fix(bios): set addr_latch even if bios is skipped

make sure to use fetch timings when fetching instructions

.gitmodules

@@ -10,3 +10,6 @@
 [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

README.md

@@ -2,13 +2,13 @@
 An in-progress Game Boy Advance Emulator written in Zig ⚡!
 
 ## Tests 
-- [ ] [jsmolka's GBA Test Collection](https://github.com/jsmolka/gba-tests)
+- [x] [jsmolka's GBA Test Collection](https://github.com/jsmolka/gba-tests)
     - [x] `arm.gba` and `thumb.gba`
     - [x] `flash64.gba`, `flash128.gba`, `none.gba`, and `sram.gba`
     - [x] `hello.gba`, `shades.gba`, and `stripes.gba`
     - [x] `memory.gba`
     - [x] `bios.gba`
-    - [ ] `nes.gba`
+    - [x] `nes.gba`
 - [ ] [DenSinH's GBA ROMs](https://github.com/DenSinH/GBARoms)
     - [x] `eeprom-test` and `flash-test`
     - [x] `midikey2freq`

build.zig

@@ -28,6 +28,9 @@ pub fn build(b: *std.build.Builder) void {
     // Argument Parsing Library
     exe.addPackagePath("clap", "lib/zig-clap/clap.zig");
 
+    // TOML Library
+    exe.addPackagePath("toml", "lib/zig-toml/src/toml.zig");
+
     // Zig SDL Bindings: https://github.com/MasterQ32/SDL.zig
     const sdk = Sdk.init(b);
     sdk.link(exe, .dynamic);

lib/zig-clap

@@ -1 +1 @@
-Subproject commit 4f4196fc3bc95c4bd3b12ce2e4a5f1050742cd3c
+Subproject commit e5d09c4b2d121025ad7195b2de704451e6306807

lib/zig-toml

@@ -0,0 +1 @@
+Subproject commit 5dfa919e03b446c66b295c04bef9bdecabd4276f

src/config.zig

@@ -0,0 +1,74 @@
+const std = @import("std");
+const toml = @import("toml");
+
+const Allocator = std.mem.Allocator;
+
+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,
+    };
+
+    // 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,
+    };
+
+    /// Settings related to debugging ZBA
+    const Debug = struct {
+        /// Enable CPU Trace logs
+        cpu_trace: bool = false,
+        /// If false and ZBA is build 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, config_path: []const u8) !void {
+    var config_file = try std.fs.cwd().openFile(config_path, .{});
+    defer config_file.close();
+
+    const contents = try config_file.readToEndAlloc(allocator, try config_file.getEndPos());
+    defer allocator.free(contents);
+
+    const table = try toml.parseContents(allocator, contents, null);
+    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 (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 (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

@@ -46,7 +46,7 @@ iwram: Iwram,
 ewram: Ewram,
 io: Io,
 
-cpu: ?*Arm7tdmi,
+cpu: *Arm7tdmi,
 sched: *Scheduler,
 
 pub fn init(self: *Self, allocator: Allocator, sched: *Scheduler, cpu: *Arm7tdmi, paths: FilePaths) !void {
@@ -82,9 +82,9 @@ pub fn dbgRead(self: *const Self, comptime T: type, address: u32) T {
         // General Internal Memory
         0x00 => blk: {
             if (address < Bios.size)
-                break :blk self.bios.dbgRead(T, self.cpu.?.r[15], aligned_addr);
+                break :blk self.bios.dbgRead(T, self.cpu.r[15], aligned_addr);
 
-            break :blk self.readOpenBus(T, address);
+            break :blk self.openBus(T, address);
         },
         0x02 => self.ewram.read(T, aligned_addr),
         0x03 => self.iwram.read(T, aligned_addr),
@@ -109,48 +109,63 @@ pub fn dbgRead(self: *const Self, comptime T: type, address: u32) T {
 
             break :blk @as(T, value) * multiplier;
         },
-        else => self.readOpenBus(T, address),
+        else => self.openBus(T, address),
     };
 }
 
 fn readIo(self: *const Self, comptime T: type, unaligned_address: u32) T {
     const maybe_value = io.read(self, T, forceAlign(T, unaligned_address));
-    return if (maybe_value) |value| value else self.readOpenBus(T, unaligned_address);
+    return if (maybe_value) |value| value else self.openBus(T, unaligned_address);
 }
 
-fn readOpenBus(self: *const Self, comptime T: type, address: u32) T {
-    const r15 = self.cpu.?.r[15];
+fn openBus(self: *const Self, comptime T: type, address: u32) T {
+    const r15 = self.cpu.r[15];
 
     const word = blk: {
-        // If u32 Open Bus, read recently fetched opcode (PC + 8)
-        if (!self.cpu.?.cpsr.t.read()) break :blk self.dbgRead(u32, r15 + 4);
+        // If Arm, get the most recently fetched instruction (PC + 8)
+        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 => {
-                // (PC + 4)
-                const halfword = self.dbgRead(u16, r15 + 2);
-
-                break :blk @as(u32, halfword) << 16 | halfword;
+                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 offset: u32 = if (address & 3 == 0b00) 2 else 0;
+                const aligned = address & 3 == 0b00;
 
-                break :blk @as(u32, self.dbgRead(u16, r15 + 2 + offset)) << 16 | self.dbgRead(u16, r15 + offset);
+                // 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 offset: u32 = if (address & 3 == 0b00) 2 else 0;
+                const aligned = address & 3 == 0b00;
 
-                break :blk @as(u32, self.dbgRead(u16, r15 + 2 - offset)) << 16 | self.dbgRead(u16, r15 + offset);
+                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");
             },
-            else => unreachable,
         }
     };
 
@@ -167,9 +182,9 @@ pub fn read(self: *Self, comptime T: type, address: u32) T {
         // General Internal Memory
         0x00 => blk: {
             if (address < Bios.size)
-                break :blk self.bios.read(T, self.cpu.?.r[15], aligned_addr);
+                break :blk self.bios.read(T, self.cpu.r[15], aligned_addr);
 
-            break :blk self.readOpenBus(T, address);
+            break :blk self.openBus(T, address);
         },
         0x02 => self.ewram.read(T, aligned_addr),
         0x03 => self.iwram.read(T, aligned_addr),
@@ -194,7 +209,7 @@ pub fn read(self: *Self, comptime T: type, address: u32) T {
 
             break :blk @as(T, value) * multiplier;
         },
-        else => self.readOpenBus(T, address),
+        else => self.openBus(T, address),
     };
 }
 

src/core/apu/Noise.zig

@@ -0,0 +1,142 @@
+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;
+    self.envelope.raw = 0;
+    self.poly.raw = 0;
+    self.cnt.raw = 0;
+
+    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 +1,138 @@
+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;
+    self.envelope.raw = 0;
+    self.freq.raw = 0;
+
+    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 +1,184 @@
+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;
+    self.sweep_dev.calc_performed = false;
+
+    self.duty.raw = 0;
+    self.envelope.raw = 0;
+    self.freq.raw = 0;
+
+    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 (self.sweep.direction.read() and !new.direction.read()) {
+        // Sweep Negate bit has been cleared
+        // If At least 1 Sweep Calculation has been made since
+        // the last trigger, the channel is immediately disabled
+
+        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 +1,132 @@
+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;
+    self.length = 0;
+    self.vol.raw = 0;
+    self.freq.raw = 0;
+
+    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;
+}
+
+/// 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
+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 +1,28 @@
+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 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 +1,18 @@
+const Self = @This();
+
+timer: u9,
+
+pub fn create() Self {
+    return .{ .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 +1,52 @@
+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 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.calc_performed) self.calc_performed = true;
+
+        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) shadow - shadow_shifted else shadow + shadow_shifted;
+    if (freq > 0x7FF) ch_enable.* = false;
+
+    return freq;
+}

src/core/apu/signal/Lfsr.zig

@@ -0,0 +1,59 @@
+const io = @import("../../bus/io.zig");
+
+/// Linear Feedback Shift Register
+const Scheduler = @import("../../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../../apu.zig").FrameSequencer;
+const Noise = @import("../Noise.zig");
+
+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 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, clogging up 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 +1,58 @@
+const std = @import("std");
+const io = @import("../../bus/io.zig");
+
+const Scheduler = @import("../../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../../apu.zig").FrameSequencer;
+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,
+    };
+}
+
+/// 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 +1,79 @@
+const std = @import("std");
+const io = @import("../../bus/io.zig");
+
+const Scheduler = @import("../../scheduler.zig").Scheduler;
+const FrameSequencer = @import("../../apu.zig").FrameSequencer;
+const Wave = @import("../Wave.zig");
+
+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,
+    };
+}
+
+/// 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

@@ -12,6 +12,36 @@ allocator: Allocator,
 
 addr_latch: u32,
 
+pub fn read(self: *Self, comptime T: type, r15: u32, addr: u32) T {
+    if (r15 < Self.size) {
+        self.addr_latch = addr;
+        return self._read(T, addr);
+    }
+
+    log.debug("Rejected read since r15=0x{X:0>8}", .{r15});
+    return @truncate(T, self._read(T, self.addr_latch + 8));
+}
+
+pub fn dbgRead(self: *const Self, comptime T: type, r15: u32, addr: u32) T {
+    if (r15 < Self.size) return self._read(T, addr);
+    return @truncate(T, self._read(T, self.addr_latch + 8));
+}
+
+/// 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 {
     const buf: ?[]u8 = if (maybe_path) |path| blk: {
         const file = try std.fs.cwd().openFile(path, .{});
@@ -31,34 +61,3 @@ pub fn deinit(self: *Self) void {
     if (self.buf) |buf| self.allocator.free(buf);
     self.* = undefined;
 }
-
-pub fn read(self: *Self, comptime T: type, r15: u32, addr: u32) T {
-    if (r15 < Self.size) {
-        self.addr_latch = addr;
-        return self.uncheckedRead(T, addr);
-    }
-
-    log.debug("Rejected read since r15=0x{X:0>8}", .{r15});
-    return @truncate(T, self.uncheckedRead(T, self.addr_latch + 8));
-}
-
-pub fn dbgRead(self: *const Self, comptime T: type, r15: u32, addr: u32) T {
-    if (r15 < Self.size) return self.uncheckedRead(T, addr);
-    return @truncate(T, self.uncheckedRead(T, self.addr_latch + 8));
-}
-
-fn uncheckedRead(self: *const Self, comptime T: type, addr: u32) T {
-    if (self.buf) |buf| {
-        return switch (T) {
-            u32, u16, u8 => std.mem.readIntSliceLittle(T, buf[addr..][0..@sizeOf(T)]),
-            else => @compileError("BIOS: Unsupported read width"),
-        };
-    }
-
-    std.debug.panic("[BIOS] ZBA tried to read {} from 0x{X:0>8} but not BIOS was present", .{ T, addr });
-}
-
-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 });
-}

src/core/bus/Ewram.zig

@@ -7,21 +7,6 @@ const Self = @This();
 buf: []u8,
 allocator: Allocator,
 
-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;
-}
-
 pub fn read(self: *const Self, comptime T: type, address: usize) T {
     const addr = address & 0x3FFFF;
 
@@ -39,3 +24,18 @@ pub fn write(self: *const Self, comptime T: type, address: usize, value: T) void
         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

@@ -1,4 +1,6 @@
 const std = @import("std");
+const config = @import("../../config.zig");
+
 const Bit = @import("bitfield").Bit;
 const Bitfield = @import("bitfield").Bitfield;
 const DateTime = @import("datetime").datetime.Datetime;
@@ -8,7 +10,6 @@ const Backup = @import("backup.zig").Backup;
 const Gpio = @import("gpio.zig").Gpio;
 const Allocator = std.mem.Allocator;
 
-const force_rtc = @import("../emu.zig").force_rtc;
 const log = std.log.scoped(.GamePak);
 
 const Self = @This();
@@ -19,78 +20,11 @@ allocator: Allocator,
 backup: Backup,
 gpio: *Gpio,
 
-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.guessKind(file_buf);
-    const device = if (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),
-    };
-}
-
-/// 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
-
-    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});
-    if (lookupMaker(maker)) |c| log.info("Maker: {s}", .{c}) else log.info("Maker Code: {s}", .{maker});
-}
-
-fn lookupMaker(slice: *const [2]u8) ?[]const u8 {
-    const id = @as(u16, slice[1]) << 8 | @as(u16, slice[0]);
-    return switch (id) {
-        0x3130 => "Nintendo",
-        else => null,
-    };
-}
-
-inline fn isLarge(self: *const Self) bool {
-    return self.buf.len > 0x100_0000;
-}
-
-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;
-}
-
 pub fn read(self: *Self, comptime T: type, address: u32) T {
     const addr = address & 0x1FF_FFFF;
 
     if (self.backup.kind == .Eeprom) {
-        if (self.isLarge()) {
+        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)
@@ -142,11 +76,19 @@ pub fn read(self: *Self, comptime T: type, address: u32) T {
     };
 }
 
+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.isLarge()) {
+        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)
@@ -161,6 +103,35 @@ pub fn dbgRead(self: *const Self, comptime T: type, address: u32) T {
         }
     }
 
+    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)),
@@ -175,7 +146,7 @@ pub fn write(self: *Self, comptime T: type, word_count: u16, address: u32, value
     if (self.backup.kind == .Eeprom) {
         const bit = @truncate(u1, value);
 
-        if (self.isLarge()) {
+        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)
@@ -213,12 +184,59 @@ pub fn write(self: *Self, comptime T: type, word_count: u16, address: u32, value
     }
 }
 
-fn get(self: *const Self, i: u32) u8 {
-    @setRuntimeSafety(false);
-    if (i < self.buf.len) return self.buf[i];
+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 lhs = i >> 1 & 0xFFFF;
-    return @truncate(u8, lhs >> 8 * @truncate(u5, i & 1));
+    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
+
+    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" {

src/core/bus/Iwram.zig

@@ -7,21 +7,6 @@ const Self = @This();
 buf: []u8,
 allocator: Allocator,
 
-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;
-}
-
 pub fn read(self: *const Self, comptime T: type, address: usize) T {
     const addr = address & 0x7FFF;
 
@@ -39,3 +24,18 @@ pub fn write(self: *const Self, comptime T: type, address: usize, value: T) void
         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

@@ -2,9 +2,13 @@ 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 span = @import("../../util.zig").span;
 
+const Needle = struct { str: []const u8, kind: Backup.Kind };
 const backup_kinds = [6]Needle{
     .{ .str = "EEPROM_V", .kind = .Eeprom },
     .{ .str = "SRAM_V", .kind = .Sram },
@@ -14,6 +18,8 @@ const backup_kinds = [6]Needle{
     .{ .str = "FLASH1M_V", .kind = .Flash1M },
 };
 
+const SaveError = error{Unsupported};
+
 pub const Backup = struct {
     const Self = @This();
 
@@ -35,122 +41,6 @@ pub const Backup = struct {
         None,
     };
 
-    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.init(),
-            .eeprom = Eeprom.init(allocator),
-        };
-
-        if (backup.save_path) |p| backup.loadSaveFromDisk(allocator, p) catch |e| log.err("Failed to load save: {}", .{e});
-        return backup;
-    }
-
-    pub fn guessKind(rom: []const u8) Kind {
-        for (backup_kinds) |needle| {
-            const needle_len = needle.str.len;
-
-            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;
-    }
-
-    pub fn deinit(self: *Self) void {
-        if (self.save_path) |path| self.writeSaveToDisk(self.allocator, path) catch |e| log.err("Failed to write save: {}", .{e});
-        self.allocator.free(self.buf);
-        self.* = undefined;
-    }
-
-    fn loadSaveFromDisk(self: *Self, allocator: Allocator, path: []const u8) !void {
-        const file_path = try self.getSaveFilePath(allocator, path);
-        defer allocator.free(file_path);
-
-        // FIXME: Don't rely on this lol
-        if (std.mem.eql(u8, file_path[file_path.len - 12 .. file_path.len], "untitled.sav")) {
-            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.UnsupportedBackupKind,
-        }
-    }
-
-    fn getSaveFilePath(self: *const Self, allocator: Allocator, path: []const u8) ![]const u8 {
-        const filename = try self.getSaveFilename(allocator);
-        defer allocator.free(filename);
-
-        return try std.fs.path.join(allocator, &[_][]const u8{ path, filename });
-    }
-
-    fn getSaveFilename(self: *const Self, allocator: Allocator) ![]const u8 {
-        const title_str = span(&escape(self.title));
-        const name = if (title_str.len != 0) title_str else "untitled";
-
-        return try std.mem.concat(allocator, u8, &[_][]const u8{ name, ".sav" });
-    }
-
-    fn writeSaveToDisk(self: Self, allocator: Allocator, path: []const u8) !void {
-        const file_path = try self.getSaveFilePath(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.UnsupportedBackupKind,
-        }
-    }
-
     pub fn read(self: *const Self, address: usize) u8 {
         const addr = address & 0xFFFF;
 
@@ -184,7 +74,7 @@ pub const Backup = struct {
         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.eraseSector(self.buf, addr);
+                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) {
@@ -209,358 +99,121 @@ pub const Backup = struct {
             .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 Needle = struct {
-    const Self = @This();
-
-    str: []const u8,
-    kind: Backup.Kind,
-
-    fn init(str: []const u8, kind: Backup.Kind) Self {
-        return .{
-            .str = str,
-            .kind = kind,
-        };
-    }
-};
-
-const SaveError = error{
-    UnsupportedBackupKind,
-};
-
-const Flash = struct {
-    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,
-    };
-
-    fn init() Self {
-        return .{
-            .state = .Ready,
-            .id_mode = false,
-            .set_bank = false,
-            .prep_erase = false,
-            .prep_write = false,
-            .bank = 0,
-        };
-    }
-
-    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 => {
+        const buf = try allocator.alloc(u8, buf_size);
         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;
-    }
-
-    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;
-    }
-
-    fn write(self: *Self, buf: []u8, idx: usize, byte: u8) void {
-        buf[self.baseAddress() + idx] = byte;
-        self.prep_write = false;
-    }
-
-    fn read(self: *const Self, buf: []u8, idx: usize) u8 {
-        return buf[self.baseAddress() + idx];
-    }
-
-    fn eraseSector(self: *Self, buf: []u8, idx: usize) void {
-        const start = self.baseAddress() + (idx & 0xF000);
-
-        std.mem.set(u8, buf[start..][0..0x1000], 0xFF);
-        self.prep_erase = false;
-        self.state = .Ready;
-    }
-
-    inline fn baseAddress(self: *const Self) usize {
-        return if (self.bank == 1) 0x10000 else @as(usize, 0);
-    }
-};
-
-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,
-    };
-
-    fn init(allocator: Allocator) Self {
-        return .{
-            .kind = .Unknown,
-            .state = .Ready,
-            .writer = Writer.init(),
-            .reader = Reader.init(),
-            .addr = 0,
+        var backup = Self{
+            .buf = buf,
             .allocator = allocator,
-        };
-    }
-
-    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,
+            .kind = kind,
+            .title = title,
+            .save_path = path,
+            .flash = Flash.create(),
+            .eeprom = Eeprom.create(allocator),
         };
 
-                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 (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;
+
+            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;
             }
         }
 
-        if (self.state == .RequestEnd) {
-            if (bit != 0) log.debug("EEPROM Request did not end in 0u1. TODO: is this ok?", .{});
-            self.state = .Ready;
-            return;
+        return .None;
     }
 
-        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
+    fn readSave(self: *Self, allocator: Allocator, path: []const u8) !void {
+        const file_path = try self.savePath(allocator, path);
+        defer allocator.free(file_path);
+
+        // FIXME: Don't rely on this lol
+        if (std.mem.eql(u8, file_path[file_path.len - 12 .. file_path.len], "untitled.sav")) {
+            return log.err("ROM header lacks title, no save loaded", .{});
         }
 
-        self.tick(buf.*);
-    }
+        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);
 
-    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]);
+            .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});
+                }
 
-                            self.reader.configure(value);
-                            self.state = .RequestEnd;
-                        }
+                log.err("{s} is {} bytes, but we expected {} bytes", .{ file_path, file_buf.len, self.buf.len });
             },
-                    .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]);
+            .Eeprom => {
+                if (file_buf.len == 0x200 or file_buf.len == 0x2000) {
+                    self.eeprom.kind = if (file_buf.len == 0x200) .Small else .Large;
 
-                            self.reader.configure(value);
-                            self.state = .RequestEnd;
+                    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,
+                });
             },
-                    else => log.err("Unable to calculate EEPROM read address. EEPROM size UNKNOWN", .{}),
+            .None => return SaveError.Unsupported,
         }
-            },
-            .Write => {
+    }
+
+    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 = span(&escape(self.title));
+        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) {
-                    .Large => {
-                        if (self.writer.len() == 14) {
-                            self.addr = @intCast(u10, self.writer.finish());
-                            self.state = .WriteTransfer;
-                        }
+            .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});
             },
-                    .Small => {
-                        if (self.writer.len() == 6) {
-                            self.addr = @intCast(u6, self.writer.finish());
-                            self.state = .WriteTransfer;
+            else => return SaveError.Unsupported,
         }
-                    },
-                    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 This = @This();
-
-        data: u64,
-        i: u8,
-        enabled: bool,
-
-        fn init() This {
-            return .{
-                .data = 0,
-                .i = 0,
-                .enabled = false,
-            };
-        }
-
-        fn configure(self: *This, value: u64) void {
-            self.data = value;
-            self.i = 0;
-            self.enabled = true;
-        }
-
-        fn read(self: *This) 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 This) 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;
     }
 };
-
-    const Writer = struct {
-        const This = @This();
-
-        data: u64,
-        i: u8,
-
-        fn init() This {
-            return .{ .data = 0, .i = 0 };
-        }
-
-        fn requestWrite(self: *This, 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: *This, 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: *This, 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 This) u8 {
-            return self.i;
-        }
-
-        fn finish(self: *This) u64 {
-            defer self.reset();
-            return self.data;
-        }
-
-        fn reset(self: *This) void {
-            self.i = 0;
-            self.data = 0;
-        }
-    };
-};

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

@@ -8,6 +8,9 @@ const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
 pub const DmaTuple = std.meta.Tuple(&[_]type{ DmaController(0), DmaController(1), DmaController(2), DmaController(3) });
 const log = std.log.scoped(.DmaTransfer);
 
+const setHi = util.setHi;
+const setLo = util.setLo;
+
 pub fn create() DmaTuple {
     return .{ DmaController(0).init(), DmaController(1).init(), DmaController(2).init(), DmaController(3).init() };
 }
@@ -40,48 +43,48 @@ pub fn write(comptime T: type, dma: *DmaTuple, addr: u32, value: T) void {
 
     switch (T) {
         u32 => switch (byte) {
-            0xB0 => dma.*[0].setSad(value),
-            0xB4 => dma.*[0].setDad(value),
-            0xB8 => dma.*[0].setCnt(value),
-            0xBC => dma.*[1].setSad(value),
-            0xC0 => dma.*[1].setDad(value),
-            0xC4 => dma.*[1].setCnt(value),
-            0xC8 => dma.*[2].setSad(value),
-            0xCC => dma.*[2].setDad(value),
-            0xD0 => dma.*[2].setCnt(value),
-            0xD4 => dma.*[3].setSad(value),
-            0xD8 => dma.*[3].setDad(value),
-            0xDC => dma.*[3].setCnt(value),
+            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) {
-            0xB0 => dma.*[0].setSad(setU32L(dma.*[0].sad, value)),
-            0xB2 => dma.*[0].setSad(setU32H(dma.*[0].sad, value)),
-            0xB4 => dma.*[0].setDad(setU32L(dma.*[0].dad, value)),
-            0xB6 => dma.*[0].setDad(setU32H(dma.*[0].dad, value)),
-            0xB8 => dma.*[0].setCntL(value),
-            0xBA => dma.*[0].setCntH(value),
+            0xB0 => dma.*[0].setDmasad(setLo(u32, dma.*[0].sad, value)),
+            0xB2 => dma.*[0].setDmasad(setHi(u32, dma.*[0].sad, value)),
+            0xB4 => dma.*[0].setDmadad(setLo(u32, dma.*[0].dad, value)),
+            0xB6 => dma.*[0].setDmadad(setHi(u32, dma.*[0].dad, value)),
+            0xB8 => dma.*[0].setDmacntL(value),
+            0xBA => dma.*[0].setDmacntH(value),
 
-            0xBC => dma.*[1].setSad(setU32L(dma.*[1].sad, value)),
-            0xBE => dma.*[1].setSad(setU32H(dma.*[1].sad, value)),
-            0xC0 => dma.*[1].setDad(setU32L(dma.*[1].dad, value)),
-            0xC2 => dma.*[1].setDad(setU32H(dma.*[1].dad, value)),
-            0xC4 => dma.*[1].setCntL(value),
-            0xC6 => dma.*[1].setCntH(value),
+            0xBC => dma.*[1].setDmasad(setLo(u32, dma.*[1].sad, value)),
+            0xBE => dma.*[1].setDmasad(setHi(u32, dma.*[1].sad, value)),
+            0xC0 => dma.*[1].setDmadad(setLo(u32, dma.*[1].dad, value)),
+            0xC2 => dma.*[1].setDmadad(setHi(u32, dma.*[1].dad, value)),
+            0xC4 => dma.*[1].setDmacntL(value),
+            0xC6 => dma.*[1].setDmacntH(value),
 
-            0xC8 => dma.*[2].setSad(setU32L(dma.*[2].sad, value)),
-            0xCA => dma.*[2].setSad(setU32H(dma.*[2].sad, value)),
-            0xCC => dma.*[2].setDad(setU32L(dma.*[2].dad, value)),
-            0xCE => dma.*[2].setDad(setU32H(dma.*[2].dad, value)),
-            0xD0 => dma.*[2].setCntL(value),
-            0xD2 => dma.*[2].setCntH(value),
+            0xC8 => dma.*[2].setDmasad(setLo(u32, dma.*[2].sad, value)),
+            0xCA => dma.*[2].setDmasad(setHi(u32, dma.*[2].sad, value)),
+            0xCC => dma.*[2].setDmadad(setLo(u32, dma.*[2].dad, value)),
+            0xCE => dma.*[2].setDmadad(setHi(u32, dma.*[2].dad, value)),
+            0xD0 => dma.*[2].setDmacntL(value),
+            0xD2 => dma.*[2].setDmacntH(value),
 
-            0xD4 => dma.*[3].setSad(setU32L(dma.*[3].sad, value)),
-            0xD6 => dma.*[3].setSad(setU32H(dma.*[3].sad, value)),
-            0xD8 => dma.*[3].setDad(setU32L(dma.*[3].dad, value)),
-            0xDA => dma.*[3].setDad(setU32H(dma.*[3].dad, value)),
-            0xDC => dma.*[3].setCntL(value),
-            0xDE => dma.*[3].setCntH(value),
+            0xD4 => dma.*[3].setDmasad(setLo(u32, dma.*[3].sad, value)),
+            0xD6 => dma.*[3].setDmasad(setHi(u32, dma.*[3].sad, value)),
+            0xD8 => dma.*[3].setDmadad(setLo(u32, dma.*[3].dad, value)),
+            0xDA => dma.*[3].setDmadad(setHi(u32, dma.*[3].dad, 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 => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, addr }),
@@ -110,15 +113,12 @@ fn DmaController(comptime id: u2) type {
         cnt: DmaControl,
 
         /// Internal. Currrent Source Address
-        _sad: u32,
+        sad_latch: u32,
         /// Internal. Current Destination Address
-        _dad: u32,
+        dad_latch: u32,
         /// Internal. Word Count
         _word_count: if (id == 3) u16 else u14,
 
-        // Internal. FIFO Word Count
-        _fifo_word_count: u8,
-
         /// 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
@@ -132,33 +132,32 @@ fn DmaController(comptime id: u2) type {
                 .cnt = .{ .raw = 0x000 },
 
                 // Internals
-                ._sad = 0,
-                ._dad = 0,
+                .sad_latch = 0,
+                .dad_latch = 0,
                 ._word_count = 0,
-                ._fifo_word_count = 4,
                 .in_progress = false,
             };
         }
 
-        pub fn setSad(self: *Self, addr: u32) void {
+        pub fn setDmasad(self: *Self, addr: u32) void {
             self.sad = addr & sad_mask;
         }
 
-        pub fn setDad(self: *Self, addr: u32) void {
+        pub fn setDmadad(self: *Self, addr: u32) void {
             self.dad = addr & dad_mask;
         }
 
-        pub fn setCntL(self: *Self, halfword: u16) void {
+        pub fn setDmacntL(self: *Self, halfword: u16) void {
             self.word_count = @truncate(@TypeOf(self.word_count), halfword);
         }
 
-        pub fn setCntH(self: *Self, halfword: u16) void {
+        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 = self.sad;
-                self._dad = self.dad;
+                self.sad_latch = self.sad;
+                self.dad_latch = self.dad;
                 self._word_count = if (self.word_count == 0) std.math.maxInt(@TypeOf(self._word_count)) else self.word_count;
 
                 // Only a Start Timing of 00 has a DMA Transfer immediately begin
@@ -168,15 +167,15 @@ fn DmaController(comptime id: u2) type {
             self.cnt.raw = halfword;
         }
 
-        pub fn setCnt(self: *Self, word: u32) void {
-            self.setCntL(@truncate(u16, word));
-            self.setCntH(@truncate(u16, word >> 16));
+        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 = Self.adjustment(self.cnt.sad_adj.read());
-            const dad_adj = if (is_fifo) .Fixed else Self.adjustment(self.cnt.dad_adj.read());
+            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);
@@ -184,22 +183,22 @@ fn DmaController(comptime id: u2) type {
             const mask = if (transfer_type) ~@as(u32, 3) else ~@as(u32, 1);
 
             if (transfer_type) {
-                cpu.bus.write(u32, self._dad & mask, cpu.bus.read(u32, self._sad & mask));
+                cpu.bus.write(u32, self.dad_latch & mask, cpu.bus.read(u32, self.sad_latch & mask));
             } else {
-                cpu.bus.write(u16, self._dad & mask, cpu.bus.read(u16, self._sad & mask));
+                cpu.bus.write(u16, self.dad_latch & mask, cpu.bus.read(u16, self.sad_latch & mask));
             }
 
             switch (sad_adj) {
-                .Increment => self._sad +%= offset,
-                .Decrement => self._sad -%= offset,
-                // TODO: Is just ignoring this ok?
+                .Increment => self.sad_latch +%= offset,
+                .Decrement => self.sad_latch -%= offset,
+                // FIXME: Is just ignoring this ok?
                 .IncrementReload => log.err("{} is a prohibited adjustment on SAD", .{sad_adj}),
                 .Fixed => {},
             }
 
             switch (dad_adj) {
-                .Increment, .IncrementReload => self._dad +%= offset,
-                .Decrement => self._dad -%= offset,
+                .Increment, .IncrementReload => self.dad_latch +%= offset,
+                .Decrement => self.dad_latch -%= offset,
                 .Fixed => {},
             }
 
@@ -227,7 +226,7 @@ fn DmaController(comptime id: u2) type {
             }
         }
 
-        pub fn pollBlankingDma(self: *Self, comptime kind: DmaKind) void {
+        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
@@ -243,11 +242,11 @@ fn DmaController(comptime id: u2) type {
             // 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 (Self.adjustment(self.cnt.dad_adj.read()) == .IncrementReload) self._dad = self.dad;
+                if (@intToEnum(Adjustment, self.cnt.dad_adj.read()) == .IncrementReload) self.dad_latch = self.dad;
             }
         }
 
-        pub fn requestSoundDma(self: *Self, _: u32) void {
+        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
 
@@ -259,23 +258,19 @@ fn DmaController(comptime id: u2) type {
             // We Assume DMACNT_L is set to 4
 
             // FIXME: Safe to just assume whatever DAD is set to is the FIFO Address?
-            // self._dad = fifo_addr;
+            // self.dad_latch = fifo_addr;
             self.cnt.repeat.set();
             self._word_count = 4;
             self.in_progress = true;
         }
-
-        fn adjustment(idx: u2) Adjustment {
-            return std.meta.intToEnum(Adjustment, idx) catch unreachable;
-        }
     };
 }
 
-pub fn pollBlankingDma(bus: *Bus, comptime kind: DmaKind) void {
-    bus.dma[0].pollBlankingDma(kind);
-    bus.dma[1].pollBlankingDma(kind);
-    bus.dma[2].pollBlankingDma(kind);
-    bus.dma[3].pollBlankingDma(kind);
+pub fn pollDmaOnBlank(bus: *Bus, comptime kind: DmaKind) void {
+    bus.dma[0].poll(kind);
+    bus.dma[1].poll(kind);
+    bus.dma[2].poll(kind);
+    bus.dma[3].poll(kind);
 }
 
 const Adjustment = enum(u2) {
@@ -291,11 +286,3 @@ const DmaKind = enum(u2) {
     VBlank,
     Special,
 };
-
-fn setU32L(left: u32, right: u16) u32 {
-    return (left & 0xFFFF_0000) | right;
-}
-
-fn setU32H(left: u32, right: u16) u32 {
-    return (left & 0x0000_FFFF) | (@as(u32, right) << 16);
-}

src/core/bus/gpio.zig

@@ -288,7 +288,7 @@ pub const Clock = struct {
         cpu.sched.push(.RealTimeClock, 1 << 24); // Every Second
     }
 
-    pub fn updateTime(self: *Self, late: u64) void {
+    pub fn onClockUpdate(self: *Self, late: u64) void {
         self.cpu.sched.push(.RealTimeClock, (1 << 24) -| late); // Reschedule
 
         const now = DateTime.now();

src/core/bus/io.zig

@@ -11,6 +11,9 @@ const Bus = @import("../Bus.zig");
 const DmaController = @import("dma.zig").DmaController;
 const Scheduler = @import("../scheduler.zig").Scheduler;
 
+const setHi = util.setLo;
+const setLo = util.setHi;
+
 const log = std.log.scoped(.@"I/O");
 
 pub const Io = struct {
@@ -233,18 +236,18 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
             0x0400_0022 => bus.ppu.aff_bg[0].pb = @bitCast(i16, value),
             0x0400_0024 => bus.ppu.aff_bg[0].pc = @bitCast(i16, value),
             0x0400_0026 => bus.ppu.aff_bg[0].pd = @bitCast(i16, value),
-            0x0400_0028 => bus.ppu.aff_bg[0].x = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[0].x) & 0xFFFF_0000 | value),
-            0x0400_002A => bus.ppu.aff_bg[0].x = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[0].x) & 0x0000_FFFF | (@as(u32, value) << 16)),
-            0x0400_002C => bus.ppu.aff_bg[0].y = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[0].y) & 0xFFFF_0000 | value),
-            0x0400_002E => bus.ppu.aff_bg[0].y = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[0].y) & 0x0000_FFFF | (@as(u32, value) << 16)),
+            0x0400_0028 => bus.ppu.aff_bg[0].x = @bitCast(i32, setLo(u32, @bitCast(u32, bus.ppu.aff_bg[0].x), value)),
+            0x0400_002A => bus.ppu.aff_bg[0].x = @bitCast(i32, setHi(u32, @bitCast(u32, bus.ppu.aff_bg[0].x), value)),
+            0x0400_002C => bus.ppu.aff_bg[0].y = @bitCast(i32, setLo(u32, @bitCast(u32, bus.ppu.aff_bg[0].y), value)),
+            0x0400_002E => bus.ppu.aff_bg[0].y = @bitCast(i32, setHi(u32, @bitCast(u32, bus.ppu.aff_bg[0].y), value)),
             0x0400_0030 => bus.ppu.aff_bg[1].pa = @bitCast(i16, value),
             0x0400_0032 => bus.ppu.aff_bg[1].pb = @bitCast(i16, value),
             0x0400_0034 => bus.ppu.aff_bg[1].pc = @bitCast(i16, value),
             0x0400_0036 => bus.ppu.aff_bg[1].pd = @bitCast(i16, value),
-            0x0400_0038 => bus.ppu.aff_bg[1].x = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].x) & 0xFFFF_0000 | value),
-            0x0400_003A => bus.ppu.aff_bg[1].x = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].x) & 0x0000_FFFF | (@as(u32, value) << 16)),
-            0x0400_003C => bus.ppu.aff_bg[1].y = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].y) & 0xFFFF_0000 | value),
-            0x0400_003E => bus.ppu.aff_bg[1].y = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].y) & 0x0000_FFFF | (@as(u32, value) << 16)),
+            0x0400_0038 => bus.ppu.aff_bg[1].x = @bitCast(i32, setLo(u32, @bitCast(u32, bus.ppu.aff_bg[1].x), value)),
+            0x0400_003A => bus.ppu.aff_bg[1].x = @bitCast(i32, setHi(u32, @bitCast(u32, bus.ppu.aff_bg[1].x), value)),
+            0x0400_003C => bus.ppu.aff_bg[1].y = @bitCast(i32, setLo(u32, @bitCast(u32, bus.ppu.aff_bg[1].y), value)),
+            0x0400_003E => bus.ppu.aff_bg[1].y = @bitCast(i32, setHi(u32, @bitCast(u32, bus.ppu.aff_bg[1].y), value)),
             0x0400_0040 => bus.ppu.win.h[0].raw = value,
             0x0400_0042 => bus.ppu.win.h[1].raw = value,
             0x0400_0044 => bus.ppu.win.v[0].raw = value,
@@ -296,16 +299,16 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
         },
         u8 => switch (address) {
             // Display
-            0x0400_0004 => bus.ppu.dispstat.raw = (bus.ppu.dispstat.raw & 0xFF00) | value,
-            0x0400_0005 => bus.ppu.dispstat.raw = (@as(u16, value) << 8) | (bus.ppu.dispstat.raw & 0xFF),
-            0x0400_0008 => bus.ppu.bg[0].cnt.raw = (bus.ppu.bg[0].cnt.raw & 0xFF00) | value,
-            0x0400_0009 => bus.ppu.bg[0].cnt.raw = (@as(u16, value) << 8) | (bus.ppu.bg[0].cnt.raw & 0xFF),
-            0x0400_000A => bus.ppu.bg[1].cnt.raw = (bus.ppu.bg[1].cnt.raw & 0xFF00) | value,
-            0x0400_000B => bus.ppu.bg[1].cnt.raw = (@as(u16, value) << 8) | (bus.ppu.bg[1].cnt.raw & 0xFF),
-            0x0400_0048 => bus.ppu.win.setInL(value),
-            0x0400_0049 => bus.ppu.win.setInH(value),
-            0x0400_004A => bus.ppu.win.setOutL(value),
-            0x0400_0054 => bus.ppu.bldy.raw = (bus.ppu.bldy.raw & 0xFF00) | value,
+            0x0400_0004 => bus.ppu.dispstat.raw = setLo(u16, bus.ppu.dispstat.raw, value),
+            0x0400_0005 => bus.ppu.dispstat.raw = setHi(u16, bus.ppu.dispstat.raw, value),
+            0x0400_0008 => bus.ppu.bg[0].cnt.raw = setLo(u16, bus.ppu.bg[0].cnt.raw, value),
+            0x0400_0009 => bus.ppu.bg[0].cnt.raw = setHi(u16, bus.ppu.bg[0].cnt.raw, value),
+            0x0400_000A => bus.ppu.bg[1].cnt.raw = setLo(u16, bus.ppu.bg[1].cnt.raw, value),
+            0x0400_000B => bus.ppu.bg[1].cnt.raw = setHi(u16, bus.ppu.bg[1].cnt.raw, value),
+            0x0400_0048 => bus.ppu.win.in.raw = setLo(u16, bus.ppu.win.in.raw, value),
+            0x0400_0049 => bus.ppu.win.in.raw = setHi(u16, bus.ppu.win.in.raw, value),
+            0x0400_004A => bus.ppu.win.out.raw = setLo(u16, bus.ppu.win.out.raw, value),
+            0x0400_0054 => bus.ppu.bldy.raw = setLo(u16, bus.ppu.bldy.raw, value),
 
             // Sound
             0x0400_0060...0x0400_00A7 => apu.write(T, &bus.apu, address, value),

src/core/bus/timer.zig

@@ -19,20 +19,20 @@ pub fn read(comptime T: type, tim: *const TimerTuple, addr: u32) ?T {
 
     return switch (T) {
         u32 => switch (nybble) {
-            0x0 => @as(T, tim.*[0].cnt.raw) << 16 | tim.*[0].getCntL(),
-            0x4 => @as(T, tim.*[1].cnt.raw) << 16 | tim.*[1].getCntL(),
-            0x8 => @as(T, tim.*[2].cnt.raw) << 16 | tim.*[2].getCntL(),
-            0xC => @as(T, tim.*[3].cnt.raw) << 16 | tim.*[3].getCntL(),
+            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.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, addr }),
         },
         u16 => switch (nybble) {
-            0x0 => tim.*[0].getCntL(),
+            0x0 => tim.*[0].timcntL(),
             0x2 => tim.*[0].cnt.raw,
-            0x4 => tim.*[1].getCntL(),
+            0x4 => tim.*[1].timcntL(),
             0x6 => tim.*[1].cnt.raw,
-            0x8 => tim.*[2].getCntL(),
+            0x8 => tim.*[2].timcntL(),
             0xA => tim.*[2].cnt.raw,
-            0xC => tim.*[3].getCntL(),
+            0xC => tim.*[3].timcntL(),
             0xE => tim.*[3].cnt.raw,
             else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, addr }),
         },
@@ -46,21 +46,21 @@ pub fn write(comptime T: type, tim: *TimerTuple, addr: u32, value: T) void {
 
     return switch (T) {
         u32 => switch (nybble) {
-            0x0 => tim.*[0].setCnt(value),
-            0x4 => tim.*[1].setCnt(value),
-            0x8 => tim.*[2].setCnt(value),
-            0xC => tim.*[3].setCnt(value),
+            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) {
-            0x0 => tim.*[0].setCntL(value),
-            0x2 => tim.*[0].setCntH(value),
-            0x4 => tim.*[1].setCntL(value),
-            0x6 => tim.*[1].setCntH(value),
-            0x8 => tim.*[2].setCntL(value),
-            0xA => tim.*[2].setCntH(value),
-            0xC => tim.*[3].setCntL(value),
-            0xE => tim.*[3].setCntH(value),
+            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 => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, addr }),
@@ -72,13 +72,13 @@ fn Timer(comptime id: u2) type {
     return struct {
         const Self = @This();
 
-        /// Read Only, Internal. Please use self.getCntL()
+        /// Read Only, Internal. Please use self.timcntL()
         _counter: u16,
 
-        /// Write Only, Internal. Please use self.setCntL()
+        /// Write Only, Internal. Please use self.setTimcntL()
         _reload: u16,
 
-        /// Write Only, Internal. Please use self.setCntH()
+        /// Write Only, Internal. Please use self.setTimcntH()
         cnt: TimerControl,
 
         /// Internal.
@@ -97,26 +97,26 @@ fn Timer(comptime id: u2) type {
             };
         }
 
-        /// TIMCNT_L
-        pub fn getCntL(self: *const Self) u16 {
+        /// 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
-        pub fn setCntL(self: *Self, halfword: u16) void {
+        /// TIMCNT_L Setter
+        pub fn setTimcntL(self: *Self, halfword: u16) void {
             self._reload = halfword;
         }
 
         /// TIMCNT_L & TIMCNT_H
-        pub fn setCnt(self: *Self, word: u32) void {
-            self.setCntL(@truncate(u16, word));
-            self.setCntH(@truncate(u16, word >> 16));
+        pub fn setTimcnt(self: *Self, word: u32) void {
+            self.setTimcntL(@truncate(u16, word));
+            self.setTimcntH(@truncate(u16, word >> 16));
         }
 
         /// TIMCNT_H
-        pub fn setCntH(self: *Self, halfword: u16) void {
+        pub fn setTimcntH(self: *Self, halfword: u16) void {
             const new = TimerControl{ .raw = halfword };
 
             // If Timer happens to be enabled, It will either be resheduled or disabled
@@ -132,12 +132,12 @@ fn Timer(comptime id: u2) type {
             if (!self.cnt.enabled.read() and new.enabled.read()) self._counter = self._reload;
 
             // If Timer is enabled and we're not cascading, we need to schedule an overflow event
-            if (new.enabled.read() and !new.cascade.read()) self.scheduleOverflow(0);
+            if (new.enabled.read() and !new.cascade.read()) self.rescheduleTimerExpire(0);
 
             self.cnt.raw = halfword;
         }
 
-        pub fn handleOverflow(self: *Self, cpu: *Arm7tdmi, late: u64) void {
+        pub fn onTimerExpire(self: *Self, cpu: *Arm7tdmi, late: u64) void {
             // Fire IRQ if enabled
             const io = &cpu.bus.io;
 
@@ -154,22 +154,22 @@ fn Timer(comptime id: u2) type {
 
             // DMA Sound Things
             if (id == 0 or id == 1) {
-                cpu.bus.apu.handleTimerOverflow(cpu, id);
+                cpu.bus.apu.onDmaAudioSampleRequest(cpu, id);
             }
 
             // Perform Cascade Behaviour
             switch (id) {
                 0 => if (cpu.bus.tim[1].cnt.cascade.read()) {
                     cpu.bus.tim[1]._counter +%= 1;
-                    if (cpu.bus.tim[1]._counter == 0) cpu.bus.tim[1].handleOverflow(cpu, late);
+                    if (cpu.bus.tim[1]._counter == 0) cpu.bus.tim[1].onTimerExpire(cpu, late);
                 },
                 1 => if (cpu.bus.tim[2].cnt.cascade.read()) {
                     cpu.bus.tim[2]._counter +%= 1;
-                    if (cpu.bus.tim[2]._counter == 0) cpu.bus.tim[2].handleOverflow(cpu, late);
+                    if (cpu.bus.tim[2]._counter == 0) cpu.bus.tim[2].onTimerExpire(cpu, late);
                 },
                 2 => if (cpu.bus.tim[3].cnt.cascade.read()) {
                     cpu.bus.tim[3]._counter +%= 1;
-                    if (cpu.bus.tim[3]._counter == 0) cpu.bus.tim[3].handleOverflow(cpu, late);
+                    if (cpu.bus.tim[3]._counter == 0) cpu.bus.tim[3].onTimerExpire(cpu, late);
                 },
                 3 => {}, // There is no Timer for TIM3 to "cascade" to,
             }
@@ -177,11 +177,11 @@ fn Timer(comptime id: u2) type {
             // Reschedule Timer if we're not cascading
             if (!self.cnt.cascade.read()) {
                 self._counter = self._reload;
-                self.scheduleOverflow(late);
+                self.rescheduleTimerExpire(late);
             }
         }
 
-        fn scheduleOverflow(self: *Self, late: u64) void {
+        fn rescheduleTimerExpire(self: *Self, late: u64) void {
             const when = (@as(u64, 0x10000) - self._counter) * self.frequency();
 
             self._start_timestamp = self.sched.now();

src/core/cpu.zig

@@ -236,13 +236,13 @@ pub const thumb = struct {
     }
 };
 
-const cpu_logging = @import("emu.zig").cpu_logging;
 const log = std.log.scoped(.Arm7Tdmi);
 
 pub const Arm7tdmi = struct {
     const Self = @This();
 
     r: [16]u32,
+    pipe: Pipeline,
     sched: *Scheduler,
     bus: *Bus,
     cpsr: PSR,
@@ -263,6 +263,7 @@ pub const Arm7tdmi = struct {
     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 },
@@ -411,29 +412,43 @@ pub const Arm7tdmi = struct {
         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[0] = 0x08000000;
+        // self.r[1] = 0x000000EA;
         self.r[13] = 0x0300_7F00;
         self.r[15] = 0x0800_0000;
 
         self.banked_r[bankedIdx(.Irq, .R13)] = 0x0300_7FA0;
         self.banked_r[bankedIdx(.Supervisor, .R13)] = 0x0300_7FE0;
 
-        self.cpsr.raw = 0x6000001F;
+        // 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 = self.fetch(u16);
-            if (cpu_logging) self.logger.?.mgbaLog(self, opcode);
+            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.fetch(u32);
-            if (cpu_logging) self.logger.?.mgbaLog(self, opcode);
+            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);
@@ -473,42 +488,41 @@ pub const Arm7tdmi = struct {
     pub fn handleInterrupt(self: *Self) void {
         const should_handle = self.bus.io.ie.raw & self.bus.io.irq.raw;
 
-        if (should_handle != 0) {
+        // 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;
-            // log.debug("An Interrupt was Fired!", .{});
 
-            // Either IME is not true or I in CPSR is true
-            // Don't handle interrupts
-            if (!self.bus.io.ime or self.cpsr.i.read()) return;
-            // log.debug("An interrupt was Handled!", .{});
-
-            // retAddr.gba says r15 on it's own is off by -04h in both ARM and THUMB mode
-            const r15 = self.r[15] + 4;
-            const cpsr = self.cpsr.raw;
+        // 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] = r15;
-            self.spsr.raw = cpsr;
-            self.r[15] = 0x000_0018;
-        }
+        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) T {
+    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)
-        defer self.r[15] += if (T == u32) 4 else 2;
 
-        // FIXME: You better hope this is optimized out
+        // 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, self.r[15] >> 24)];
+        defer self.sched.tick = tick_cache + Bus.fetch_timings[@boolToInt(T == u32)][@truncate(u4, address >> 24)];
 
-        return self.bus.read(T, self.r[15]);
-    }
-
-    pub fn fakePC(self: *const Self) u32 {
-        return self.r[15] + 4;
+        return self.bus.read(T, address);
     }
 
     pub fn panic(self: *const Self, comptime format: []const u8, args: anytype) noreturn {
@@ -525,6 +539,8 @@ pub const Arm7tdmi = struct {
         std.debug.print("spsr: 0x{X:0>8} ", .{self.spsr.raw});
         prettyPrintPsr(&self.spsr);
 
+        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);
@@ -588,7 +604,7 @@ pub const Arm7tdmi = struct {
         const r12 = self.r[12];
         const r13 = self.r[13];
         const r14 = self.r[14];
-        const r15 = self.r[15];
+        const r15 = self.r[15] -| if (self.cpsr.t.read()) 2 else @as(u32, 4);
 
         const c_psr = self.cpsr.raw;
 
@@ -596,7 +612,7 @@ pub const Arm7tdmi = struct {
         if (self.cpsr.t.read()) {
             if (opcode >> 11 == 0x1E) {
                 // Instruction 1 of a BL Opcode, print in ARM mode
-                const other_half = self.bus.dbgRead(u16, self.r[15]);
+                const other_half = self.bus.debugRead(u16, self.r[15] - 2);
                 const bl_opcode = @as(u32, opcode) << 16 | other_half;
 
                 log_str = try std.fmt.bufPrint(&buf, arm_fmt, .{ r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15, c_psr, bl_opcode });
@@ -632,6 +648,49 @@ pub fn checkCond(cpsr: PSR, cond: u4) bool {
     };
 }
 
+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);
+
+        // FIXME: https://github.com/ziglang/zig/issues/12642
+        var 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),

src/core/cpu/arm/block_data_transfer.zig

@@ -55,8 +55,10 @@ pub fn blockDataTransfer(comptime P: bool, comptime U: bool, comptime S: bool, c
 
                 if (L) {
                     cpu.r[15] = bus.read(u32, und_addr);
+                    cpu.pipe.reload(cpu);
                 } else {
-                    bus.write(u32, und_addr, cpu.r[15] + 8);
+                    // FIXME: Should r15 on write be +12 ahead?
+                    bus.write(u32, und_addr, cpu.r[15] + 4);
                 }
 
                 cpu.r[rn] = if (U) cpu.r[rn] + 0x40 else cpu.r[rn] - 0x40;
@@ -86,17 +88,23 @@ pub fn blockDataTransfer(comptime P: bool, comptime U: bool, comptime S: bool, c
                     cpu.setUserModeRegister(i, bus.read(u32, address));
                 } else {
                     const value = bus.read(u32, address);
-                    cpu.r[i] = if (i == 0xF) value & 0xFFFF_FFFC else value;
-                    if (S and i == 0xF) cpu.setCpsr(cpu.spsr.raw);
+
+                    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) 8 else @as(u32, 0)); // PC is already 4 ahead to make 12
+                    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) 8 else @as(u32, 0));
+                    bus.write(u32, address, cpu.r[i] + if (i == 0xF) 4 else @as(u32, 0));
                 }
             }
         }

src/core/cpu/arm/branch.zig

@@ -9,14 +9,20 @@ 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];
-            cpu.r[15] = cpu.fakePC() +% (sext(u32, u24, opcode) << 2);
+            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;
-    cpu.cpsr.t.write(cpu.r[rn] & 1 == 1);
-    cpu.r[15] = cpu.r[rn] & 0xFFFF_FFFE;
+
+    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

@@ -2,10 +2,10 @@ const Bus = @import("../../Bus.zig");
 const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
 const InstrFn = @import("../../cpu.zig").arm.InstrFn;
 
-const rotateRight = @import("../barrel_shifter.zig").rotateRight;
-const execute = @import("../barrel_shifter.zig").execute;
+const exec = @import("../barrel_shifter.zig").exec;
+const ror = @import("../barrel_shifter.zig").ror;
 
-pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime instrKind: u4) InstrFn {
+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);
@@ -13,269 +13,168 @@ pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime instrKind: u4
             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 op1 = if (rn == 0xF) cpu.fakePC() else cpu.r[rn];
-
-            var op2: u32 = undefined;
-            if (I) {
             const amount = @truncate(u8, (opcode >> 8 & 0xF) << 1);
-                op2 = rotateRight(S, &cpu.cpsr, opcode & 0xFF, amount);
-            } else {
-                op2 = execute(S, cpu, opcode);
-            }
+            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;
 
-            switch (instrKind) {
-                0x0 => {
-                    // AND
-                    const result = op1 & op2;
-                    cpu.r[rd] = result;
-                    setArmLogicOpFlags(S, cpu, rd, result);
-                },
-                0x1 => {
-                    // EOR
-                    const result = op1 ^ op2;
-                    cpu.r[rd] = result;
-                    setArmLogicOpFlags(S, cpu, rd, result);
-                },
-                0x2 => {
-                    // SUB
-                    cpu.r[rd] = armSub(S, cpu, rd, op1, op2);
-                },
-                0x3 => {
-                    // RSB
-                    cpu.r[rd] = armSub(S, cpu, rd, op2, op1);
-                },
-                0x4 => {
-                    // ADD
-                    cpu.r[rd] = armAdd(S, cpu, rd, op1, op2);
-                },
-                0x5 => {
-                    // ADC
-                    cpu.r[rd] = armAdc(S, cpu, rd, op1, op2, old_carry);
-                },
-                0x6 => {
-                    // SBC
-                    cpu.r[rd] = armSbc(S, cpu, rd, op1, op2, old_carry);
-                },
-                0x7 => {
-                    // RSC
-                    cpu.r[rd] = armSbc(S, cpu, rd, op2, op1, old_carry);
-                },
+            var result: u32 = undefined;
+            var overflow: bool = 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) {
-                        undefinedTestBehaviour(cpu);
-                        return;
-                    }
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
 
-                    const result = op1 & op2;
-                    setTestOpFlags(S, cpu, opcode, result);
+                    result = op1 & op2;
                 },
                 0x9 => {
                     // TEQ
-                    if (rd == 0xF) {
-                        undefinedTestBehaviour(cpu);
-                        return;
-                    }
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
 
-                    const result = op1 ^ op2;
-                    setTestOpFlags(S, cpu, opcode, result);
+                    result = op1 ^ op2;
                 },
                 0xA => {
                     // CMP
-                    if (rd == 0xF) {
-                        undefinedTestBehaviour(cpu);
-                        return;
-                    }
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
 
-                    cmp(cpu, op1, op2);
+                    result = op1 -% op2;
                 },
                 0xB => {
                     // CMN
-                    if (rd == 0xF) {
-                        undefinedTestBehaviour(cpu);
-                        return;
+                    if (rd == 0xF)
+                        return undefinedTestBehaviour(cpu);
+
+                    overflow = @addWithOverflow(u32, op1, op2, &result);
+                },
+                0xC => result = op1 | op2, // ORR
+                0xD => result = op2, // MOV
+                0xE => result = op1 & ~op2, // BIC
+                0xF => result = ~op2, // MVN
             }
 
-                    cmn(cpu, op1, op2);
-                },
-                0xC => {
-                    // ORR
-                    const result = op1 | op2;
+            // Write to Destination Register
+            switch (kind) {
+                0x8, 0x9, 0xA, 0xB => {}, // Test Operations
+                else => {
                     cpu.r[rd] = result;
-                    setArmLogicOpFlags(S, cpu, rd, result);
+                    if (rd == 0xF) {
+                        if (S) cpu.setCpsr(cpu.spsr.raw);
+                        cpu.pipe.reload(cpu);
+                    }
                 },
-                0xD => {
-                    // MOV
-                    cpu.r[rd] = op2;
-                    setArmLogicOpFlags(S, cpu, rd, op2);
+            }
+
+            // 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
+
                 },
-                0xE => {
-                    // BIC
-                    const result = op1 & ~op2;
-                    cpu.r[rd] = result;
-                    setArmLogicOpFlags(S, cpu, rd, result);
+                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);
+                    }
                 },
-                0xF => {
-                    // MVN
-                    const result = ~op2;
-                    cpu.r[rd] = result;
-                    setArmLogicOpFlags(S, cpu, rd, result);
+                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);
+                    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);
+                        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;
 }
 
-fn armSbc(comptime S: bool, cpu: *Arm7tdmi, rd: u4, left: u32, right: u32, old_carry: u1) u32 {
-    var result: u32 = undefined;
-    if (S and rd == 0xF) {
-        result = sbc(false, cpu, left, right, old_carry);
-        cpu.setCpsr(cpu.spsr.raw);
-    } else {
-        result = sbc(S, cpu, left, right, old_carry);
-    }
-
-    return result;
-}
-
-pub fn sbc(comptime S: bool, cpu: *Arm7tdmi, left: u32, right: u32, old_carry: u1) u32 {
+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 result = @truncate(u32, left -% subtrahend);
+    const ret = @truncate(u32, left -% subtrahend);
 
-    if (S) {
-        cpu.cpsr.n.write(result >> 31 & 1 == 1);
-        cpu.cpsr.z.write(result == 0);
-        cpu.cpsr.c.write(subtrahend <= left);
-        cpu.cpsr.v.write(((left ^ result) & (~right ^ result)) >> 31 & 1 == 1);
+    return ret;
 }
 
-    return result;
+pub fn add(overflow: *bool, left: u32, right: u32) u32 {
+    var ret: u32 = undefined;
+    overflow.* = @addWithOverflow(u32, left, right, &ret);
+    return ret;
 }
 
-fn armSub(comptime S: bool, cpu: *Arm7tdmi, rd: u4, left: u32, right: u32) u32 {
-    var result: u32 = undefined;
-    if (S and rd == 0xF) {
-        result = sub(false, cpu, left, right);
-        cpu.setCpsr(cpu.spsr.raw);
-    } else {
-        result = sub(S, cpu, left, right);
-    }
+pub fn adc(overflow: *bool, left: u32, right: u32, old_carry: u1) u32 {
+    var ret: u32 = undefined;
+    const first = @addWithOverflow(u32, left, right, &ret);
+    const second = @addWithOverflow(u32, ret, old_carry, &ret);
 
-    return result;
-}
-
-pub fn sub(comptime S: bool, cpu: *Arm7tdmi, left: u32, right: u32) u32 {
-    const result = left -% right;
-
-    if (S) {
-        cpu.cpsr.n.write(result >> 31 & 1 == 1);
-        cpu.cpsr.z.write(result == 0);
-        cpu.cpsr.c.write(right <= left);
-        cpu.cpsr.v.write(((left ^ result) & (~right ^ result)) >> 31 & 1 == 1);
-    }
-
-    return result;
-}
-
-fn armAdd(comptime S: bool, cpu: *Arm7tdmi, rd: u4, left: u32, right: u32) u32 {
-    var result: u32 = undefined;
-    if (S and rd == 0xF) {
-        result = add(false, cpu, left, right);
-        cpu.setCpsr(cpu.spsr.raw);
-    } else {
-        result = add(S, cpu, left, right);
-    }
-
-    return result;
-}
-
-pub fn add(comptime S: bool, cpu: *Arm7tdmi, left: u32, right: u32) u32 {
-    var result: u32 = undefined;
-    const didOverflow = @addWithOverflow(u32, left, right, &result);
-
-    if (S) {
-        cpu.cpsr.n.write(result >> 31 & 1 == 1);
-        cpu.cpsr.z.write(result == 0);
-        cpu.cpsr.c.write(didOverflow);
-        cpu.cpsr.v.write(((left ^ result) & (right ^ result)) >> 31 & 1 == 1);
-    }
-
-    return result;
-}
-
-fn armAdc(comptime S: bool, cpu: *Arm7tdmi, rd: u4, left: u32, right: u32, old_carry: u1) u32 {
-    var result: u32 = undefined;
-    if (S and rd == 0xF) {
-        result = adc(false, cpu, left, right, old_carry);
-        cpu.setCpsr(cpu.spsr.raw);
-    } else {
-        result = adc(S, cpu, left, right, old_carry);
-    }
-
-    return result;
-}
-
-pub fn adc(comptime S: bool, cpu: *Arm7tdmi, left: u32, right: u32, old_carry: u1) u32 {
-    var result: u32 = undefined;
-    const did = @addWithOverflow(u32, left, right, &result);
-    const overflow = @addWithOverflow(u32, result, old_carry, &result);
-
-    if (S) {
-        cpu.cpsr.n.write(result >> 31 & 1 == 1);
-        cpu.cpsr.z.write(result == 0);
-        cpu.cpsr.c.write(did or overflow);
-        cpu.cpsr.v.write(((left ^ result) & (right ^ result)) >> 31 & 1 == 1);
-    }
-
-    return result;
-}
-
-pub fn cmp(cpu: *Arm7tdmi, left: u32, right: u32) void {
-    const result = left -% right;
-
-    cpu.cpsr.n.write(result >> 31 & 1 == 1);
-    cpu.cpsr.z.write(result == 0);
-    cpu.cpsr.c.write(right <= left);
-    cpu.cpsr.v.write(((left ^ result) & (~right ^ result)) >> 31 & 1 == 1);
-}
-
-pub fn cmn(cpu: *Arm7tdmi, left: u32, right: u32) void {
-    var result: u32 = undefined;
-    const didOverflow = @addWithOverflow(u32, left, right, &result);
-
-    cpu.cpsr.n.write(result >> 31 & 1 == 1);
-    cpu.cpsr.z.write(result == 0);
-    cpu.cpsr.c.write(didOverflow);
-    cpu.cpsr.v.write(((left ^ result) & (right ^ result)) >> 31 & 1 == 1);
-}
-
-fn setArmLogicOpFlags(comptime S: bool, cpu: *Arm7tdmi, rd: u4, result: u32) void {
-    if (S and rd == 0xF) {
-        cpu.setCpsr(cpu.spsr.raw);
-    } else {
-        setLogicOpFlags(S, cpu, result);
-    }
-}
-
-pub fn setLogicOpFlags(comptime S: bool, cpu: *Arm7tdmi, result: u32) void {
-    if (S) {
-        cpu.cpsr.n.write(result >> 31 & 1 == 1);
-        cpu.cpsr.z.write(result == 0);
-        // C set by Barrel Shifter, V is unaffected
-    }
-}
-
-fn setTestOpFlags(comptime S: bool, cpu: *Arm7tdmi, opcode: u32, result: u32) void {
-    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) _ = execute(true, cpu, opcode);
+    overflow.* = first or second;
+    return ret;
 }
 
 fn undefinedTestBehaviour(cpu: *Arm7tdmi) void {

src/core/cpu/arm/half_signed_data_transfer.zig

@@ -15,20 +15,8 @@ pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I:
             const rm = opcode & 0xF;
             const imm_offset_high = opcode >> 8 & 0xF;
 
-            var base: u32 = undefined;
-            if (rn == 0xF) {
-                base = cpu.fakePC();
-                if (!L) base += 4;
-            } else {
-                base = cpu.r[rn];
-            }
-
-            var offset: u32 = undefined;
-            if (I) {
-                offset = imm_offset_high << 4 | rm;
-            } else {
-                offset = cpu.r[rm];
-            }
+            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;

src/core/cpu/arm/single_data_transfer.zig

@@ -14,15 +14,10 @@ pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool,
             const rn = opcode >> 16 & 0xF;
             const rd = opcode >> 12 & 0xF;
 
-            var base: u32 = undefined;
-            if (rn == 0xF) {
-                base = cpu.fakePC();
-                if (!L) base += 4; // Offset of 12
-            } else {
-                base = cpu.r[rn];
-            }
+            // rn is r15 and L is not set, the PC is 12 ahead
+            const base = cpu.r[rn] + if (!L and rn == 0xF) 4 else @as(u32, 0);
 
-            const offset = if (I) shifter.immShift(false, cpu, opcode) else opcode & 0xFFF;
+            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;
@@ -40,18 +35,26 @@ pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool,
             } else {
                 if (B) {
                     // STRB
-                    const value = if (rd == 0xF) cpu.r[rd] + 8 else cpu.r[rd];
+                    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 = if (rd == 0xF) cpu.r[rd] + 8 else cpu.r[rd];
+                    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 (L) cpu.r[rd] = result; // This emulates the LDR rd == rn behaviour
+            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

@@ -6,7 +6,7 @@ pub fn armSoftwareInterrupt() InstrFn {
     return struct {
         fn inner(cpu: *Arm7tdmi, _: *Bus, _: u32) void {
             // Copy Values from Current Mode
-            const r15 = cpu.r[15];
+            const ret_addr = cpu.r[15] - 4;
             const cpsr = cpu.cpsr.raw;
 
             // Switch Mode
@@ -14,9 +14,10 @@ pub fn armSoftwareInterrupt() InstrFn {
             cpu.cpsr.t.write(false); // Force ARM Mode
             cpu.cpsr.i.write(true); // Disable normal interrupts
 
-            cpu.r[14] = r15; // Resume Execution
+            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

@@ -5,37 +5,33 @@ const CPSR = @import("../cpu.zig").PSR;
 
 const rotr = @import("../../util.zig").rotr;
 
-pub fn execute(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
+pub fn exec(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
     var result: u32 = undefined;
     if (opcode >> 4 & 1 == 1) {
-        result = registerShift(S, cpu, opcode);
+        result = register(S, cpu, opcode);
     } else {
-        result = immShift(S, cpu, opcode);
+        result = immediate(S, cpu, opcode);
     }
 
     return result;
 }
 
-fn registerShift(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
+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]);
 
-    const rm_idx = opcode & 0xF;
-    const rm = if (rm_idx == 0xF) cpu.fakePC() else cpu.r[rm_idx];
-
     return switch (@truncate(u2, opcode >> 5)) {
-        0b00 => logicalLeft(S, &cpu.cpsr, rm, rs),
-        0b01 => logicalRight(S, &cpu.cpsr, rm, rs),
-        0b10 => arithmeticRight(S, &cpu.cpsr, rm, rs),
-        0b11 => rotateRight(S, &cpu.cpsr, rm, rs),
+        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 immShift(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
+pub fn immediate(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
     const amount = @truncate(u8, opcode >> 7 & 0x1F);
-
-    const rm_idx = opcode & 0xF;
-    const rm = if (rm_idx == 0xF) cpu.fakePC() else cpu.r[rm_idx];
+    const rm = cpu.r[opcode & 0xF];
 
     var result: u32 = undefined;
     if (amount == 0) {
@@ -64,17 +60,17 @@ pub fn immShift(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
         }
     } else {
         switch (@truncate(u2, opcode >> 5)) {
-            0b00 => result = logicalLeft(S, &cpu.cpsr, rm, amount),
-            0b01 => result = logicalRight(S, &cpu.cpsr, rm, amount),
-            0b10 => result = arithmeticRight(S, &cpu.cpsr, rm, amount),
-            0b11 => result = rotateRight(S, &cpu.cpsr, rm, amount),
+            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 logicalLeft(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
+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;
 
@@ -101,7 +97,7 @@ pub fn logicalLeft(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32
     return result;
 }
 
-pub fn logicalRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u32) u32 {
+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;
 
@@ -125,7 +121,7 @@ pub fn logicalRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u32) u
     return result;
 }
 
-pub fn arithmeticRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
+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;
 
@@ -142,7 +138,7 @@ pub fn arithmeticRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8)
     return result;
 }
 
-pub fn rotateRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
+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) {

src/core/cpu/thumb/alu.zig

@@ -4,16 +4,11 @@ const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
 
 const adc = @import("../arm/data_processing.zig").adc;
 const sbc = @import("../arm/data_processing.zig").sbc;
-const sub = @import("../arm/data_processing.zig").sub;
-const cmp = @import("../arm/data_processing.zig").cmp;
-const cmn = @import("../arm/data_processing.zig").cmn;
-const setTestOpFlags = @import("../arm/data_processing.zig").setTestOpFlags;
-const setLogicOpFlags = @import("../arm/data_processing.zig").setLogicOpFlags;
 
-const logicalLeft = @import("../barrel_shifter.zig").logicalLeft;
-const logicalRight = @import("../barrel_shifter.zig").logicalRight;
-const arithmeticRight = @import("../barrel_shifter.zig").arithmeticRight;
-const rotateRight = @import("../barrel_shifter.zig").rotateRight;
+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 {
@@ -22,96 +17,85 @@ pub fn fmt4(comptime op: u4) InstrFn {
             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: bool = undefined;
             switch (op) {
-                0x0 => {
-                    // AND
-                    const result = cpu.r[rd] & cpu.r[rs];
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
+                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 => overflow = @addWithOverflow(u32, op1, op2, &result), // CMN
+                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
                 },
-                0x1 => {
-                    // EOR
-                    const result = cpu.r[rd] ^ cpu.r[rs];
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
+                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);
+                    }
                 },
-                0x2 => {
-                    // LSL
-                    const result = logicalLeft(true, &cpu.cpsr, cpu.r[rd], @truncate(u8, cpu.r[rs]));
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
-                },
-                0x3 => {
-                    // LSR
-                    const result = logicalRight(true, &cpu.cpsr, cpu.r[rd], @truncate(u8, cpu.r[rs]));
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
-                },
-                0x4 => {
-                    // ASR
-                    const result = arithmeticRight(true, &cpu.cpsr, cpu.r[rd], @truncate(u8, cpu.r[rs]));
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
-                },
-                0x5 => {
-                    // ADC
-                    cpu.r[rd] = adc(true, cpu, cpu.r[rd], cpu.r[rs], carry);
+                0x5, 0xB => {
+                    // ADC, CMN
+                    cpu.cpsr.n.write(result >> 31 & 1 == 1);
+                    cpu.cpsr.z.write(result == 0);
+                    cpu.cpsr.c.write(overflow);
+                    cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                 },
                 0x6 => {
                     // SBC
-                    cpu.r[rd] = sbc(true, cpu, cpu.r[rd], cpu.r[rs], carry);
-                },
-                0x7 => {
-                    // ROR
-                    const result = rotateRight(true, &cpu.cpsr, cpu.r[rd], @truncate(u8, cpu.r[rs]));
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
-                },
-                0x8 => {
-                    // TST
-                    const result = cpu.r[rd] & cpu.r[rs];
-                    setLogicOpFlags(true, cpu, result);
+                    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.r[rd] = sub(true, cpu, 0, cpu.r[rs]);
-                },
-                0xA => {
-                    // CMP
-                    cmp(cpu, cpu.r[rd], cpu.r[rs]);
-                },
-                0xB => {
-                    // CMN
-                    cmn(cpu, cpu.r[rd], cpu.r[rs]);
-                },
-                0xC => {
-                    // ORR
-                    const result = cpu.r[rd] | cpu.r[rs];
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
+                    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 => {
-                    // MUL
-                    const temp = @as(u64, cpu.r[rs]) * @as(u64, cpu.r[rd]);
-                    const result = @truncate(u32, temp);
-                    cpu.r[rd] = result;
-
+                    // 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
                 },
-                0xE => {
-                    // BIC
-                    const result = cpu.r[rd] & ~cpu.r[rs];
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
-                },
-                0xF => {
-                    // MVN
-                    const result = ~cpu.r[rs];
-                    cpu.r[rd] = result;
-                    setLogicOpFlags(true, cpu, result);
-                },
             }
         }
     }.inner;

src/core/cpu/thumb/block_data_transfer.zig

@@ -33,7 +33,8 @@ pub fn fmt14(comptime L: bool, comptime R: bool) InstrFn {
             if (R) {
                 if (L) {
                     const value = bus.read(u32, address);
-                    cpu.r[15] = value & 0xFFFF_FFFE;
+                    cpu.r[15] = value & ~@as(u32, 1);
+                    cpu.pipe.reload(cpu);
                 } else {
                     bus.write(u32, address, cpu.r[14]);
                 }
@@ -52,7 +53,13 @@ pub fn fmt15(comptime L: bool, comptime rb: u3) InstrFn {
             const end_address = cpu.r[rb] + 4 * countRlist(opcode);
 
             if (opcode & 0xFF == 0) {
-                if (L) cpu.r[15] = bus.read(u32, address) else bus.write(u32, address, cpu.r[15] + 4);
+                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;
             }

src/core/cpu/thumb/branch.zig

@@ -9,16 +9,13 @@ pub fn fmt16(comptime cond: u4) InstrFn {
     return struct {
         fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
             // B
-            const offset = sext(u32, u8, opcode & 0xFF) << 1;
+            if (cond == 0xE or cond == 0xF)
+                cpu.panic("[CPU/THUMB.16] Undefined conditional branch with condition {}", .{cond});
 
-            const should_execute = switch (cond) {
-                0xE, 0xF => cpu.panic("[CPU/THUMB.16] Undefined conditional branch with condition {}", .{cond}),
-                else => checkCond(cpu.cpsr, cond),
-            };
+            if (!checkCond(cpu.cpsr, cond)) return;
 
-            if (should_execute) {
-                cpu.r[15] = (cpu.r[15] + 2) +% offset;
-            }
+            cpu.r[15] +%= sext(u32, u8, opcode & 0xFF) << 1;
+            cpu.pipe.reload(cpu);
         }
     }.inner;
 }
@@ -27,8 +24,8 @@ pub fn fmt18() InstrFn {
     return struct {
         // B but conditional
         fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
-            const offset = sext(u32, u11, opcode & 0x7FF) << 1;
-            cpu.r[15] = (cpu.r[15] + 2) +% offset;
+            cpu.r[15] +%= sext(u32, u11, opcode & 0x7FF) << 1;
+            cpu.pipe.reload(cpu);
         }
     }.inner;
 }
@@ -41,13 +38,16 @@ pub fn fmt19(comptime is_low: bool) InstrFn {
 
             if (is_low) {
                 // Instruction 2
-                const old_pc = cpu.r[15];
+                const next_opcode = cpu.r[15] - 2;
 
                 cpu.r[15] = cpu.r[14] +% (offset << 1);
-                cpu.r[14] = old_pc | 1;
+                cpu.r[14] = next_opcode | 1;
+
+                cpu.pipe.reload(cpu);
             } else {
                 // Instruction 1
-                cpu.r[14] = (cpu.r[15] + 2) +% (sext(u32, u11, offset) << 12);
+                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

@@ -3,14 +3,12 @@ const std = @import("std");
 const Bus = @import("../../Bus.zig");
 const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
 const InstrFn = @import("../../cpu.zig").thumb.InstrFn;
-const shifter = @import("../barrel_shifter.zig");
 
 const add = @import("../arm/data_processing.zig").add;
-const sub = @import("../arm/data_processing.zig").sub;
-const cmp = @import("../arm/data_processing.zig").cmp;
-const setLogicOpFlags = @import("../arm/data_processing.zig").setLogicOpFlags;
 
-const log = std.log.scoped(.Thumb1);
+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 {
@@ -24,7 +22,7 @@ pub fn fmt1(comptime op: u2, comptime offset: u5) InstrFn {
                     if (offset == 0) {
                         break :blk cpu.r[rs];
                     } else {
-                        break :blk shifter.logicalLeft(true, &cpu.cpsr, cpu.r[rs], offset);
+                        break :blk lsl(true, &cpu.cpsr, cpu.r[rs], offset);
                     }
                 },
                 0b01 => blk: {
@@ -33,7 +31,7 @@ pub fn fmt1(comptime op: u2, comptime offset: u5) InstrFn {
                         cpu.cpsr.c.write(cpu.r[rs] >> 31 & 1 == 1);
                         break :blk @as(u32, 0);
                     } else {
-                        break :blk shifter.logicalRight(true, &cpu.cpsr, cpu.r[rs], offset);
+                        break :blk lsr(true, &cpu.cpsr, cpu.r[rs], offset);
                     }
                 },
                 0b10 => blk: {
@@ -42,7 +40,7 @@ pub fn fmt1(comptime op: u2, comptime offset: u5) InstrFn {
                         cpu.cpsr.c.write(cpu.r[rs] >> 31 & 1 == 1);
                         break :blk @bitCast(u32, @bitCast(i32, cpu.r[rs]) >> 31);
                     } else {
-                        break :blk shifter.arithmeticRight(true, &cpu.cpsr, cpu.r[rs], offset);
+                        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}),
@@ -50,7 +48,10 @@ pub fn fmt1(comptime op: u2, comptime offset: u5) InstrFn {
 
             // Equivalent to an ARM MOVS
             cpu.r[rd] = result;
-            setLogicOpFlags(true, cpu, result);
+
+            // Write Flags
+            cpu.cpsr.n.write(result >> 31 & 1 == 1);
+            cpu.cpsr.z.write(result == 0);
         }
     }.inner;
 }
@@ -58,28 +59,51 @@ pub fn fmt1(comptime op: u2, comptime offset: u5) InstrFn {
 pub fn fmt5(comptime op: u2, comptime h1: u1, comptime h2: u1) InstrFn {
     return struct {
         fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
-            const src_idx = @as(u4, h2) << 3 | (opcode >> 3 & 0x7);
-            const dst_idx = @as(u4, h1) << 3 | (opcode & 0x7);
+            const rs = @as(u4, h2) << 3 | (opcode >> 3 & 0x7);
+            const rd = @as(u4, h1) << 3 | (opcode & 0x7);
 
-            const src = if (src_idx == 0xF) (cpu.r[src_idx] + 2) & 0xFFFF_FFFE else cpu.r[src_idx];
-            const dst = if (dst_idx == 0xF) (cpu.r[dst_idx] + 2) & 0xFFFF_FFFE else cpu.r[dst_idx];
+            const op1 = cpu.r[rd];
+            const op2 = cpu.r[rs];
 
+            var result: u32 = undefined;
+            var overflow: bool = undefined;
             switch (op) {
-                0b00 => {
-                    // ADD
-                    const sum = add(false, cpu, dst, src);
-                    cpu.r[dst_idx] = if (dst_idx == 0xF) sum & 0xFFFF_FFFE else sum;
-                },
-                0b01 => cmp(cpu, dst, src), // CMP
-                0b10 => {
-                    // MOV
-                    cpu.r[dst_idx] = if (dst_idx == 0xF) src & 0xFFFF_FFFE else src;
-                },
+                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
-                    cpu.cpsr.t.write(src & 1 == 1);
-                    cpu.r[15] = src & 0xFFFF_FFFE;
+                    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;
@@ -90,21 +114,28 @@ pub fn fmt2(comptime I: bool, is_sub: bool, rn: u3) InstrFn {
         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
-                cpu.r[rd] = if (I) blk: {
-                    break :blk sub(true, cpu, cpu.r[rs], rn);
-                } else blk: {
-                    break :blk sub(true, cpu, cpu.r[rs], cpu.r[rn]);
-                };
+                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
-                cpu.r[rd] = if (I) blk: {
-                    break :blk add(true, cpu, cpu.r[rs], rn);
-                } else blk: {
-                    break :blk add(true, cpu, cpu.r[rs], cpu.r[rn]);
-                };
+                var overflow: bool = 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);
+                cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
             }
         }
     }.inner;
@@ -113,17 +144,36 @@ pub fn fmt2(comptime I: bool, is_sub: bool, rn: u3) InstrFn {
 pub fn fmt3(comptime op: u2, comptime rd: u3) InstrFn {
     return struct {
         fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
-            const offset = @truncate(u8, opcode);
+            const op1 = cpu.r[rd];
+            const op2: u32 = opcode & 0xFF; // Offset
+
+            var overflow: bool = 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
-                    cpu.r[rd] = offset;
-                    setLogicOpFlags(true, cpu, offset);
+                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);
+                    cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                 },
-                0b01 => cmp(cpu, cpu.r[rd], offset), // CMP
-                0b10 => cpu.r[rd] = add(true, cpu, cpu.r[rd], offset), // ADD
-                0b11 => cpu.r[rd] = sub(true, cpu, cpu.r[rd], offset), // SUB
             }
         }
     }.inner;
@@ -133,10 +183,9 @@ 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] + 2) & 0xFFFF_FFFD;
+            const left = if (isSP) cpu.r[13] else cpu.r[15] & ~@as(u32, 2);
             const right = (opcode & 0xFF) << 2;
-            const result = left + right;
-            cpu.r[rd] = result;
+            cpu.r[rd] = left + right;
         }
     }.inner;
 }

src/core/cpu/thumb/data_transfer.zig

@@ -12,7 +12,9 @@ pub fn fmt6(comptime rd: u3) InstrFn {
         fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
             // LDR
             const offset = (opcode & 0xFF) << 2;
-            cpu.r[rd] = bus.read(u32, (cpu.r[15] + 2 & 0xFFFF_FFFD) + offset);
+
+            // Bit 1 of the PC intentionally ignored
+            cpu.r[rd] = bus.read(u32, (cpu.r[15] & ~@as(u32, 2)) + offset);
         }
     }.inner;
 }

src/core/cpu/thumb/software_interrupt.zig

@@ -6,7 +6,7 @@ pub fn fmt17() InstrFn {
     return struct {
         fn inner(cpu: *Arm7tdmi, _: *Bus, _: u16) void {
             // Copy Values from Current Mode
-            const r15 = cpu.r[15];
+            const ret_addr = cpu.r[15] - 2;
             const cpsr = cpu.cpsr.raw;
 
             // Switch Mode
@@ -14,9 +14,10 @@ pub fn fmt17() InstrFn {
             cpu.cpsr.t.write(false); // Force ARM Mode
             cpu.cpsr.i.write(true); // Disable normal interrupts
 
-            cpu.r[14] = r15; // Resume Execution
+            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

@@ -1,5 +1,6 @@
 const std = @import("std");
 const SDL = @import("sdl2");
+const config = @import("../config.zig");
 
 const Bus = @import("Bus.zig");
 const Scheduler = @import("scheduler.zig").Scheduler;
@@ -12,14 +13,6 @@ const Thread = std.Thread;
 const Atomic = std.atomic.Atomic;
 const Allocator = std.mem.Allocator;
 
-// TODO: Move these to a TOML File
-const sync_audio = false; // Enable Audio Sync
-const sync_video: RunKind = .LimitedFPS; // Configure Video Sync
-pub const win_scale = 4; // 1x, 2x, 3x, etc. Window Scaling
-pub const cpu_logging = false; // Enable detailed CPU logging
-pub const allow_unhandled_io = true; // Only relevant in Debug Builds
-pub const force_rtc = false;
-
 // 228 Lines which consist of 308 dots (which are 4 cycles long)
 const cycles_per_frame: u64 = 228 * (308 * 4); //280896
 const clock_rate: u64 = 1 << 24; // 16.78MHz
@@ -40,18 +33,57 @@ const RunKind = enum {
     UnlimitedFPS,
     Limited,
     LimitedFPS,
-    LimitedBusy,
 };
 
-pub fn run(quit: *Atomic(bool), fps: *FpsTracker, sched: *Scheduler, cpu: *Arm7tdmi) void {
-    if (sync_audio) log.info("Audio sync enabled", .{});
+pub fn run(quit: *Atomic(bool), scheduler: *Scheduler, cpu: *Arm7tdmi, tracker: *FpsTracker) void {
+    const audio_sync = config.config().guest.audio_sync;
+    if (audio_sync) log.info("Audio sync enabled", .{});
 
-    switch (sync_video) {
-        .Unlimited => runUnsynchronized(quit, sched, cpu, null),
-        .Limited => runSynchronized(quit, sched, cpu, null),
-        .UnlimitedFPS => runUnsynchronized(quit, sched, cpu, fps),
-        .LimitedFPS => runSynchronized(quit, sched, cpu, fps),
-        .LimitedBusy => runBusyLoop(quit, sched, cpu),
+    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(.SeqCst)) {
+                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(.SeqCst)) {
+                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();
+            }
+        },
     }
 }
 
@@ -72,7 +104,7 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
     }
 }
 
-fn syncToAudio(stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
+fn audioSync(audio_sync: bool, stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
     const sample_size = 2 * @sizeOf(u16);
     const max_buf_size: c_int = 0x400;
 
@@ -85,90 +117,20 @@ fn syncToAudio(stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
 
     while (true) {
         still_full = SDL.SDL_AudioStreamAvailable(stream) > sample_size * max_buf_size >> 1;
-        if (!sync_audio or !still_full) break;
+        if (!audio_sync or !still_full) break;
     }
 }
 
-pub fn runUnsynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi, fps: ?*FpsTracker) void {
-    log.info("Emulation thread w/out video sync", .{});
-
-    if (fps) |tracker| {
-        log.info("FPS Tracking Enabled", .{});
-
-        while (!quit.load(.SeqCst)) {
-            runFrame(sched, cpu);
-            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
-
-            tracker.tick();
-        }
-    } else {
-        while (!quit.load(.SeqCst)) {
-            runFrame(sched, cpu);
-            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
-        }
-    }
-}
-
-pub fn runSynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi, fps: ?*FpsTracker) void {
-    log.info("Emulation thread w/ video sync", .{});
-    var timer = Timer.start() catch std.debug.panic("Failed to initialize std.timer.Timer", .{});
-    var wake_time: u64 = frame_period;
-
-    if (fps) |tracker| {
-        log.info("FPS Tracking Enabled", .{});
-
-        while (!quit.load(.SeqCst)) {
-            runFrame(sched, cpu);
-            const new_wake_time = blockOnVideo(&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
-            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
-            if (!sync_audio) spinLoop(&timer, wake_time);
-            wake_time = new_wake_time;
-
-            tracker.tick();
-        }
-    } else {
-        while (!quit.load(.SeqCst)) {
-            runFrame(sched, cpu);
-            const new_wake_time = blockOnVideo(&timer, wake_time);
-
-            // see above comment
-            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
-            if (!sync_audio) spinLoop(&timer, wake_time);
-            wake_time = new_wake_time;
-        }
-    }
-}
-
-inline fn blockOnVideo(timer: *Timer, wake_time: u64) u64 {
+fn videoSync(timer: *Timer, wake_time: u64) u64 {
     // Use the OS scheduler to put the emulation thread to sleep
-    const maybe_recalc_wake_time = sleep(timer, wake_time);
+    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 if (maybe_recalc_wake_time) |recalc| recalc else wake_time + frame_period;
-}
-
-pub fn runBusyLoop(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi) void {
-    log.info("Emulation thread with video sync using busy loop", .{});
-    var timer = Timer.start() catch unreachable;
-    var wake_time: u64 = frame_period;
-
-    while (!quit.load(.SeqCst)) {
-        runFrame(sched, cpu);
-        spinLoop(&timer, wake_time);
-
-        syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
-
-        // Update to the new wake time
-        wake_time += frame_period;
-    }
+    return recalculated orelse wake_time + frame_period;
 }
 
+// TODO: Better sleep impl?
 fn sleep(timer: *Timer, wake_time: u64) ?u64 {
     // const step = std.time.ns_per_ms * 10; // 10ms
     const timestamp = timer.read();

src/core/ppu.zig

@@ -10,7 +10,7 @@ const Bitfield = @import("bitfield").Bitfield;
 
 const Allocator = std.mem.Allocator;
 const log = std.log.scoped(.PPU);
-const pollBlankingDma = @import("bus/dma.zig").pollBlankingDma;
+const pollDmaOnBlank = @import("bus/dma.zig").pollDmaOnBlank;
 
 /// This is used to generate byuu / Talurabi's Color Correction algorithm
 const COLOUR_LUT = genColourLut();
@@ -562,7 +562,7 @@ pub const Ppu = struct {
         };
     }
 
-    pub fn handleHDrawEnd(self: *Self, cpu: *Arm7tdmi, late: u64) void {
+    pub fn onHdrawEnd(self: *Self, cpu: *Arm7tdmi, late: u64) void {
         // Transitioning to a Hblank
         if (self.dispstat.hblank_irq.read()) {
             cpu.bus.io.irq.hblank.set();
@@ -572,13 +572,13 @@ pub const Ppu = struct {
         // See if HBlank DMA is present and not enabled
 
         if (!self.dispstat.vblank.read())
-            pollBlankingDma(cpu.bus, .HBlank);
+            pollDmaOnBlank(cpu.bus, .HBlank);
 
         self.dispstat.hblank.set();
         self.sched.push(.HBlank, 68 * 4 -| late);
     }
 
-    pub fn handleHBlankEnd(self: *Self, cpu: *Arm7tdmi, late: u64) void {
+    pub fn onHblankEnd(self: *Self, cpu: *Arm7tdmi, late: u64) void {
         // The End of a Hblank (During Draw or Vblank)
         const old_scanline = self.vcount.scanline.read();
         const scanline = (old_scanline + 1) % 228;
@@ -614,7 +614,7 @@ pub const Ppu = struct {
                 self.aff_bg[1].latchRefPoints();
 
                 // See if Vblank DMA is present and not enabled
-                pollBlankingDma(cpu.bus, .VBlank);
+                pollDmaOnBlank(cpu.bus, .VBlank);
             }
 
             if (scanline == 227) self.dispstat.vblank.unset();
@@ -808,18 +808,6 @@ const Window = struct {
         self.in.raw = @truncate(u16, value);
         self.out.raw = @truncate(u16, value >> 16);
     }
-
-    pub fn setInL(self: *Self, value: u8) void {
-        self.in.raw = (self.in.raw & 0xFF00) | value;
-    }
-
-    pub fn setInH(self: *Self, value: u8) void {
-        self.in.raw = (self.in.raw & 0x00FF) | (@as(u16, value) << 8);
-    }
-
-    pub fn setOutL(self: *Self, value: u8) void {
-        self.out.raw = (self.out.raw & 0xFF00) | value;
-    }
 };
 
 const Background = struct {

src/core/scheduler.zig

@@ -43,22 +43,22 @@ pub const Scheduler = struct {
                 .Draw => {
                     // The end of a VDraw
                     cpu.bus.ppu.drawScanline();
-                    cpu.bus.ppu.handleHDrawEnd(cpu, late);
+                    cpu.bus.ppu.onHdrawEnd(cpu, late);
                 },
                 .TimerOverflow => |id| {
                     switch (id) {
-                        0 => cpu.bus.tim[0].handleOverflow(cpu, late),
-                        1 => cpu.bus.tim[1].handleOverflow(cpu, late),
-                        2 => cpu.bus.tim[2].handleOverflow(cpu, late),
-                        3 => cpu.bus.tim[3].handleOverflow(cpu, late),
+                        0 => cpu.bus.tim[0].onTimerExpire(cpu, late),
+                        1 => cpu.bus.tim[1].onTimerExpire(cpu, late),
+                        2 => cpu.bus.tim[2].onTimerExpire(cpu, late),
+                        3 => cpu.bus.tim[3].onTimerExpire(cpu, late),
                     }
                 },
                 .ApuChannel => |id| {
                     switch (id) {
-                        0 => cpu.bus.apu.ch1.channelTimerOverflow(late),
-                        1 => cpu.bus.apu.ch2.channelTimerOverflow(late),
-                        2 => cpu.bus.apu.ch3.channelTimerOverflow(late),
-                        3 => cpu.bus.apu.ch4.channelTimerOverflow(late),
+                        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 => {
@@ -66,12 +66,12 @@ pub const Scheduler = struct {
                     if (device.kind != .Rtc or device.ptr == null) return;
 
                     const clock = @ptrCast(*Clock, @alignCast(@alignOf(*Clock), device.ptr.?));
-                    clock.updateTime(late);
+                    clock.onClockUpdate(late);
                 },
-                .FrameSequencer => cpu.bus.apu.tickFrameSequencer(late),
+                .FrameSequencer => cpu.bus.apu.onSequencerTick(late),
                 .SampleAudio => cpu.bus.apu.sampleAudio(late),
-                .HBlank => cpu.bus.ppu.handleHBlankEnd(cpu, late), // The end of a HBlank
-                .VBlank => cpu.bus.ppu.handleHDrawEnd(cpu, late), // The end of a VBlank
+                .HBlank => cpu.bus.ppu.onHblankEnd(cpu, late), // The end of a HBlank
+                .VBlank => cpu.bus.ppu.onHdrawEnd(cpu, late), // The end of a VBlank
             }
         }
     }

src/main.zig

@@ -1,9 +1,10 @@
 const std = @import("std");
 const builtin = @import("builtin");
-
 const known_folders = @import("known_folders");
 const clap = @import("clap");
 
+const config = @import("config.zig");
+
 const Gui = @import("platform.zig").Gui;
 const Bus = @import("core/Bus.zig");
 const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
@@ -14,11 +15,8 @@ const Allocator = std.mem.Allocator;
 const log = std.log.scoped(.Cli);
 const width = @import("core/ppu.zig").width;
 const height = @import("core/ppu.zig").height;
-const cpu_logging = @import("core/emu.zig").cpu_logging;
 pub const log_level = if (builtin.mode != .Debug) .info else std.log.default_level;
 
-// TODO: Reimpl Logging
-
 // CLI Arguments + Help Text
 const params = clap.parseParamsComptime(
     \\-h, --help            Display this help and exit.
@@ -33,14 +31,27 @@ pub fn main() anyerror!void {
     defer std.debug.assert(!gpa.deinit());
     const allocator = gpa.allocator();
 
+    // TODO: Make Error message not Linux Specific
+    const data_path = try known_folders.getPath(allocator, .data) orelse exit("Unable to Determine XDG Data Path", .{});
+    defer allocator.free(data_path);
+
+    const config_path = try configFilePath(allocator, data_path);
+    defer allocator.free(config_path);
+
+    const save_path = try savePath(allocator, data_path);
+    defer allocator.free(save_path);
+
+    try config.load(allocator, config_path);
+
     // Handle CLI Input
     const result = try clap.parse(clap.Help, &params, clap.parsers.default, .{});
     defer result.deinit();
 
-    const paths = try handleArguments(allocator, &result);
+    const paths = try handleArguments(allocator, data_path, &result);
     defer if (paths.save) |path| allocator.free(path);
 
-    const log_file: ?std.fs.File = if (cpu_logging) try std.fs.cwd().createFile("zba.log", .{}) else null;
+    const cpu_trace = config.config().debug.cpu_trace;
+    const log_file: ?std.fs.File = if (cpu_trace) try std.fs.cwd().createFile("zba.log", .{}) else null;
     defer if (log_file) |file| file.close();
 
     // TODO: Take Emulator Init Code out of main.zig
@@ -49,50 +60,27 @@ pub fn main() anyerror!void {
 
     var bus: Bus = undefined;
     var cpu = Arm7tdmi.init(&scheduler, &bus, log_file);
-    if (paths.bios == null) cpu.fastBoot();
 
     try bus.init(allocator, &scheduler, &cpu, paths);
     defer bus.deinit();
 
+    if (paths.bios == null) cpu.fastBoot();
+
     var gui = Gui.init(&bus.pak.title, &bus.apu, width, height);
     defer gui.deinit();
 
     try gui.run(&cpu, &scheduler);
 }
 
-fn getSavePath(allocator: Allocator) !?[]const u8 {
-    const save_subpath = "zba" ++ [_]u8{std.fs.path.sep} ++ "save";
-
-    const maybe_data_path = try known_folders.getPath(allocator, .data);
-    defer if (maybe_data_path) |path| allocator.free(path);
-
-    const save_path = if (maybe_data_path) |base| try std.fs.path.join(allocator, &[_][]const u8{ base, "zba", "save" }) else null;
-
-    if (save_path) |_| {
-        // If we've determined what our save path should be, ensure the prereq directories
-        // are present so that we can successfully write to the path when necessary
-        const maybe_data_dir = try known_folders.open(allocator, .data, .{});
-        if (maybe_data_dir) |data_dir| try data_dir.makePath(save_subpath);
-    }
-
-    return save_path;
-}
-
-fn getRomPath(result: *const clap.Result(clap.Help, &params, clap.parsers.default)) ![]const u8 {
-    return switch (result.positionals.len) {
-        1 => result.positionals[0],
-        0 => std.debug.panic("ZBA requires a positional path to a GamePak ROM.\n", .{}),
-        else => std.debug.panic("ZBA received too many arguments.\n", .{}),
-    };
-}
-
-pub fn handleArguments(allocator: Allocator, result: *const clap.Result(clap.Help, &params, clap.parsers.default)) !FilePaths {
-    const rom_path = try getRomPath(result);
+pub 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.info("No BIOS provided", .{});
-    const save_path = try getSavePath(allocator);
-    if (save_path) |path| log.info("Save path: {s}", .{path});
+
+    const save_path = try savePath(allocator, data_path);
+    log.info("Save path: {s}", .{save_path});
 
     return FilePaths{
         .rom = rom_path,
@@ -100,3 +88,42 @@ pub fn handleArguments(allocator: Allocator, result: *const clap.Result(clap.Hel
         .save = save_path,
     };
 }
+
+fn configFilePath(allocator: Allocator, data_path: []const u8) ![]const u8 {
+    const path = try std.fs.path.join(allocator, &[_][]const u8{ data_path, "zba", "config.toml" });
+
+    // 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 {
+        const file_handle = try std.fs.createFileAbsolute(path, .{});
+        file_handle.close();
+    };
+
+    return path;
+}
+
+fn savePath(allocator: Allocator, data_path: []const u8) ![]const u8 {
+    var dir = try std.fs.openDirAbsolute(data_path, .{});
+    defer dir.close();
+
+    // Will either make the path recursively, or just exit early since it already exists
+    try dir.makePath("zba" ++ [_]u8{std.fs.path.sep} ++ "save");
+
+    // FIXME: Do we have to allocate? :sad:
+    return try std.fs.path.join(allocator, &[_][]const u8{ data_path, "zba", "save" });
+}
+
+fn romPath(result: *const clap.Result(clap.Help, &params, clap.parsers.default)) []const u8 {
+    return switch (result.positionals.len) {
+        1 => result.positionals[0],
+        0 => exit("ZBA requires a path to a GamePak ROM\n", .{}),
+        else => exit("ZBA received too many positional arguments. \n", .{}),
+    };
+}
+
+fn exit(comptime format: []const u8, args: anytype) noreturn {
+    const stderr = std.io.getStdErr().writer();
+    stderr.print(format, args) catch {}; // Just exit already...
+    std.os.exit(1);
+}

src/platform.zig

@@ -1,6 +1,7 @@
 const std = @import("std");
 const SDL = @import("sdl2");
 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;
@@ -10,8 +11,6 @@ const FpsTracker = @import("util.zig").FpsTracker;
 const span = @import("util.zig").span;
 
 const pitch = @import("core/ppu.zig").framebuf_pitch;
-const scale = @import("core/emu.zig").win_scale;
-
 const default_title: []const u8 = "ZBA";
 
 pub const Gui = struct {
@@ -28,16 +27,19 @@ pub const Gui = struct {
         const ret = SDL.SDL_Init(SDL.SDL_INIT_VIDEO | SDL.SDL_INIT_EVENTS | SDL.SDL_INIT_AUDIO | SDL.SDL_INIT_GAMECONTROLLER);
         if (ret < 0) panic();
 
+        const win_scale = @intCast(c_int, config.config().host.win_scale);
+
         const window = SDL.SDL_CreateWindow(
             default_title.ptr,
             SDL.SDL_WINDOWPOS_CENTERED,
             SDL.SDL_WINDOWPOS_CENTERED,
-            @as(c_int, width * scale),
-            @as(c_int, height * scale),
+            @as(c_int, width * win_scale),
+            @as(c_int, height * win_scale),
             SDL.SDL_WINDOW_SHOWN,
         ) orelse panic();
 
-        const renderer = SDL.SDL_CreateRenderer(window, -1, SDL.SDL_RENDERER_ACCELERATED | SDL.SDL_RENDERER_PRESENTVSYNC) orelse panic();
+        const renderer_flags = SDL.SDL_RENDERER_ACCELERATED | if (config.config().host.vsync) SDL.SDL_RENDERER_PRESENTVSYNC else 0;
+        const renderer = SDL.SDL_CreateRenderer(window, -1, @bitCast(u32, renderer_flags)) orelse panic();
 
         const texture = SDL.SDL_CreateTexture(
             renderer,
@@ -58,9 +60,9 @@ pub const Gui = struct {
 
     pub fn run(self: *Self, cpu: *Arm7tdmi, scheduler: *Scheduler) !void {
         var quit = std.atomic.Atomic(bool).init(false);
-        var frame_rate = FpsTracker.init();
+        var tracker = FpsTracker.init();
 
-        const thread = try std.Thread.spawn(.{}, emu.run, .{ &quit, &frame_rate, scheduler, cpu });
+        const thread = try std.Thread.spawn(.{}, emu.run, .{ &quit, scheduler, cpu, &tracker });
         defer thread.join();
 
         var title_buf: [0x100]u8 = [_]u8{0} ** 0x100;
@@ -127,7 +129,7 @@ pub const Gui = struct {
             _ = SDL.SDL_RenderCopy(self.renderer, self.texture, null, null);
             SDL.SDL_RenderPresent(self.renderer);
 
-            const dyn_title = std.fmt.bufPrint(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, frame_rate.value() }) catch unreachable;
+            const dyn_title = std.fmt.bufPrint(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, tracker.value() }) catch unreachable;
             SDL.SDL_SetWindowTitle(self.window, dyn_title.ptr);
         }
 

src/util.zig

@@ -1,10 +1,10 @@
 const std = @import("std");
 const builtin = @import("builtin");
+const config = @import("config.zig");
+
 const Log2Int = std.math.Log2Int;
 const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
 
-const allow_unhandled_io = @import("core/emu.zig").allow_unhandled_io;
-
 // 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
@@ -144,8 +144,10 @@ pub const io = struct {
         }
 
         pub fn undef(comptime T: type, log: anytype, comptime format: []const u8, args: anytype) ?T {
+            const unhandled_io = config.config().debug.unhandled_io;
+
             log.warn(format, args);
-            if (builtin.mode == .Debug and !allow_unhandled_io) std.debug.panic("TODO: Implement I/O Register", .{});
+            if (builtin.mode == .Debug and !unhandled_io) std.debug.panic("TODO: Implement I/O Register", .{});
 
             return null;
         }
@@ -153,22 +155,13 @@ pub const io = struct {
 
     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 !allow_unhandled_io) std.debug.panic("TODO: Implement I/O Register", .{});
+            if (builtin.mode == .Debug and !unhandled_io) std.debug.panic("TODO: Implement I/O Register", .{});
         }
     };
 };
-pub fn readUndefined(log: anytype, comptime format: []const u8, args: anytype) u8 {
-    log.warn(format, args);
-    if (builtin.mode == .Debug) std.debug.panic("TODO: Implement I/O Register", .{});
-
-    return 0;
-}
-
-pub fn writeUndefined(log: anytype, comptime format: []const u8, args: anytype) void {
-    log.warn(format, args);
-    if (builtin.mode == .Debug) std.debug.panic("TODO: Implement I/O Register", .{});
-}
 
 pub const Logger = struct {
     const Self = @This();
@@ -183,6 +176,7 @@ pub const Logger = struct {
 
     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 {
@@ -222,7 +216,7 @@ pub const Logger = struct {
             cpu.r[12],
             cpu.r[13],
             cpu.r[14],
-            cpu.r[15],
+            cpu.r[15] - if (cpu.cpsr.t.read()) 2 else @as(u32, 4),
             cpu.cpsr.raw,
             opcode,
         };
@@ -230,3 +224,81 @@ pub const Logger = struct {
 };
 
 const FmtArgTuple = std.meta.Tuple(&.{ u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32 });
+
+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 the high bits of an integer to a value
+pub inline fn setHi(comptime T: type, left: T, right: HalfInt(T)) T {
+    return switch (T) {
+        u32 => (left & 0xFFFF_0000) | right,
+        u16 => (left & 0xFF00) | right,
+        u8 => (left & 0xF0) | right,
+        else => @compileError("unsupported type"),
+    };
+}
+
+/// sets the low bits of an integer to a value
+pub inline fn setLo(comptime T: type, left: T, right: HalfInt(T)) T {
+    return switch (T) {
+        u32 => (left & 0x0000_FFFF) | @as(u32, right) << 16,
+        u16 => (left & 0x00FF) | @as(u16, right) << 8,
+        u8 => (left & 0x0F) | @as(u8, right) << 4,
+        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);
+}