
+                        cpu.spsr.raw = fieldMask(&cpu.spsr, field_mask, right);
+                    } else {
+                        if (cpu.isPrivileged()) cpu.setCpsr(fieldMask(&cpu.cpsr, field_mask, right));
+                    }
+                },
+                else => cpu.panic("[CPU/PSR Transfer] Bits 21:220 of {X:0>8} are undefined", .{opcode}),
+            }
+        }
+    }.inner;
+}
+
+fn fieldMask(psr: *const PSR, field_mask: u4, right: u32) u32 {
+    // This bitwise ORs bits 3 and 0 of the field mask into a u2
+    // We do this because we only care about bits 7:0 and 31:28 of the CPSR
+    const bits = @truncate(u2, (field_mask >> 2 & 0x2) | (field_mask & 1));
+
+    const mask: u32 = switch (bits) {
+        0b00 => 0x0000_0000,
+        0b01 => 0x0000_00FF,
+        0b10 => 0xF000_0000,
+        0b11 => 0xF000_00FF,
+    };
+
+    return (psr.raw & ~mask) | (right & mask);
+}

src/core/cpu/arm/single_data_swap.zig

@@ -0,0 +1,29 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+const rotr = @import("../../../util.zig").rotr;
+
+pub fn singleDataSwap(comptime B: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
+            const rn = opcode >> 16 & 0xF;
+            const rd = opcode >> 12 & 0xF;
+            const rm = opcode & 0xF;
+
+            const address = cpu.r[rn];
+
+            if (B) {
+                // SWPB
+                const value = bus.read(u8, address);
+                bus.write(u8, address, @truncate(u8, cpu.r[rm]));
+                cpu.r[rd] = value;
+            } else {
+                // SWP
+                const value = rotr(u32, bus.read(u32, address), 8 * (address & 0x3));
+                bus.write(u32, address, cpu.r[rm]);
+                cpu.r[rd] = value;
+            }
+        }
+    }.inner;
+}

src/core/cpu/arm/single_data_transfer.zig

@@ -0,0 +1,55 @@
+const shifter = @import("../barrel_shifter.zig");
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+const rotr = @import("../../../util.zig").rotr;
+
+pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool, comptime B: bool, comptime W: bool, comptime L: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
+            const rn = opcode >> 16 & 0xF;
+            const rd = opcode >> 12 & 0xF;
+
+            const base = cpu.r[rn];
+            const offset = if (I) shifter.immediate(false, cpu, opcode) else opcode & 0xFFF;
+
+            const modified_base = if (U) base +% offset else base -% offset;
+            var address = if (P) modified_base else base;
+
+            var result: u32 = undefined;
+            if (L) {
+                if (B) {
+                    // LDRB
+                    result = bus.read(u8, address);
+                } else {
+                    // LDR
+                    const value = bus.read(u32, address);
+                    result = rotr(u32, value, 8 * (address & 0x3));
+                }
+            } else {
+                if (B) {
+                    // STRB
+                    const value = cpu.r[rd] + if (rd == 0xF) 4 else @as(u32, 0); // PC is 12 ahead
+                    bus.write(u8, address, @truncate(u8, value));
+                } else {
+                    // STR
+                    const value = cpu.r[rd] + if (rd == 0xF) 4 else @as(u32, 0);
+                    bus.write(u32, address, value);
+                }
+            }
+
+            address = modified_base;
+            if (W and P or !P) {
+                cpu.r[rn] = address;
+                if (rn == 0xF) cpu.pipe.reload(cpu);
+            }
+
+            if (L) {
+                // This emulates the LDR rd == rn behaviour
+                cpu.r[rd] = result;
+                if (rd == 0xF) cpu.pipe.reload(cpu);
+            }
+        }
+    }.inner;
+}

src/core/cpu/arm/software_interrupt.zig

@@ -0,0 +1,23 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").arm.InstrFn;
+
+pub fn armSoftwareInterrupt() InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, _: u32) void {
+            // Copy Values from Current Mode
+            const ret_addr = cpu.r[15] - 4;
+            const cpsr = cpu.cpsr.raw;
+
+            // Switch Mode
+            cpu.changeMode(.Supervisor);
+            cpu.cpsr.t.write(false); // Force ARM Mode
+            cpu.cpsr.i.write(true); // Disable normal interrupts
+
+            cpu.r[14] = ret_addr; // Resume Execution
+            cpu.spsr.raw = cpsr; // Previous mode CPSR
+            cpu.r[15] = 0x0000_0008;
+            cpu.pipe.reload(cpu);
+        }
+    }.inner;
+}

src/core/cpu/barrel_shifter.zig

@@ -0,0 +1,147 @@
+const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
+const CPSR = @import("../cpu.zig").PSR;
+
+const rotr = @import("../../util.zig").rotr;
+
+pub fn exec(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
+    var result: u32 = undefined;
+    if (opcode >> 4 & 1 == 1) {
+        result = register(S, cpu, opcode);
+    } else {
+        result = immediate(S, cpu, opcode);
+    }
+
+    return result;
+}
+
+fn register(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
+    const rs_idx = opcode >> 8 & 0xF;
+    const rm = cpu.r[opcode & 0xF];
+    const rs = @truncate(u8, cpu.r[rs_idx]);
+
+    return switch (@truncate(u2, opcode >> 5)) {
+        0b00 => lsl(S, &cpu.cpsr, rm, rs),
+        0b01 => lsr(S, &cpu.cpsr, rm, rs),
+        0b10 => asr(S, &cpu.cpsr, rm, rs),
+        0b11 => ror(S, &cpu.cpsr, rm, rs),
+    };
+}
+
+pub fn immediate(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
+    const amount = @truncate(u8, opcode >> 7 & 0x1F);
+    const rm = cpu.r[opcode & 0xF];
+
+    var result: u32 = undefined;
+    if (amount == 0) {
+        switch (@truncate(u2, opcode >> 5)) {
+            0b00 => {
+                // LSL #0
+                result = rm;
+            },
+            0b01 => {
+                // LSR #0 aka LSR #32
+                if (S) cpu.cpsr.c.write(rm >> 31 & 1 == 1);
+                result = 0x0000_0000;
+            },
+            0b10 => {
+                // ASR #0 aka ASR #32
+                result = @bitCast(u32, @bitCast(i32, rm) >> 31);
+                if (S) cpu.cpsr.c.write(result >> 31 & 1 == 1);
+            },
+            0b11 => {
+                // ROR #0 aka RRX
+                const carry: u32 = @boolToInt(cpu.cpsr.c.read());
+                if (S) cpu.cpsr.c.write(rm & 1 == 1);
+
+                result = (carry << 31) | (rm >> 1);
+            },
+        }
+    } else {
+        switch (@truncate(u2, opcode >> 5)) {
+            0b00 => result = lsl(S, &cpu.cpsr, rm, amount),
+            0b01 => result = lsr(S, &cpu.cpsr, rm, amount),
+            0b10 => result = asr(S, &cpu.cpsr, rm, amount),
+            0b11 => result = ror(S, &cpu.cpsr, rm, amount),
+        }
+    }
+
+    return result;
+}
+
+pub fn lsl(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
+    const amount = @truncate(u5, total_amount);
+    const bit_count: u8 = @typeInfo(u32).Int.bits;
+
+    var result: u32 = 0x0000_0000;
+    if (total_amount < bit_count) {
+        // We can perform a well-defined shift here
+        result = rm << amount;
+
+        if (S and total_amount != 0) {
+            const carry_bit = @truncate(u5, bit_count - amount);
+            cpsr.c.write(rm >> carry_bit & 1 == 1);
+        }
+    } else {
+        if (S) {
+            if (total_amount == bit_count) {
+                // Shifted all bits out, carry bit is bit 0 of rm
+                cpsr.c.write(rm & 1 == 1);
+            } else {
+                cpsr.c.write(false);
+            }
+        }
+    }
+
+    return result;
+}
+
+pub fn lsr(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u32) u32 {
+    const amount = @truncate(u5, total_amount);
+    const bit_count: u8 = @typeInfo(u32).Int.bits;
+
+    var result: u32 = 0x0000_0000;
+    if (total_amount < bit_count) {
+        // We can perform a well-defined shift
+        result = rm >> amount;
+        if (S and total_amount != 0) cpsr.c.write(rm >> (amount - 1) & 1 == 1);
+    } else {
+        if (S) {
+            if (total_amount == bit_count) {
+                // LSR #32
+                cpsr.c.write(rm >> 31 & 1 == 1);
+            } else {
+                // All bits have been shifted out, including carry bit
+                cpsr.c.write(false);
+            }
+        }
+    }
+
+    return result;
+}
+
+pub fn asr(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
+    const amount = @truncate(u5, total_amount);
+    const bit_count: u8 = @typeInfo(u32).Int.bits;
+
+    var result: u32 = 0x0000_0000;
+    if (total_amount < bit_count) {
+        result = @bitCast(u32, @bitCast(i32, rm) >> amount);
+        if (S and total_amount != 0) cpsr.c.write(rm >> (amount - 1) & 1 == 1);
+    } else {
+        // ASR #32 and ASR #>32 have the same result
+        result = @bitCast(u32, @bitCast(i32, rm) >> 31);
+        if (S) cpsr.c.write(result >> 31 & 1 == 1);
+    }
+
+    return result;
+}
+
+pub fn ror(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
+    const result = rotr(u32, rm, total_amount);
+
+    if (S and total_amount != 0) {
+        cpsr.c.write(result >> 31 & 1 == 1);
+    }
+
+    return result;
+}

src/core/cpu/thumb/alu.zig

@@ -0,0 +1,108 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
+
+const adc = @import("../arm/data_processing.zig").adc;
+const sbc = @import("../arm/data_processing.zig").sbc;
+
+const lsl = @import("../barrel_shifter.zig").lsl;
+const lsr = @import("../barrel_shifter.zig").lsr;
+const asr = @import("../barrel_shifter.zig").asr;
+const ror = @import("../barrel_shifter.zig").ror;
+
+pub fn fmt4(comptime op: u4) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            const rs = opcode >> 3 & 0x7;
+            const rd = opcode & 0x7;
+            const carry = @boolToInt(cpu.cpsr.c.read());
+
+            const op1 = cpu.r[rd];
+            const op2 = cpu.r[rs];
+
+            var result: u32 = undefined;
+            var overflow: u1 = undefined;
+
+            switch (op) {
+                0x0 => result = op1 & op2, // AND
+                0x1 => result = op1 ^ op2, // EOR
+                0x2 => result = lsl(true, &cpu.cpsr, op1, @truncate(u8, op2)), // LSL
+                0x3 => result = lsr(true, &cpu.cpsr, op1, @truncate(u8, op2)), // LSR
+                0x4 => result = asr(true, &cpu.cpsr, op1, @truncate(u8, op2)), // ASR
+                0x5 => result = adc(&overflow, op1, op2, carry), // ADC
+                0x6 => result = sbc(op1, op2, carry), // SBC
+                0x7 => result = ror(true, &cpu.cpsr, op1, @truncate(u8, op2)), // ROR
+                0x8 => result = op1 & op2, // TST
+                0x9 => result = 0 -% op2, // NEG
+                0xA => result = op1 -% op2, // CMP
+                0xB => {
+                    // CMN
+                    const tmp = @addWithOverflow(op1, op2);
+                    result = tmp[0];
+                    overflow = tmp[1];
+                },
+                0xC => result = op1 | op2, // ORR
+                0xD => result = @truncate(u32, @as(u64, op2) * @as(u64, op1)),
+                0xE => result = op1 & ~op2,
+                0xF => result = ~op2,
+            }
+
+            // Write to Destination Register
+            switch (op) {
+                0x8, 0xA, 0xB => {},
+                else => cpu.r[rd] = result,
+            }
+
+            // Write Flags
+            switch (op) {
+                0x0, 0x1, 0x2, 0x3, 0x4, 0x7, 0xC, 0xE, 0xF => {
+                    // Logic Operations
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    // C set by Barrel Shifter, V is unaffected
+                },
+                0x8, 0xA => {
+                    // Test Flags
+                    // CMN (0xB) is handled with ADC
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+
+                    if (op == 0xA) {
+                        // CMP specific
+                        cpu.cpsr.c.write(op2 <= op1);
+                        cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                    }
+                },
+                0x5, 0xB => {
+                    // ADC, CMN
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    cpu.cpsr.c.write(overflow == 0b1);
+                    cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
+                },
+                0x6 => {
+                    // SBC
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+
+                    const subtrahend = @as(u64, op2) -% carry +% 1;
+                    cpu.cpsr.c.write(subtrahend <= op1);
+                    cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                },
+                0x9 => {
+                    // NEG
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    cpu.cpsr.c.write(op2 <= 0);
+                    cpu.cpsr.v.write(((0 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                },
+                0xD => {
+                    // Multiplication
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    // V is unaffected, assuming similar behaviour to ARMv4 MUL C is undefined
+                },
+            }
+        }
+    }.inner;
+}

src/core/cpu/thumb/block_data_transfer.zig

@@ -0,0 +1,100 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
+
+pub fn fmt14(comptime L: bool, comptime R: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            const count = @boolToInt(R) + countRlist(opcode);
+            const start = cpu.r[13] - if (!L) count * 4 else 0;
+
+            var end = cpu.r[13];
+            if (L) {
+                end += count * 4;
+            } else {
+                end -= 4;
+            }
+
+            var address = start;
+
+            var i: u4 = 0;
+            while (i < 8) : (i += 1) {
+                if (opcode >> i & 1 == 1) {
+                    if (L) {
+                        cpu.r[i] = bus.read(u32, address);
+                    } else {
+                        bus.write(u32, address, cpu.r[i]);
+                    }
+
+                    address += 4;
+                }
+            }
+
+            if (R) {
+                if (L) {
+                    const value = bus.read(u32, address);
+                    cpu.r[15] = value & ~@as(u32, 1);
+                    cpu.pipe.reload(cpu);
+                } else {
+                    bus.write(u32, address, cpu.r[14]);
+                }
+                address += 4;
+            }
+
+            cpu.r[13] = if (L) end else start;
+        }
+    }.inner;
+}
+
+pub fn fmt15(comptime L: bool, comptime rb: u3) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            var address = cpu.r[rb];
+            const end_address = cpu.r[rb] + 4 * countRlist(opcode);
+
+            if (opcode & 0xFF == 0) {
+                if (L) {
+                    cpu.r[15] = bus.read(u32, address);
+                    cpu.pipe.reload(cpu);
+                } else {
+                    bus.write(u32, address, cpu.r[15] + 2);
+                }
+
+                cpu.r[rb] += 0x40;
+                return;
+            }
+
+            var i: u4 = 0;
+            var first_write = true;
+
+            while (i < 8) : (i += 1) {
+                if (opcode >> i & 1 == 1) {
+                    if (L) {
+                        cpu.r[i] = bus.read(u32, address);
+                    } else {
+                        bus.write(u32, address, cpu.r[i]);
+                    }
+
+                    if (!L and first_write) {
+                        cpu.r[rb] = end_address;
+                        first_write = false;
+                    }
+
+                    address += 4;
+                }
+            }
+
+            if (L and opcode >> rb & 1 != 1) cpu.r[rb] = address;
+        }
+    }.inner;
+}
+
+inline fn countRlist(opcode: u16) u32 {
+    var count: u32 = 0;
+
+    inline for (0..8) |i| {
+        if (opcode >> (7 - i) & 1 == 1) count += 1;
+    }
+
+    return count;
+}

src/core/cpu/thumb/branch.zig

@@ -0,0 +1,54 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
+
+const checkCond = @import("../../cpu.zig").checkCond;
+const sext = @import("../../../util.zig").sext;
+
+pub fn fmt16(comptime cond: u4) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            // B
+            if (cond == 0xE or cond == 0xF)
+                cpu.panic("[CPU/THUMB.16] Undefined conditional branch with condition {}", .{cond});
+
+            if (!checkCond(cpu.cpsr, cond)) return;
+
+            cpu.r[15] +%= sext(u32, u8, opcode & 0xFF) << 1;
+            cpu.pipe.reload(cpu);
+        }
+    }.inner;
+}
+
+pub fn fmt18() InstrFn {
+    return struct {
+        // B but conditional
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            cpu.r[15] +%= sext(u32, u11, opcode & 0x7FF) << 1;
+            cpu.pipe.reload(cpu);
+        }
+    }.inner;
+}
+
+pub fn fmt19(comptime is_low: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            // BL
+            const offset = opcode & 0x7FF;
+
+            if (is_low) {
+                // Instruction 2
+                const next_opcode = cpu.r[15] - 2;
+
+                cpu.r[15] = cpu.r[14] +% (offset << 1);
+                cpu.r[14] = next_opcode | 1;
+
+                cpu.pipe.reload(cpu);
+            } else {
+                // Instruction 1
+                const lr_offset = sext(u32, u11, offset) << 12;
+                cpu.r[14] = (cpu.r[15] +% lr_offset) & ~@as(u32, 1);
+            }
+        }
+    }.inner;
+}

src/core/cpu/thumb/data_processing.zig

@@ -0,0 +1,199 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
+
+const add = @import("../arm/data_processing.zig").add;
+
+const lsl = @import("../barrel_shifter.zig").lsl;
+const lsr = @import("../barrel_shifter.zig").lsr;
+const asr = @import("../barrel_shifter.zig").asr;
+
+pub fn fmt1(comptime op: u2, comptime offset: u5) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            const rs = opcode >> 3 & 0x7;
+            const rd = opcode & 0x7;
+
+            const result = switch (op) {
+                0b00 => blk: {
+                    // LSL
+                    if (offset == 0) {
+                        break :blk cpu.r[rs];
+                    } else {
+                        break :blk lsl(true, &cpu.cpsr, cpu.r[rs], offset);
+                    }
+                },
+                0b01 => blk: {
+                    // LSR
+                    if (offset == 0) {
+                        cpu.cpsr.c.write(cpu.r[rs] >> 31 & 1 == 1);
+                        break :blk @as(u32, 0);
+                    } else {
+                        break :blk lsr(true, &cpu.cpsr, cpu.r[rs], offset);
+                    }
+                },
+                0b10 => blk: {
+                    // ASR
+                    if (offset == 0) {
+                        cpu.cpsr.c.write(cpu.r[rs] >> 31 & 1 == 1);
+                        break :blk @bitCast(u32, @bitCast(i32, cpu.r[rs]) >> 31);
+                    } else {
+                        break :blk asr(true, &cpu.cpsr, cpu.r[rs], offset);
+                    }
+                },
+                else => cpu.panic("[CPU/THUMB.1] 0b{b:0>2} is not a valid op", .{op}),
+            };
+
+            // Equivalent to an ARM MOVS
+            cpu.r[rd] = result;
+
+            // Write Flags
+            cpu.cpsr.n.write(result >> 31 & 1 == 1);
+            cpu.cpsr.z.write(result == 0);
+        }
+    }.inner;
+}
+
+pub fn fmt5(comptime op: u2, comptime h1: u1, comptime h2: u1) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            const rs = @as(u4, h2) << 3 | (opcode >> 3 & 0x7);
+            const rd = @as(u4, h1) << 3 | (opcode & 0x7);
+
+            const op1 = cpu.r[rd];
+            const op2 = cpu.r[rs];
+
+            var result: u32 = undefined;
+            var overflow: u1 = undefined;
+            switch (op) {
+                0b00 => result = add(&overflow, op1, op2), // ADD
+                0b01 => result = op1 -% op2, // CMP
+                0b10 => result = op2, // MOV
+                0b11 => {},
+            }
+
+            // Write to Destination Register
+            switch (op) {
+                0b01 => {}, // Test Instruction
+                0b11 => {
+                    // BX
+                    const is_thumb = op2 & 1 == 1;
+                    cpu.r[15] = op2 & ~@as(u32, 1);
+
+                    cpu.cpsr.t.write(is_thumb);
+                    cpu.pipe.reload(cpu);
+                },
+                else => {
+                    cpu.r[rd] = result;
+                    if (rd == 0xF) {
+                        cpu.r[15] &= ~@as(u32, 1);
+                        cpu.pipe.reload(cpu);
+                    }
+                },
+            }
+
+            // Write Flags
+            switch (op) {
+                0b01 => {
+                    // CMP
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    cpu.cpsr.c.write(op2 <= op1);
+                    cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                },
+                0b00, 0b10, 0b11 => {}, // MOV and Branch Instruction
+            }
+        }
+    }.inner;
+}
+
+pub fn fmt2(comptime I: bool, is_sub: bool, rn: u3) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            const rs = opcode >> 3 & 0x7;
+            const rd = @truncate(u3, opcode);
+            const op1 = cpu.r[rs];
+            const op2: u32 = if (I) rn else cpu.r[rn];
+
+            if (is_sub) {
+                // SUB
+                const result = op1 -% op2;
+                cpu.r[rd] = result;
+
+                cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                cpu.cpsr.z.write(result == 0);
+                cpu.cpsr.c.write(op2 <= op1);
+                cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+            } else {
+                // ADD
+                var overflow: u1 = undefined;
+                const result = add(&overflow, op1, op2);
+                cpu.r[rd] = result;
+
+                cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                cpu.cpsr.z.write(result == 0);
+                cpu.cpsr.c.write(overflow == 0b1);
+                cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
+            }
+        }
+    }.inner;
+}
+
+pub fn fmt3(comptime op: u2, comptime rd: u3) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            const op1 = cpu.r[rd];
+            const op2: u32 = opcode & 0xFF; // Offset
+
+            var overflow: u1 = undefined;
+            const result: u32 = switch (op) {
+                0b00 => op2, // MOV
+                0b01 => op1 -% op2, // CMP
+                0b10 => add(&overflow, op1, op2), // ADD
+                0b11 => op1 -% op2, // SUB
+            };
+
+            // Write to Register
+            if (op != 0b01) cpu.r[rd] = result;
+
+            // Write Flags
+            cpu.cpsr.n.write(result >> 31 & 1 == 1);
+            cpu.cpsr.z.write(result == 0);
+
+            switch (op) {
+                0b00 => {}, // MOV | C set by Barrel Shifter, V is unaffected
+                0b01, 0b11 => {
+                    // SUB, CMP
+                    cpu.cpsr.c.write(op2 <= op1);
+                    cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
+                },
+                0b10 => {
+                    // ADD
+                    cpu.cpsr.c.write(overflow == 0b1);
+                    cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
+                },
+            }
+        }
+    }.inner;
+}
+
+pub fn fmt12(comptime isSP: bool, comptime rd: u3) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            // ADD
+            const left = if (isSP) cpu.r[13] else cpu.r[15] & ~@as(u32, 2);
+            const right = (opcode & 0xFF) << 2;
+            cpu.r[rd] = left + right;
+        }
+    }.inner;
+}
+
+pub fn fmt13(comptime S: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
+            // ADD
+            const offset = (opcode & 0x7F) << 2;
+            cpu.r[13] = if (S) cpu.r[13] - offset else cpu.r[13] + offset;
+        }
+    }.inner;
+}

src/core/cpu/thumb/data_transfer.zig

@@ -0,0 +1,145 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
+
+const rotr = @import("../../../util.zig").rotr;
+const sext = @import("../../../util.zig").sext;
+
+pub fn fmt6(comptime rd: u3) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            // LDR
+            const offset = (opcode & 0xFF) << 2;
+
+            // Bit 1 of the PC intentionally ignored
+            cpu.r[rd] = bus.read(u32, (cpu.r[15] & ~@as(u32, 2)) + offset);
+        }
+    }.inner;
+}
+
+pub fn fmt78(comptime op: u2, comptime T: bool) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            const ro = opcode >> 6 & 0x7;
+            const rb = opcode >> 3 & 0x7;
+            const rd = opcode & 0x7;
+
+            const address = cpu.r[rb] +% cpu.r[ro];
+
+            if (T) {
+                // Format 8
+                switch (op) {
+                    0b00 => {
+                        // STRH
+                        bus.write(u16, address, @truncate(u16, cpu.r[rd]));
+                    },
+                    0b01 => {
+                        // LDSB
+                        cpu.r[rd] = sext(u32, u8, bus.read(u8, address));
+                    },
+                    0b10 => {
+                        // LDRH
+                        const value = bus.read(u16, address);
+                        cpu.r[rd] = rotr(u32, value, 8 * (address & 1));
+                    },
+                    0b11 => {
+                        // LDRSH
+                        const value = bus.read(u16, address);
+                        cpu.r[rd] = if (address & 1 == 1) sext(u32, u8, @truncate(u8, value >> 8)) else sext(u32, u16, value);
+                    },
+                }
+            } else {
+                // Format 7
+                switch (op) {
+                    0b00 => {
+                        // STR
+                        bus.write(u32, address, cpu.r[rd]);
+                    },
+                    0b01 => {
+                        // STRB
+                        bus.write(u8, address, @truncate(u8, cpu.r[rd]));
+                    },
+                    0b10 => {
+                        // LDR
+                        const value = bus.read(u32, address);
+                        cpu.r[rd] = rotr(u32, value, 8 * (address & 0x3));
+                    },
+                    0b11 => {
+                        // LDRB
+                        cpu.r[rd] = bus.read(u8, address);
+                    },
+                }
+            }
+        }
+    }.inner;
+}
+
+pub fn fmt9(comptime B: bool, comptime L: bool, comptime offset: u5) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            const rb = opcode >> 3 & 0x7;
+            const rd = opcode & 0x7;
+
+            if (L) {
+                if (B) {
+                    // LDRB
+                    const address = cpu.r[rb] + offset;
+                    cpu.r[rd] = bus.read(u8, address);
+                } else {
+                    // LDR
+                    const address = cpu.r[rb] + (@as(u32, offset) << 2);
+                    const value = bus.read(u32, address);
+                    cpu.r[rd] = rotr(u32, value, 8 * (address & 0x3));
+                }
+            } else {
+                if (B) {
+                    // STRB
+                    const address = cpu.r[rb] + offset;
+                    bus.write(u8, address, @truncate(u8, cpu.r[rd]));
+                } else {
+                    // STR
+                    const address = cpu.r[rb] + (@as(u32, offset) << 2);
+                    bus.write(u32, address, cpu.r[rd]);
+                }
+            }
+        }
+    }.inner;
+}
+
+pub fn fmt10(comptime L: bool, comptime offset: u5) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            const rb = opcode >> 3 & 0x7;
+            const rd = opcode & 0x7;
+
+            const address = cpu.r[rb] + (@as(u6, offset) << 1);
+
+            if (L) {
+                // LDRH
+                const value = bus.read(u16, address);
+                cpu.r[rd] = rotr(u32, value, 8 * (address & 1));
+            } else {
+                // STRH
+                bus.write(u16, address, @truncate(u16, cpu.r[rd]));
+            }
+        }
+    }.inner;
+}
+
+pub fn fmt11(comptime L: bool, comptime rd: u3) InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
+            const offset = (opcode & 0xFF) << 2;
+            const address = cpu.r[13] + offset;
+
+            if (L) {
+                // LDR
+                const value = bus.read(u32, address);
+                cpu.r[rd] = rotr(u32, value, 8 * (address & 0x3));
+            } else {
+                // STR
+                bus.write(u32, address, cpu.r[rd]);
+            }
+        }
+    }.inner;
+}

src/core/cpu/thumb/software_interrupt.zig

@@ -0,0 +1,23 @@
+const Bus = @import("../../Bus.zig");
+const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
+const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
+
+pub fn fmt17() InstrFn {
+    return struct {
+        fn inner(cpu: *Arm7tdmi, _: *Bus, _: u16) void {
+            // Copy Values from Current Mode
+            const ret_addr = cpu.r[15] - 2;
+            const cpsr = cpu.cpsr.raw;
+
+            // Switch Mode
+            cpu.changeMode(.Supervisor);
+            cpu.cpsr.t.write(false); // Force ARM Mode
+            cpu.cpsr.i.write(true); // Disable normal interrupts
+
+            cpu.r[14] = ret_addr; // Resume Execution
+            cpu.spsr.raw = cpsr; // Previous mode CPSR
+            cpu.r[15] = 0x0000_0008;
+            cpu.pipe.reload(cpu);
+        }
+    }.inner;
+}

src/core/emu.zig

@@ -0,0 +1,220 @@
+const std = @import("std");
+const SDL = @import("sdl2");
+const config = @import("../config.zig");
+
+const Scheduler = @import("scheduler.zig").Scheduler;
+const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
+const FpsTracker = @import("../util.zig").FpsTracker;
+
+const Timer = std.time.Timer;
+const Atomic = std.atomic.Atomic;
+
+/// 4 Cycles in 1 dot
+const cycles_per_dot = 4;
+
+/// The GBA draws 228 Horizontal which each consist 308 dots
+/// (note: not all lines are visible)
+const cycles_per_frame = 228 * (308 * cycles_per_dot); //280896
+
+/// The GBA ARM7TDMI runs at 2^24 Hz
+const clock_rate = 1 << 24; // 16.78MHz
+
+/// The # of nanoseconds a frame should take
+const frame_period = (std.time.ns_per_s * cycles_per_frame) / clock_rate;
+
+/// Exact Value:  59.7275005696Hz
+/// The inverse of the frame period
+pub const frame_rate: f64 = @intToFloat(f64, clock_rate) / cycles_per_frame;
+
+const log = std.log.scoped(.Emulation);
+
+const RunKind = enum {
+    Unlimited,
+    UnlimitedFPS,
+    Limited,
+    LimitedFPS,
+};
+
+pub fn run(quit: *Atomic(bool), scheduler: *Scheduler, cpu: *Arm7tdmi, tracker: *FpsTracker) void {
+    const audio_sync = config.config().guest.audio_sync and !config.config().host.mute;
+    if (audio_sync) log.info("Audio sync enabled", .{});
+
+    if (config.config().guest.video_sync) {
+        inner(.LimitedFPS, audio_sync, quit, scheduler, cpu, tracker);
+    } else {
+        inner(.UnlimitedFPS, audio_sync, quit, scheduler, cpu, tracker);
+    }
+}
+
+fn inner(comptime kind: RunKind, audio_sync: bool, quit: *Atomic(bool), scheduler: *Scheduler, cpu: *Arm7tdmi, tracker: ?*FpsTracker) void {
+    if (kind == .UnlimitedFPS or kind == .LimitedFPS) {
+        std.debug.assert(tracker != null);
+        log.info("FPS tracking enabled", .{});
+    }
+
+    switch (kind) {
+        .Unlimited, .UnlimitedFPS => {
+            log.info("Emulation w/out video sync", .{});
+
+            while (!quit.load(.Monotonic)) {
+                runFrame(scheduler, cpu);
+                audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+
+                if (kind == .UnlimitedFPS) tracker.?.tick();
+            }
+        },
+        .Limited, .LimitedFPS => {
+            log.info("Emulation w/ video sync", .{});
+            var timer = Timer.start() catch @panic("failed to initalize std.timer.Timer");
+            var wake_time: u64 = frame_period;
+
+            while (!quit.load(.Monotonic)) {
+                runFrame(scheduler, cpu);
+                const new_wake_time = videoSync(&timer, wake_time);
+
+                // Spin to make up the difference of OS scheduler innacuracies
+                // If we happen to also be syncing to audio, we choose to spin on
+                // the amount of time needed for audio to catch up rather than
+                // our expected wake-up time
+
+                audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+                if (!audio_sync) spinLoop(&timer, wake_time);
+                wake_time = new_wake_time;
+
+                if (kind == .LimitedFPS) tracker.?.tick();
+            }
+        },
+    }
+}
+
+pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
+    const frame_end = sched.tick + cycles_per_frame;
+
+    while (sched.tick < frame_end) {
+        if (!cpu.stepDmaTransfer()) {
+            if (cpu.isHalted()) {
+                // Fast-forward to next Event
+                sched.tick = sched.nextTimestamp();
+            } else {
+                cpu.step();
+            }
+        }
+
+        if (sched.tick >= sched.nextTimestamp()) sched.handleEvent(cpu);
+    }
+}
+
+fn audioSync(audio_sync: bool, stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
+    comptime std.debug.assert(@import("../platform.zig").sample_format == SDL.AUDIO_U16);
+    const sample_size = 2 * @sizeOf(u16);
+    const max_buf_size: c_int = 0x400;
+
+    // Determine whether the APU is busy right at this moment
+    var still_full: bool = SDL.SDL_AudioStreamAvailable(stream) > sample_size * if (is_buffer_full.*) max_buf_size >> 1 else max_buf_size;
+    defer is_buffer_full.* = still_full; // Update APU Busy status right before exiting scope
+
+    // If Busy is false, there's no need to sync here
+    if (!still_full) return;
+
+    while (true) {
+        still_full = SDL.SDL_AudioStreamAvailable(stream) > sample_size * max_buf_size >> 1;
+        if (!audio_sync or !still_full) break;
+    }
+}
+
+fn videoSync(timer: *Timer, wake_time: u64) u64 {
+    // Use the OS scheduler to put the emulation thread to sleep
+    const recalculated = sleep(timer, wake_time);
+
+    // If sleep() determined we need to adjust our wake up time, do so
+    // otherwise predict our next wake up time according to the frame period
+    return recalculated orelse wake_time + frame_period;
+}
+
+// TODO: Better sleep impl?
+fn sleep(timer: *Timer, wake_time: u64) ?u64 {
+    const timestamp = timer.read();
+
+    // ns_late is non zero if we are late.
+    var ns_late = timestamp -| wake_time;
+
+    // If we're more than a frame late, skip the rest of this loop
+    // Recalculate what our new wake time should be so that we can
+    // get "back on track"
+    if (ns_late > frame_period) return timestamp + frame_period;
+    const sleep_for = frame_period - ns_late;
+
+    const step = 2 * std.time.ns_per_ms; // Granularity of 2ms
+    const times = sleep_for / step;
+
+    for (0..times) |_| {
+        std.time.sleep(step);
+
+        // Upon wakeup, check to see if this particular sleep was longer than expected
+        // if so we should exit early, but probably not skip a whole frame period
+        ns_late = timer.read() -| wake_time;
+        if (ns_late > frame_period) return null;
+    }
+
+    return null;
+}
+
+fn spinLoop(timer: *Timer, wake_time: u64) void {
+    while (true) if (timer.read() > wake_time) break;
+}
+
+pub const EmuThing = struct {
+    const Self = @This();
+    const Interface = @import("gdbstub").Emulator;
+    const Allocator = std.mem.Allocator;
+
+    cpu: *Arm7tdmi,
+    scheduler: *Scheduler,
+
+    pub fn init(cpu: *Arm7tdmi, scheduler: *Scheduler) Self {
+        return .{ .cpu = cpu, .scheduler = scheduler };
+    }
+
+    pub fn interface(self: *Self, allocator: Allocator) Interface {
+        return Interface.init(allocator, self);
+    }
+
+    pub fn read(self: *const Self, addr: u32) u8 {
+        return self.cpu.bus.dbgRead(u8, addr);
+    }
+
+    pub fn write(self: *Self, addr: u32, value: u8) void {
+        self.cpu.bus.dbgWrite(u8, addr, value);
+    }
+
+    pub fn registers(self: *const Self) *[16]u32 {
+        return &self.cpu.r;
+    }
+
+    pub fn cpsr(self: *const Self) u32 {
+        return self.cpu.cpsr.raw;
+    }
+
+    pub fn step(self: *Self) void {
+        const cpu = self.cpu;
+        const sched = self.scheduler;
+
+        // Is true when we have executed one (1) instruction
+        var did_step: bool = false;
+
+        // TODO: How can I make it easier to keep this in lock-step with runFrame?
+        while (!did_step) {
+            if (!cpu.stepDmaTransfer()) {
+                if (cpu.isHalted()) {
+                    // Fast-forward to next Event
+                    sched.tick = sched.queue.peek().?.tick;
+                } else {
+                    cpu.step();
+                    did_step = true;
+                }
+            }
+
+            if (sched.tick >= sched.nextTimestamp()) sched.handleEvent(cpu);
+        }
+    }
+};

src/core/ppu/Oam.zig

@@ -0,0 +1,40 @@
+const std = @import("std");
+
+const Allocator = std.mem.Allocator;
+
+const buf_len = 0x400;
+const Self = @This();
+
+buf: []u8,
+allocator: Allocator,
+
+pub fn read(self: *const Self, comptime T: type, address: usize) T {
+    const addr = address & 0x3FF;
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.readIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)]),
+        else => @compileError("OAM: Unsupported read width"),
+    };
+}
+
+pub fn write(self: *Self, comptime T: type, address: usize, value: T) void {
+    const addr = address & 0x3FF;
+
+    switch (T) {
+        u32, u16 => std.mem.writeIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)], value),
+        u8 => return, // 8-bit writes are explicitly ignored
+        else => @compileError("OAM: Unsupported write width"),
+    }
+}
+
+pub fn init(allocator: Allocator) !Self {
+    const buf = try allocator.alloc(u8, buf_len);
+    std.mem.set(u8, buf, 0);
+
+    return Self{ .buf = buf, .allocator = allocator };
+}
+
+pub fn deinit(self: *Self) void {
+    self.allocator.free(self.buf);
+    self.* = undefined;
+}

src/core/ppu/Palette.zig

@@ -0,0 +1,47 @@
+const std = @import("std");
+
+const Allocator = std.mem.Allocator;
+
+const buf_len = 0x400;
+const Self = @This();
+
+buf: []u8,
+allocator: Allocator,
+
+pub fn read(self: *const Self, comptime T: type, address: usize) T {
+    const addr = address & 0x3FF;
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.readIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)]),
+        else => @compileError("PALRAM: Unsupported read width"),
+    };
+}
+
+pub fn write(self: *Self, comptime T: type, address: usize, value: T) void {
+    const addr = address & 0x3FF;
+
+    switch (T) {
+        u32, u16 => std.mem.writeIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)], value),
+        u8 => {
+            const align_addr = addr & ~@as(u32, 1); // Aligned to Halfword boundary
+            std.mem.writeIntSliceLittle(u16, self.buf[align_addr..][0..@sizeOf(u16)], @as(u16, value) * 0x101);
+        },
+        else => @compileError("PALRAM: Unsupported write width"),
+    }
+}
+
+pub fn init(allocator: Allocator) !Self {
+    const buf = try allocator.alloc(u8, buf_len);
+    std.mem.set(u8, buf, 0);
+
+    return Self{ .buf = buf, .allocator = allocator };
+}
+
+pub fn deinit(self: *Self) void {
+    self.allocator.free(self.buf);
+    self.* = undefined;
+}
+
+pub inline fn backdrop(self: *const Self) u16 {
+    return std.mem.readIntNative(u16, self.buf[0..2]);
+}

src/core/ppu/Vram.zig

@@ -0,0 +1,60 @@
+const std = @import("std");
+const io = @import("../bus/io.zig");
+
+const Allocator = std.mem.Allocator;
+
+const buf_len = 0x18000;
+const Self = @This();
+
+buf: []u8,
+allocator: Allocator,
+
+pub fn read(self: *const Self, comptime T: type, address: usize) T {
+    const addr = Self.mirror(address);
+
+    return switch (T) {
+        u32, u16, u8 => std.mem.readIntSliceLittle(T, self.buf[addr..][0..@sizeOf(T)]),
+        else => @compileError("VRAM: Unsupported read width"),
+    };
+}
+
+pub fn write(self: *Self, comptime T: type, dispcnt: io.DisplayControl, address: usize, value: T) void {
+    const mode: u3 = dispcnt.bg_mode.read();
+    const idx = Self.mirror(address);
+
+    switch (T) {
+        u32, u16 => std.mem.writeIntSliceLittle(T, self.buf[idx..][0..@sizeOf(T)], value),
+        u8 => {
+            // Ignore write if it falls within the boundaries of OBJ VRAM
+            switch (mode) {
+                0, 1, 2 => if (0x0001_0000 <= idx) return,
+                else => if (0x0001_4000 <= idx) return,
+            }
+
+            const align_idx = idx & ~@as(u32, 1); // Aligned to a halfword boundary
+            std.mem.writeIntSliceLittle(u16, self.buf[align_idx..][0..@sizeOf(u16)], @as(u16, value) * 0x101);
+        },
+        else => @compileError("VRAM: Unsupported write width"),
+    }
+}
+
+pub fn init(allocator: Allocator) !Self {
+    const buf = try allocator.alloc(u8, buf_len);
+    std.mem.set(u8, buf, 0);
+
+    return Self{ .buf = buf, .allocator = allocator };
+}
+
+pub fn deinit(self: *Self) void {
+    self.allocator.free(self.buf);
+    self.* = undefined;
+}
+
+pub fn mirror(address: usize) usize {
+    // Mirrored in steps of 128K (64K + 32K + 32K) (abcc)
+    const addr = address & 0x1FFFF;
+
+    // If the address is within 96K we don't do anything,
+    // otherwise we want to mirror the last 32K (addresses between 64K and 96K)
+    return if (addr < buf_len) addr else 0x10000 + (addr & 0x7FFF);
+}

src/core/scheduler.zig

@@ -0,0 +1,120 @@
+const std = @import("std");
+
+const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
+const Clock = @import("bus/gpio.zig").Clock;
+
+const Order = std.math.Order;
+const PriorityQueue = std.PriorityQueue;
+const Allocator = std.mem.Allocator;
+const log = std.log.scoped(.Scheduler);
+
+pub const Scheduler = struct {
+    const Self = @This();
+
+    tick: u64,
+    queue: PriorityQueue(Event, void, lessThan),
+
+    pub fn init(allocator: Allocator) Self {
+        var sched = Self{ .tick = 0, .queue = PriorityQueue(Event, void, lessThan).init(allocator, {}) };
+        sched.queue.add(.{ .kind = .HeatDeath, .tick = std.math.maxInt(u64) }) catch unreachable;
+
+        return sched;
+    }
+
+    pub fn deinit(self: *Self) void {
+        self.queue.deinit();
+        self.* = undefined;
+    }
+
+    pub inline fn now(self: *const Self) u64 {
+        return self.tick;
+    }
+
+    pub fn handleEvent(self: *Self, cpu: *Arm7tdmi) void {
+        const event = self.queue.remove();
+        const late = self.tick - event.tick;
+
+        switch (event.kind) {
+            .HeatDeath => {
+                log.err("u64 overflow. This *actually* should never happen.", .{});
+                unreachable;
+            },
+            .Draw => {
+                // The end of a VDraw
+                cpu.bus.ppu.drawScanline();
+                cpu.bus.ppu.onHdrawEnd(cpu, late);
+            },
+            .TimerOverflow => |id| {
+                switch (id) {
+                    inline 0...3 => |idx| cpu.bus.tim[idx].onTimerExpire(cpu, late),
+                }
+            },
+            .ApuChannel => |id| {
+                switch (id) {
+                    0 => cpu.bus.apu.ch1.onToneSweepEvent(late),
+                    1 => cpu.bus.apu.ch2.onToneEvent(late),
+                    2 => cpu.bus.apu.ch3.onWaveEvent(late),
+                    3 => cpu.bus.apu.ch4.onNoiseEvent(late),
+                }
+            },
+            .RealTimeClock => {
+                const device = &cpu.bus.pak.gpio.device;
+                if (device.kind != .Rtc or device.ptr == null) return;
+
+                const clock = @ptrCast(*Clock, @alignCast(@alignOf(*Clock), device.ptr.?));
+                clock.onClockUpdate(late);
+            },
+            .FrameSequencer => cpu.bus.apu.onSequencerTick(late),
+            .SampleAudio => cpu.bus.apu.sampleAudio(late),
+            .HBlank => cpu.bus.ppu.onHblankEnd(cpu, late), // The end of a HBlank
+            .VBlank => cpu.bus.ppu.onHdrawEnd(cpu, late), // The end of a VBlank
+        }
+    }
+
+    /// Removes the **first** scheduled event of type `needle`
+    pub fn removeScheduledEvent(self: *Self, needle: EventKind) void {
+        for (self.queue.items, 0..) |event, i| {
+            if (std.meta.eql(event.kind, needle)) {
+
+                // invalidates the slice we're iterating over
+                _ = self.queue.removeIndex(i);
+
+                log.debug("Removed {?}@{}", .{ event.kind, event.tick });
+                break;
+            }
+        }
+    }
+
+    pub fn push(self: *Self, kind: EventKind, end: u64) void {
+        self.queue.add(.{ .kind = kind, .tick = self.now() + end }) catch unreachable;
+    }
+
+    pub inline fn nextTimestamp(self: *const Self) u64 {
+        @setRuntimeSafety(false);
+
+        // Typically you'd use PriorityQueue.peek here, but there's always at least a HeatDeath
+        // event in the PQ so we can just do this instead. Should be faster in ReleaseSafe
+        return self.queue.items[0].tick;
+    }
+};
+
+pub const Event = struct {
+    kind: EventKind,
+    tick: u64,
+};
+
+fn lessThan(_: void, a: Event, b: Event) Order {
+    return std.math.order(a.tick, b.tick);
+}
+
+pub const EventKind = union(enum) {
+    HeatDeath,
+    HBlank,
+    VBlank,
+    Draw,
+    TimerOverflow: u2,
+    SampleAudio,
+    FrameSequencer,
+    ApuChannel: u2,
+    RealTimeClock,
+};

src/cpu.zig

@@ -1,189 +0,0 @@
-const std = @import("std");
-const util = @import("util.zig");
-
-const BarrelShifter = @import("cpu/barrel_shifter.zig");
-const Bus = @import("Bus.zig");
-const Bit = @import("bitfield").Bit;
-const Bitfield = @import("bitfield").Bitfield;
-const Scheduler = @import("scheduler.zig").Scheduler;
-
-const dataProcessing = @import("cpu/data_processing.zig").dataProcessing;
-const singleDataTransfer = @import("cpu/single_data_transfer.zig").singleDataTransfer;
-const halfAndSignedDataTransfer = @import("cpu/half_signed_data_transfer.zig").halfAndSignedDataTransfer;
-const branch = @import("cpu/branch.zig").branch;
-
-pub const InstrFn = fn (*Arm7tdmi, *Bus, u32) void;
-const arm_lut: [0x1000]InstrFn = populate();
-
-pub const Arm7tdmi = struct {
-    r: [16]u32,
-    sched: *Scheduler,
-    bus: *Bus,
-    cpsr: CPSR,
-
-    pub fn init(sched: *Scheduler, bus: *Bus) @This() {
-        return .{
-            .r = [_]u32{0x00} ** 16,
-            .sched = sched,
-            .bus = bus,
-            .cpsr = .{ .raw = 0x0000_00DF },
-        };
-    }
-
-    pub fn skipBios(self: *@This()) void {
-        self.r[0] = 0x08000000;
-        self.r[1] = 0x000000EA;
-        // GPRs 2 -> 12 *should* already be 0 initialized
-        self.r[13] = 0x0300_7F00;
-        self.r[14] = 0x0000_0000;
-        self.r[15] = 0x0800_0000;
-
-        // TODO: Set sp_irq = 0x0300_7FA0, sp_svc = 0x0300_7FE0
-
-        self.cpsr.raw = 0x6000001F;
-    }
-
-    pub inline fn step(self: *@This()) u64 {
-        const opcode = self.fetch();
-        // self.mgbaLog(opcode);
-
-        if (checkCond(&self.cpsr, opcode)) arm_lut[armIdx(opcode)](self, self.bus, opcode);
-        return 1;
-    }
-
-    fn fetch(self: *@This()) u32 {
-        const word = self.bus.read32(self.r[15]);
-        self.r[15] += 4;
-        return word;
-    }
-
-    pub fn fakePC(self: *const @This()) u32 {
-        return self.r[15] + 4;
-    }
-
-    fn mgbaLog(self: *const @This(), opcode: u32) void {
-        const stderr = std.io.getStdErr().writer();
-        std.debug.getStderrMutex().lock();
-        defer std.debug.getStderrMutex().unlock();
-
-        const r0 = self.r[0];
-        const r1 = self.r[1];
-        const r2 = self.r[2];
-        const r3 = self.r[3];
-        const r4 = self.r[4];
-        const r5 = self.r[5];
-        const r6 = self.r[6];
-        const r7 = self.r[7];
-        const r8 = self.r[8];
-        const r9 = self.r[9];
-        const r10 = self.r[10];
-        const r11 = self.r[11];
-        const r12 = self.r[12];
-        const r13 = self.r[13];
-        const r14 = self.r[14];
-        const r15 = self.r[15];
-
-        const cpsr = self.cpsr.raw;
-
-        nosuspend stderr.print("{X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} cpsr: {X:0>8} | {X:0>8}:\n", .{ r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15, cpsr, opcode }) catch return;
-    }
-};
-
-fn armIdx(opcode: u32) u12 {
-    return @truncate(u12, opcode >> 20 & 0xFF) << 4 | @truncate(u12, opcode >> 4 & 0xF);
-}
-
-fn checkCond(cpsr: *const CPSR, opcode: u32) bool {
-    // TODO: Should I implement an enum?
-    return switch (@truncate(u4, opcode >> 28)) {
-        0x0 => cpsr.z.read(), // EQ - Equal
-        0x1 => !cpsr.z.read(), // NEQ - Not equal
-        0x2 => cpsr.c.read(), // CS - Unsigned higher or same
-        0x3 => !cpsr.c.read(), // CC - Unsigned lower
-        0x4 => cpsr.n.read(), // MI - Negative
-        0x5 => !cpsr.n.read(), // PL - Positive or zero
-        0x6 => cpsr.v.read(), // VS - Overflow
-        0x7 => !cpsr.v.read(), // VC - No overflow
-        0x8 => cpsr.c.read() and !cpsr.z.read(), // HI - unsigned higher
-        0x9 => !cpsr.c.read() and cpsr.z.read(), // LS - unsigned lower or same
-        0xA => cpsr.n.read() == cpsr.v.read(), // GE - Greater or equal
-        0xB => cpsr.n.read() != cpsr.v.read(), // LT - Less than
-        0xC => !cpsr.z.read() and (cpsr.n.read() == cpsr.z.read()), // GT - Greater than
-        0xD => cpsr.z.read() or (cpsr.n.read() != cpsr.v.read()), // LE - Less than or equal
-        0xE => true, // AL - Always
-        0xF => std.debug.panic("[CPU] 0xF is a reserved condition field", .{}),
-    };
-}
-
-fn populate() [0x1000]InstrFn {
-    return comptime {
-        @setEvalBranchQuota(0x5000); // TODO: Figure out exact size
-        var lut = [_]InstrFn{undefinedInstruction} ** 0x1000;
-
-        var i: usize = 0;
-        while (i < lut.len) : (i += 1) {
-            if (i >> 10 & 0x3 == 0b00) {
-                const I = i >> 9 & 1 == 1;
-                const S = i >> 4 & 1 == 1;
-                const instrKind = i >> 5 & 0xF;
-
-                lut[i] = dataProcessing(I, S, instrKind);
-            }
-
-            if (i >> 9 & 0x7 == 0b000 and i >> 3 & 1 == 1 and i & 1 == 1) {
-                const P = i >> 8 & 1 == 1;
-                const U = i >> 7 & 1 == 1;
-                const I = i >> 6 & 1 == 1;
-                const W = i >> 5 & 1 == 1;
-                const L = i >> 4 & 1 == 1;
-
-                lut[i] = halfAndSignedDataTransfer(P, U, I, W, L);
-            }
-
-            if (i >> 10 & 0x3 == 0b01) {
-                const I = i >> 9 & 1 == 1;
-                const P = i >> 8 & 1 == 1;
-                const U = i >> 7 & 1 == 1;
-                const B = i >> 6 & 1 == 1;
-                const W = i >> 5 & 1 == 1;
-                const L = i >> 4 & 1 == 1;
-
-                lut[i] = singleDataTransfer(I, P, U, B, W, L);
-            }
-
-            if (i >> 9 & 0x7 == 0b101) {
-                const L = i >> 8 & 1 == 1;
-                lut[i] = branch(L);
-            }
-        }
-
-        return lut;
-    };
-}
-
-pub const CPSR = extern union {
-    mode: Bitfield(u32, 0, 5),
-    t: Bit(u32, 5),
-    f: Bit(u32, 6),
-    i: Bit(u32, 7),
-    v: Bit(u32, 28),
-    c: Bit(u32, 29),
-    z: Bit(u32, 30),
-    n: Bit(u32, 31),
-    raw: u32,
-};
-
-const Mode = enum(u5) {
-    User = 0b10000,
-    FIQ = 0b10001,
-    IRQ = 0b10010,
-    Supervisor = 0b10011,
-    Abort = 0b10111,
-    Undefined = 0b11011,
-    System = 0b11111,
-};
-
-fn undefinedInstruction(_: *Arm7tdmi, _: *Bus, opcode: u32) void {
-    const id = armIdx(opcode);
-    std.debug.panic("[CPU] {{0x{X:}}} 0x{X:} is an illegal opcode", .{ id, opcode });
-}

src/cpu/barrel_shifter.zig

@@ -1,92 +0,0 @@
-const std = @import("std");
-
-const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
-const CPSR = @import("../cpu.zig").CPSR;
-
-pub inline fn exec(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
-    var shift_amt: u8 = undefined;
-    if (opcode >> 4 & 1 == 1) {
-        shift_amt = @truncate(u8, cpu.r[opcode >> 8 & 0xF]);
-    } else {
-        shift_amt = @truncate(u8, opcode >> 7 & 0x1F);
-    }
-
-    const rm = cpu.r[opcode & 0xF];
-
-    if (S) {
-        return switch (@truncate(u2, opcode >> 5)) {
-            0b00 => logical_left(&cpu.cpsr, rm, shift_amt),
-            0b01 => logical_right(&cpu.cpsr, rm, shift_amt),
-            0b10 => arithmetic_right(&cpu.cpsr, rm, shift_amt),
-            0b11 => rotate_right(&cpu.cpsr, rm, shift_amt),
-        };
-    } else {
-        var dummy = CPSR{ .raw = 0x0000_0000 };
-        return switch (@truncate(u2, opcode >> 5)) {
-            0b00 => logical_left(&dummy, rm, shift_amt),
-            0b01 => logical_right(&dummy, rm, shift_amt),
-            0b10 => arithmetic_right(&dummy, rm, shift_amt),
-            0b11 => rotate_right(&dummy, rm, shift_amt),
-        };
-    }
-}
-
-pub inline fn logical_left(cpsr: *CPSR, rm: u32, shift_byte: u8) u32 {
-    const shift_amt = @truncate(u5, shift_byte);
-    const bit_count: u8 = @typeInfo(u32).Int.bits;
-
-    var result: u32 = 0x0000_0000;
-
-    if (shift_byte < bit_count) {
-        // We can perform a well-defined shift here
-
-        // FIXME: We assume cpu.r[rs] == 0 and imm_shift == 0 are equivalent
-        if (shift_amt != 0) {
-            const carry_bit = @truncate(u5, bit_count - shift_amt);
-            cpsr.c.write(rm >> carry_bit & 1 == 1);
-        }
-
-        result = rm << shift_amt;
-    } else if (shift_byte == bit_count) {
-        // Shifted all bits out, carry bit is bit 0 of rm
-        cpsr.c.write(rm & 1 == 1);
-    } else {
-        // Shifted all bits out, carry bit has also been shifted out
-        cpsr.c.write(false);
-    }
-
-    return result;
-}
-
-pub inline fn logical_right(cpsr: *CPSR, rm: u32, shift_byte: u8) u32 {
-    const shift_amt = @truncate(u5, shift_byte);
-    const bit_count: u8 = @typeInfo(u32).Int.bits;
-
-    var result: u32 = 0x0000_0000;
-
-    if (shift_byte == 0 or shift_byte == bit_count) {
-        // Actualy LSR #32
-        cpsr.c.write(rm >> 31 & 1 == 1);
-    } else if (shift_byte < bit_count) {
-        // We can perform a well-defined shift
-        const carry_bit = shift_amt - 1;
-        cpsr.c.write(rm >> carry_bit & 1 == 1);
-
-        result = rm >> shift_amt;
-    } else {
-        // All bits have been shifted out, including carry bit
-        cpsr.c.write(false);
-    }
-
-    return result;
-}
-
-pub fn arithmetic_right(_: *CPSR, _: u32, _: u8) u32 {
-    // @bitCast(u32, @bitCast(i32, r_val) >> @truncate(u5, amount))
-    std.debug.panic("[BarrelShifter] implement arithmetic shift right", .{});
-}
-
-pub fn rotate_right(_: *CPSR, _: u32, _: u8) u32 {
-    // std.math.rotr(u32, r_val, amount)
-    std.debug.panic("[BarrelShifter] implement rotate right", .{});
-}

src/cpu/branch.zig

@@ -1,17 +0,0 @@
-const util = @import("../util.zig");
-
-const Bus = @import("../Bus.zig");
-const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
-const InstrFn = @import("../cpu.zig").InstrFn;
-
-pub fn branch(comptime L: bool) InstrFn {
-    return struct {
-        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
-            if (L) {
-                cpu.r[14] = cpu.r[15] - 4;
-            }
-
-            cpu.r[15] = cpu.fakePC() +% util.u32SignExtend(24, opcode << 2);
-        }
-    }.inner;
-}

src/cpu/data_processing.zig

@@ -1,71 +0,0 @@
-const std = @import("std");
-
-const BarrelShifter = @import("barrel_shifter.zig");
-const Bus = @import("../Bus.zig");
-const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
-const InstrFn = @import("../cpu.zig").InstrFn;
-
-pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime instrKind: u4) InstrFn {
-    return struct {
-        fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
-            const rd = opcode >> 12 & 0xF;
-            const op1 = opcode >> 16 & 0xF;
-
-            var op2: u32 = undefined;
-            if (I) {
-                op2 = std.math.rotr(u32, opcode & 0xFF, (opcode >> 8 & 0xF) << 1);
-            } else {
-                if (S and rd == 0xF) {
-                    std.debug.panic("[CPU] Data Processing Instruction w/ S set and Rd == 15", .{});
-                } else {
-                    op2 = BarrelShifter.exec(S, cpu, opcode);
-                }
-            }
-
-            switch (instrKind) {
-                0x4 => {
-                    // ADD
-                    var result: u32 = undefined;
-                    const didOverflow = @addWithOverflow(u32, cpu.r[op1], op2, &result);
-                    cpu.r[rd] = result;
-
-                    if (S and rd != 0xF) {
-                        cpu.cpsr.n.write(result >> 31 & 1 == 1);
-                        cpu.cpsr.z.write(result == 0);
-                        cpu.cpsr.c.write(didOverflow);
-                        cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
-                    }
-                },
-                0x8 => {
-                    // TST
-                    const result = cpu.r[op1] & op2;
-
-                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
-                    cpu.cpsr.z.write(result == 0);
-                    // Barrel Shifter should always calc CPSR C in TST
-                    if (!S) _ = BarrelShifter.exec(true, cpu, opcode);
-                },
-                0xD => {
-                    // MOV
-                    cpu.r[rd] = op2;
-
-                    if (S and rd != 0xF) {
-                        cpu.cpsr.n.write(op2 >> 31 & 1 == 1);
-                        cpu.cpsr.z.write(op2 == 0);
-                        // C set by Barr0x15el Shifter, V is unnafected
-                    }
-                },
-                0xA => {
-                    // CMP
-                    const result = cpu.r[op1] -% op2;
-
-                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
-                    cpu.cpsr.z.write(result == 0);
-                    cpu.cpsr.c.write(op2 <= cpu.r[op1]);
-                    cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
-                },
-                else => std.debug.panic("[CPU] TODO: implement data processing type {}", .{instrKind}),
-            }
-        }
-    }.inner;
-}

src/cpu/half_signed_data_transfer.zig

@@ -1,62 +0,0 @@
-const std = @import("std");
-const util = @import("../util.zig");
-
-const Bus = @import("../Bus.zig");
-const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
-const InstrFn = @import("../cpu.zig").InstrFn;
-
-pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I: bool, comptime W: bool, comptime L: bool) InstrFn {
-    return struct {
-        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
-            const rn = opcode >> 16 & 0xF;
-            const rd = opcode >> 12 & 0xF;
-            const rm = opcode & 0xF;
-            const imm_offset_high = opcode >> 8 & 0xF;
-
-            const base = cpu.r[rn];
-
-            var offset: u32 = undefined;
-            if (I) {
-                offset = imm_offset_high << 4 | rm;
-            } else {
-                offset = cpu.r[rm];
-            }
-
-            const modified_base = if (U) base + offset else base - offset;
-            var address = if (P) modified_base else base;
-
-            if (L) {
-                switch (@truncate(u2, opcode >> 5)) {
-                    0b00 => {
-                        // SWP
-                        std.debug.panic("[CPU] TODO: Implement SWP", .{});
-                    },
-                    0b01 => {
-                        // LDRH
-                        cpu.r[rd] = bus.read16(address);
-                    },
-                    0b10 => {
-                        // LDRSB
-                        cpu.r[rd] = util.u32SignExtend(8, @as(u32, bus.read8(address)));
-                        std.debug.panic("TODO: Affect the CPSR", .{});
-                    },
-                    0b11 => {
-                        // LDRSH
-                        cpu.r[rd] = util.u32SignExtend(16, @as(u32, bus.read16(address)));
-                        std.debug.panic("TODO: Affect the CPSR", .{});
-                    },
-                }
-            } else {
-                if (opcode >> 5 & 0x01 == 0x01) {
-                    // STRH
-                    bus.write16(address, @truncate(u16, cpu.r[rd]));
-                } else {
-                    std.debug.panic("[CPU] TODO: Figure out if this is also SWP", .{});
-                }
-            }
-
-            address = modified_base;
-            if (W and P or !P) cpu.r[rn] = address;
-        }
-    }.inner;
-}

src/cpu/single_data_transfer.zig

@@ -1,64 +0,0 @@
-const std = @import("std");
-const util = @import("../util.zig");
-
-const BarrelShifter = @import("barrel_shifter.zig");
-const Bus = @import("../Bus.zig");
-const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
-const CPSR = @import("../cpu.zig").CPSR;
-const InstrFn = @import("../cpu.zig").InstrFn;
-
-pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool, comptime B: bool, comptime W: bool, comptime L: bool) InstrFn {
-    return struct {
-        fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
-            const rn = opcode >> 16 & 0xF;
-            const rd = opcode >> 12 & 0xF;
-
-            const base = cpu.r[rn];
-            const offset = if (I) registerOffset(cpu, opcode) else opcode & 0xFFF;
-
-            const modified_base = if (U) base + offset else base - offset;
-            var address = if (P) modified_base else base;
-
-            if (L) {
-                if (B) {
-                    // LDRB
-                    cpu.r[rd] = bus.read8(address);
-                } else {
-                    // LDR
-
-                    // FIXME: Unsure about how I calculate the boundary offset
-                    cpu.r[rd] = std.math.rotl(u32, bus.read32(address), address % 4);
-                }
-            } else {
-                if (B) {
-                    // STRB
-                    bus.write8(address, @truncate(u8, cpu.r[rd]));
-                } else {
-                    // STR
-                    const force_aligned = address & 0xFFFF_FFFC;
-                    bus.write32(force_aligned, cpu.r[rd]);
-                }
-            }
-
-            address = modified_base;
-            if (W and P or !P) cpu.r[rn] = address;
-
-            // TODO: W-bit forces non-privledged mode for the transfer
-        }
-    }.inner;
-}
-
-fn registerOffset(cpu: *Arm7tdmi, opcode: u32) u32 {
-    const shift_byte = @truncate(u8, opcode >> 7 & 0x1F);
-
-    const rm = cpu.r[opcode & 0xF];
-
-    var dummy = CPSR{ .raw = 0x0000_0000 };
-
-    return switch (@truncate(u2, opcode >> 5)) {
-        0b00 => BarrelShifter.logical_left(&dummy, rm, shift_byte),
-        0b01 => BarrelShifter.logical_right(&dummy, rm, shift_byte),
-        0b10 => BarrelShifter.arithmetic_right(&dummy, rm, shift_byte),
-        0b11 => BarrelShifter.rotate_right(&dummy, rm, shift_byte),
-    };
-}

src/emu.zig

@@ -1,17 +0,0 @@
-const Bus = @import("Bus.zig");
-const Scheduler = @import("scheduler.zig").Scheduler;
-const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
-
-const cycles_per_frame: u64 = 100; // TODO: How many cycles actually?
-
-pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi, bus: *Bus) void {
-    var cycles: u64 = 0;
-    while (cycles < cycles_per_frame) : (cycles += 1) {
-        sched.tick += 1;
-        _ = cpu.step();
-
-        while (sched.tick >= sched.nextTimestamp()) {
-            sched.handleEvent(cpu, bus);
-        }
-    }
-}

src/main.zig

@@ -1,83 +1,200 @@
 const std = @import("std");
-const SDL = @import("sdl2");
+const builtin = @import("builtin");
+const known_folders = @import("known_folders");
+const clap = @import("clap");
 
-const emu = @import("emu.zig");
-const Bus = @import("Bus.zig");
-const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
-const Scheduler = @import("scheduler.zig").Scheduler;
+const config = @import("config.zig");
+const emu = @import("core/emu.zig");
 
-const buf_pitch = @import("ppu.zig").buf_pitch;
+const Gui = @import("platform.zig").Gui;
+const Bus = @import("core/Bus.zig");
+const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
+const Scheduler = @import("core/scheduler.zig").Scheduler;
+const FilePaths = @import("util.zig").FilePaths;
+const FpsTracker = @import("util.zig").FpsTracker;
+const Allocator = std.mem.Allocator;
+const Atomic = std.atomic.Atomic;
 
-pub fn main() anyerror!void {
-    // Allocator for Emulator + CLI Aruments
+const log = std.log.scoped(.Cli);
+const width = @import("core/ppu.zig").width;
+const height = @import("core/ppu.zig").height;
+pub const log_level = if (builtin.mode != .Debug) .info else std.log.default_level;
+
+// CLI Arguments + Help Text
+const params = clap.parseParamsComptime(
+    \\-h, --help            Display this help and exit.
+    \\-s, --skip            Skip BIOS.
+    \\-b, --bios <str>      Optional path to a GBA BIOS ROM.
+    \\ --gdb                Run ZBA from the context of a GDB Server
+    \\<str>                 Path to the GBA GamePak ROM.
+    \\
+);
+
+pub fn main() void {
+    // Main Allocator for ZBA
     var gpa = std.heap.GeneralPurposeAllocator(.{}){};
-    const alloc = gpa.allocator();
     defer std.debug.assert(!gpa.deinit());
 
-    // Handle CLI Arguments
-    const args = try std.process.argsAlloc(alloc);
-    defer std.process.argsFree(alloc, args);
+    const allocator = gpa.allocator();
 
-    const zba_args: []const []const u8 = args[1..];
+    // Determine the Data Directory (stores saves)
+    const data_path = blk: {
+        const result = known_folders.getPath(allocator, .data);
+        const option = result catch |e| exitln("interrupted while determining the data folder: {}", .{e});
+        const path = option orelse exitln("no valid data folder found", .{});
+        ensureDataDirsExist(path) catch |e| exitln("failed to create folders under \"{s}\": {}", .{ path, e });
 
-    if (zba_args.len == 0) {
-        std.log.err("Expected PATH to Gameboy Advance ROM as a CLI argument", .{});
-        return;
-    } else if (zba_args.len > 1) {
-        std.log.err("Too many CLI arguments were provided", .{});
-        return;
-    }
+        break :blk path;
+    };
+    defer allocator.free(data_path);
 
-    // Initialize Emulator
-    var scheduler = Scheduler.init(alloc);
+    // Determine the Config Directory
+    const config_path = blk: {
+        const result = known_folders.getPath(allocator, .roaming_configuration);
+        const option = result catch |e| exitln("interreupted while determining the config folder: {}", .{e});
+        const path = option orelse exitln("no valid config folder found", .{});
+        ensureConfigDirExists(path) catch |e| exitln("failed to create required folder \"{s}\": {}", .{ path, e });
+
+        break :blk path;
+    };
+    defer allocator.free(config_path);
+
+    // Parse CLI
+    const result = clap.parse(clap.Help, &params, clap.parsers.default, .{}) catch |e| exitln("failed to parse cli: {}", .{e});
+    defer result.deinit();
+
+    // TODO: Move config file to XDG Config directory?
+    const cfg_file_path = configFilePath(allocator, config_path) catch |e| exitln("failed to ready config file for access: {}", .{e});
+    defer allocator.free(cfg_file_path);
+
+    config.load(allocator, cfg_file_path) catch |e| exitln("failed to load config file: {}", .{e});
+
+    const paths = handleArguments(allocator, data_path, &result) catch |e| exitln("failed to handle cli arguments: {}", .{e});
+    defer if (paths.save) |path| allocator.free(path);
+
+    const log_file = if (config.config().debug.cpu_trace) blk: {
+        break :blk std.fs.cwd().createFile("zba.log", .{}) catch |e| exitln("failed to create trace log file: {}", .{e});
+    } else null;
+    defer if (log_file) |file| file.close();
+
+    // TODO: Take Emulator Init Code out of main.zig
+    var scheduler = Scheduler.init(allocator);
     defer scheduler.deinit();
 
-    var bus = try Bus.init(alloc, &scheduler, zba_args[0]);
+    var bus: Bus = undefined;
+    var cpu = Arm7tdmi.init(&scheduler, &bus, log_file);
+
+    bus.init(allocator, &scheduler, &cpu, paths) catch |e| exitln("failed to init zba bus: {}", .{e});
     defer bus.deinit();
 
-    var cpu = Arm7tdmi.init(&scheduler, &bus);
-    cpu.skipBios();
+    if (config.config().guest.skip_bios or result.args.skip or paths.bios == null) {
+        cpu.fastBoot();
+    }
 
-    // Initialize SDL
-    const status = SDL.SDL_Init(SDL.SDL_INIT_VIDEO | SDL.SDL_INIT_EVENTS | SDL.SDL_INIT_AUDIO);
-    if (status < 0) sdlPanic();
-    defer SDL.SDL_Quit();
+    var quit = Atomic(bool).init(false);
+    var gui = Gui.init(&bus.pak.title, &bus.apu, width, height) catch |e| exitln("failed to init gui: {}", .{e});
+    defer gui.deinit();
 
-    var window = SDL.SDL_CreateWindow(
-        "Gameboy Advance Emulator",
-        SDL.SDL_WINDOWPOS_CENTERED,
-        SDL.SDL_WINDOWPOS_CENTERED,
-        240 * 3,
-        160 * 3,
-        SDL.SDL_WINDOW_SHOWN,
-    ) orelse sdlPanic();
-    defer SDL.SDL_DestroyWindow(window);
+    if (result.args.gdb) {
+        const Server = @import("gdbstub").Server;
+        const EmuThing = @import("core/emu.zig").EmuThing;
 
-    var renderer = SDL.SDL_CreateRenderer(window, -1, SDL.SDL_RENDERER_ACCELERATED) orelse sdlPanic();
-    defer SDL.SDL_DestroyRenderer(renderer);
+        var wrapper = EmuThing.init(&cpu, &scheduler);
+        var emulator = wrapper.interface(allocator);
+        defer emulator.deinit();
 
-    const texture = SDL.SDL_CreateTexture(renderer, SDL.SDL_PIXELFORMAT_BGR555, SDL.SDL_TEXTUREACCESS_STREAMING, 240, 160) orelse sdlPanic();
-    defer SDL.SDL_DestroyTexture(texture);
+        log.info("Ready to connect", .{});
 
-    emu_loop: while (true) {
-        emu.runFrame(&scheduler, &cpu, &bus);
+        var server = Server.init(emulator) catch |e| exitln("failed to init gdb server: {}", .{e});
+        defer server.deinit(allocator);
 
-        var event: SDL.SDL_Event = undefined;
-        _ = SDL.SDL_PollEvent(&event);
+        log.info("Starting GDB Server Thread", .{});
 
-        switch (event.type) {
-            SDL.SDL_QUIT => break :emu_loop,
-            else => {},
-        }
+        const thread = std.Thread.spawn(.{}, Server.run, .{ &server, allocator, &quit }) catch |e| exitln("gdb server thread crashed: {}", .{e});
+        defer thread.join();
 
-        const buf_ptr = bus.ppu.frame_buf.ptr;
-        _ = SDL.SDL_UpdateTexture(texture, null, buf_ptr, buf_pitch);
-        _ = SDL.SDL_RenderCopy(renderer, texture, null, null);
-        SDL.SDL_RenderPresent(renderer);
+        gui.run(.{
+            .cpu = &cpu,
+            .scheduler = &scheduler,
+            .quit = &quit,
+        }) catch |e| exitln("main thread panicked: {}", .{e});
+    } else {
+        var tracker = FpsTracker.init();
+
+        const thread = std.Thread.spawn(.{}, emu.run, .{ &quit, &scheduler, &cpu, &tracker }) catch |e| exitln("emu thread panicked: {}", .{e});
+        defer thread.join();
+
+        gui.run(.{
+            .cpu = &cpu,
+            .scheduler = &scheduler,
+            .tracker = &tracker,
+            .quit = &quit,
+        }) catch |e| exitln("main thread panicked: {}", .{e});
     }
 }
 
-fn sdlPanic() noreturn {
-    const str = @as(?[*:0]const u8, SDL.SDL_GetError()) orelse "unknown error";
-    @panic(std.mem.sliceTo(str, 0));
+fn handleArguments(allocator: Allocator, data_path: []const u8, result: *const clap.Result(clap.Help, &params, clap.parsers.default)) !FilePaths {
+    const rom_path = romPath(result);
+    log.info("ROM path: {s}", .{rom_path});
+
+    const bios_path = result.args.bios;
+    if (bios_path) |path| log.info("BIOS path: {s}", .{path}) else log.warn("No BIOS provided", .{});
+
+    const save_path = try std.fs.path.join(allocator, &[_][]const u8{ data_path, "zba", "save" });
+    log.info("Save path: {s}", .{save_path});
+
+    return .{
+        .rom = rom_path,
+        .bios = bios_path,
+        .save = save_path,
+    };
+}
+
+fn configFilePath(allocator: Allocator, config_path: []const u8) ![]const u8 {
+    const path = try std.fs.path.join(allocator, &[_][]const u8{ config_path, "zba", "config.toml" });
+    errdefer allocator.free(path);
+
+    // We try to create the file exclusively, meaning that we err out if the file already exists.
+    // All we care about is a file being there so we can just ignore that error in particular and
+    // continue down the happy pathj
+    std.fs.accessAbsolute(path, .{}) catch |e| {
+        if (e != error.FileNotFound) return e;
+
+        const config_file = std.fs.createFileAbsolute(path, .{}) catch |err| exitln("failed to create \"{s}\": {}", .{ path, err });
+        defer config_file.close();
+
+        try config_file.writeAll(@embedFile("../example.toml"));
+    };
+
+    return path;
+}
+
+fn ensureDataDirsExist(data_path: []const u8) !void {
+    var dir = try std.fs.openDirAbsolute(data_path, .{});
+    defer dir.close();
+
+    // Will recursively create directories
+    try dir.makePath("zba" ++ std.fs.path.sep_str ++ "save");
+}
+
+fn ensureConfigDirExists(config_path: []const u8) !void {
+    var dir = try std.fs.openDirAbsolute(config_path, .{});
+    defer dir.close();
+
+    try dir.makePath("zba");
+}
+
+fn romPath(result: *const clap.Result(clap.Help, &params, clap.parsers.default)) []const u8 {
+    return switch (result.positionals.len) {
+        1 => result.positionals[0],
+        0 => exitln("ZBA requires a path to a GamePak ROM", .{}),
+        else => exitln("ZBA received too many positional arguments.", .{}),
+    };
+}
+
+fn exitln(comptime format: []const u8, args: anytype) noreturn {
+    const stderr = std.io.getStdErr().writer();
+    stderr.print(format, args) catch {}; // Just exit already...
+    stderr.writeByte('\n') catch {};
+    std.os.exit(1);
 }

src/platform.zig

@@ -0,0 +1,343 @@
+const std = @import("std");
+const SDL = @import("sdl2");
+const gl = @import("gl");
+const emu = @import("core/emu.zig");
+const config = @import("config.zig");
+
+const Apu = @import("core/apu.zig").Apu;
+const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
+const Scheduler = @import("core/scheduler.zig").Scheduler;
+const FpsTracker = @import("util.zig").FpsTracker;
+
+const gba_width = @import("core/ppu.zig").width;
+const gba_height = @import("core/ppu.zig").height;
+
+pub const sample_rate = 1 << 15;
+pub const sample_format = SDL.AUDIO_U16;
+
+const default_title = "ZBA";
+
+pub const Gui = struct {
+    const Self = @This();
+    const SDL_GLContext = *anyopaque; // SDL.SDL_GLContext is a ?*anyopaque
+    const log = std.log.scoped(.Gui);
+
+    // zig fmt: off
+    const vertices: [32]f32 = [_]f32{
+        // Positions        // Colours      // Texture Coords
+         1.0, -1.0, 0.0,    1.0, 0.0, 0.0,  1.0, 1.0, // Top Right
+         1.0,  1.0, 0.0,    0.0, 1.0, 0.0,  1.0, 0.0, // Bottom Right
+        -1.0,  1.0, 0.0,    0.0, 0.0, 1.0,  0.0, 0.0, // Bottom Left
+        -1.0, -1.0, 0.0,    1.0, 1.0, 0.0,  0.0, 1.0, // Top Left
+    };
+
+    const indices: [6]u32 = [_]u32{
+        0, 1, 3, // First Triangle
+        1, 2, 3, // Second Triangle
+    };
+    // zig fmt: on
+
+    window: *SDL.SDL_Window,
+    ctx: SDL_GLContext,
+    title: []const u8,
+    audio: Audio,
+
+    program_id: gl.GLuint,
+
+    pub fn init(title: *const [12]u8, apu: *Apu, width: i32, height: i32) !Self {
+        if (SDL.SDL_Init(SDL.SDL_INIT_VIDEO | SDL.SDL_INIT_EVENTS | SDL.SDL_INIT_AUDIO) < 0) panic();
+        if (SDL.SDL_GL_SetAttribute(SDL.SDL_GL_CONTEXT_PROFILE_MASK, SDL.SDL_GL_CONTEXT_PROFILE_CORE) < 0) panic();
+        if (SDL.SDL_GL_SetAttribute(SDL.SDL_GL_CONTEXT_MAJOR_VERSION, 3) < 0) panic();
+        if (SDL.SDL_GL_SetAttribute(SDL.SDL_GL_CONTEXT_MAJOR_VERSION, 3) < 0) panic();
+
+        const win_scale = @intCast(c_int, config.config().host.win_scale);
+
+        const window = SDL.SDL_CreateWindow(
+            default_title,
+            SDL.SDL_WINDOWPOS_CENTERED,
+            SDL.SDL_WINDOWPOS_CENTERED,
+            @as(c_int, width * win_scale),
+            @as(c_int, height * win_scale),
+            SDL.SDL_WINDOW_OPENGL | SDL.SDL_WINDOW_SHOWN,
+        ) orelse panic();
+
+        const ctx = SDL.SDL_GL_CreateContext(window) orelse panic();
+        if (SDL.SDL_GL_MakeCurrent(window, ctx) < 0) panic();
+
+        gl.load(ctx, Self.glGetProcAddress) catch {};
+        if (SDL.SDL_GL_SetSwapInterval(@boolToInt(config.config().host.vsync)) < 0) panic();
+
+        const program_id = try compileShaders();
+
+        return Self{
+            .window = window,
+            .title = std.mem.sliceTo(title, 0),
+            .ctx = ctx,
+            .program_id = program_id,
+            .audio = Audio.init(apu),
+        };
+    }
+
+    fn compileShaders() !gl.GLuint {
+        // TODO: Panic on Shader Compiler Failure + Error Message
+        const vert_shader = @embedFile("shader/pixelbuf.vert");
+        const frag_shader = @embedFile("shader/pixelbuf.frag");
+
+        const vs = gl.createShader(gl.VERTEX_SHADER);
+        defer gl.deleteShader(vs);
+
+        gl.shaderSource(vs, 1, &[_][*c]const u8{vert_shader}, 0);
+        gl.compileShader(vs);
+
+        if (!shader.didCompile(vs)) return error.VertexCompileError;
+
+        const fs = gl.createShader(gl.FRAGMENT_SHADER);
+        defer gl.deleteShader(fs);
+
+        gl.shaderSource(fs, 1, &[_][*c]const u8{frag_shader}, 0);
+        gl.compileShader(fs);
+
+        if (!shader.didCompile(fs)) return error.FragmentCompileError;
+
+        const program = gl.createProgram();
+        gl.attachShader(program, vs);
+        gl.attachShader(program, fs);
+        gl.linkProgram(program);
+
+        return program;
+    }
+
+    // Returns the VAO ID since it's used in run()
+    fn generateBuffers() struct { c_uint, c_uint, c_uint } {
+        var vao_id: c_uint = undefined;
+        var vbo_id: c_uint = undefined;
+        var ebo_id: c_uint = undefined;
+        gl.genVertexArrays(1, &vao_id);
+        gl.genBuffers(1, &vbo_id);
+        gl.genBuffers(1, &ebo_id);
+
+        gl.bindVertexArray(vao_id);
+
+        gl.bindBuffer(gl.ARRAY_BUFFER, vbo_id);
+        gl.bufferData(gl.ARRAY_BUFFER, @sizeOf(@TypeOf(vertices)), &vertices, gl.STATIC_DRAW);
+
+        gl.bindBuffer(gl.ELEMENT_ARRAY_BUFFER, ebo_id);
+        gl.bufferData(gl.ELEMENT_ARRAY_BUFFER, @sizeOf(@TypeOf(indices)), &indices, gl.STATIC_DRAW);
+
+        // Position
+        gl.vertexAttribPointer(0, 3, gl.FLOAT, gl.FALSE, 8 * @sizeOf(f32), @intToPtr(?*anyopaque, 0)); // lmao
+        gl.enableVertexAttribArray(0);
+        // Colour
+        gl.vertexAttribPointer(1, 3, gl.FLOAT, gl.FALSE, 8 * @sizeOf(f32), @intToPtr(?*anyopaque, (3 * @sizeOf(f32))));
+        gl.enableVertexAttribArray(1);
+        // Texture Coord
+        gl.vertexAttribPointer(2, 2, gl.FLOAT, gl.FALSE, 8 * @sizeOf(f32), @intToPtr(?*anyopaque, (6 * @sizeOf(f32))));
+        gl.enableVertexAttribArray(2);
+
+        return .{ vao_id, vbo_id, ebo_id };
+    }
+
+    fn generateTexture(buf: []const u8) c_uint {
+        var tex_id: c_uint = undefined;
+        gl.genTextures(1, &tex_id);
+        gl.bindTexture(gl.TEXTURE_2D, tex_id);
+
+        // gl.texParameteri(gl.TEXTURE_2D, gl.TEXTURE_WRAP_S, gl.CLAMP_TO_EDGE);
+        // gl.texParameteri(gl.TEXTURE_2D, gl.TEXTURE_WRAP_T, gl.CLAMP_TO_EDGE);
+
+        gl.texParameteri(gl.TEXTURE_2D, gl.TEXTURE_MIN_FILTER, gl.NEAREST);
+        gl.texParameteri(gl.TEXTURE_2D, gl.TEXTURE_MAG_FILTER, gl.NEAREST);
+
+        gl.texImage2D(gl.TEXTURE_2D, 0, gl.RGBA, gba_width, gba_height, 0, gl.RGBA, gl.UNSIGNED_INT_8_8_8_8, buf.ptr);
+        // gl.generateMipmap(gl.TEXTURE_2D); // TODO: Remove?
+
+        return tex_id;
+    }
+
+    const RunOptions = struct {
+        quit: *std.atomic.Atomic(bool),
+        tracker: ?*FpsTracker = null,
+        cpu: *Arm7tdmi,
+        scheduler: *Scheduler,
+    };
+
+    pub fn run(self: *Self, opt: RunOptions) !void {
+        const cpu = opt.cpu;
+        const tracker = opt.tracker;
+        const quit = opt.quit;
+
+        var buffer_ids = Self.generateBuffers();
+        defer {
+            gl.deleteBuffers(1, &buffer_ids[2]); // EBO
+            gl.deleteBuffers(1, &buffer_ids[1]); // VBO
+            gl.deleteVertexArrays(1, &buffer_ids[0]); // VAO
+        }
+        const vao_id = buffer_ids[0];
+
+        const tex_id = Self.generateTexture(cpu.bus.ppu.framebuf.get(.Renderer));
+        defer gl.deleteTextures(1, &tex_id);
+
+        var title_buf: [0x100]u8 = undefined;
+
+        emu_loop: while (true) {
+            var event: SDL.SDL_Event = undefined;
+
+            // This might be true if the emu is running via a gdbstub server
+            // and the gdb stub exits first
+            if (quit.load(.Monotonic)) break :emu_loop;
+
+            while (SDL.SDL_PollEvent(&event) != 0) {
+                switch (event.type) {
+                    SDL.SDL_QUIT => break :emu_loop,
+                    SDL.SDL_KEYDOWN => {
+                        const key_code = event.key.keysym.sym;
+                        var keyinput = cpu.bus.io.keyinput.load(.Monotonic);
+
+                        switch (key_code) {
+                            SDL.SDLK_UP => keyinput.up.unset(),
+                            SDL.SDLK_DOWN => keyinput.down.unset(),
+                            SDL.SDLK_LEFT => keyinput.left.unset(),
+                            SDL.SDLK_RIGHT => keyinput.right.unset(),
+                            SDL.SDLK_x => keyinput.a.unset(),
+                            SDL.SDLK_z => keyinput.b.unset(),
+                            SDL.SDLK_a => keyinput.shoulder_l.unset(),
+                            SDL.SDLK_s => keyinput.shoulder_r.unset(),
+                            SDL.SDLK_RETURN => keyinput.start.unset(),
+                            SDL.SDLK_RSHIFT => keyinput.select.unset(),
+                            else => {},
+                        }
+
+                        cpu.bus.io.keyinput.store(keyinput.raw, .Monotonic);
+                    },
+                    SDL.SDL_KEYUP => {
+                        const key_code = event.key.keysym.sym;
+                        var keyinput = cpu.bus.io.keyinput.load(.Monotonic);
+
+                        switch (key_code) {
+                            SDL.SDLK_UP => keyinput.up.set(),
+                            SDL.SDLK_DOWN => keyinput.down.set(),
+                            SDL.SDLK_LEFT => keyinput.left.set(),
+                            SDL.SDLK_RIGHT => keyinput.right.set(),
+                            SDL.SDLK_x => keyinput.a.set(),
+                            SDL.SDLK_z => keyinput.b.set(),
+                            SDL.SDLK_a => keyinput.shoulder_l.set(),
+                            SDL.SDLK_s => keyinput.shoulder_r.set(),
+                            SDL.SDLK_RETURN => keyinput.start.set(),
+                            SDL.SDLK_RSHIFT => keyinput.select.set(),
+                            SDL.SDLK_i => {
+                                comptime std.debug.assert(sample_format == SDL.AUDIO_U16);
+                                log.err("Sample Count: {}", .{@intCast(u32, SDL.SDL_AudioStreamAvailable(cpu.bus.apu.stream)) / (2 * @sizeOf(u16))});
+                            },
+                            // SDL.SDLK_j => log.err("Scheduler Capacity: {} | Scheduler Event Count: {}", .{ scheduler.queue.capacity(), scheduler.queue.count() }),
+                            SDL.SDLK_k => {},
+                            else => {},
+                        }
+
+                        cpu.bus.io.keyinput.store(keyinput.raw, .Monotonic);
+                    },
+                    else => {},
+                }
+            }
+
+            // Emulator has an internal Double Buffer
+            const framebuf = cpu.bus.ppu.framebuf.get(.Renderer);
+            gl.texSubImage2D(gl.TEXTURE_2D, 0, 0, 0, gba_width, gba_height, gl.RGBA, gl.UNSIGNED_INT_8_8_8_8, framebuf.ptr);
+
+            gl.useProgram(self.program_id);
+            gl.bindVertexArray(vao_id);
+            gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_INT, null);
+            SDL.SDL_GL_SwapWindow(self.window);
+
+            if (tracker) |t| {
+                const dyn_title = std.fmt.bufPrintZ(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, t.value() }) catch unreachable;
+                SDL.SDL_SetWindowTitle(self.window, dyn_title.ptr);
+            }
+        }
+
+        quit.store(true, .Monotonic); // Terminate Emulator Thread
+    }
+
+    pub fn deinit(self: *Self) void {
+        self.audio.deinit();
+        gl.deleteProgram(self.program_id);
+        SDL.SDL_GL_DeleteContext(self.ctx);
+        SDL.SDL_DestroyWindow(self.window);
+        SDL.SDL_Quit();
+        self.* = undefined;
+    }
+
+    fn glGetProcAddress(ctx: SDL.SDL_GLContext, proc: [:0]const u8) ?*anyopaque {
+        _ = ctx;
+        return SDL.SDL_GL_GetProcAddress(proc.ptr);
+    }
+};
+
+const Audio = struct {
+    const Self = @This();
+    const log = std.log.scoped(.PlatformAudio);
+
+    device: SDL.SDL_AudioDeviceID,
+
+    fn init(apu: *Apu) Self {
+        var have: SDL.SDL_AudioSpec = undefined;
+        var want: SDL.SDL_AudioSpec = std.mem.zeroes(SDL.SDL_AudioSpec);
+        want.freq = sample_rate;
+        want.format = sample_format;
+        want.channels = 2;
+        want.samples = 0x100;
+        want.callback = Self.callback;
+        want.userdata = apu;
+
+        std.debug.assert(sample_format == SDL.AUDIO_U16);
+        log.info("Host Sample Rate: {}Hz, Host Format: SDL.AUDIO_U16", .{sample_rate});
+
+        const device = SDL.SDL_OpenAudioDevice(null, 0, &want, &have, 0);
+        if (device == 0) panic();
+
+        if (!config.config().host.mute) {
+            SDL.SDL_PauseAudioDevice(device, 0); // Unpause Audio
+            log.info("Unpaused Device", .{});
+        }
+
+        return .{ .device = device };
+    }
+
+    fn deinit(self: *Self) void {
+        SDL.SDL_CloseAudioDevice(self.device);
+        self.* = undefined;
+    }
+
+    export fn callback(userdata: ?*anyopaque, stream: [*c]u8, len: c_int) void {
+        const T = *Apu;
+        const apu = @ptrCast(T, @alignCast(@alignOf(T), userdata));
+
+        _ = SDL.SDL_AudioStreamGet(apu.stream, stream, len);
+    }
+};
+
+const shader = struct {
+    const Kind = enum { vertex, fragment };
+    const log = std.log.scoped(.Shader);
+
+    fn didCompile(id: gl.GLuint) bool {
+        var success: gl.GLint = undefined;
+        gl.getShaderiv(id, gl.COMPILE_STATUS, &success);
+
+        if (success == 0) err(id);
+
+        return success == 1;
+    }
+
+    fn err(id: gl.GLuint) void {
+        const buf_len = 512;
+        var error_msg: [buf_len]u8 = undefined;
+
+        gl.getShaderInfoLog(id, buf_len, 0, &error_msg);
+        log.err("{s}", .{std.mem.sliceTo(&error_msg, 0)});
+    }
+};
+
+fn panic() noreturn {
+    const str = @as(?[*:0]const u8, SDL.SDL_GetError()) orelse "unknown error";
+    @panic(std.mem.sliceTo(str, 0));
+}

src/ppu.zig

@@ -1,130 +0,0 @@
-const std = @import("std");
-
-const EventKind = @import("scheduler.zig").EventKind;
-const Io = @import("bus/io.zig").Io;
-const Scheduler = @import("scheduler.zig").Scheduler;
-
-const Allocator = std.mem.Allocator;
-const width = 240;
-const height = 160;
-pub const buf_pitch = width * @sizeOf(u16);
-const buf_len = buf_pitch * height;
-
-pub const Ppu = struct {
-    vram: Vram,
-    palette: Palette,
-    sched: *Scheduler,
-    frame_buf: []u8,
-    alloc: Allocator,
-
-    pub fn init(alloc: Allocator, sched: *Scheduler) !@This() {
-        // Queue first Hblank
-        sched.push(.{ .kind = .Draw, .tick = sched.tick + 240 * 4 });
-
-        return @This(){
-            .vram = try Vram.init(alloc),
-            .palette = try Palette.init(alloc),
-            .sched = sched,
-            .frame_buf = try alloc.alloc(u8, buf_len),
-            .alloc = alloc,
-        };
-    }
-
-    pub fn deinit(self: @This()) void {
-        self.alloc.free(self.frame_buf);
-        self.vram.deinit();
-        self.palette.deinit();
-    }
-
-    pub fn drawScanline(self: *@This(), io: *const Io) void {
-        const bg_mode = io.dispcnt.bg_mode.read();
-        const scanline = io.vcount.scanline.read();
-
-        switch (bg_mode) {
-            0x3 => {
-                // Mode 3
-                const start = buf_pitch * @as(usize, scanline);
-                const end = start + buf_pitch;
-
-                std.mem.copy(u8, self.frame_buf[start..end], self.vram.buf[start..end]);
-            },
-            else => std.debug.panic("[PPU] TODO: Implement BG Mode {}", .{bg_mode}),
-        }
-    }
-};
-
-const Palette = struct {
-    buf: []u8,
-    alloc: Allocator,
-
-    fn init(alloc: Allocator) !@This() {
-        return @This(){
-            .buf = try alloc.alloc(u8, 0x400),
-            .alloc = alloc,
-        };
-    }
-
-    fn deinit(self: @This()) void {
-        self.alloc.free(self.buf);
-    }
-
-    pub inline fn get32(self: *const @This(), idx: usize) u32 {
-        return (@as(u32, self.get16(idx + 2)) << 16) | @as(u32, self.get16(idx));
-    }
-
-    pub inline fn set32(self: *@This(), idx: usize, word: u32) void {
-        self.set16(idx + 2, @truncate(u16, word >> 16));
-        self.set16(idx, @truncate(u16, word));
-    }
-
-    pub inline fn get16(self: *const @This(), idx: usize) u16 {
-        return (@as(u16, self.buf[idx + 1]) << 8) | @as(u16, self.buf[idx]);
-    }
-
-    pub inline fn set16(self: *@This(), idx: usize, halfword: u16) void {
-        self.buf[idx + 1] = @truncate(u8, halfword >> 8);
-        self.buf[idx] = @truncate(u8, halfword);
-    }
-
-    pub inline fn get8(self: *const @This(), idx: usize) u8 {
-        return self.buf[idx];
-    }
-};
-
-const Vram = struct {
-    buf: []u8,
-    alloc: Allocator,
-
-    fn init(alloc: Allocator) !@This() {
-        return @This(){
-            .buf = try alloc.alloc(u8, 0x18000),
-            .alloc = alloc,
-        };
-    }
-
-    fn deinit(self: @This()) void {
-        self.alloc.free(self.buf);
-    }
-
-    pub inline fn get32(self: *const @This(), idx: usize) u32 {
-        return (@as(u32, self.get16(idx + 2)) << 16) | @as(u32, self.get16(idx));
-    }
-
-    pub inline fn set32(self: *@This(), idx: usize, word: u32) void {
-        self.set16(idx + 2, @truncate(u16, word >> 16));
-        self.set16(idx, @truncate(u16, word));
-    }
-
-    pub inline fn get16(self: *const @This(), idx: usize) u16 {
-        return (@as(u16, self.buf[idx + 1]) << 8) | @as(u16, self.buf[idx]);
-    }
-
-    pub inline fn set16(self: *@This(), idx: usize, halfword: u16) void {
-        self.buf[idx + 1] = @truncate(u8, halfword >> 8);
-        self.buf[idx] = @truncate(u8, halfword);
-    }
-
-    pub inline fn get8(self: *const @This(), idx: usize) u8 {
-        return self.buf[idx];
-    }
-};

src/scheduler.zig

@@ -1,113 +0,0 @@
-const std = @import("std");
-
-const Bus = @import("Bus.zig");
-const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
-
-const Order = std.math.Order;
-const PriorityQueue = std.PriorityQueue;
-const Allocator = std.mem.Allocator;
-
-pub const Scheduler = struct {
-    tick: u64,
-    queue: PriorityQueue(Event, void, lessThan),
-
-    pub fn init(alloc: Allocator) @This() {
-        var scheduler = Scheduler{ .tick = 0, .queue = PriorityQueue(Event, void, lessThan).init(alloc, {}) };
-
-        scheduler.queue.add(.{
-            .kind = EventKind.HeatDeath,
-            .tick = std.math.maxInt(u64),
-        }) catch unreachable;
-
-        return scheduler;
-    }
-
-    pub fn deinit(self: @This()) void {
-        self.queue.deinit();
-    }
-
-    pub fn handleEvent(self: *@This(), _: *Arm7tdmi, bus: *Bus) void {
-        const should_handle = if (self.queue.peek()) |e| self.tick >= e.tick else false;
-
-        if (should_handle) {
-            const event = self.queue.remove();
-
-            switch (event.kind) {
-                .HeatDeath => {
-                    std.debug.panic("[Scheduler] Somehow, a u64 overflowed", .{});
-                },
-                .HBlank => {
-                    // The End of a Hblank
-                    const scanline = bus.io.vcount.scanline.read();
-                    const new_scanline = scanline + 1;
-
-                    // TODO: Should this be done @ end of Draw instead of end of Hblank?
-                    bus.ppu.drawScanline(&bus.io);
-
-                    bus.io.vcount.scanline.write(new_scanline);
-                    bus.io.dispstat.hblank.unset();
-
-                    if (new_scanline < 160) {
-                        // Transitioning to another Draw
-                        self.push(.{ .kind = .Draw, .tick = self.tick + (240 * 4) });
-                    } else {
-                        // Transitioning to a Vblank
-                        bus.io.dispstat.vblank.set();
-                        self.push(.{ .kind = .VBlank, .tick = self.tick + 68 * (308 * 4) });
-                    }
-                },
-                .Draw => {
-                    // The end of a Draw
-
-                    // Transitioning to a Hblank
-                    bus.io.dispstat.hblank.set();
-                    self.push(.{ .kind = .HBlank, .tick = self.tick + (68 * 4) });
-                },
-                .VBlank => {
-                    // The end of a Vblank
-
-                    const scanline = bus.io.vcount.scanline.read();
-                    const new_scanline = scanline + 1;
-                    bus.io.vcount.scanline.write(new_scanline);
-
-                    if (new_scanline < 228) {
-                        // Transition to another Vblank
-                        self.push(.{ .kind = .VBlank, .tick = self.tick + 68 * (308 * 4) });
-                    } else {
-                        // Transition to another Draw
-                        bus.io.vcount.scanline.write(0); // Reset Scanline
-
-                        bus.io.dispstat.vblank.unset();
-                        self.push(.{ .kind = .Draw, .tick = self.tick + (240 * 4) });
-                    }
-                },
-            }
-        }
-    }
-
-    pub inline fn push(self: *@This(), event: Event) void {
-        self.queue.add(event) catch unreachable;
-    }
-
-    pub inline fn nextTimestamp(self: *@This()) u64 {
-        if (self.queue.peek()) |e| {
-            return e.tick;
-        } else unreachable;
-    }
-};
-
-pub const Event = struct {
-    kind: EventKind,
-    tick: u64,
-};
-
-fn lessThan(_: void, a: Event, b: Event) Order {
-    return std.math.order(a.tick, b.tick);
-}
-
-pub const EventKind = enum {
-    HeatDeath,
-    HBlank,
-    VBlank,
-    Draw,
-};

src/shader/pixelbuf.frag

@@ -0,0 +1,25 @@
+#version 330 core
+out vec4 frag_color;
+
+in vec3 color;
+in vec2 uv;
+
+uniform sampler2D screen;
+
+void main() {
+	// https://near.sh/video/color-emulation
+	// Thanks to Talarubi + Near for the Colour Correction
+	// Thanks to fleur + mattrb for the Shader Impl
+
+	vec4 color = texture(screen, uv);
+	color.rgb = pow(color.rgb, vec3(4.0)); // LCD Gamma
+  
+	frag_color = vec4(
+		pow(vec3(
+		  	  0 * color.b +  50 * color.g + 255 * color.r,
+	     	 30 * color.b + 230 * color.g +  10 * color.r,
+			220 * color.b +  10 * color.g +  50 * color.r
+		) / 255, vec3(1.0 / 2.2)), // Out Gamma
+	1.0); 
+}
+

src/shader/pixelbuf.vert

@@ -0,0 +1,13 @@
+#version 330 core
+layout (location = 0) in vec3 pos;
+layout (location = 1) in vec3 in_color;
+layout (location = 2) in vec2 in_uv;
+
+out vec3 color;
+out vec2 uv;
+
+void main() {
+	color = in_color;
+	uv = in_uv;
+	gl_Position = vec4(pos, 1.0);
+}

src/util.zig

@@ -1,29 +1,319 @@
 const std = @import("std");
+const builtin = @import("builtin");
+const config = @import("config.zig");
 
-pub fn signExtend(comptime T: type, comptime bits: usize, value: anytype) T {
-    const ValT = comptime @TypeOf(value);
-    comptime std.debug.assert(isInteger(ValT));
-    comptime std.debug.assert(isSigned(ValT));
+const Log2Int = std.math.Log2Int;
+const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
 
-    const value_bits = @typeInfo(ValT).Int.bits;
-    comptime std.debug.assert(value_bits >= bits);
+const Allocator = std.mem.Allocator;
 
-    const bit_diff = value_bits - bits;
+// Sign-Extend value of type `T` to type `U`
+pub fn sext(comptime T: type, comptime U: type, value: T) T {
+    // U must have less bits than T
+    comptime std.debug.assert(@typeInfo(U).Int.bits <= @typeInfo(T).Int.bits);
 
-    // (1 << bits) -1 is a mask that will take values like 0x100 and make them 0xFF
-    // value & mask so that only the relevant bits are sign extended
-    // therefore, value & ((1 << bits) - 1) is the isolation of the relevant bits
-    return ((value & ((1 << bits) - 1)) << bit_diff) >> bit_diff;
+    const iT = std.meta.Int(.signed, @typeInfo(T).Int.bits);
+    const ExtU = if (@typeInfo(U).Int.signedness == .unsigned) T else iT;
+    const shift_amt = @intCast(Log2Int(T), @typeInfo(T).Int.bits - @typeInfo(U).Int.bits);
+
+    return @bitCast(T, @bitCast(iT, @as(ExtU, @truncate(U, value)) << shift_amt) >> shift_amt);
 }
 
-pub fn u32SignExtend(comptime bits: usize, value: u32) u32 {
-    return @bitCast(u32, signExtend(i32, bits, @bitCast(i32, value)));
+/// See https://godbolt.org/z/W3en9Eche
+pub inline fn rotr(comptime T: type, x: T, r: anytype) T {
+    if (@typeInfo(T).Int.signedness == .signed)
+        @compileError("cannot rotate signed integer");
+
+    const ar = @intCast(Log2Int(T), @mod(r, @typeInfo(T).Int.bits));
+    return x >> ar | x << (1 +% ~ar);
 }
 
-fn isInteger(comptime T: type) bool {
-    return @typeInfo(T) == .Int;
+pub const FpsTracker = struct {
+    const Self = @This();
+
+    fps: u32,
+    count: std.atomic.Atomic(u32),
+    timer: std.time.Timer,
+
+    pub fn init() Self {
+        return .{
+            .fps = 0,
+            .count = std.atomic.Atomic(u32).init(0),
+            .timer = std.time.Timer.start() catch unreachable,
+        };
+    }
+
+    pub fn tick(self: *Self) void {
+        _ = self.count.fetchAdd(1, .Monotonic);
+    }
+
+    pub fn value(self: *Self) u32 {
+        if (self.timer.read() >= std.time.ns_per_s) {
+            self.fps = self.count.swap(0, .Monotonic);
+            self.timer.reset();
+        }
+
+        return self.fps;
+    }
+};
+
+pub fn intToBytes(comptime T: type, value: anytype) [@sizeOf(T)]u8 {
+    comptime std.debug.assert(@typeInfo(T) == .Int);
+
+    var result: [@sizeOf(T)]u8 = undefined;
+
+    var i: Log2Int(T) = 0;
+    while (i < result.len) : (i += 1) result[i] = @truncate(u8, value >> i * @bitSizeOf(u8));
+
+    return result;
 }
 
-fn isSigned(comptime T: type) bool {
-    return @typeInfo(T).Int.signedness == .signed;
+/// Creates a copy of a title with all Filesystem-invalid characters replaced
+///
+/// e.g. POKEPIN R/S to POKEPIN R_S
+pub fn escape(title: [12]u8) [12]u8 {
+    var ret: [12]u8 = title;
+
+    //TODO: Add more replacements
+    std.mem.replaceScalar(u8, &ret, '/', '_');
+    std.mem.replaceScalar(u8, &ret, '\\', '_');
+
+    return ret;
 }
+
+pub const FilePaths = struct {
+    rom: []const u8,
+    bios: ?[]const u8,
+    save: ?[]const u8,
+};
+
+pub const io = struct {
+    pub const read = struct {
+        pub fn todo(comptime log: anytype, comptime format: []const u8, args: anytype) u8 {
+            log.debug(format, args);
+            return 0;
+        }
+
+        pub fn undef(comptime T: type, comptime log: anytype, comptime format: []const u8, args: anytype) ?T {
+            @setCold(true);
+
+            const unhandled_io = config.config().debug.unhandled_io;
+
+            log.warn(format, args);
+            if (builtin.mode == .Debug and !unhandled_io) std.debug.panic("TODO: Implement I/O Register", .{});
+
+            return null;
+        }
+
+        pub fn err(comptime T: type, comptime log: anytype, comptime format: []const u8, args: anytype) ?T {
+            @setCold(true);
+
+            log.err(format, args);
+            return null;
+        }
+    };
+
+    pub const write = struct {
+        pub fn undef(log: anytype, comptime format: []const u8, args: anytype) void {
+            const unhandled_io = config.config().debug.unhandled_io;
+
+            log.warn(format, args);
+            if (builtin.mode == .Debug and !unhandled_io) std.debug.panic("TODO: Implement I/O Register", .{});
+        }
+    };
+};
+
+pub const Logger = struct {
+    const Self = @This();
+    const FmtArgTuple = std.meta.Tuple(&.{ u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32 });
+
+    buf: std.io.BufferedWriter(4096 << 2, std.fs.File.Writer),
+
+    pub fn init(file: std.fs.File) Self {
+        return .{
+            .buf = .{ .unbuffered_writer = file.writer() },
+        };
+    }
+
+    pub fn print(self: *Self, comptime format: []const u8, args: anytype) !void {
+        try self.buf.writer().print(format, args);
+        try self.buf.flush(); // FIXME: On panics, whatever is in the buffer isn't written to file
+    }
+
+    pub fn mgbaLog(self: *Self, cpu: *const Arm7tdmi, opcode: u32) void {
+        const fmt_base = "{X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} cpsr: {X:0>8} | ";
+        const thumb_fmt = fmt_base ++ "{X:0>4}:\n";
+        const arm_fmt = fmt_base ++ "{X:0>8}:\n";
+
+        if (cpu.cpsr.t.read()) {
+            if (opcode >> 11 == 0x1E) {
+                // Instruction 1 of a BL Opcode, print in ARM mode
+                const low = cpu.bus.dbgRead(u16, cpu.r[15] - 2);
+                const bl_opcode = @as(u32, opcode) << 16 | low;
+
+                self.print(arm_fmt, Self.fmtArgs(cpu, bl_opcode)) catch @panic("failed to write to log file");
+            } else {
+                self.print(thumb_fmt, Self.fmtArgs(cpu, opcode)) catch @panic("failed to write to log file");
+            }
+        } else {
+            self.print(arm_fmt, Self.fmtArgs(cpu, opcode)) catch @panic("failed to write to log file");
+        }
+    }
+
+    fn fmtArgs(cpu: *const Arm7tdmi, opcode: u32) FmtArgTuple {
+        return .{
+            cpu.r[0],
+            cpu.r[1],
+            cpu.r[2],
+            cpu.r[3],
+            cpu.r[4],
+            cpu.r[5],
+            cpu.r[6],
+            cpu.r[7],
+            cpu.r[8],
+            cpu.r[9],
+            cpu.r[10],
+            cpu.r[11],
+            cpu.r[12],
+            cpu.r[13],
+            cpu.r[14],
+            cpu.r[15] - if (cpu.cpsr.t.read()) 2 else @as(u32, 4),
+            cpu.cpsr.raw,
+            opcode,
+        };
+    }
+};
+
+pub const audio = struct {
+    const _io = @import("core/bus/io.zig");
+
+    const ToneSweep = @import("core/apu/ToneSweep.zig");
+    const Tone = @import("core/apu/Tone.zig");
+    const Wave = @import("core/apu/Wave.zig");
+    const Noise = @import("core/apu/Noise.zig");
+
+    pub const length = struct {
+        const FrameSequencer = @import("core/apu.zig").FrameSequencer;
+
+        /// Update State of Ch1, Ch2 and Ch3 length timer
+        pub fn update(comptime T: type, self: *T, fs: *const FrameSequencer, nrx34: _io.Frequency) void {
+            comptime std.debug.assert(T == ToneSweep or T == Tone or T == Wave);
+
+            // Write to NRx4 when FS's next step is not one that clocks the length counter
+            if (!fs.isLengthNext()) {
+                // If length_enable was disabled but is now enabled and length timer is not 0 already,
+                // decrement the length timer
+
+                if (!self.freq.length_enable.read() and nrx34.length_enable.read() and self.len_dev.timer != 0) {
+                    self.len_dev.timer -= 1;
+
+                    // If Length Timer is now 0 and trigger is clear, disable the channel
+                    if (self.len_dev.timer == 0 and !nrx34.trigger.read()) self.enabled = false;
+                }
+            }
+        }
+
+        pub const ch4 = struct {
+            /// update state of ch4 length timer
+            pub fn update(self: *Noise, fs: *const FrameSequencer, nr44: _io.NoiseControl) void {
+                // Write to NRx4 when FS's next step is not one that clocks the length counter
+                if (!fs.isLengthNext()) {
+                    // If length_enable was disabled but is now enabled and length timer is not 0 already,
+                    // decrement the length timer
+
+                    if (!self.cnt.length_enable.read() and nr44.length_enable.read() and self.len_dev.timer != 0) {
+                        self.len_dev.timer -= 1;
+
+                        // If Length Timer is now 0 and trigger is clear, disable the channel
+                        if (self.len_dev.timer == 0 and !nr44.trigger.read()) self.enabled = false;
+                    }
+                }
+            }
+        };
+    };
+};
+
+/// Sets a quarter (8) of the bits of the u32 `left` to the value of u8 `right`
+pub inline fn setQuart(left: u32, addr: u8, right: u8) u32 {
+    const offset = @truncate(u2, addr);
+
+    return switch (offset) {
+        0b00 => (left & 0xFFFF_FF00) | right,
+        0b01 => (left & 0xFFFF_00FF) | @as(u32, right) << 8,
+        0b10 => (left & 0xFF00_FFFF) | @as(u32, right) << 16,
+        0b11 => (left & 0x00FF_FFFF) | @as(u32, right) << 24,
+    };
+}
+
+/// Calculates the correct shift offset for an aligned/unaligned u8 read
+///
+/// TODO: Support u16 reads of u32 values?
+pub inline fn getHalf(byte: u8) u4 {
+    return @truncate(u4, byte & 1) << 3;
+}
+
+pub inline fn setHalf(comptime T: type, left: T, addr: u8, right: HalfInt(T)) T {
+    const offset = @truncate(u1, addr >> if (T == u32) 1 else 0);
+
+    return switch (T) {
+        u32 => switch (offset) {
+            0b0 => (left & 0xFFFF_0000) | right,
+            0b1 => (left & 0x0000_FFFF) | @as(u32, right) << 16,
+        },
+        u16 => switch (offset) {
+            0b0 => (left & 0xFF00) | right,
+            0b1 => (left & 0x00FF) | @as(u16, right) << 8,
+        },
+        else => @compileError("unsupported type"),
+    };
+}
+
+/// The Integer type which corresponds to T with exactly half the amount of bits
+fn HalfInt(comptime T: type) type {
+    const type_info = @typeInfo(T);
+    comptime std.debug.assert(type_info == .Int); // Type must be an integer
+    comptime std.debug.assert(type_info.Int.bits % 2 == 0); // Type must have an even amount of bits
+
+    return std.meta.Int(type_info.Int.signedness, type_info.Int.bits >> 1);
+}
+
+/// Double Buffering Implementation
+pub const FrameBuffer = struct {
+    const Self = @This();
+
+    layers: [2][]u8,
+    buf: []u8,
+    current: u1,
+
+    allocator: Allocator,
+
+    // TODO: Rename
+    const Device = enum { Emulator, Renderer };
+
+    pub fn init(allocator: Allocator, comptime len: comptime_int) !Self {
+        const buf = try allocator.alloc(u8, len * 2);
+        std.mem.set(u8, buf, 0);
+
+        return .{
+            // Front and Back Framebuffers
+            .layers = [_][]u8{ buf[0..][0..len], buf[len..][0..len] },
+            .buf = buf,
+            .current = 0,
+
+            .allocator = allocator,
+        };
+    }
+
+    pub fn deinit(self: *Self) void {
+        self.allocator.free(self.buf);
+        self.* = undefined;
+    }
+
+    pub fn swap(self: *Self) void {
+        self.current = ~self.current;
+    }
+
+    pub fn get(self: *Self, comptime dev: Device) []u8 {
+        return self.layers[if (dev == .Emulator) self.current else ~self.current];
+    }
+};