ci: ensure that submodules are updated recursively

Reviewed-on: #6

.github/workflows/main.yml

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

.gitmodules

@@ -13,3 +13,6 @@
 [submodule "lib/zig-toml"]
 	path = lib/zig-toml
 	url = https://github.com/aeronavery/zig-toml
+[submodule "lib/zba-gdbstub"]
+	path = lib/zba-gdbstub
+	url = https://git.musuka.dev/paoda/zba-gdbstub

.vscode/extensions.json

@@ -1,8 +0,0 @@
-{
-    "recommendations": [
-        "augusterame.zls-vscode",
-        "usernamehw.errorlens",
-        "vadimcn.vscode-lldb",
-        "dan-c-underwood.arm"
-    ]
-}

README.md

@@ -1,80 +1,110 @@
 # ZBA (working title)
+
 A Game Boy Advance Emulator written in Zig ⚡!
 
 ## Scope
-I'm hardly the first to write a Game Boy Advance Emulator nor will I be the last. This project isn't going to compete with the GOATs like 
-[mGBA](https://github.com/mgba-emu) or [NanoBoyAdvance](https://github.com/nba-emu/NanoBoyAdvance). There aren't any interesting
-ideas either like in [DSHBA](https://github.com/DenSinH/DSHBA). 
 
-This is a simple (read: incomplete) for-fun long-term project. I hope to get "mostly there", which to me means that I'm not missing any major hardware
+I'm hardly the first to write a Game Boy Advance Emulator nor will I be the last. This project isn't going to compete with the GOATs like [mGBA](https://github.com/mgba-emu) or [NanoBoyAdvance](https://github.com/nba-emu/NanoBoyAdvance). There aren't any interesting ideas either like in [DSHBA](https://github.com/DenSinH/DSHBA).
-features and the set of possible improvements would be in memory timing or in UI/UX. With respect to that goal, here's what's outstanding: 
+
+This is a simple (read: incomplete) for-fun long-term project. I hope to get "mostly there", which to me means that I'm not missing any major hardware features and the set of possible improvements would be in memory timing or in UI/UX. With respect to that goal, here's what's outstanding:
 
 ### TODO
-- [ ] Affine Sprites
+
+- [x] Affine Sprites
 - [ ] Windowing (see [this branch](https://git.musuka.dev/paoda/zba/src/branch/window))
 - [ ] Audio Resampler (Having issues with SDL2's)
-- [ ] Immediate Mode GUI
+- [x] Immediate Mode GUI (see [this branch](https://git.musuka.dev/paoda/zba/src/branch/imgui))
 - [ ] Refactoring for easy-ish perf boosts
 
+## Usage
+
+As it currently exists, ZBA is run from the terminal. In your console of choice, type `./zba --help` to see what you can do.
+
+I typically find myself typing `./zba -b ./bin/bios.bin ./bin/test/suite.gba` to see how badly my "cool new feature" broke everything else.
+
+Need a BIOS? Why not try using the open-source [Cult-Of-GBA BIOS](https://github.com/Cult-of-GBA/BIOS) written by [fleroviux](https://github.com/fleroviux) and [DenSinH](https://github.com/DenSinH)?
+
+Finally it's worth noting that ZBA uses a TOML config file it'll store in your OS's data directory. See `example.toml` to learn about the defaults and what exactly you can mess around with.
+
 ## Tests
-- [x] [jsmolka's GBA Test Collection](https://github.com/jsmolka/gba-tests)
+
-    - [x] `arm.gba` and `thumb.gba`
+GBA Tests | [jsmolka](https://github.com/jsmolka/)
-    - [x] `flash64.gba`, `flash128.gba`, `none.gba`, and `sram.gba`
+--- | ---
-    - [x] `hello.gba`, `shades.gba`, and `stripes.gba`
+`arm.gba`,  `thumb.gba` | PASS
-    - [x] `memory.gba`
+`memory.gba`, `bios.gba` | PASS
-    - [x] `bios.gba`
+`flash64.gba`, `flash128.gba` | PASS
-    - [x] `nes.gba`
+`sram.gba` | PASS
-- [ ] [DenSinH's GBA ROMs](https://github.com/DenSinH/GBARoms)
+`none.gba` | PASS
-    - [x] `eeprom-test` and `flash-test`
+`hello.gba`, `shades.gba`, `stripes.gba` | PASS
-    - [x] `midikey2freq`
+`nes.gba` | PASS
-    - [ ] `swi-tests-random`
+
-- [ ] [destoer's GBA Tests](https://github.com/destoer/gba_tests)
+GBARoms | [DenSinH](https://github.com/DenSinH/)
-    - [x] `cond_invalid.gba`
+--- | ---
-    - [x] `dma_priority.gba`
+`eeprom-test`, `flash-test` | PASS
-    - [x] `hello_world.gba`
+`midikey2freq` | PASS
-    - [x] `if_ack.gba`
+`swi-tests-random` | FAIL
-    - [ ] `line_timing.gba`
+
-    - [ ] `lyc_midline.gba`
+gba_tests | [destoer](https://github.com/destoer/)
-    - [ ] `window_midframe.gba`
+--- | ---
-- [x] [ladystarbreeze's GBA Test Collection](https://github.com/ladystarbreeze/GBA-Test-Collection)
+`cond_invalid.gba` | PASS
-    - [x] `retAddr.gba`
+`dma_priority.gba` | PASS
-    - [x] `helloWorld.gba`
+`hello_world.gba` | PASS
-    - [x] `helloAudio.gba`
+`if_ack.gba` | PASS
-- [x] [`armwrestler-gba-fixed.gba`](https://github.com/destoer/armwrestler-gba-fixed)
+`line_timing.gba` | FAIL
-- [x] [FuzzARM](https://github.com/DenSinH/FuzzARM)
+`lyc_midline.gba` | FAIL
+`window_midframe.gba` | FAIL
+
+GBA Test Collection | [ladystarbreeze](https://github.com/ladystarbreeze)
+--- | ---
+`retAddr.gba` | PASS
+`helloWorld.gba` | PASS
+`helloAudio.gba` | PASS
+
+FuzzARM | [DenSinH](https://github.com/DenSinH/)
+--- | ---
+`main.gba` | PASS
+
+arm7wrestler GBA Fixed | [destoer](https://github.com/destoer)
+--- | ---
+`armwrestler-gba-fixed.gba` | PASS
 
 ## Resources
-* [GBATEK](https://problemkaputt.de/gbatek.htm)
+
-* [TONC](https://coranac.com/tonc/text/toc.htm)
+- [GBATEK](https://problemkaputt.de/gbatek.htm)
-* [ARM Architecture Reference Manual](https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/third-party/ddi0100e_arm_arm.pdf)
+- [TONC](https://coranac.com/tonc/text/toc.htm)
-* [ARM7TDMI Data Sheet](https://www.dca.fee.unicamp.br/cursos/EA871/references/ARM/ARM7TDMIDataSheet.pdf)
+- [ARM Architecture Reference Manual](https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/third-party/ddi0100e_arm_arm.pdf)
+- [ARM7TDMI Data Sheet](https://www.dca.fee.unicamp.br/cursos/EA871/references/ARM/ARM7TDMIDataSheet.pdf)
 
 ## Compiling
-Most recently built on Zig [0.10.0-dev.4474+b41b35f57](https://github.com/ziglang/zig/tree/b41b35f57)
+
+Most recently built on Zig [v0.11.0-dev.1580+a5b34a61a](https://github.com/ziglang/zig/tree/a5b34a61a)
 
 ### Dependencies
-* [SDL.zig](https://github.com/MasterQ32/SDL.zig)
-    * [SDL2](https://www.libsdl.org/download-2.0.php)
-* [zig-clap](https://github.com/Hejsil/zig-clap)
-* [known-folders](https://github.com/ziglibs/known-folders)
-* [zig-toml](https://github.com/aeronavery/zig-toml)
-* [zig-datetime](https://github.com/frmdstryr/zig-datetime)
-* [`bitfields.zig`](https://github.com/FlorenceOS/Florence/blob/aaa5a9e568/lib/util/bitfields.zig)
 
-`bitfields.zig` from [FlorenceOS](https://github.com/FlorenceOS) is included under `lib/util/bitfield.zig`.
+Dependency | Source
+--- | ---
+SDL.zig | <https://github.com/MasterQ32/SDL.zig>
+zig-clap | <https://github.com/Hejsil/zig-clap>
+known-folders | <https://github.com/ziglibs/known-folders>
+zig-toml | <https://github.com/aeronavery/zig-toml>
+zig-datetime | <https://github.com/frmdstryr/zig-datetime>
+`bitfields.zig` | [https://github.com/FlorenceOS/Florence](https://github.com/FlorenceOS/Florence/blob/aaa5a9e568/lib/util/bitfields.zig)
+`gl.zig` | <https://github.com/MasterQ32/zig-opengl>
 
-Use `git submodule update --init` from the project root to pull the git submodules `SDL.zig`, `zig-clap`, `known-folders`, `zig-toml` and `zig-datetime`
+Use `git submodule update --init` from the project root to pull the git relevant git submodules
 
 Be sure to provide SDL2 using:
-* Linux: Your distro's package manager
+
-* MacOS: ¯\\\_(ツ)_/¯
+- Linux: Your distro's package manager
-* Windows: [`vcpkg`](https://github.com/Microsoft/vcpkg) (install `sdl2:x64-windows`)
+- macOS: ¯\\\_(ツ)_/¯ (try [this formula](https://formulae.brew.sh/formula/sdl2)?)
+- Windows: [`vcpkg`](https://github.com/Microsoft/vcpkg) (install `sdl2:x64-windows`)
 
 `SDL.zig` will provide a helpful compile error if the zig compiler is unable to find SDL2.
 
-Once you've got all the dependencies, execute `zig build -Drelease-fast`. The executable is located at `zig-out/bin/`. 
+Once you've got all the dependencies, execute `zig build -Doptimize=ReleaseSafe`. The executable is located at `zig-out/bin/`.
 
 ## Controls
+
 Key | Button
 --- | ---
 <kbd>X</kbd> | A

build.zig

@@ -1,46 +1,55 @@
 const std = @import("std");
+const builtin = @import("builtin");
+
 const Sdk = @import("lib/SDL.zig/Sdk.zig");
+const Gdbstub = @import("lib/zba-gdbstub/build.zig");
 
 pub fn build(b: *std.build.Builder) void {
-    // Standard target options allows the person running `zig build` to choose
+    // Minimum Zig Version
-    // what target to build for. Here we do not override the defaults, which
+    const min_ver = std.SemanticVersion.parse("0.11.0-dev.1580+a5b34a61a") catch return; // https://github.com/ziglang/zig/commit/a5b34a61a
-    // means any target is allowed, and the default is native. Other options
+    if (builtin.zig_version.order(min_ver).compare(.lt)) {
-    // for restricting supported target set are available.
+        std.log.err("{s}", .{b.fmt("Zig v{} does not meet the minimum version requirement. (Zig v{})", .{ builtin.zig_version, min_ver })});
+        std.os.exit(1);
+    }
+
     const target = b.standardTargetOptions(.{});
+    const optimize = b.standardOptimizeOption(.{});
 
-    // Standard release options allow the person running `zig build` to select
+    const exe = b.addExecutable(.{
-    // between Debug, ReleaseSafe, ReleaseFast, and ReleaseSmall.
+        .name = "zba",
-    const mode = b.standardReleaseOptions();
+        .root_source_file = .{ .path = "src/main.zig" },
+        .target = target,
+        .optimize = optimize,
+    });
 
-    const exe = b.addExecutable("zba", "src/main.zig");
     exe.setMainPkgPath("."); // Necessary so that src/main.zig can embed example.toml
-    exe.setTarget(target);
 
     // Known Folders (%APPDATA%, XDG, etc.)
-    exe.addPackagePath("known_folders", "lib/known-folders/known-folders.zig");
+    exe.addAnonymousModule("known_folders", .{ .source_file = .{ .path = "lib/known-folders/known-folders.zig" } });
 
     // DateTime Library
-    exe.addPackagePath("datetime", "lib/zig-datetime/src/main.zig");
+    exe.addAnonymousModule("datetime", .{ .source_file = .{ .path = "lib/zig-datetime/src/main.zig" } });
 
     // Bitfield type from FlorenceOS: https://github.com/FlorenceOS/
-    // exe.addPackage(.{ .name = "bitfield", .path = .{ .path = "lib/util/bitfield.zig" } });
+    exe.addAnonymousModule("bitfield", .{ .source_file = .{ .path = "lib/util/bitfield.zig" } });
-    exe.addPackagePath("bitfield", "lib/util/bitfield.zig");
 
     // Argument Parsing Library
-    exe.addPackagePath("clap", "lib/zig-clap/clap.zig");
+    exe.addAnonymousModule("clap", .{ .source_file = .{ .path = "lib/zig-clap/clap.zig" } });
 
     // TOML Library
-    exe.addPackagePath("toml", "lib/zig-toml/src/toml.zig");
+    exe.addAnonymousModule("toml", .{ .source_file = .{ .path = "lib/zig-toml/src/toml.zig" } });
 
     // OpenGL 3.3 Bindings
-    exe.addPackagePath("gl", "lib/gl.zig");
+    exe.addAnonymousModule("gl", .{ .source_file = .{ .path = "lib/gl.zig" } });
+
+    // gdbstub
+    Gdbstub.link(exe);
 
     // Zig SDL Bindings: https://github.com/MasterQ32/SDL.zig
-    const sdk = Sdk.init(b);
+    const sdk = Sdk.init(b, null);
     sdk.link(exe, .dynamic);
-    exe.addPackage(sdk.getNativePackage("sdl2"));
+    exe.addModule("sdl2", sdk.getNativeModule());
 
-    exe.setBuildMode(mode);
     exe.install();
 
     const run_cmd = exe.run();
@@ -52,9 +61,11 @@ pub fn build(b: *std.build.Builder) void {
     const run_step = b.step("run", "Run the app");
     run_step.dependOn(&run_cmd.step);
 
-    const exe_tests = b.addTest("src/main.zig");
+    const exe_tests = b.addTest(.{
-    exe_tests.setTarget(target);
+        .root_source_file = .{ .path = "src/main.zig" },
-    exe_tests.setBuildMode(mode);
+        .target = target,
+        .optimize = optimize,
+    });
 
     const test_step = b.step("test", "Run unit tests");
     test_step.dependOn(&exe_tests.step);

src/config.zig

@@ -49,16 +49,19 @@ pub fn config() *const Config {
 }
 
 /// Reads a config file and then loads it into the global state
-pub fn load(allocator: Allocator, config_path: []const u8) !void {
+pub fn load(allocator: Allocator, file_path: []const u8) !void {
-    var config_file = try std.fs.cwd().openFile(config_path, .{});
+    var config_file = try std.fs.cwd().openFile(file_path, .{});
     defer config_file.close();
 
-    log.info("loaded from {s}", .{config_path});
+    log.info("loaded from {s}", .{file_path});
 
     const contents = try config_file.readToEndAlloc(allocator, try config_file.getEndPos());
     defer allocator.free(contents);
 
-    const table = try toml.parseContents(allocator, contents, null);
+    var parser = try toml.parseFile(allocator, file_path);
+    defer parser.deinit();
+
+    const table = try parser.parse();
     defer table.deinit();
 
     // TODO: Report unknown config options

src/core/Bus.zig

@@ -33,6 +33,11 @@ pub const fetch_timings: [2][0x10]u8 = [_][0x10]u8{
     [_]u8{ 1, 1, 6, 1, 1, 2, 2, 1, 4, 4, 4, 4, 4, 4, 8, 8 }, // 32-bit
 };
 
+// Fastmem Related
+const page_size = 1 * 0x400; // 1KiB
+const address_space_size = 0x1000_0000;
+const table_len = address_space_size / page_size;
+
 const Self = @This();
 
 pak: GamePak,
@@ -48,7 +53,16 @@ io: Io,
 cpu: *Arm7tdmi,
 sched: *Scheduler,
 
+read_table: *const [table_len]?*const anyopaque,
+write_tables: [2]*const [table_len]?*anyopaque,
+allocator: Allocator,
+
 pub fn init(self: *Self, allocator: Allocator, sched: *Scheduler, cpu: *Arm7tdmi, paths: FilePaths) !void {
+    const tables = try allocator.alloc(?*anyopaque, 3 * table_len); // Allocate all tables
+
+    const read_table = tables[0..table_len];
+    const write_tables = .{ tables[table_len .. 2 * table_len], tables[2 * table_len .. 3 * table_len] };
+
     self.* = .{
         .pak = try GamePak.init(allocator, cpu, paths.rom, paths.save),
         .bios = try Bios.init(allocator, paths.bios),
@@ -61,7 +75,17 @@ pub fn init(self: *Self, allocator: Allocator, sched: *Scheduler, cpu: *Arm7tdmi
         .io = Io.init(),
         .cpu = cpu,
         .sched = sched,
+
+        .read_table = read_table,
+        .write_tables = write_tables,
+        .allocator = allocator,
     };
+
+    self.fillReadTable(read_table);
+
+    // Internal Display Memory behaves differently on 8-bit reads
+    self.fillWriteTable(u32, write_tables[0]);
+    self.fillWriteTable(u8, write_tables[1]);
 }
 
 pub fn deinit(self: *Self) void {
@@ -70,59 +94,126 @@ pub fn deinit(self: *Self) void {
     self.pak.deinit();
     self.bios.deinit();
     self.ppu.deinit();
+
+    // This is so I can deallocate the original `allocator.alloc`. I have to re-make the type
+    // since I'm not keeping it around, This is very jank and bad though
+    // FIXME: please figure out another way
+    self.allocator.free(@ptrCast([*]const ?*anyopaque, self.read_table[0..])[0 .. 3 * table_len]);
     self.* = undefined;
 }
 
-pub fn dbgRead(self: *const Self, comptime T: type, address: u32) T {
+fn fillReadTable(self: *Self, table: *[table_len]?*const anyopaque) void {
-    const page = @truncate(u8, address >> 24);
+    const vramMirror = @import("ppu/Vram.zig").mirror;
-    const aligned_addr = forceAlign(T, address);
 
-    return switch (page) {
+    for (table, 0..) |*ptr, i| {
-        // General Internal Memory
+        const addr = @intCast(u32, page_size * i);
-        0x00 => blk: {
-            if (address < Bios.size)
-                break :blk self.bios.dbgRead(T, self.cpu.r[15], aligned_addr);
 
-            break :blk self.openBus(T, address);
+        ptr.* = switch (addr) {
-        },
+            // General Internal Memory
-        0x02 => self.ewram.read(T, aligned_addr),
+            0x0000_0000...0x0000_3FFF => null, // BIOS has it's own checks
-        0x03 => self.iwram.read(T, aligned_addr),
+            0x0200_0000...0x02FF_FFFF => &self.ewram.buf[addr & 0x3FFFF],
-        0x04 => self.readIo(T, address),
+            0x0300_0000...0x03FF_FFFF => &self.iwram.buf[addr & 0x7FFF],
+            0x0400_0000...0x0400_03FF => null, // I/O
 
-        // Internal Display Memory
+            // Internal Display Memory
-        0x05 => self.ppu.palette.read(T, aligned_addr),
+            0x0500_0000...0x05FF_FFFF => &self.ppu.palette.buf[addr & 0x3FF],
-        0x06 => self.ppu.vram.read(T, aligned_addr),
+            0x0600_0000...0x06FF_FFFF => &self.ppu.vram.buf[vramMirror(addr)],
-        0x07 => self.ppu.oam.read(T, aligned_addr),
+            0x0700_0000...0x07FF_FFFF => &self.ppu.oam.buf[addr & 0x3FF],
 
-        // External Memory (Game Pak)
+            // External Memory (Game Pak)
-        0x08...0x0D => self.pak.dbgRead(T, aligned_addr),
+            0x0800_0000...0x0DFF_FFFF => self.fillReadTableExternal(addr),
-        0x0E...0x0F => blk: {
+            0x0E00_0000...0x0FFF_FFFF => null, // SRAM
-            const value = self.pak.backup.read(address);
+            else => null,
-
+        };
-            const multiplier = switch (T) {
+    }
-                u32 => 0x01010101,
-                u16 => 0x0101,
-                u8 => 1,
-                else => @compileError("Backup: Unsupported read width"),
-            };
-
-            break :blk @as(T, value) * multiplier;
-        },
-        else => self.openBus(T, address),
-    };
 }
 
-/// TODO: Should open bus read addresses be force-aligned?
+fn fillWriteTable(self: *Self, comptime T: type, table: *[table_len]?*const anyopaque) void {
-fn readIo(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    comptime std.debug.assert(T == u32 or T == u16 or T == u8);
-    const maybe_value = io.read(self, T, forceAlign(T, unaligned_address));
+    const vramMirror = @import("ppu/Vram.zig").mirror;
-    return if (maybe_value) |value| value else self.openBus(T, unaligned_address);
+
+    for (table, 0..) |*ptr, i| {
+        const addr = @intCast(u32, page_size * i);
+
+        ptr.* = switch (addr) {
+            // General Internal Memory
+            0x0000_0000...0x0000_3FFF => null, // BIOS has it's own checks
+            0x0200_0000...0x02FF_FFFF => &self.ewram.buf[addr & 0x3FFFF],
+            0x0300_0000...0x03FF_FFFF => &self.iwram.buf[addr & 0x7FFF],
+            0x0400_0000...0x0400_03FF => null, // I/O
+
+            // Internal Display Memory
+            0x0500_0000...0x05FF_FFFF => if (T != u8) &self.ppu.palette.buf[addr & 0x3FF] else null,
+            0x0600_0000...0x06FF_FFFF => if (T != u8) &self.ppu.vram.buf[vramMirror(addr)] else null,
+            0x0700_0000...0x07FF_FFFF => if (T != u8) &self.ppu.oam.buf[addr & 0x3FF] else null,
+
+            // External Memory (Game Pak)
+            0x0800_0000...0x0DFF_FFFF => null, // ROM
+            0x0E00_0000...0x0FFF_FFFF => null, // SRAM
+            else => null,
+        };
+    }
+}
+
+fn fillReadTableExternal(self: *Self, addr: u32) ?*anyopaque {
+    // see `GamePak.zig` for more information about what conditions need to be true
+    // so that a simple pointer dereference isn't possible
+
+    std.debug.assert(addr & @as(u32, page_size - 1) == 0); // addr is guaranteed to be page-aligned
+
+    const start_addr = addr;
+    const end_addr = start_addr + page_size;
+
+    {
+        const data = start_addr <= 0x0800_00C4 and 0x0800_00C4 < end_addr; // GPIO Data
+        const direction = start_addr <= 0x0800_00C6 and 0x0800_00C6 < end_addr; // GPIO Direction
+        const control = start_addr <= 0x0800_00C8 and 0x0800_00C8 < end_addr; // GPIO Control
+
+        const has_gpio = data or direction or control;
+        const gpio_kind = self.pak.gpio.device.kind;
+
+        // There is a GPIO Device, and the current page contains at least one memory-mapped GPIO register
+        if (gpio_kind != .None and has_gpio) return null;
+    }
+
+    if (self.pak.backup.kind == .Eeprom) {
+        if (self.pak.buf.len > 0x100_000) {
+            // We are using a "large" EEPROM which means that if the below check is true
+            // this page has an address that's reserved for the EEPROM and therefore must
+            // be handled in slowmem
+            if (addr & 0x1FF_FFFF > 0x1FF_FEFF) return null;
+        } else {
+            // We are using a "small" EEPROM which means that if the below check is true
+            // (that is, we're in the 0xD address page) then we must handle at least one
+            // address in this page in slowmem
+            if (@truncate(u4, addr >> 24) == 0xD) return null;
+        }
+    }
+
+    // Finally, the GamePak has some unique behaviour for reads past the end of the ROM,
+    // so those will be handled by slowmem as well
+    const masked_addr = addr & 0x1FF_FFFF;
+    if (masked_addr >= self.pak.buf.len) return null;
+
+    return &self.pak.buf[masked_addr];
+}
+
+fn readIo(self: *const Self, comptime T: type, address: u32) T {
+    return io.read(self, T, address) orelse self.openBus(T, address);
 }
 
 fn openBus(self: *const Self, comptime T: type, address: u32) T {
+    @setCold(true);
     const r15 = self.cpu.r[15];
 
     const word = blk: {
         // If Arm, get the most recently fetched instruction (PC + 8)
+        //
+        // FIXME: This is most likely a faulty assumption.
+        // I think what *actually* happens is that the Bus has a latch for the most
+        // recently fetched piece of data, which is then returned during Open Bus (also DMA open bus?)
+        // I can "get away" with this because it's very statistically likely that the most recently latched value is
+        // the most recently fetched instruction by the pipeline
         if (!self.cpu.cpsr.t.read()) break :blk self.cpu.pipe.stage[1].?;
 
         const page = @truncate(u8, r15 >> 24);
@@ -172,85 +263,237 @@ fn openBus(self: *const Self, comptime T: type, address: u32) T {
     return @truncate(T, word);
 }
 
-pub fn read(self: *Self, comptime T: type, address: u32) T {
+pub fn read(self: *Self, comptime T: type, unaligned_address: u32) T {
-    const page = @truncate(u8, address >> 24);
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
-    const aligned_addr = forceAlign(T, address);
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
 
-    self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, page)];
+    // whether or not we do this in slowmem or fastmem, we should advance the scheduler
+    self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, unaligned_address >> 24)];
+
+    // We're doing some serious out-of-bounds open-bus reads
+    if (page >= table_len) return self.openBus(T, unaligned_address);
+
+    if (self.read_table[page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]const T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        return ptr[forceAlign(T, offset) / @sizeOf(T)];
+    }
+
+    return self.slowRead(T, unaligned_address);
+}
+
+pub fn dbgRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // We're doing some serious out-of-bounds open-bus reads
+    if (page >= table_len) return self.openBus(T, unaligned_address);
+
+    if (self.read_table[page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]const T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        return ptr[forceAlign(T, offset) / @sizeOf(T)];
+    }
+
+    return self.dbgSlowRead(T, unaligned_address);
+}
+
+fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T {
+    @setCold(true);
+
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
 
     return switch (page) {
         // General Internal Memory
         0x00 => blk: {
             if (address < Bios.size)
-                break :blk self.bios.read(T, self.cpu.r[15], aligned_addr);
+                break :blk self.bios.read(T, self.cpu.r[15], unaligned_address);
 
             break :blk self.openBus(T, address);
         },
-        0x02 => self.ewram.read(T, aligned_addr),
+        0x02 => unreachable, // completely handled by fastmeme
-        0x03 => self.iwram.read(T, aligned_addr),
+        0x03 => unreachable, // completely handled by fastmeme
         0x04 => self.readIo(T, address),
 
         // Internal Display Memory
-        0x05 => self.ppu.palette.read(T, aligned_addr),
+        0x05 => unreachable, // completely handled by fastmeme
-        0x06 => self.ppu.vram.read(T, aligned_addr),
+        0x06 => unreachable, // completely handled by fastmeme
-        0x07 => self.ppu.oam.read(T, aligned_addr),
+        0x07 => unreachable, // completely handled by fastmeme
 
         // External Memory (Game Pak)
-        0x08...0x0D => self.pak.read(T, aligned_addr),
+        0x08...0x0D => self.pak.read(T, address),
-        0x0E...0x0F => blk: {
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
-            const value = self.pak.backup.read(address);
-
-            const multiplier = switch (T) {
-                u32 => 0x01010101,
-                u16 => 0x0101,
-                u8 => 1,
-                else => @compileError("Backup: Unsupported read width"),
-            };
-
-            break :blk @as(T, value) * multiplier;
-        },
         else => self.openBus(T, address),
     };
 }
 
-pub fn write(self: *Self, comptime T: type, address: u32, value: T) void {
+fn dbgSlowRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
-    const page = @truncate(u8, address >> 24);
+    const page = @truncate(u8, unaligned_address >> 24);
-    const aligned_addr = forceAlign(T, address);
+    const address = forceAlign(T, unaligned_address);
 
-    self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, page)];
+    return switch (page) {
+        // General Internal Memory
+        0x00 => blk: {
+            if (address < Bios.size)
+                break :blk self.bios.dbgRead(T, self.cpu.r[15], unaligned_address);
+
+            break :blk self.openBus(T, address);
+        },
+        0x02 => unreachable, // handled by fastmem
+        0x03 => unreachable, // handled by fastmem
+        0x04 => self.readIo(T, address),
+
+        // Internal Display Memory
+        0x05 => unreachable, // handled by fastmem
+        0x06 => unreachable, // handled by fastmem
+        0x07 => unreachable, // handled by fastmem
+
+        // External Memory (Game Pak)
+        0x08...0x0D => self.pak.dbgRead(T, address),
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
+        else => self.openBus(T, address),
+    };
+}
+
+fn readBackup(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const value = self.pak.backup.read(unaligned_address);
+
+    const multiplier = switch (T) {
+        u32 => 0x01010101,
+        u16 => 0x0101,
+        u8 => 1,
+        else => @compileError("Backup: Unsupported read width"),
+    };
+
+    return @as(T, value) * multiplier;
+}
+
+pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // whether or not we do this in slowmem or fastmem, we should advance the scheduler
+    self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, unaligned_address >> 24)];
+
+    // We're doing some serious out-of-bounds open-bus writes, they do nothing though
+    if (page >= table_len) return;
+
+    if (self.write_tables[@boolToInt(T == u8)][page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        ptr[forceAlign(T, offset) / @sizeOf(T)] = value;
+    } else {
+        // we can return early if this is an 8-bit OAM write
+        if (T == u8 and @truncate(u8, unaligned_address >> 24) == 0x07) return;
+
+        self.slowWrite(T, unaligned_address, value);
+    }
+}
+
+/// Mostly Identical to `Bus.write`, slowmeme is handled by `Bus.dbgSlowWrite`
+pub fn dbgWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
+    const page = unaligned_address >> bits;
+    const offset = unaligned_address & (page_size - 1);
+
+    // We're doing some serious out-of-bounds open-bus writes, they do nothing though
+    if (page >= table_len) return;
+
+    if (self.write_tables[@boolToInt(T == u8)][page]) |some_ptr| {
+        // We have a pointer to a page, cast the pointer to it's underlying type
+        const Ptr = [*]T;
+        const ptr = @ptrCast(Ptr, @alignCast(@alignOf(std.meta.Child(Ptr)), some_ptr));
+
+        // Note: We don't check array length, since we force align the
+        // lower bits of the address as the GBA would
+        ptr[forceAlign(T, offset) / @sizeOf(T)] = value;
+    } else {
+        // we can return early if this is an 8-bit OAM write
+        if (T == u8 and @truncate(u8, unaligned_address >> 24) == 0x07) return;
+
+        self.dbgSlowWrite(T, unaligned_address, value);
+    }
+}
+
+fn slowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    @setCold(true);
+
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
 
     switch (page) {
         // General Internal Memory
-        0x00 => self.bios.write(T, aligned_addr, value),
+        0x00 => self.bios.write(T, address, value),
-        0x02 => self.ewram.write(T, aligned_addr, value),
+        0x02 => unreachable, // completely handled by fastmem
-        0x03 => self.iwram.write(T, aligned_addr, value),
+        0x03 => unreachable, // completely handled by fastmem
-        0x04 => io.write(self, T, aligned_addr, value),
+        0x04 => io.write(self, T, address, value),
 
         // Internal Display Memory
-        0x05 => self.ppu.palette.write(T, aligned_addr, value),
+        0x05 => self.ppu.palette.write(T, address, value),
-        0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, aligned_addr, value),
+        0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, address, value),
-        0x07 => self.ppu.oam.write(T, aligned_addr, value),
+        0x07 => unreachable, // completely handled by fastmem
 
         // External Memory (Game Pak)
-        0x08...0x0D => self.pak.write(T, self.dma[3].word_count, aligned_addr, value),
+        0x08...0x0D => self.pak.write(T, self.dma[3].word_count, address, value),
-        0x0E...0x0F => {
+        0x0E...0x0F => self.pak.backup.write(unaligned_address, @truncate(u8, rotr(T, value, 8 * rotateBy(T, unaligned_address)))),
-            const rotate_by = switch (T) {
-                u32 => address & 3,
-                u16 => address & 1,
-                u8 => 0,
-                else => @compileError("Backup: Unsupported write width"),
-            };
-
-            self.pak.backup.write(address, @truncate(u8, rotr(T, value, 8 * rotate_by)));
-        },
         else => {},
     }
 }
 
-fn forceAlign(comptime T: type, address: u32) u32 {
+fn dbgSlowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
+    @setCold(true);
+
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    switch (page) {
+        // General Internal Memory
+        0x00 => self.bios.write(T, address, value),
+        0x02 => unreachable, // completely handled by fastmem
+        0x03 => unreachable, // completely handled by fastmem
+        0x04 => return, // FIXME: Let debug writes mess with I/O
+
+        // Internal Display Memory
+        0x05 => self.ppu.palette.write(T, address, value),
+        0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, address, value),
+        0x07 => unreachable, // completely handled by fastmem
+
+        // External Memory (Game Pak)
+        0x08...0x0D => return, // FIXME: Debug Write to Backup/GPIO w/out messing with state
+        0x0E...0x0F => return, // FIXME: Debug Write to Backup w/out messing with state
+        else => {},
+    }
+}
+
+inline fn rotateBy(comptime T: type, address: u32) u32 {
     return switch (T) {
-        u32 => address & 0xFFFF_FFFC,
+        u32 => address & 3,
-        u16 => address & 0xFFFF_FFFE,
+        u16 => address & 1,
+        u8 => 0,
+        else => @compileError("Unsupported write width"),
+    };
+}
+
+pub inline fn forceAlign(comptime T: type, address: u32) u32 {
+    return switch (T) {
+        u32 => address & ~@as(u32, 3),
+        u16 => address & ~@as(u32, 1),
         u8 => address,
         else => @compileError("Bus: Invalid read/write type"),
     };

src/core/apu.zig

@@ -108,106 +108,121 @@ pub fn read(comptime T: type, apu: *const Apu, addr: u32) ?T {
 pub fn write(comptime T: type, apu: *Apu, addr: u32, value: T) void {
     const byte_addr = @truncate(u8, addr);
 
+    if (byte_addr <= 0x81 and !apu.cnt.apu_enable.read()) return;
+
     switch (T) {
-        u32 => switch (byte_addr) {
+        u32 => {
-            0x60 => apu.ch1.setSound1Cnt(value),
+            // 0x80 and 0x81 handled in setSoundCnt
-            0x64 => apu.ch1.setSound1CntX(&apu.fs, @truncate(u16, value)),
+            if (byte_addr < 0x80 and !apu.cnt.apu_enable.read()) return;
 
-            0x68 => apu.ch2.setSound2CntL(@truncate(u16, value)),
+            switch (byte_addr) {
-            0x6C => apu.ch2.setSound2CntH(&apu.fs, @truncate(u16, value)),
+                0x60 => apu.ch1.setSound1Cnt(value),
+                0x64 => apu.ch1.setSound1CntX(&apu.fs, @truncate(u16, value)),
 
-            0x70 => apu.ch3.setSound3Cnt(value),
+                0x68 => apu.ch2.setSound2CntL(@truncate(u16, value)),
-            0x74 => apu.ch3.setSound3CntX(&apu.fs, @truncate(u16, value)),
+                0x6C => apu.ch2.setSound2CntH(&apu.fs, @truncate(u16, value)),
 
-            0x78 => apu.ch4.setSound4CntL(@truncate(u16, value)),
+                0x70 => apu.ch3.setSound3Cnt(value),
-            0x7C => apu.ch4.setSound4CntH(&apu.fs, @truncate(u16, value)),
+                0x74 => apu.ch3.setSound3CntX(&apu.fs, @truncate(u16, value)),
 
-            0x80 => apu.setSoundCnt(value),
+                0x78 => apu.ch4.setSound4CntL(@truncate(u16, value)),
-            0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
+                0x7C => apu.ch4.setSound4CntH(&apu.fs, @truncate(u16, value)),
-            0x88 => apu.bias.raw = @truncate(u16, value),
-            0x8C => {},
 
-            0x90, 0x94, 0x98, 0x9C => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0x80 => apu.setSoundCnt(value),
-            0xA0 => apu.chA.push(value), // FIFO_A
+                0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
-            0xA4 => apu.chB.push(value), // FIFO_B
+                0x88 => apu.bias.raw = @truncate(u16, value),
-            else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, addr }),
+                0x8C => {},
+
+                0x90, 0x94, 0x98, 0x9C => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0xA0 => apu.chA.push(value), // FIFO_A
+                0xA4 => apu.chB.push(value), // FIFO_B
+                else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, addr }),
+            }
         },
-        u16 => switch (byte_addr) {
+        u16 => {
-            0x60 => apu.ch1.setSound1CntL(@truncate(u8, value)), // SOUND1CNT_L
+            if (byte_addr <= 0x81 and !apu.cnt.apu_enable.read()) return;
-            0x62 => apu.ch1.setSound1CntH(value),
-            0x64 => apu.ch1.setSound1CntX(&apu.fs, value),
-            0x66 => {},
 
-            0x68 => apu.ch2.setSound2CntL(value),
+            switch (byte_addr) {
-            0x6A => {},
+                0x60 => apu.ch1.setSound1CntL(@truncate(u8, value)), // SOUND1CNT_L
-            0x6C => apu.ch2.setSound2CntH(&apu.fs, value),
+                0x62 => apu.ch1.setSound1CntH(value),
-            0x6E => {},
+                0x64 => apu.ch1.setSound1CntX(&apu.fs, value),
+                0x66 => {},
 
-            0x70 => apu.ch3.setSound3CntL(@truncate(u8, value)),
+                0x68 => apu.ch2.setSound2CntL(value),
-            0x72 => apu.ch3.setSound3CntH(value),
+                0x6A => {},
-            0x74 => apu.ch3.setSound3CntX(&apu.fs, value),
+                0x6C => apu.ch2.setSound2CntH(&apu.fs, value),
-            0x76 => {},
+                0x6E => {},
 
-            0x78 => apu.ch4.setSound4CntL(value),
+                0x70 => apu.ch3.setSound3CntL(@truncate(u8, value)),
-            0x7A => {},
+                0x72 => apu.ch3.setSound3CntH(value),
-            0x7C => apu.ch4.setSound4CntH(&apu.fs, value),
+                0x74 => apu.ch3.setSound3CntX(&apu.fs, value),
-            0x7E => {},
+                0x76 => {},
 
-            0x80 => apu.setSoundCntL(value),
+                0x78 => apu.ch4.setSound4CntL(value),
-            0x82 => apu.setSoundCntH(value),
+                0x7A => {},
-            0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
+                0x7C => apu.ch4.setSound4CntH(&apu.fs, value),
-            0x86 => {},
+                0x7E => {},
-            0x88 => apu.bias.raw = value, // SOUNDBIAS
-            0x8A, 0x8C, 0x8E => {},
 
-            0x90, 0x92, 0x94, 0x96, 0x98, 0x9A, 0x9C, 0x9E => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0x80 => apu.setSoundCntL(value),
-            0xA0, 0xA2 => log.err("Tried to write 0x{X:0>4}{} to FIFO_A", .{ value, T }),
+                0x82 => apu.setSoundCntH(value),
-            0xA4, 0xA6 => log.err("Tried to write 0x{X:0>4}{} to FIFO_B", .{ value, T }),
+                0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
-            else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, addr }),
+                0x86 => {},
+                0x88 => apu.bias.raw = value, // SOUNDBIAS
+                0x8A, 0x8C, 0x8E => {},
+
+                0x90, 0x92, 0x94, 0x96, 0x98, 0x9A, 0x9C, 0x9E => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0xA0, 0xA2 => log.err("Tried to write 0x{X:0>4}{} to FIFO_A", .{ value, T }),
+                0xA4, 0xA6 => log.err("Tried to write 0x{X:0>4}{} to FIFO_B", .{ value, T }),
+                else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, addr }),
+            }
         },
-        u8 => switch (byte_addr) {
+        u8 => {
-            0x60 => apu.ch1.setSound1CntL(value),
+            if (byte_addr <= 0x81 and !apu.cnt.apu_enable.read()) return;
-            0x61 => {},
-            0x62 => apu.ch1.setNr11(value),
-            0x63 => apu.ch1.setNr12(value),
-            0x64 => apu.ch1.setNr13(value),
-            0x65 => apu.ch1.setNr14(&apu.fs, value),
-            0x66, 0x67 => {},
 
-            0x68 => apu.ch2.setNr21(value),
+            switch (byte_addr) {
-            0x69 => apu.ch2.setNr22(value),
+                0x60 => apu.ch1.setSound1CntL(value),
-            0x6A, 0x6B => {},
+                0x61 => {},
-            0x6C => apu.ch2.setNr23(value),
+                0x62 => apu.ch1.setNr11(value),
-            0x6D => apu.ch2.setNr24(&apu.fs, value),
+                0x63 => apu.ch1.setNr12(value),
-            0x6E, 0x6F => {},
+                0x64 => apu.ch1.setNr13(value),
+                0x65 => apu.ch1.setNr14(&apu.fs, value),
+                0x66, 0x67 => {},
 
-            0x70 => apu.ch3.setSound3CntL(value), // NR30
+                0x68 => apu.ch2.setNr21(value),
-            0x71 => {},
+                0x69 => apu.ch2.setNr22(value),
-            0x72 => apu.ch3.setNr31(value),
+                0x6A, 0x6B => {},
-            0x73 => apu.ch3.vol.raw = value, // NR32
+                0x6C => apu.ch2.setNr23(value),
-            0x74 => apu.ch3.setNr33(value),
+                0x6D => apu.ch2.setNr24(&apu.fs, value),
-            0x75 => apu.ch3.setNr34(&apu.fs, value),
+                0x6E, 0x6F => {},
-            0x76, 0x77 => {},
 
-            0x78 => apu.ch4.setNr41(value),
+                0x70 => apu.ch3.setSound3CntL(value), // NR30
-            0x79 => apu.ch4.setNr42(value),
+                0x71 => {},
-            0x7A, 0x7B => {},
+                0x72 => apu.ch3.setNr31(value),
-            0x7C => apu.ch4.poly.raw = value, // NR 43
+                0x73 => apu.ch3.vol.raw = value, // NR32
-            0x7D => apu.ch4.setNr44(&apu.fs, value),
+                0x74 => apu.ch3.setNr33(value),
-            0x7E, 0x7F => {},
+                0x75 => apu.ch3.setNr34(&apu.fs, value),
+                0x76, 0x77 => {},
 
-            0x80, 0x81 => apu.setSoundCntL(setHalf(u16, apu.psg_cnt.raw, byte_addr, value)),
+                0x78 => apu.ch4.setNr41(value),
-            0x82, 0x83 => apu.setSoundCntH(setHalf(u16, apu.dma_cnt.raw, byte_addr, value)),
+                0x79 => apu.ch4.setNr42(value),
-            0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
+                0x7A, 0x7B => {},
-            0x85 => {},
+                0x7C => apu.ch4.poly.raw = value, // NR 43
-            0x86, 0x87 => {},
+                0x7D => apu.ch4.setNr44(&apu.fs, value),
-            0x88, 0x89 => apu.bias.raw = setHalf(u16, apu.bias.raw, byte_addr, value), // SOUNDBIAS
+                0x7E, 0x7F => {},
-            0x8A...0x8F => {},
 
-            0x90...0x9F => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0x80, 0x81 => apu.setSoundCntL(setHalf(u16, apu.psg_cnt.raw, byte_addr, value)),
-            0xA0...0xA3 => log.err("Tried to write 0x{X:0>2}{} to FIFO_A", .{ value, T }),
+                0x82, 0x83 => apu.setSoundCntH(setHalf(u16, apu.dma_cnt.raw, byte_addr, value)),
-            0xA4...0xA7 => log.err("Tried to write 0x{X:0>2}{} to FIFO_B", .{ value, T }),
+                0x84 => apu.setSoundCntX(value >> 7 & 1 == 1),
-            else => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, addr }),
+                0x85 => {},
+                0x86, 0x87 => {},
+                0x88, 0x89 => apu.bias.raw = setHalf(u16, apu.bias.raw, byte_addr, value), // SOUNDBIAS
+                0x8A...0x8F => {},
+
+                0x90...0x9F => apu.ch3.wave_dev.write(T, apu.ch3.select, addr, value),
+                0xA0...0xA3 => log.err("Tried to write 0x{X:0>2}{} to FIFO_A", .{ value, T }),
+                0xA4...0xA7 => log.err("Tried to write 0x{X:0>2}{} to FIFO_B", .{ value, T }),
+                else => util.io.write.undef(log, "Tried to write 0x{X:0>2}{} to 0x{X:0>8}", .{ value, T, addr }),
+            }
         },
         else => @compileError("APU: Unsupported write width"),
     }
@@ -275,15 +290,20 @@ pub const Apu = struct {
     }
 
     fn reset(self: *Self) void {
+        // All PSG Registers between 0x0400_0060..0x0400_0081 are zeroed
+        // 0x0400_0082 and 0x0400_0088 retain their values
         self.ch1.reset();
         self.ch2.reset();
         self.ch3.reset();
         self.ch4.reset();
+
+        // GBATEK says 4000060h..4000081h I take this to mean inclusive
+        self.psg_cnt.raw = 0x0000;
     }
 
     /// SOUNDCNT
     fn setSoundCnt(self: *Self, value: u32) void {
-        self.setSoundCntL(@truncate(u16, value));
+        if (self.cnt.apu_enable.read()) self.setSoundCntL(@truncate(u16, value));
         self.setSoundCntH(@truncate(u16, value >> 16));
     }
 
@@ -322,11 +342,14 @@ pub const Apu = struct {
             self.fs.step = 0; // Reset Frame Sequencer
 
             // Reset Square Wave Offsets
-            self.ch1.square.pos = 0;
+            self.ch1.square.reset();
-            self.ch2.square.pos = 0;
+            self.ch2.square.reset();
 
-            // Reset Wave Device Offsets
+            // Reset Wave
-            self.ch3.wave_dev.offset = 0;
+            self.ch3.wave_dev.reset();
+
+            // Rest Noise
+            self.ch4.lfsr.reset();
         } else {
             self.reset();
         }
@@ -395,8 +418,8 @@ pub const Apu = struct {
         right += if (self.dma_cnt.chB_right.read()) chB_sample else 0;
 
         // Add SOUNDBIAS
-        // FIXME: Is SOUNDBIAS 9-bit or 10-bit?
+        // FIXME: SOUNDBIAS is 10-bit but The waveform is centered around 0 if I treat it as 11-bit
-        const bias = @as(i16, self.bias.level.read()) << 1;
+        const bias = @as(i16, self.bias.level.read()) << 2;
         left += bias;
         right += bias;
 
@@ -407,7 +430,6 @@ pub const Apu = struct {
         const ext_left = (clamped_left << 5) | (clamped_left >> 6);
         const ext_right = (clamped_right << 5) | (clamped_right >> 6);
 
-        // FIXME: This rarely happens
         if (self.sampling_cycle != self.bias.sampling_cycle.read()) self.replaceSDLResampler();
 
         _ = SDL.SDL_AudioStreamPut(self.stream, &[2]u16{ ext_left, ext_right }, 2 * @sizeOf(u16));
@@ -473,11 +495,15 @@ pub const Apu = struct {
         if (!self.cnt.apu_enable.read()) return;
 
         if (@boolToInt(self.dma_cnt.chA_timer.read()) == tim_id) {
+            if (!self.chA.enabled) return;
+
             self.chA.updateSample();
             if (self.chA.len() <= 15) cpu.bus.dma[1].requestAudio(0x0400_00A0);
         }
 
         if (@boolToInt(self.dma_cnt.chB_timer.read()) == tim_id) {
+            if (!self.chB.enabled) return;
+
             self.chB.updateSample();
             if (self.chB.len() <= 15) cpu.bus.dma[2].requestAudio(0x0400_00A4);
         }
@@ -491,19 +517,28 @@ pub fn DmaSound(comptime kind: DmaSoundKind) type {
         fifo: SoundFifo,
         kind: DmaSoundKind,
         sample: i8,
+        enabled: bool,
 
         fn init() Self {
             return .{
                 .fifo = SoundFifo.init(),
                 .kind = kind,
                 .sample = 0,
+                .enabled = false,
             };
         }
 
         pub fn push(self: *Self, value: u32) void {
+            if (!self.enabled) self.enable();
+
             self.fifo.write(&intToBytes(u32, value)) catch |e| log.err("{} Error: {}", .{ kind, e });
         }
 
+        fn enable(self: *Self) void {
+            @setCold(true);
+            self.enabled = true;
+        }
+
         pub fn len(self: *const Self) usize {
             return self.fifo.readableLength();
         }

src/core/apu/Noise.zig

@@ -49,10 +49,13 @@ pub fn init(sched: *Scheduler) Self {
 }
 
 pub fn reset(self: *Self) void {
-    self.len = 0;
+    self.len = 0; // NR41
-    self.envelope.raw = 0;
+    self.envelope.raw = 0; // NR42
-    self.poly.raw = 0;
+    self.poly.raw = 0; // NR43
-    self.cnt.raw = 0;
+    self.cnt.raw = 0; // NR44
+
+    self.len_dev.reset();
+    self.env_dev.reset();
 
     self.sample = 0;
     self.enabled = false;

src/core/apu/Tone.zig

@@ -43,9 +43,12 @@ pub fn init(sched: *Scheduler) Self {
 }
 
 pub fn reset(self: *Self) void {
-    self.duty.raw = 0;
+    self.duty.raw = 0; // NR21
-    self.envelope.raw = 0;
+    self.envelope.raw = 0; // NR22
-    self.freq.raw = 0;
+    self.freq.raw = 0; // NR32, NR24
+
+    self.len_dev.reset();
+    self.env_dev.reset();
 
     self.sample = 0;
     self.enabled = false;

src/core/apu/ToneSweep.zig

@@ -50,12 +50,14 @@ pub fn init(sched: *Scheduler) Self {
 }
 
 pub fn reset(self: *Self) void {
-    self.sweep.raw = 0;
+    self.sweep.raw = 0; // NR10
-    self.sweep_dev.calc_performed = false;
+    self.duty.raw = 0; // NR11
+    self.envelope.raw = 0; // NR12
+    self.freq.raw = 0; // NR13, NR14
 
-    self.duty.raw = 0;
+    self.len_dev.reset();
-    self.envelope.raw = 0;
+    self.sweep_dev.reset();
-    self.freq.raw = 0;
+    self.env_dev.reset();
 
     self.sample = 0;
     self.enabled = false;
@@ -92,10 +94,9 @@ pub fn sound1CntL(self: *const Self) u8 {
 pub fn setSound1CntL(self: *Self, value: u8) void {
     const new = io.Sweep{ .raw = value };
 
-    if (self.sweep.direction.read() and !new.direction.read()) {
+    if (!new.direction.read()) {
-        // Sweep Negate bit has been cleared
+        // If at least one (1) sweep calculation has been made with
-        // If At least 1 Sweep Calculation has been made since
+        // the negate bit set (since last trigger), disable the channel
-        // the last trigger, the channel is immediately disabled
 
         if (self.sweep_dev.calc_performed) self.enabled = false;
     }

src/core/apu/Wave.zig

@@ -42,10 +42,13 @@ pub fn init(sched: *Scheduler) Self {
 }
 
 pub fn reset(self: *Self) void {
-    self.select.raw = 0;
+    self.select.raw = 0; // NR30
-    self.length = 0;
+    self.length = 0; // NR31
-    self.vol.raw = 0;
+    self.vol.raw = 0; // NR32
-    self.freq.raw = 0;
+    self.freq.raw = 0; // NR33, NR34
+
+    self.len_dev.reset();
+    self.wave_dev.reset();
 
     self.sample = 0;
     self.enabled = false;

src/core/apu/device/Envelope.zig

@@ -11,6 +11,11 @@ pub fn create() Self {
     return .{ .timer = 0, .vol = 0 };
 }
 
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.vol = 0;
+}
+
 pub fn tick(self: *Self, nrx2: io.Envelope) void {
     if (nrx2.period.read() != 0) {
         if (self.timer != 0) self.timer -= 1;

src/core/apu/device/Length.zig

@@ -6,6 +6,10 @@ pub fn create() Self {
     return .{ .timer = 0 };
 }
 
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+}
+
 pub fn tick(self: *Self, enabled: bool, ch_enable: *bool) void {
     if (enabled) {
         if (self.timer == 0) return;

src/core/apu/device/Sweep.zig

@@ -18,13 +18,19 @@ pub fn create() Self {
     };
 }
 
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.enabled = false;
+    self.shadow = 0;
+    self.calc_performed = false;
+}
+
 pub fn tick(self: *Self, ch1: *ToneSweep) void {
     if (self.timer != 0) self.timer -= 1;
 
     if (self.timer == 0) {
         const period = ch1.sweep.period.read();
         self.timer = if (period == 0) 8 else period;
-        if (!self.calc_performed) self.calc_performed = true;
 
         if (self.enabled and period != 0) {
             const new_freq = self.calculate(ch1.sweep, &ch1.enabled);
@@ -45,7 +51,10 @@ pub fn calculate(self: *Self, sweep: io.Sweep, ch_enable: *bool) u12 {
     const shadow_shifted = shadow >> sweep.shift.read();
     const decrease = sweep.direction.read();
 
-    const freq = if (decrease) shadow - shadow_shifted else shadow + shadow_shifted;
+    const freq = if (decrease) blk: {
+        self.calc_performed = true;
+        break :blk shadow - shadow_shifted;
+    } else shadow + shadow_shifted;
     if (freq > 0x7FF) ch_enable.* = false;
 
     return freq;

src/core/apu/signal/Lfsr.zig

@@ -19,6 +19,11 @@ pub fn create(sched: *Scheduler) Self {
     };
 }
 
+pub fn reset(self: *Self) void {
+    self.shift = 0;
+    self.timer = 0;
+}
+
 pub fn sample(self: *const Self) i8 {
     return if ((~self.shift & 1) == 1) 1 else -1;
 }
@@ -33,7 +38,7 @@ pub fn reload(self: *Self, poly: io.PolyCounter) void {
 }
 
 /// Scheduler Event Handler for LFSR Timer Expire
-/// FIXME: This gets called a lot, clogging up the Scheduler
+/// FIXME: This gets called a lot, slowing down the scheduler
 pub fn onLfsrTimerExpire(self: *Self, poly: io.PolyCounter, late: u64) void {
     // Obscure: "Using a noise channel clock shift of 14 or 15
     // results in the LFSR receiving no clocks."

src/core/apu/signal/Square.zig

@@ -20,6 +20,11 @@ pub fn init(sched: *Scheduler) Self {
     };
 }
 
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.pos = 0;
+}
+
 /// Scheduler Event Handler for Square Synth Timer Expire
 pub fn onSquareTimerExpire(self: *Self, comptime T: type, nrx34: io.Frequency, late: u64) void {
     comptime std.debug.assert(T == ToneSweep or T == Tone);

src/core/apu/signal/Wave.zig

@@ -38,6 +38,13 @@ pub fn init(sched: *Scheduler) Self {
     };
 }
 
+pub fn reset(self: *Self) void {
+    self.timer = 0;
+    self.offset = 0;
+
+    // sample buffer isn't reset because it's outside of the range of what NR52{7}'s effects
+}
+
 /// Reload internal Wave Timer
 pub fn reload(self: *Self, value: u11) void {
     self.sched.removeScheduledEvent(.{ .ApuChannel = 2 });

src/core/bus/Bios.zig

@@ -3,6 +3,9 @@ const std = @import("std");
 const Allocator = std.mem.Allocator;
 const log = std.log.scoped(.Bios);
 
+const rotr = @import("../../util.zig").rotr;
+const forceAlign = @import("../Bus.zig").forceAlign;
+
 /// Size of the BIOS in bytes
 pub const size = 0x4000;
 const Self = @This();
@@ -10,21 +13,37 @@ const Self = @This();
 buf: ?[]u8,
 allocator: Allocator,
 
-addr_latch: u32,
+addr_latch: u32 = 0,
 
-pub fn read(self: *Self, comptime T: type, r15: u32, addr: u32) T {
+// https://github.com/ITotalJustice/notorious_beeg/issues/106
+pub fn read(self: *Self, comptime T: type, r15: u32, address: u32) T {
     if (r15 < Self.size) {
+        const addr = forceAlign(T, address);
+
         self.addr_latch = addr;
         return self._read(T, addr);
     }
 
-    log.debug("Rejected read since r15=0x{X:0>8}", .{r15});
+    log.warn("Open Bus! Read from 0x{X:0>8}, but PC was 0x{X:0>8}", .{ address, r15 });
-    return @truncate(T, self._read(T, self.addr_latch));
+    const value = self._read(u32, self.addr_latch);
+
+    return @truncate(T, rotr(u32, value, 8 * rotateBy(T, address)));
 }
 
-pub fn dbgRead(self: *const Self, comptime T: type, r15: u32, addr: u32) T {
+fn rotateBy(comptime T: type, address: u32) u32 {
-    if (r15 < Self.size) return self._read(T, addr);
+    return switch (T) {
-    return @truncate(T, self._read(T, self.addr_latch + 8));
+        u8 => address & 3,
+        u16 => address & 2,
+        u32 => 0,
+        else => @compileError("bios: unsupported read width"),
+    };
+}
+
+pub fn dbgRead(self: *const Self, comptime T: type, r15: u32, address: u32) T {
+    if (r15 < Self.size) return self._read(T, forceAlign(T, address));
+
+    const value = self._read(u32, self.addr_latch);
+    return @truncate(T, rotr(u32, value, 8 * rotateBy(T, address)));
 }
 
 /// Read without the GBA safety checks
@@ -43,18 +62,19 @@ pub fn write(_: *Self, comptime T: type, addr: u32, value: T) void {
 }
 
 pub fn init(allocator: Allocator, maybe_path: ?[]const u8) !Self {
-    const buf: ?[]u8 = if (maybe_path) |path| blk: {
+    if (maybe_path == null) return .{ .buf = null, .allocator = allocator };
-        const file = try std.fs.cwd().openFile(path, .{});
+    const path = maybe_path.?;
-        defer file.close();
 
-        break :blk try file.readToEndAlloc(allocator, try file.getEndPos());
+    const buf = try allocator.alloc(u8, Self.size);
-    } else null;
+    errdefer allocator.free(buf);
 
-    return Self{
+    const file = try std.fs.cwd().openFile(path, .{});
-        .buf = buf,
+    defer file.close();
-        .allocator = allocator,
+
-        .addr_latch = 0,
+    const file_len = try file.readAll(buf);
-    };
+    if (file_len != Self.size) log.err("Expected BIOS to be {}B, was {}B", .{ Self.size, file_len });
+
+    return Self{ .buf = buf, .allocator = allocator };
 }
 
 pub fn deinit(self: *Self) void {

src/core/bus/GamePak.zig

@@ -105,14 +105,13 @@ pub fn dbgRead(self: *const Self, comptime T: type, address: u32) T {
 
         switch (T) {
             u32 => switch (address) {
-                // TODO: Do I even need to implement these?
+                // FIXME: Do I even need to implement these?
                 0x0800_00C4 => std.debug.panic("Handle 32-bit GPIO Data/Direction Reads", .{}),
                 0x0800_00C6 => std.debug.panic("Handle 32-bit GPIO Direction/Control Reads", .{}),
                 0x0800_00C8 => std.debug.panic("Handle 32-bit GPIO Control Reads", .{}),
                 else => {},
             },
             u16 => switch (address) {
-                // 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),
@@ -214,6 +213,7 @@ fn guessDevice(buf: []const u8) Gpio.Device.Kind {
     // Try to Guess if ROM uses RTC
     const needle = "RTC_V"; // I was told SIIRTC_V, though Pokemen Firered (USA) is a false negative
 
+    // TODO: Use new for loop syntax?
     var i: usize = 0;
     while ((i + needle.len) < buf.len) : (i += 1) {
         if (std.mem.eql(u8, needle, buf[i..(i + needle.len)])) return .Rtc;

src/core/bus/backup.zig

@@ -6,7 +6,6 @@ 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{
@@ -138,6 +137,7 @@ pub const Backup = struct {
         for (backup_kinds) |needle| {
             const needle_len = needle.str.len;
 
+            // TODO: Use new for loop syntax?
             var i: usize = 0;
             while ((i + needle_len) < rom.len) : (i += 1) {
                 if (std.mem.eql(u8, needle.str, rom[i..][0..needle_len])) return needle.kind;
@@ -151,8 +151,8 @@ pub const Backup = struct {
         const file_path = try self.savePath(allocator, path);
         defer allocator.free(file_path);
 
-        // FIXME: Don't rely on this lol
+        const expected = "untitled.sav";
-        if (std.mem.eql(u8, file_path[file_path.len - 12 .. file_path.len], "untitled.sav")) {
+        if (std.mem.eql(u8, file_path[file_path.len - expected.len .. file_path.len], expected)) {
             return log.err("ROM header lacks title, no save loaded", .{});
         }
 
@@ -195,7 +195,7 @@ pub const Backup = struct {
     }
 
     fn saveName(self: *const Self, allocator: Allocator) ![]const u8 {
-        const title_str = span(&escape(self.title));
+        const title_str = std.mem.sliceTo(&escape(self.title), 0);
         const name = if (title_str.len != 0) title_str else "untitled";
 
         return try std.mem.concat(allocator, u8, &[_][]const u8{ name, ".sav" });

src/core/bus/backup/eeprom.zig

@@ -63,7 +63,7 @@ pub const Eeprom = struct {
         }
 
         if (self.state == .RequestEnd) {
-            if (bit != 0) log.debug("EEPROM Request did not end in 0u1. TODO: is this ok?", .{});
+            // if (bit != 0) log.debug("EEPROM Request did not end in 0u1. TODO: is this ok?", .{});
             self.state = .Ready;
             return;
         }

src/core/bus/dma.zig

@@ -5,7 +5,7 @@ const DmaControl = @import("io.zig").DmaControl;
 const Bus = @import("../Bus.zig");
 const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
 
-pub const DmaTuple = std.meta.Tuple(&[_]type{ DmaController(0), DmaController(1), DmaController(2), DmaController(3) });
+pub const DmaTuple = struct { DmaController(0), DmaController(1), DmaController(2), DmaController(3) };
 const log = std.log.scoped(.DmaTransfer);
 
 const getHalf = util.getHalf;
@@ -256,12 +256,16 @@ fn DmaController(comptime id: u2) type {
                 cpu.bus.write(u16, dad_addr, @truncate(u16, rotr(u32, self.data_latch, 8 * (dad_addr & 3))));
             }
 
-            switch (sad_adj) {
+            switch (@truncate(u8, sad_addr >> 24)) {
-                .Increment => self.sad_latch +%= offset,
+                // according to fleroviux, DMAs with a source address in ROM misbehave
-                .Decrement => self.sad_latch -%= offset,
+                // the resultant behaviour is that the source address will increment despite what DMAXCNT says
-                // FIXME: Is just ignoring this ok?
+                0x08...0x0D => self.sad_latch +%= offset, // obscure behaviour
-                .IncrementReload => log.err("{} is a prohibited adjustment on SAD", .{sad_adj}),
+                else => switch (sad_adj) {
-                .Fixed => {},
+                    .Increment => self.sad_latch +%= offset,
+                    .Decrement => self.sad_latch -%= offset,
+                    .IncrementReload => log.err("{} is a prohibited adjustment on SAD", .{sad_adj}),
+                    .Fixed => {},
+                },
             }
 
             switch (dad_adj) {
@@ -334,11 +338,8 @@ fn DmaController(comptime id: u2) type {
     };
 }
 
-pub fn pollDmaOnBlank(bus: *Bus, comptime kind: DmaKind) void {
+pub fn onBlanking(bus: *Bus, comptime kind: DmaKind) void {
-    bus.dma[0].poll(kind);
+    inline for (0..4) |i| bus.dma[i].poll(kind);
-    bus.dma[1].poll(kind);
-    bus.dma[2].poll(kind);
-    bus.dma[3].poll(kind);
 }
 
 const Adjustment = enum(u2) {

src/core/bus/gpio.zig

@@ -71,7 +71,7 @@ pub const Gpio = struct {
 
         self.* = .{
             .data = 0b0000,
-            .direction = 0b1111, // TODO: What is GPIO DIrection set to by default?
+            .direction = 0b1111, // TODO: What is GPIO Direction set to by default?
             .cnt = 0b0,
 
             .device = switch (kind) {
@@ -293,13 +293,13 @@ pub const Clock = struct {
         self.cpu.sched.push(.RealTimeClock, (1 << 24) -| late); // Reschedule
 
         const now = DateTime.now();
-        self.year = bcd(u8, @intCast(u8, now.date.year - 2000));
+        self.year = bcd(@intCast(u8, now.date.year - 2000));
-        self.month = bcd(u5, now.date.month);
+        self.month = @truncate(u5, bcd(now.date.month));
-        self.day = bcd(u6, now.date.day);
+        self.day = @truncate(u6, bcd(now.date.day));
-        self.weekday = bcd(u3, (now.date.weekday() + 1) % 7); // API is Monday = 0, Sunday = 6. We want Sunday = 0, Saturday = 6
+        self.weekday = @truncate(u3, bcd((now.date.weekday() + 1) % 7)); // API is Monday = 0, Sunday = 6. We want Sunday = 0, Saturday = 6
-        self.hour = bcd(u6, now.time.hour);
+        self.hour = @truncate(u6, bcd(now.time.hour));
-        self.minute = bcd(u7, now.time.minute);
+        self.minute = @truncate(u7, bcd(now.time.minute));
-        self.second = bcd(u7, now.time.second);
+        self.second = @truncate(u7, bcd(now.time.second));
     }
 
     fn step(self: *Self, value: Data) u4 {
@@ -449,16 +449,8 @@ pub const Clock = struct {
     }
 };
 
-fn bcd(comptime T: type, value: u8) T {
+/// Converts an 8-bit unsigned integer to its BCD representation.
-    var input = value;
+/// Note: Algorithm only works for values between 0 and 99 inclusive.
-    var ret: u8 = 0;
+fn bcd(value: u8) u8 {
-    var shift: u3 = 0;
+    return ((value / 10) << 4) + (value % 10);
-
-    while (input > 0) {
-        ret |= (input % 10) << (shift << 2);
-        shift += 1;
-        input /= 10;
-    }
-
-    return @truncate(T, ret);
 }

src/core/bus/io.zig

@@ -9,6 +9,9 @@ const Bit = @import("bitfield").Bit;
 const Bitfield = @import("bitfield").Bitfield;
 const Bus = @import("../Bus.zig");
 
+const getHalf = util.getHalf;
+const setHalf = util.setHalf;
+
 const log = std.log.scoped(.@"I/O");
 
 pub const Io = struct {
@@ -19,15 +22,17 @@ pub const Io = struct {
     ie: InterruptEnable,
     irq: InterruptRequest,
     postflg: PostFlag,
+    waitcnt: WaitControl,
     haltcnt: HaltControl,
-    keyinput: KeyInput,
+    keyinput: AtomicKeyInput,
 
     pub fn init() Self {
         return .{
             .ime = false,
             .ie = .{ .raw = 0x0000 },
             .irq = .{ .raw = 0x0000 },
-            .keyinput = .{ .raw = 0x03FF },
+            .keyinput = AtomicKeyInput.init(.{ .raw = 0x03FF }),
+            .waitcnt = .{ .raw = 0x0000_0000 }, // Bit 15 == 0 for GBA
             .postflg = .FirstBoot,
             .haltcnt = .Execute,
         };
@@ -64,8 +69,10 @@ pub fn read(bus: *const Bus, comptime T: type, address: u32) ?T {
             0x0400_0150 => util.io.read.todo(log, "Read {} from JOY_RECV", .{T}),
 
             // Interrupts
-            0x0400_0200 => @as(T, bus.io.irq.raw) << 16 | bus.io.ie.raw,
+            0x0400_0200 => @as(u32, bus.io.irq.raw) << 16 | bus.io.ie.raw,
+            0x0400_0204 => bus.io.waitcnt.raw,
             0x0400_0208 => @boolToInt(bus.io.ime),
+            0x0400_0300 => @enumToInt(bus.io.postflg),
             else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, address }),
         },
         u16 => switch (address) {
@@ -85,16 +92,23 @@ pub fn read(bus: *const Bus, comptime T: type, address: u32) ?T {
             0x0400_0128 => util.io.read.todo(log, "Read {} from SIOCNT", .{T}),
 
             // Keypad Input
-            0x0400_0130 => bus.io.keyinput.raw,
+            0x0400_0130 => bus.io.keyinput.load(.Monotonic).raw,
 
             // Serial Communication 2
             0x0400_0134 => util.io.read.todo(log, "Read {} from RCNT", .{T}),
+            0x0400_0136 => 0x0000,
+            0x0400_0142 => 0x0000,
+            0x0400_015A => 0x0000,
 
             // Interrupts
             0x0400_0200 => bus.io.ie.raw,
             0x0400_0202 => bus.io.irq.raw,
-            0x0400_0204 => util.io.read.todo(log, "Read {} from WAITCNT", .{T}),
+            0x0400_0204 => bus.io.waitcnt.raw,
+            0x0400_0206 => 0x0000,
             0x0400_0208 => @boolToInt(bus.io.ime),
+            0x0400_020A => 0x0000,
+            0x0400_0300 => @enumToInt(bus.io.postflg),
+            0x0400_0302 => 0x0000,
             else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, address }),
         },
         u8 => return switch (address) {
@@ -118,10 +132,20 @@ pub fn read(bus: *const Bus, comptime T: type, address: u32) ?T {
 
             // Serial Communication 2
             0x0400_0135 => util.io.read.todo(log, "Read {} from RCNT_H", .{T}),
+            0x0400_0136, 0x0400_0137 => 0x00,
+            0x0400_0142, 0x0400_0143 => 0x00,
+            0x0400_015A, 0x0400_015B => 0x00,
 
             // Interrupts
-            0x0400_0200 => @truncate(T, bus.io.ie.raw),
+            0x0400_0200, 0x0400_0201 => @truncate(T, bus.io.ie.raw >> getHalf(@truncate(u8, address))),
+            0x0400_0202, 0x0400_0203 => @truncate(T, bus.io.irq.raw >> getHalf(@truncate(u8, address))),
+            0x0400_0204, 0x0400_0205 => @truncate(T, bus.io.waitcnt.raw >> getHalf(@truncate(u8, address))),
+            0x0400_0206, 0x0400_0207 => 0x00,
+            0x0400_0208, 0x0400_0209 => @truncate(T, @as(u16, @boolToInt(bus.io.ime)) >> getHalf(@truncate(u8, address))),
+            0x0400_020A, 0x0400_020B => 0x00,
             0x0400_0300 => @enumToInt(bus.io.postflg),
+            0x0400_0301 => null,
+            0x0400_0302, 0x0400_0303 => 0x00,
             else => util.io.read.undef(T, log, "Tried to perform a {} read to 0x{X:0>8}", .{ T, address }),
         },
         else => @compileError("I/O: Unsupported read width"),
@@ -168,9 +192,12 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
 
             // Interrupts
             0x0400_0200 => bus.io.setIrqs(value),
-            0x0400_0204 => log.debug("Wrote 0x{X:0>8} to WAITCNT", .{value}),
+            0x0400_0204 => bus.io.waitcnt.set(@truncate(u16, value)),
             0x0400_0208 => bus.io.ime = value & 1 == 1,
-            0x0400_020C...0x0400_021C => {}, // Unused
+            0x0400_0300 => {
+                bus.io.postflg = @intToEnum(PostFlag, value & 1);
+                bus.io.haltcnt = if (value >> 15 & 1 == 0) .Halt else @panic("TODO: Implement STOP");
+            },
             else => util.io.write.undef(log, "Tried to write 0x{X:0>8}{} to 0x{X:0>8}", .{ value, T, address }),
         },
         u16 => switch (address) {
@@ -186,7 +213,7 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
 
             // Timers
             0x0400_0100...0x0400_010E => timer.write(T, &bus.tim, address, value),
-            0x0400_0114 => {}, // TODO: Gyakuten Saiban writes 0x8000 to 0x0400_0114
+            0x0400_0114 => {},
             0x0400_0110 => {}, // Not Used,
 
             // Serial Communication 1
@@ -210,9 +237,14 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
             // Interrupts
             0x0400_0200 => bus.io.ie.raw = value,
             0x0400_0202 => bus.io.irq.raw &= ~value,
-            0x0400_0204 => log.debug("Wrote 0x{X:0>4} to WAITCNT", .{value}),
+            0x0400_0204 => bus.io.waitcnt.set(value),
+            0x0400_0206 => {},
             0x0400_0208 => bus.io.ime = value & 1 == 1,
-            0x0400_0206, 0x0400_020A => {}, // Not Used
+            0x0400_020A => {},
+            0x0400_0300 => {
+                bus.io.postflg = @intToEnum(PostFlag, value & 1);
+                bus.io.haltcnt = if (value >> 15 & 1 == 0) .Halt else @panic("TODO: Implement STOP");
+            },
             else => util.io.write.undef(log, "Tried to write 0x{X:0>4}{} to 0x{X:0>8}", .{ value, T, address }),
         },
         u8 => switch (address) {
@@ -237,9 +269,16 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
             0x0400_0140 => log.debug("Wrote 0x{X:0>2} to JOYCNT_L", .{value}),
 
             // Interrupts
+            0x0400_0200, 0x0400_0201 => bus.io.ie.raw = setHalf(u16, bus.io.ie.raw, @truncate(u8, address), value),
             0x0400_0202 => bus.io.irq.raw &= ~@as(u16, value),
+            0x0400_0203 => bus.io.irq.raw &= ~@as(u16, value) << 8, // TODO: Is this good?
+            0x0400_0204, 0x0400_0205 => bus.io.waitcnt.set(setHalf(u16, @truncate(u16, bus.io.waitcnt.raw), @truncate(u8, address), value)),
+            0x0400_0206, 0x0400_0207 => {},
             0x0400_0208 => bus.io.ime = value & 1 == 1,
-            0x0400_0300 => bus.io.postflg = std.meta.intToEnum(PostFlag, value & 1) catch unreachable,
+            0x0400_0209 => {},
+            0x0400_020A, 0x0400_020B => {},
+
+            0x0400_0300 => bus.io.postflg = @intToEnum(PostFlag, value & 1),
             0x0400_0301 => bus.io.haltcnt = if (value >> 7 & 1 == 0) .Halt else std.debug.panic("TODO: Implement STOP", .{}),
 
             0x0400_0410 => log.debug("Wrote 0x{X:0>2} to the common yet undocumented 0x{X:0>8}", .{ value, address }),
@@ -278,14 +317,22 @@ pub const DisplayControl = extern union {
 
 /// Read / Write
 pub const DisplayStatus = extern union {
+    /// read-only
     vblank: Bit(u16, 0),
+    /// read-only
     hblank: Bit(u16, 1),
+    // read-only
     coincidence: Bit(u16, 2),
     vblank_irq: Bit(u16, 3),
     hblank_irq: Bit(u16, 4),
     vcount_irq: Bit(u16, 5),
     vcount_trigger: Bitfield(u16, 8, 8),
     raw: u16,
+
+    pub fn set(self: *DisplayStatus, value: u16) void {
+        const mask: u16 = 0x00C7; // set bits are read-only
+        self.raw = (self.raw & mask) | (value & ~mask);
+    }
 };
 
 /// Read Only
@@ -329,6 +376,31 @@ const KeyInput = extern union {
     raw: u16,
 };
 
+const AtomicKeyInput = struct {
+    const Self = @This();
+    const Ordering = std.atomic.Ordering;
+
+    inner: KeyInput,
+
+    pub fn init(value: KeyInput) Self {
+        return .{ .inner = value };
+    }
+
+    pub inline fn load(self: *const Self, comptime ordering: Ordering) KeyInput {
+        return .{ .raw = switch (ordering) {
+            .AcqRel, .Release => @compileError("not supported for atomic loads"),
+            else => @atomicLoad(u16, &self.inner.raw, ordering),
+        } };
+    }
+
+    pub inline fn store(self: *Self, value: u16, comptime ordering: Ordering) void {
+        switch (ordering) {
+            .AcqRel, .Acquire => @compileError("not supported for atomic stores"),
+            else => @atomicStore(u16, &self.inner.raw, value, ordering),
+        }
+    }
+};
+
 // Read / Write
 pub const BackgroundControl = extern union {
     priority: Bitfield(u16, 0, 2),
@@ -377,6 +449,8 @@ pub const BldY = extern union {
     raw: u16,
 };
 
+const u8WriteKind = enum { Hi, Lo };
+
 /// Write-only
 pub const WinH = extern union {
     x2: Bitfield(u16, 0, 8),
@@ -386,6 +460,8 @@ pub const WinH = extern union {
 
 /// Write-only
 pub const WinV = extern union {
+    const Self = @This();
+
     y2: Bitfield(u16, 0, 8),
     y1: Bitfield(u16, 8, 8),
     raw: u16,
@@ -394,20 +470,20 @@ pub const WinV = extern union {
 pub const WinIn = extern union {
     w0_bg: Bitfield(u16, 0, 4),
     w0_obj: Bit(u16, 4),
-    w0_colour: Bit(u16, 5),
+    w0_bld: Bit(u16, 5),
     w1_bg: Bitfield(u16, 8, 4),
     w1_obj: Bit(u16, 12),
-    w1_colour: Bit(u16, 13),
+    w1_bld: Bit(u16, 13),
     raw: u16,
 };
 
 pub const WinOut = extern union {
     out_bg: Bitfield(u16, 0, 4),
     out_obj: Bit(u16, 4),
-    out_colour: Bit(u16, 5),
+    out_bld: Bit(u16, 5),
     obj_bg: Bitfield(u16, 8, 4),
     obj_obj: Bit(u16, 12),
-    obj_colour: Bit(u16, 13),
+    obj_bld: Bit(u16, 13),
     raw: u16,
 };
 
@@ -576,3 +652,24 @@ pub const SoundBias = extern union {
     sampling_cycle: Bitfield(u16, 14, 2),
     raw: u16,
 };
+
+/// Read / Write
+pub const WaitControl = extern union {
+    sram_cnt: Bitfield(u16, 0, 2),
+    s0_first: Bitfield(u16, 2, 2),
+    s0_second: Bit(u16, 4),
+    s1_first: Bitfield(u16, 5, 2),
+    s1_second: Bit(u16, 7),
+    s2_first: Bitfield(u16, 8, 2),
+    s2_second: Bit(u16, 10),
+    phi_out: Bitfield(u16, 11, 2),
+
+    prefetch_enable: Bit(u16, 14),
+    pak_kind: Bit(u16, 15),
+    raw: u16,
+
+    pub fn set(self: *WaitControl, value: u16) void {
+        const mask: u16 = 0x8000; // set bits are read-only
+        self.raw = (self.raw & mask) | (value & ~mask);
+    }
+};

src/core/bus/timer.zig

@@ -5,7 +5,7 @@ const TimerControl = @import("io.zig").TimerControl;
 const Scheduler = @import("../scheduler.zig").Scheduler;
 const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
 
-pub const TimerTuple = std.meta.Tuple(&[_]type{ Timer(0), Timer(1), Timer(2), Timer(3) });
+pub const TimerTuple = struct { Timer(0), Timer(1), Timer(2), Timer(3) };
 const log = std.log.scoped(.Timer);
 
 const getHalf = util.getHalf;
@@ -150,21 +150,36 @@ fn Timer(comptime id: u2) type {
         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
+            if (self.cnt.enabled.read()) {
-            self.sched.removeScheduledEvent(.{ .TimerOverflow = id });
+                // timer was already enabled
 
-            if (self.cnt.enabled.read() and (new.cascade.read() or !new.enabled.read())) {
+                // If enabled falling edge or cascade falling edge, timer is paused
-                // Either through the cascade bit or the enable bit, the timer has effectively been disabled
+                if (!new.enabled.read() or (!self.cnt.cascade.read() and new.cascade.read())) {
-                // The Counter should hold whatever value it should have been at when it was disabled
+                    self.sched.removeScheduledEvent(.{ .TimerOverflow = id });
-                self._counter +%= @truncate(u16, (self.sched.now() - self._start_timestamp) / self.frequency());
+
+                    // Counter should hold the value it stopped at meaning we have to calculate it now
+                    self._counter +%= @truncate(u16, (self.sched.now() - self._start_timestamp) / self.frequency());
+                }
+
+                // the timer has always been enabled, but the cascade bit which was blocking the timer has been unset
+                if (new.enabled.read() and (self.cnt.cascade.read() and !new.cascade.read())) {
+                    // we want to reschedule the timer event, however we won't reload the counter.
+                    // the invariant here is that self._counter holds the already calculated paused value
+
+                    self.rescheduleTimerExpire(0);
+                }
+            } else {
+                // the timer was previously disabeld
+
+                if (new.enabled.read()) {
+                    // timer should start counting (with a reloaded counter value)
+                    self._counter = self._reload;
+
+                    // if cascade happens to be set, the timer doesn't actually do anything though
+                    if (!new.cascade.read()) self.rescheduleTimerExpire(0);
+                }
             }
 
-            // The counter is only reloaded on the rising edge of the enable bit
-            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.rescheduleTimerExpire(0);
-
             self.cnt.raw = halfword;
         }
 
@@ -190,23 +205,20 @@ fn Timer(comptime id: u2) type {
 
             // Perform Cascade Behaviour
             switch (id) {
-                0 => if (cpu.bus.tim[1].cnt.cascade.read()) {
+                inline 0, 1, 2 => |idx| {
-                    cpu.bus.tim[1]._counter +%= 1;
+                    const next = idx + 1;
-                    if (cpu.bus.tim[1]._counter == 0) cpu.bus.tim[1].onTimerExpire(cpu, late);
+
+                    if (cpu.bus.tim[next].cnt.cascade.read()) {
+                        cpu.bus.tim[next]._counter +%= 1;
+                        if (cpu.bus.tim[next]._counter == 0) cpu.bus.tim[next].onTimerExpire(cpu, late);
+                    }
                 },
-                1 => if (cpu.bus.tim[2].cnt.cascade.read()) {
+                3 => {}, // THere is no timer for TIM3 to cascade to
-                    cpu.bus.tim[2]._counter +%= 1;
-                    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].onTimerExpire(cpu, late);
-                },
-                3 => {}, // There is no Timer for TIM3 to "cascade" to,
             }
 
             // Reschedule Timer if we're not cascading
-            if (!self.cnt.cascade.read()) {
+            // TIM0 cascade value is N/A
+            if (id == 0 or !self.cnt.cascade.read()) {
                 self._counter = self._reload;
                 self.rescheduleTimerExpire(late);
             }

src/core/cpu.zig

@@ -7,6 +7,7 @@ const Scheduler = @import("scheduler.zig").Scheduler;
 const Logger = @import("../util.zig").Logger;
 
 const File = std.fs.File;
+const log = std.log.scoped(.Arm7Tdmi);
 
 // ARM Instructions
 pub const arm = struct {
@@ -38,13 +39,12 @@ pub const arm = struct {
     }
 
     fn populate() [0x1000]InstrFn {
-        return comptime {
+        comptime {
             @setEvalBranchQuota(0xE000);
-            var ret = [_]InstrFn{und} ** 0x1000;
+            var table = [_]InstrFn{und} ** 0x1000;
 
-            var i: usize = 0;
+            for (&table, 0..) |*handler, i| {
-            while (i < ret.len) : (i += 1) {
+                handler.* = switch (@as(u2, i >> 10)) {
-                ret[i] = switch (@as(u2, i >> 10)) {
                     0b00 => if (i == 0x121) blk: {
                         break :blk branchExchange;
                     } else if (i & 0xFCF == 0x009) blk: {
@@ -106,8 +106,8 @@ pub const arm = struct {
                 };
             }
 
-            return ret;
+            return table;
-        };
+        }
     }
 };
 
@@ -135,13 +135,12 @@ pub const thumb = struct {
     }
 
     fn populate() [0x400]InstrFn {
-        return comptime {
+        comptime {
             @setEvalBranchQuota(5025); // This is exact
-            var ret = [_]InstrFn{und} ** 0x400;
+            var table = [_]InstrFn{und} ** 0x400;
 
-            var i: usize = 0;
+            for (&table, 0..) |*handler, i| {
-            while (i < ret.len) : (i += 1) {
+                handler.* = switch (@as(u3, i >> 7 & 0x7)) {
-                ret[i] = switch (@as(u3, i >> 7 & 0x7)) {
                     0b000 => if (i >> 5 & 0x3 == 0b11) blk: {
                         const I = i >> 4 & 1 == 1;
                         const is_sub = i >> 3 & 1 == 1;
@@ -229,13 +228,11 @@ pub const thumb = struct {
                 };
             }
 
-            return ret;
+            return table;
-        };
+        }
     }
 };
 
-const log = std.log.scoped(.Arm7Tdmi);
-
 pub const Arm7tdmi = struct {
     const Self = @This();
 
@@ -246,18 +243,64 @@ pub const Arm7tdmi = struct {
     cpsr: PSR,
     spsr: PSR,
 
-    /// Storage  for R8_fiq -> R12_fiq and their normal counterparts
+    bank: Bank,
-    /// e.g [r[0 + 8], fiq_r[0 + 8], r[1 + 8], fiq_r[1 + 8]...]
-    banked_fiq: [2 * 5]u32,
-
-    /// Storage for r13_<mode>, r14_<mode>
-    /// e.g. [r13, r14, r13_svc, r14_svc]
-    banked_r: [2 * 6]u32,
-
-    banked_spsr: [5]PSR,
 
     logger: ?Logger,
 
+    /// Bank of Registers from other CPU Modes
+    const Bank = struct {
+        /// Storage for r13_<mode>, r14_<mode>
+        /// e.g. [r13, r14, r13_svc, r14_svc]
+        r: [2 * 6]u32,
+
+        /// Storage  for R8_fiq -> R12_fiq and their normal counterparts
+        /// e.g [r[0 + 8], fiq_r[0 + 8], r[1 + 8], fiq_r[1 + 8]...]
+        fiq: [2 * 5]u32,
+
+        spsr: [5]PSR,
+
+        const Kind = enum(u1) {
+            R13 = 0,
+            R14,
+        };
+
+        pub fn create() Bank {
+            return .{
+                .r = [_]u32{0x00} ** 12,
+                .fiq = [_]u32{0x00} ** 10,
+                .spsr = [_]PSR{.{ .raw = 0x0000_0000 }} ** 5,
+            };
+        }
+
+        inline fn regIdx(mode: Mode, kind: Kind) usize {
+            const idx: usize = switch (mode) {
+                .User, .System => 0,
+                .Supervisor => 1,
+                .Abort => 2,
+                .Undefined => 3,
+                .Irq => 4,
+                .Fiq => 5,
+            };
+
+            return (idx * 2) + if (kind == .R14) @as(usize, 1) else 0;
+        }
+
+        inline fn spsrIdx(mode: Mode) usize {
+            return switch (mode) {
+                .Supervisor => 0,
+                .Abort => 1,
+                .Undefined => 2,
+                .Irq => 3,
+                .Fiq => 4,
+                else => std.debug.panic("[CPU/Mode] {} does not have a SPSR Register", .{mode}),
+            };
+        }
+
+        inline fn fiqIdx(i: usize, mode: Mode) usize {
+            return (i * 2) + if (mode == .Fiq) @as(usize, 1) else 0;
+        }
+    };
+
     pub fn init(sched: *Scheduler, bus: *Bus, log_file: ?std.fs.File) Self {
         return Self{
             .r = [_]u32{0x00} ** 16,
@@ -266,41 +309,11 @@ pub const Arm7tdmi = struct {
             .bus = bus,
             .cpsr = .{ .raw = 0x0000_001F },
             .spsr = .{ .raw = 0x0000_0000 },
-            .banked_fiq = [_]u32{0x00} ** 10,
+            .bank = Bank.create(),
-            .banked_r = [_]u32{0x00} ** 12,
-            .banked_spsr = [_]PSR{.{ .raw = 0x0000_0000 }} ** 5,
             .logger = if (log_file) |file| Logger.init(file) else null,
         };
     }
 
-    inline fn bankedIdx(mode: Mode, kind: BankedKind) usize {
-        const idx: usize = switch (mode) {
-            .User, .System => 0,
-            .Supervisor => 1,
-            .Abort => 2,
-            .Undefined => 3,
-            .Irq => 4,
-            .Fiq => 5,
-        };
-
-        return (idx * 2) + if (kind == .R14) @as(usize, 1) else 0;
-    }
-
-    inline fn bankedSpsrIndex(mode: Mode) usize {
-        return switch (mode) {
-            .Supervisor => 0,
-            .Abort => 1,
-            .Undefined => 2,
-            .Irq => 3,
-            .Fiq => 4,
-            else => std.debug.panic("[CPU/Mode] {} does not have a SPSR Register", .{mode}),
-        };
-    }
-
-    inline fn bankedFiqIdx(i: usize, mode: Mode) usize {
-        return (i * 2) + if (mode == .Fiq) @as(usize, 1) else 0;
-    }
-
     pub inline fn hasSPSR(self: *const Self) bool {
         const mode = getModeChecked(self, self.cpsr.mode.read());
         return switch (mode) {
@@ -336,14 +349,14 @@ pub const Arm7tdmi = struct {
         switch (idx) {
             8...12 => {
                 if (current == .Fiq) {
-                    self.banked_fiq[bankedFiqIdx(idx - 8, .User)] = value;
+                    self.bank.fiq[Bank.fiqIdx(idx - 8, .User)] = value;
                 } else self.r[idx] = value;
             },
             13, 14 => switch (current) {
                 .User, .System => self.r[idx] = value,
                 else => {
-                    const kind = std.meta.intToEnum(BankedKind, idx - 13) catch unreachable;
+                    const kind = std.meta.intToEnum(Bank.Kind, idx - 13) catch unreachable;
-                    self.banked_r[bankedIdx(.User, kind)] = value;
+                    self.bank.r[Bank.regIdx(.User, kind)] = value;
                 },
             },
             else => self.r[idx] = value, // R0 -> R7  and R15
@@ -354,12 +367,12 @@ pub const Arm7tdmi = struct {
         const current = getModeChecked(self, self.cpsr.mode.read());
 
         return switch (idx) {
-            8...12 => if (current == .Fiq) self.banked_fiq[bankedFiqIdx(idx - 8, .User)] else self.r[idx],
+            8...12 => if (current == .Fiq) self.bank.fiq[Bank.fiqIdx(idx - 8, .User)] else self.r[idx],
             13, 14 => switch (current) {
                 .User, .System => self.r[idx],
                 else => blk: {
-                    const kind = std.meta.intToEnum(BankedKind, idx - 13) catch unreachable;
+                    const kind = std.meta.intToEnum(Bank.Kind, idx - 13) catch unreachable;
-                    break :blk self.banked_r[bankedIdx(.User, kind)];
+                    break :blk self.bank.r[Bank.regIdx(.User, kind)];
                 },
             },
             else => self.r[idx], // R0 -> R7  and R15
@@ -370,40 +383,38 @@ pub const Arm7tdmi = struct {
         const now = getModeChecked(self, self.cpsr.mode.read());
 
         // Bank R8 -> r12
-        var i: usize = 0;
+        for (0..5) |i| {
-        while (i < 5) : (i += 1) {
+            self.bank.fiq[Bank.fiqIdx(i, now)] = self.r[8 + i];
-            self.banked_fiq[bankedFiqIdx(i, now)] = self.r[8 + i];
         }
 
         // Bank r13, r14, SPSR
         switch (now) {
             .User, .System => {
-                self.banked_r[bankedIdx(now, .R13)] = self.r[13];
+                self.bank.r[Bank.regIdx(now, .R13)] = self.r[13];
-                self.banked_r[bankedIdx(now, .R14)] = self.r[14];
+                self.bank.r[Bank.regIdx(now, .R14)] = self.r[14];
             },
             else => {
-                self.banked_r[bankedIdx(now, .R13)] = self.r[13];
+                self.bank.r[Bank.regIdx(now, .R13)] = self.r[13];
-                self.banked_r[bankedIdx(now, .R14)] = self.r[14];
+                self.bank.r[Bank.regIdx(now, .R14)] = self.r[14];
-                self.banked_spsr[bankedSpsrIndex(now)] = self.spsr;
+                self.bank.spsr[Bank.spsrIdx(now)] = self.spsr;
             },
         }
 
         // Grab R8 -> R12
-        i = 0;
+        for (0..5) |i| {
-        while (i < 5) : (i += 1) {
+            self.r[8 + i] = self.bank.fiq[Bank.fiqIdx(i, next)];
-            self.r[8 + i] = self.banked_fiq[bankedFiqIdx(i, next)];
         }
 
         // Grab r13, r14, SPSR
         switch (next) {
             .User, .System => {
-                self.r[13] = self.banked_r[bankedIdx(next, .R13)];
+                self.r[13] = self.bank.r[Bank.regIdx(next, .R13)];
-                self.r[14] = self.banked_r[bankedIdx(next, .R14)];
+                self.r[14] = self.bank.r[Bank.regIdx(next, .R14)];
             },
             else => {
-                self.r[13] = self.banked_r[bankedIdx(next, .R13)];
+                self.r[13] = self.bank.r[Bank.regIdx(next, .R13)];
-                self.r[14] = self.banked_r[bankedIdx(next, .R14)];
+                self.r[14] = self.bank.r[Bank.regIdx(next, .R14)];
-                self.spsr = self.banked_spsr[bankedSpsrIndex(next)];
+                self.spsr = self.bank.spsr[Bank.spsrIdx(next)];
             },
         }
 
@@ -424,8 +435,8 @@ pub const Arm7tdmi = struct {
         self.r[13] = 0x0300_7F00;
         self.r[15] = 0x0800_0000;
 
-        self.banked_r[bankedIdx(.Irq, .R13)] = 0x0300_7FA0;
+        self.bank.r[Bank.regIdx(.Irq, .R13)] = 0x0300_7FA0;
-        self.banked_r[bankedIdx(.Supervisor, .R13)] = 0x0300_7FE0;
+        self.bank.r[Bank.regIdx(.Supervisor, .R13)] = 0x0300_7FE0;
 
         // self.cpsr.raw = 0x6000001F;
         self.cpsr.raw = 0x0000_001F;
@@ -455,29 +466,11 @@ pub const Arm7tdmi = struct {
     }
 
     pub fn stepDmaTransfer(self: *Self) bool {
-        const dma0 = &self.bus.dma[0];
+        inline for (0..4) |i| {
-        const dma1 = &self.bus.dma[1];
+            if (self.bus.dma[i].in_progress) {
-        const dma2 = &self.bus.dma[2];
+                self.bus.dma[i].step(self);
-        const dma3 = &self.bus.dma[3];
+                return true;
-
+            }
-        if (dma0.in_progress) {
-            dma0.step(self);
-            return true;
-        }
-
-        if (dma1.in_progress) {
-            dma1.step(self);
-            return true;
-        }
-
-        if (dma2.in_progress) {
-            dma2.step(self);
-            return true;
-        }
-
-        if (dma3.in_progress) {
-            dma3.step(self);
-            return true;
         }
 
         return false;
@@ -532,10 +525,10 @@ pub const Arm7tdmi = struct {
             std.debug.print("R{}: 0x{X:0>8}\tR{}: 0x{X:0>8}\tR{}: 0x{X:0>8}\tR{}: 0x{X:0>8}\n", .{ i, self.r[i], i_1, self.r[i_1], i_2, self.r[i_2], i_3, self.r[i_3] });
         }
         std.debug.print("cpsr: 0x{X:0>8} ", .{self.cpsr.raw});
-        prettyPrintPsr(&self.cpsr);
+        self.cpsr.toString();
 
         std.debug.print("spsr: 0x{X:0>8} ", .{self.spsr.raw});
-        prettyPrintPsr(&self.spsr);
+        self.spsr.toString();
 
         std.debug.print("pipeline: {??X:0>8}\n", .{self.pipe.stage});
 
@@ -553,97 +546,31 @@ pub const Arm7tdmi = struct {
 
         std.debug.panic(format, args);
     }
-
-    fn prettyPrintPsr(psr: *const PSR) void {
-        std.debug.print("[", .{});
-
-        if (psr.n.read()) std.debug.print("N", .{}) else std.debug.print("-", .{});
-        if (psr.z.read()) std.debug.print("Z", .{}) else std.debug.print("-", .{});
-        if (psr.c.read()) std.debug.print("C", .{}) else std.debug.print("-", .{});
-        if (psr.v.read()) std.debug.print("V", .{}) else std.debug.print("-", .{});
-        if (psr.i.read()) std.debug.print("I", .{}) else std.debug.print("-", .{});
-        if (psr.f.read()) std.debug.print("F", .{}) else std.debug.print("-", .{});
-        if (psr.t.read()) std.debug.print("T", .{}) else std.debug.print("-", .{});
-        std.debug.print("|", .{});
-        if (getMode(psr.mode.read())) |mode| std.debug.print("{s}", .{modeString(mode)}) else std.debug.print("---", .{});
-
-        std.debug.print("]\n", .{});
-    }
-
-    fn modeString(mode: Mode) []const u8 {
-        return switch (mode) {
-            .User => "usr",
-            .Fiq => "fiq",
-            .Irq => "irq",
-            .Supervisor => "svc",
-            .Abort => "abt",
-            .Undefined => "und",
-            .System => "sys",
-        };
-    }
-
-    fn mgbaLog(self: *const Self, file: *const File, opcode: u32) !void {
-        const thumb_fmt = "{X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} cpsr: {X:0>8} | {X:0>4}:\n";
-        const arm_fmt = "{X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} {X:0>8} cpsr: {X:0>8} | {X:0>8}:\n";
-        var buf: [0x100]u8 = [_]u8{0x00} ** 0x100; // this is larger than it needs to be
-
-        const r0 = self.r[0];
-        const r1 = self.r[1];
-        const r2 = self.r[2];
-        const r3 = self.r[3];
-        const r4 = self.r[4];
-        const r5 = self.r[5];
-        const r6 = self.r[6];
-        const r7 = self.r[7];
-        const r8 = self.r[8];
-        const r9 = self.r[9];
-        const r10 = self.r[10];
-        const r11 = self.r[11];
-        const r12 = self.r[12];
-        const r13 = self.r[13];
-        const r14 = self.r[14];
-        const r15 = self.r[15] -| if (self.cpsr.t.read()) 2 else @as(u32, 4);
-
-        const c_psr = self.cpsr.raw;
-
-        var log_str: []u8 = undefined;
-        if (self.cpsr.t.read()) {
-            if (opcode >> 11 == 0x1E) {
-                // Instruction 1 of a BL Opcode, print in ARM mode
-                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 });
-            } else {
-                log_str = try std.fmt.bufPrint(&buf, thumb_fmt, .{ r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15, c_psr, opcode });
-            }
-        } else {
-            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, opcode });
-        }
-
-        _ = try file.writeAll(log_str);
-    }
 };
 
-pub fn checkCond(cpsr: PSR, cond: u4) bool {
+const condition_lut = [_]u16{
-    return switch (cond) {
+    0xF0F0, // EQ - Equal
-        0x0 => cpsr.z.read(), // EQ - Equal
+    0x0F0F, // NE - Not Equal
-        0x1 => !cpsr.z.read(), // NE - Not equal
+    0xCCCC, // CS - Unsigned higher or same
-        0x2 => cpsr.c.read(), // CS - Unsigned higher or same
+    0x3333, // CC - Unsigned lower
-        0x3 => !cpsr.c.read(), // CC - Unsigned lower
+    0xFF00, // MI - Negative
-        0x4 => cpsr.n.read(), // MI - Negative
+    0x00FF, // PL - Positive or Zero
-        0x5 => !cpsr.n.read(), // PL - Positive or zero
+    0xAAAA, // VS - Overflow
-        0x6 => cpsr.v.read(), // VS - Overflow
+    0x5555, // VC - No Overflow
-        0x7 => !cpsr.v.read(), // VC - No overflow
+    0x0C0C, // HI - unsigned hierh
-        0x8 => cpsr.c.read() and !cpsr.z.read(), // HI - unsigned higher
+    0xF3F3, // LS - unsigned lower or same
-        0x9 => !cpsr.c.read() or cpsr.z.read(), // LS - unsigned lower or same
+    0xAA55, // GE - greater or equal
-        0xA => cpsr.n.read() == cpsr.v.read(), // GE - Greater or equal
+    0x55AA, // LT - less than
-        0xB => cpsr.n.read() != cpsr.v.read(), // LT - Less than
+    0x0A05, // GT - greater than
-        0xC => !cpsr.z.read() and (cpsr.n.read() == cpsr.v.read()), // GT - Greater than
+    0xF5FA, // LE - less than or equal
-        0xD => cpsr.z.read() or (cpsr.n.read() != cpsr.v.read()), // LE - Less than or equal
+    0xFFFF, // AL - always
-        0xE => true, // AL - Always
+    0x0000, // NV - never
-        0xF => false, // NV - Never (reserved in ARMv3 and up, but seems to have not changed?)
+};
-    };
+
+pub inline fn checkCond(cpsr: PSR, cond: u4) bool {
+    const flags = @truncate(u4, cpsr.raw >> 28);
+
+    return condition_lut[cond] & (@as(u16, 1) << flags) != 0;
 }
 
 const Pipeline = struct {
@@ -665,9 +592,7 @@ const Pipeline = struct {
     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
+        const opcode = self.stage[0];
-        var opcode = self.stage[0];
-
         self.stage[0] = self.stage[1];
         self.stage[1] = cpu.fetch(T, cpu.r[15]);
 
@@ -699,6 +624,22 @@ pub const PSR = extern union {
     z: Bit(u32, 30),
     n: Bit(u32, 31),
     raw: u32,
+
+    fn toString(self: PSR) void {
+        std.debug.print("[", .{});
+
+        if (self.n.read()) std.debug.print("N", .{}) else std.debug.print("-", .{});
+        if (self.z.read()) std.debug.print("Z", .{}) else std.debug.print("-", .{});
+        if (self.c.read()) std.debug.print("C", .{}) else std.debug.print("-", .{});
+        if (self.v.read()) std.debug.print("V", .{}) else std.debug.print("-", .{});
+        if (self.i.read()) std.debug.print("I", .{}) else std.debug.print("-", .{});
+        if (self.f.read()) std.debug.print("F", .{}) else std.debug.print("-", .{});
+        if (self.t.read()) std.debug.print("T", .{}) else std.debug.print("-", .{});
+        std.debug.print("|", .{});
+        if (getMode(self.mode.read())) |m| std.debug.print("{s}", .{m.toString()}) else std.debug.print("---", .{});
+
+        std.debug.print("]\n", .{});
+    }
 };
 
 const Mode = enum(u5) {
@@ -709,11 +650,18 @@ const Mode = enum(u5) {
     Abort = 0b10111,
     Undefined = 0b11011,
     System = 0b11111,
-};
 
-const BankedKind = enum(u1) {
+    fn toString(self: Mode) []const u8 {
-    R13 = 0,
+        return switch (self) {
-    R14,
+            .User => "usr",
+            .Fiq => "fiq",
+            .Irq => "irq",
+            .Supervisor => "svc",
+            .Abort => "abt",
+            .Undefined => "und",
+            .System => "sys",
+        };
+    }
 };
 
 fn getMode(bits: u5) ?Mode {

src/core/cpu/arm/block_data_transfer.zig

@@ -57,7 +57,6 @@ pub fn blockDataTransfer(comptime P: bool, comptime U: bool, comptime S: bool, c
                     cpu.r[15] = bus.read(u32, und_addr);
                     cpu.pipe.reload(cpu);
                 } else {
-                    // FIXME: Should r15 on write be +12 ahead?
                     bus.write(u32, und_addr, cpu.r[15] + 4);
                 }
 

src/core/cpu/arm/data_processing.zig

@@ -24,7 +24,7 @@ pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime kind: u4) Ins
             if (!I and opcode >> 4 & 1 == 1) cpu.r[15] -= 4;
 
             var result: u32 = undefined;
-            var overflow: bool = undefined;
+            var overflow: u1 = undefined;
 
             // Perform Data Processing Logic
             switch (kind) {
@@ -62,7 +62,9 @@ pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime kind: u4) Ins
                     if (rd == 0xF)
                         return undefinedTestBehaviour(cpu);
 
-                    overflow = @addWithOverflow(u32, op1, op2, &result);
+                    const tmp = @addWithOverflow(op1, op2);
+                    result = tmp[0];
+                    overflow = tmp[1];
                 },
                 0xC => result = op1 | op2, // ORR
                 0xD => result = op2, // MOV
@@ -110,7 +112,7 @@ pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime kind: u4) Ins
                     // ADD, ADC Flags
                     cpu.cpsr.n.write(result >> 31 & 1 == 1);
                     cpu.cpsr.z.write(result == 0);
-                    cpu.cpsr.c.write(overflow);
+                    cpu.cpsr.c.write(overflow == 0b1);
                     cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                 },
                 0x6, 0x7 => if (S and rd != 0xF) {
@@ -141,7 +143,7 @@ pub fn dataProcessing(comptime I: bool, comptime S: bool, comptime kind: u4) Ins
                         cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
                     } else if (kind == 0xB) {
                         // CMN specific
-                        cpu.cpsr.c.write(overflow);
+                        cpu.cpsr.c.write(overflow == 0b1);
                         cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                     } else {
                         // TST, TEQ specific
@@ -162,19 +164,19 @@ pub fn sbc(left: u32, right: u32, old_carry: u1) u32 {
     return ret;
 }
 
-pub fn add(overflow: *bool, left: u32, right: u32) u32 {
+pub fn add(overflow: *u1, left: u32, right: u32) u32 {
-    var ret: u32 = undefined;
+    const ret = @addWithOverflow(left, right);
-    overflow.* = @addWithOverflow(u32, left, right, &ret);
+    overflow.* = ret[1];
-    return ret;
+
+    return ret[0];
 }
 
-pub fn adc(overflow: *bool, left: u32, right: u32, old_carry: u1) u32 {
+pub fn adc(overflow: *u1, left: u32, right: u32, old_carry: u1) u32 {
-    var ret: u32 = undefined;
+    const tmp = @addWithOverflow(left, right);
-    const first = @addWithOverflow(u32, left, right, &ret);
+    const ret = @addWithOverflow(tmp[0], old_carry);
-    const second = @addWithOverflow(u32, ret, old_carry, &ret);
+    overflow.* = tmp[1] | ret[1];
 
-    overflow.* = first or second;
+    return ret[0];
-    return ret;
 }
 
 fn undefinedTestBehaviour(cpu: *Arm7tdmi) void {

src/core/cpu/arm/single_data_transfer.zig

@@ -11,9 +11,7 @@ pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool,
             const rn = opcode >> 16 & 0xF;
             const rd = opcode >> 12 & 0xF;
 
-            // rn is r15 and L is not set, the PC is 12 ahead
+            const base = cpu.r[rn];
-            const base = cpu.r[rn] + if (!L and rn == 0xF) 4 else @as(u32, 0);
-
             const offset = if (I) shifter.immediate(false, cpu, opcode) else opcode & 0xFFF;
 
             const modified_base = if (U) base +% offset else base -% offset;

src/core/cpu/thumb/alu.zig

@@ -21,7 +21,8 @@ pub fn fmt4(comptime op: u4) InstrFn {
             const op2 = cpu.r[rs];
 
             var result: u32 = undefined;
-            var overflow: bool = undefined;
+            var overflow: u1 = undefined;
+
             switch (op) {
                 0x0 => result = op1 & op2, // AND
                 0x1 => result = op1 ^ op2, // EOR
@@ -34,7 +35,12 @@ pub fn fmt4(comptime op: u4) InstrFn {
                 0x8 => result = op1 & op2, // TST
                 0x9 => result = 0 -% op2, // NEG
                 0xA => result = op1 -% op2, // CMP
-                0xB => overflow = @addWithOverflow(u32, op1, op2, &result), // CMN
+                0xB => {
+                    // CMN
+                    const tmp = @addWithOverflow(op1, op2);
+                    result = tmp[0];
+                    overflow = tmp[1];
+                },
                 0xC => result = op1 | op2, // ORR
                 0xD => result = @truncate(u32, @as(u64, op2) * @as(u64, op1)),
                 0xE => result = op1 & ~op2,
@@ -71,7 +77,7 @@ pub fn fmt4(comptime op: u4) InstrFn {
                     // ADC, CMN
                     cpu.cpsr.n.write(result >> 31 & 1 == 1);
                     cpu.cpsr.z.write(result == 0);
-                    cpu.cpsr.c.write(overflow);
+                    cpu.cpsr.c.write(overflow == 0b1);
                     cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                 },
                 0x6 => {

src/core/cpu/thumb/block_data_transfer.zig

@@ -92,8 +92,7 @@ pub fn fmt15(comptime L: bool, comptime rb: u3) InstrFn {
 inline fn countRlist(opcode: u16) u32 {
     var count: u32 = 0;
 
-    comptime var i: u4 = 0;
+    inline for (0..8) |i| {
-    inline while (i < 8) : (i += 1) {
         if (opcode >> (7 - i) & 1 == 1) count += 1;
     }
 

src/core/cpu/thumb/data_processing.zig

@@ -64,7 +64,7 @@ pub fn fmt5(comptime op: u2, comptime h1: u1, comptime h2: u1) InstrFn {
             const op2 = cpu.r[rs];
 
             var result: u32 = undefined;
-            var overflow: bool = undefined;
+            var overflow: u1 = undefined;
             switch (op) {
                 0b00 => result = add(&overflow, op1, op2), // ADD
                 0b01 => result = op1 -% op2, // CMP
@@ -126,13 +126,13 @@ pub fn fmt2(comptime I: bool, is_sub: bool, rn: u3) InstrFn {
                 cpu.cpsr.v.write(((op1 ^ result) & (~op2 ^ result)) >> 31 & 1 == 1);
             } else {
                 // ADD
-                var overflow: bool = undefined;
+                var overflow: u1 = undefined;
                 const result = add(&overflow, op1, op2);
                 cpu.r[rd] = result;
 
                 cpu.cpsr.n.write(result >> 31 & 1 == 1);
                 cpu.cpsr.z.write(result == 0);
-                cpu.cpsr.c.write(overflow);
+                cpu.cpsr.c.write(overflow == 0b1);
                 cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
             }
         }
@@ -145,7 +145,7 @@ pub fn fmt3(comptime op: u2, comptime rd: u3) InstrFn {
             const op1 = cpu.r[rd];
             const op2: u32 = opcode & 0xFF; // Offset
 
-            var overflow: bool = undefined;
+            var overflow: u1 = undefined;
             const result: u32 = switch (op) {
                 0b00 => op2, // MOV
                 0b01 => op1 -% op2, // CMP
@@ -169,7 +169,7 @@ pub fn fmt3(comptime op: u2, comptime rd: u3) InstrFn {
                 },
                 0b10 => {
                     // ADD
-                    cpu.cpsr.c.write(overflow);
+                    cpu.cpsr.c.write(overflow == 0b1);
                     cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                 },
             }

src/core/emu.zig

@@ -56,7 +56,7 @@ fn inner(comptime kind: RunKind, audio_sync: bool, quit: *Atomic(bool), schedule
         .Unlimited, .UnlimitedFPS => {
             log.info("Emulation w/out video sync", .{});
 
-            while (!quit.load(.SeqCst)) {
+            while (!quit.load(.Monotonic)) {
                 runFrame(scheduler, cpu);
                 audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
 
@@ -68,7 +68,7 @@ fn inner(comptime kind: RunKind, audio_sync: bool, quit: *Atomic(bool), schedule
             var timer = Timer.start() catch @panic("failed to initalize std.timer.Timer");
             var wake_time: u64 = frame_period;
 
-            while (!quit.load(.SeqCst)) {
+            while (!quit.load(.Monotonic)) {
                 runFrame(scheduler, cpu);
                 const new_wake_time = videoSync(&timer, wake_time);
 
@@ -94,7 +94,7 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
         if (!cpu.stepDmaTransfer()) {
             if (cpu.isHalted()) {
                 // Fast-forward to next Event
-                sched.tick = sched.queue.peek().?.tick;
+                sched.tick = sched.nextTimestamp();
             } else {
                 cpu.step();
             }
@@ -105,6 +105,7 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
 }
 
 fn audioSync(audio_sync: bool, stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
+    comptime std.debug.assert(@import("../platform.zig").sample_format == SDL.AUDIO_U16);
     const sample_size = 2 * @sizeOf(u16);
     const max_buf_size: c_int = 0x400;
 
@@ -145,9 +146,8 @@ fn sleep(timer: *Timer, wake_time: u64) ?u64 {
 
     const step = 2 * std.time.ns_per_ms; // Granularity of 2ms
     const times = sleep_for / step;
-    var i: usize = 0;
 
-    while (i < times) : (i += 1) {
+    for (0..times) |_| {
         std.time.sleep(step);
 
         // Upon wakeup, check to see if this particular sleep was longer than expected
@@ -162,3 +162,59 @@ fn sleep(timer: *Timer, wake_time: u64) ?u64 {
 fn spinLoop(timer: *Timer, wake_time: u64) void {
     while (true) if (timer.read() > wake_time) break;
 }
+
+pub const EmuThing = struct {
+    const Self = @This();
+    const Interface = @import("gdbstub").Emulator;
+    const Allocator = std.mem.Allocator;
+
+    cpu: *Arm7tdmi,
+    scheduler: *Scheduler,
+
+    pub fn init(cpu: *Arm7tdmi, scheduler: *Scheduler) Self {
+        return .{ .cpu = cpu, .scheduler = scheduler };
+    }
+
+    pub fn interface(self: *Self, allocator: Allocator) Interface {
+        return Interface.init(allocator, self);
+    }
+
+    pub fn read(self: *const Self, addr: u32) u8 {
+        return self.cpu.bus.dbgRead(u8, addr);
+    }
+
+    pub fn write(self: *Self, addr: u32, value: u8) void {
+        self.cpu.bus.dbgWrite(u8, addr, value);
+    }
+
+    pub fn registers(self: *const Self) *[16]u32 {
+        return &self.cpu.r;
+    }
+
+    pub fn cpsr(self: *const Self) u32 {
+        return self.cpu.cpsr.raw;
+    }
+
+    pub fn step(self: *Self) void {
+        const cpu = self.cpu;
+        const sched = self.scheduler;
+
+        // Is true when we have executed one (1) instruction
+        var did_step: bool = false;
+
+        // TODO: How can I make it easier to keep this in lock-step with runFrame?
+        while (!did_step) {
+            if (!cpu.stepDmaTransfer()) {
+                if (cpu.isHalted()) {
+                    // Fast-forward to next Event
+                    sched.tick = sched.queue.peek().?.tick;
+                } else {
+                    cpu.step();
+                    did_step = true;
+                }
+            }
+
+            if (sched.tick >= sched.nextTimestamp()) sched.handleEvent(cpu);
+        }
+    }
+};

src/core/ppu/Oam.zig

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

src/core/ppu/Palette.zig

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

src/core/ppu/Vram.zig

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

src/core/scheduler.zig

@@ -31,64 +31,55 @@ pub const Scheduler = struct {
     }
 
     pub fn handleEvent(self: *Self, cpu: *Arm7tdmi) void {
-        if (self.queue.removeOrNull()) |event| {
+        const event = self.queue.remove();
-            const late = self.tick - event.tick;
+        const late = self.tick - event.tick;
 
-            switch (event.kind) {
+        switch (event.kind) {
-                .HeatDeath => {
+            .HeatDeath => {
-                    log.err("u64 overflow. This *actually* should never happen.", .{});
+                log.err("u64 overflow. This *actually* should never happen.", .{});
-                    unreachable;
+                unreachable;
-                },
+            },
-                .Draw => {
+            .Draw => {
-                    // The end of a VDraw
+                // The end of a VDraw
-                    cpu.bus.ppu.drawScanline();
+                cpu.bus.ppu.drawScanline();
-                    cpu.bus.ppu.onHdrawEnd(cpu, late);
+                cpu.bus.ppu.onHdrawEnd(cpu, late);
-                },
+            },
-                .TimerOverflow => |id| {
+            .TimerOverflow => |id| {
-                    switch (id) {
+                switch (id) {
-                        0 => cpu.bus.tim[0].onTimerExpire(cpu, late),
+                    inline 0...3 => |idx| cpu.bus.tim[idx].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.onToneSweepEvent(late),
-                .ApuChannel => |id| {
+                    1 => cpu.bus.apu.ch2.onToneEvent(late),
-                    switch (id) {
+                    2 => cpu.bus.apu.ch3.onWaveEvent(late),
-                        0 => cpu.bus.apu.ch1.onToneSweepEvent(late),
+                    3 => cpu.bus.apu.ch4.onNoiseEvent(late),
-                        1 => cpu.bus.apu.ch2.onToneEvent(late),
+                }
-                        2 => cpu.bus.apu.ch3.onWaveEvent(late),
+            },
-                        3 => cpu.bus.apu.ch4.onNoiseEvent(late),
+            .RealTimeClock => {
-                    }
+                const device = &cpu.bus.pak.gpio.device;
-                },
+                if (device.kind != .Rtc or device.ptr == null) return;
-                .RealTimeClock => {
-                    const device = &cpu.bus.pak.gpio.device;
-                    if (device.kind != .Rtc or device.ptr == null) return;
 
-                    const clock = @ptrCast(*Clock, @alignCast(@alignOf(*Clock), device.ptr.?));
+                const clock = @ptrCast(*Clock, @alignCast(@alignOf(*Clock), device.ptr.?));
-                    clock.onClockUpdate(late);
+                clock.onClockUpdate(late);
-                },
+            },
-                .FrameSequencer => cpu.bus.apu.onSequencerTick(late),
+            .FrameSequencer => cpu.bus.apu.onSequencerTick(late),
-                .SampleAudio => cpu.bus.apu.sampleAudio(late),
+            .SampleAudio => cpu.bus.apu.sampleAudio(late),
-                .HBlank => cpu.bus.ppu.onHblankEnd(cpu, late), // The end of a HBlank
+            .HBlank => cpu.bus.ppu.onHblankEnd(cpu, late), // The end of a HBlank
-                .VBlank => cpu.bus.ppu.onHdrawEnd(cpu, late), // The end of a VBlank
+            .VBlank => cpu.bus.ppu.onHdrawEnd(cpu, late), // The end of a VBlank
-            }
         }
     }
 
     /// Removes the **first** scheduled event of type `needle`
     pub fn removeScheduledEvent(self: *Self, needle: EventKind) void {
-        var it = self.queue.iterator();
+        for (self.queue.items, 0..) |event, i| {
-
-        var i: usize = 0;
-        while (it.next()) |event| : (i += 1) {
             if (std.meta.eql(event.kind, needle)) {
 
-                // This invalidates the iterator
+                // invalidates the slice we're iterating over
                 _ = self.queue.removeIndex(i);
 
-                // Since removing something from the PQ invalidates the iterator,
+                log.debug("Removed {?}@{}", .{ event.kind, event.tick });
-                // this implementation can safely only remove the first instance of
-                // a Scheduled Event. Exit Early
                 break;
             }
         }

src/main.zig

@@ -4,14 +4,17 @@ const known_folders = @import("known_folders");
 const clap = @import("clap");
 
 const config = @import("config.zig");
+const emu = @import("core/emu.zig");
 
 const Gui = @import("platform.zig").Gui;
 const Bus = @import("core/Bus.zig");
 const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
 const Scheduler = @import("core/scheduler.zig").Scheduler;
 const FilePaths = @import("util.zig").FilePaths;
-
+const FpsTracker = @import("util.zig").FpsTracker;
 const Allocator = std.mem.Allocator;
+const Atomic = std.atomic.Atomic;
+
 const log = std.log.scoped(.Cli);
 const width = @import("core/ppu.zig").width;
 const height = @import("core/ppu.zig").height;
@@ -22,37 +25,49 @@ const params = clap.parseParamsComptime(
     \\-h, --help            Display this help and exit.
     \\-s, --skip            Skip BIOS.
     \\-b, --bios <str>      Optional path to a GBA BIOS ROM.
+    \\ --gdb                Run ZBA from the context of a GDB Server
     \\<str>                 Path to the GBA GamePak ROM.
     \\
 );
 
-pub fn main() anyerror!void {
+pub fn main() void {
     // Main Allocator for ZBA
     var gpa = std.heap.GeneralPurposeAllocator(.{}){};
     defer std.debug.assert(!gpa.deinit());
 
     const allocator = gpa.allocator();
 
-    // Determine the Data Directory (stores saves, config file, etc.)
+    // Determine the Data Directory (stores saves)
     const data_path = blk: {
         const result = known_folders.getPath(allocator, .data);
-        const option = result catch |e| exitln("interrupted while attempting to find a data directory: {}", .{e});
+        const option = result catch |e| exitln("interrupted while determining the data folder: {}", .{e});
-        const path = option orelse exitln("no valid data directory could be found", .{});
+        const path = option orelse exitln("no valid data folder found", .{});
-        ensureDirectoriesExist(path) catch |e| exitln("failed to create directories under \"{s}\": {}", .{ path, e });
+        ensureDataDirsExist(path) catch |e| exitln("failed to create folders under \"{s}\": {}", .{ path, e });
 
         break :blk path;
     };
     defer allocator.free(data_path);
 
+    // Determine the Config Directory
+    const config_path = blk: {
+        const result = known_folders.getPath(allocator, .roaming_configuration);
+        const option = result catch |e| exitln("interreupted while determining the config folder: {}", .{e});
+        const path = option orelse exitln("no valid config folder found", .{});
+        ensureConfigDirExists(path) catch |e| exitln("failed to create required folder \"{s}\": {}", .{ path, e });
+
+        break :blk path;
+    };
+    defer allocator.free(config_path);
+
     // Parse CLI
     const result = clap.parse(clap.Help, &params, clap.parsers.default, .{}) catch |e| exitln("failed to parse cli: {}", .{e});
     defer result.deinit();
 
     // TODO: Move config file to XDG Config directory?
-    const config_path = configFilePath(allocator, data_path) catch |e| exitln("failed to determine the config file path for ZBA: {}", .{e});
+    const cfg_file_path = configFilePath(allocator, config_path) catch |e| exitln("failed to ready config file for access: {}", .{e});
-    defer allocator.free(config_path);
+    defer allocator.free(cfg_file_path);
 
-    config.load(allocator, config_path) catch |e| exitln("failed to read config file: {}", .{e});
+    config.load(allocator, cfg_file_path) catch |e| exitln("failed to load config file: {}", .{e});
 
     const paths = handleArguments(allocator, data_path, &result) catch |e| exitln("failed to handle cli arguments: {}", .{e});
     defer if (paths.save) |path| allocator.free(path);
@@ -76,13 +91,49 @@ pub fn main() anyerror!void {
         cpu.fastBoot();
     }
 
-    var gui = Gui.init(&bus.pak.title, &bus.apu, width, height);
+    var quit = Atomic(bool).init(false);
+    var gui = Gui.init(&bus.pak.title, &bus.apu, width, height) catch |e| exitln("failed to init gui: {}", .{e});
     defer gui.deinit();
 
-    gui.run(&cpu, &scheduler) catch |e| exitln("failed to run gui thread: {}", .{e});
+    if (result.args.gdb) {
+        const Server = @import("gdbstub").Server;
+        const EmuThing = @import("core/emu.zig").EmuThing;
+
+        var wrapper = EmuThing.init(&cpu, &scheduler);
+        var emulator = wrapper.interface(allocator);
+        defer emulator.deinit();
+
+        log.info("Ready to connect", .{});
+
+        var server = Server.init(emulator) catch |e| exitln("failed to init gdb server: {}", .{e});
+        defer server.deinit(allocator);
+
+        log.info("Starting GDB Server Thread", .{});
+
+        const thread = std.Thread.spawn(.{}, Server.run, .{ &server, allocator, &quit }) catch |e| exitln("gdb server thread crashed: {}", .{e});
+        defer thread.join();
+
+        gui.run(.{
+            .cpu = &cpu,
+            .scheduler = &scheduler,
+            .quit = &quit,
+        }) catch |e| exitln("main thread panicked: {}", .{e});
+    } else {
+        var tracker = FpsTracker.init();
+
+        const thread = std.Thread.spawn(.{}, emu.run, .{ &quit, &scheduler, &cpu, &tracker }) catch |e| exitln("emu thread panicked: {}", .{e});
+        defer thread.join();
+
+        gui.run(.{
+            .cpu = &cpu,
+            .scheduler = &scheduler,
+            .tracker = &tracker,
+            .quit = &quit,
+        }) catch |e| exitln("main thread panicked: {}", .{e});
+    }
 }
 
-pub fn handleArguments(allocator: Allocator, data_path: []const u8, result: *const clap.Result(clap.Help, &params, clap.parsers.default)) !FilePaths {
+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});
 
@@ -99,8 +150,8 @@ pub fn handleArguments(allocator: Allocator, data_path: []const u8, result: *con
     };
 }
 
-fn configFilePath(allocator: Allocator, data_path: []const u8) ![]const u8 {
+fn configFilePath(allocator: Allocator, config_path: []const u8) ![]const u8 {
-    const path = try std.fs.path.join(allocator, &[_][]const u8{ data_path, "zba", "config.toml" });
+    const path = try std.fs.path.join(allocator, &[_][]const u8{ config_path, "zba", "config.toml" });
     errdefer allocator.free(path);
 
     // We try to create the file exclusively, meaning that we err out if the file already exists.
@@ -109,7 +160,7 @@ fn configFilePath(allocator: Allocator, data_path: []const u8) ![]const u8 {
     std.fs.accessAbsolute(path, .{}) catch |e| {
         if (e != error.FileNotFound) return e;
 
-        const config_file = try std.fs.createFileAbsolute(path, .{});
+        const config_file = std.fs.createFileAbsolute(path, .{}) catch |err| exitln("failed to create \"{s}\": {}", .{ path, err });
         defer config_file.close();
 
         try config_file.writeAll(@embedFile("../example.toml"));
@@ -118,17 +169,21 @@ fn configFilePath(allocator: Allocator, data_path: []const u8) ![]const u8 {
     return path;
 }
 
-fn ensureDirectoriesExist(data_path: []const u8) !void {
+fn ensureDataDirsExist(data_path: []const u8) !void {
     var dir = try std.fs.openDirAbsolute(data_path, .{});
     defer dir.close();
 
-    // We want to make sure: %APPDATA%/zba and %APPDATA%/zba/save exist
-    // (~/.local/share/zba/save for linux, ??? for macOS)
-
     // Will recursively create directories
     try dir.makePath("zba" ++ std.fs.path.sep_str ++ "save");
 }
 
+fn ensureConfigDirExists(config_path: []const u8) !void {
+    var dir = try std.fs.openDirAbsolute(config_path, .{});
+    defer dir.close();
+
+    try dir.makePath("zba");
+}
+
 fn romPath(result: *const clap.Result(clap.Help, &params, clap.parsers.default)) []const u8 {
     return switch (result.positionals.len) {
         1 => result.positionals[0],

src/platform.zig

@@ -9,15 +9,13 @@ const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
 const Scheduler = @import("core/scheduler.zig").Scheduler;
 const FpsTracker = @import("util.zig").FpsTracker;
 
-const span = @import("util.zig").span;
-
 const gba_width = @import("core/ppu.zig").width;
 const gba_height = @import("core/ppu.zig").height;
 
-pub const sample_rate = 44100;
+pub const sample_rate = 1 << 15;
-pub const sample_format = SDL.AUDIO_F32;
+pub const sample_format = SDL.AUDIO_U16;
 
-const default_title: []const u8 = "ZBA";
+const default_title = "ZBA";
 
 pub const Gui = struct {
     const Self = @This();
@@ -46,7 +44,7 @@ pub const Gui = struct {
 
     program_id: gl.GLuint,
 
-    pub fn init(title: *const [12]u8, apu: *Apu, width: i32, height: i32) Self {
+    pub fn init(title: *const [12]u8, apu: *Apu, width: i32, height: i32) !Self {
         if (SDL.SDL_Init(SDL.SDL_INIT_VIDEO | SDL.SDL_INIT_EVENTS | SDL.SDL_INIT_AUDIO) < 0) panic();
         if (SDL.SDL_GL_SetAttribute(SDL.SDL_GL_CONTEXT_PROFILE_MASK, SDL.SDL_GL_CONTEXT_PROFILE_CORE) < 0) panic();
         if (SDL.SDL_GL_SetAttribute(SDL.SDL_GL_CONTEXT_MAJOR_VERSION, 3) < 0) panic();
@@ -55,7 +53,7 @@ pub const Gui = struct {
         const win_scale = @intCast(c_int, config.config().host.win_scale);
 
         const window = SDL.SDL_CreateWindow(
-            default_title.ptr,
+            default_title,
             SDL.SDL_WINDOWPOS_CENTERED,
             SDL.SDL_WINDOWPOS_CENTERED,
             @as(c_int, width * win_scale),
@@ -66,21 +64,21 @@ pub const Gui = struct {
         const ctx = SDL.SDL_GL_CreateContext(window) orelse panic();
         if (SDL.SDL_GL_MakeCurrent(window, ctx) < 0) panic();
 
-        gl.load(ctx, Self.glGetProcAddress) catch @panic("gl.load failed");
+        gl.load(ctx, Self.glGetProcAddress) catch {};
         if (SDL.SDL_GL_SetSwapInterval(@boolToInt(config.config().host.vsync)) < 0) panic();
 
-        const program_id = compileShaders();
+        const program_id = try compileShaders();
 
         return Self{
             .window = window,
-            .title = span(title),
+            .title = std.mem.sliceTo(title, 0),
             .ctx = ctx,
             .program_id = program_id,
             .audio = Audio.init(apu),
         };
     }
 
-    fn compileShaders() gl.GLuint {
+    fn compileShaders() !gl.GLuint {
         // TODO: Panic on Shader Compiler Failure + Error Message
         const vert_shader = @embedFile("shader/pixelbuf.vert");
         const frag_shader = @embedFile("shader/pixelbuf.frag");
@@ -91,12 +89,16 @@ pub const Gui = struct {
         gl.shaderSource(vs, 1, &[_][*c]const u8{vert_shader}, 0);
         gl.compileShader(vs);
 
+        if (!shader.didCompile(vs)) return error.VertexCompileError;
+
         const fs = gl.createShader(gl.FRAGMENT_SHADER);
         defer gl.deleteShader(fs);
 
         gl.shaderSource(fs, 1, &[_][*c]const u8{frag_shader}, 0);
         gl.compileShader(fs);
 
+        if (!shader.didCompile(fs)) return error.FragmentCompileError;
+
         const program = gl.createProgram();
         gl.attachShader(program, vs);
         gl.attachShader(program, fs);
@@ -106,7 +108,7 @@ pub const Gui = struct {
     }
 
     // Returns the VAO ID since it's used in run()
-    fn generateBuffers() [3]c_uint {
+    fn generateBuffers() struct { c_uint, c_uint, c_uint } {
         var vao_id: c_uint = undefined;
         var vbo_id: c_uint = undefined;
         var ebo_id: c_uint = undefined;
@@ -152,69 +154,86 @@ pub const Gui = struct {
         return tex_id;
     }
 
-    pub fn run(self: *Self, cpu: *Arm7tdmi, scheduler: *Scheduler) !void {
+    const RunOptions = struct {
-        var quit = std.atomic.Atomic(bool).init(false);
+        quit: *std.atomic.Atomic(bool),
-        var tracker = FpsTracker.init();
+        tracker: ?*FpsTracker = null,
+        cpu: *Arm7tdmi,
+        scheduler: *Scheduler,
+    };
 
-        const thread = try std.Thread.spawn(.{}, emu.run, .{ &quit, scheduler, cpu, &tracker });
+    pub fn run(self: *Self, opt: RunOptions) !void {
-        defer thread.join();
+        const cpu = opt.cpu;
+        const tracker = opt.tracker;
+        const quit = opt.quit;
 
-        var title_buf: [0x100]u8 = [_]u8{0} ** 0x100;
+        var buffer_ids = Self.generateBuffers();
+        defer {
+            gl.deleteBuffers(1, &buffer_ids[2]); // EBO
+            gl.deleteBuffers(1, &buffer_ids[1]); // VBO
+            gl.deleteVertexArrays(1, &buffer_ids[0]); // VAO
+        }
+        const vao_id = buffer_ids[0];
 
-        const vao_id = Self.generateBuffers()[0];
+        const tex_id = Self.generateTexture(cpu.bus.ppu.framebuf.get(.Renderer));
-        _ = Self.generateTexture(cpu.bus.ppu.framebuf.get(.Renderer));
+        defer gl.deleteTextures(1, &tex_id);
+
+        var title_buf: [0x100]u8 = undefined;
 
         emu_loop: while (true) {
             var event: SDL.SDL_Event = undefined;
+
+            // This might be true if the emu is running via a gdbstub server
+            // and the gdb stub exits first
+            if (quit.load(.Monotonic)) break :emu_loop;
+
             while (SDL.SDL_PollEvent(&event) != 0) {
                 switch (event.type) {
                     SDL.SDL_QUIT => break :emu_loop,
                     SDL.SDL_KEYDOWN => {
-                        const io = &cpu.bus.io;
                         const key_code = event.key.keysym.sym;
+                        var keyinput = cpu.bus.io.keyinput.load(.Monotonic);
 
                         switch (key_code) {
-                            SDL.SDLK_UP => io.keyinput.up.unset(),
+                            SDL.SDLK_UP => keyinput.up.unset(),
-                            SDL.SDLK_DOWN => io.keyinput.down.unset(),
+                            SDL.SDLK_DOWN => keyinput.down.unset(),
-                            SDL.SDLK_LEFT => io.keyinput.left.unset(),
+                            SDL.SDLK_LEFT => keyinput.left.unset(),
-                            SDL.SDLK_RIGHT => io.keyinput.right.unset(),
+                            SDL.SDLK_RIGHT => keyinput.right.unset(),
-                            SDL.SDLK_x => io.keyinput.a.unset(),
+                            SDL.SDLK_x => keyinput.a.unset(),
-                            SDL.SDLK_z => io.keyinput.b.unset(),
+                            SDL.SDLK_z => keyinput.b.unset(),
-                            SDL.SDLK_a => io.keyinput.shoulder_l.unset(),
+                            SDL.SDLK_a => keyinput.shoulder_l.unset(),
-                            SDL.SDLK_s => io.keyinput.shoulder_r.unset(),
+                            SDL.SDLK_s => keyinput.shoulder_r.unset(),
-                            SDL.SDLK_RETURN => io.keyinput.start.unset(),
+                            SDL.SDLK_RETURN => keyinput.start.unset(),
-                            SDL.SDLK_RSHIFT => io.keyinput.select.unset(),
+                            SDL.SDLK_RSHIFT => keyinput.select.unset(),
                             else => {},
                         }
+
+                        cpu.bus.io.keyinput.store(keyinput.raw, .Monotonic);
                     },
                     SDL.SDL_KEYUP => {
-                        const io = &cpu.bus.io;
                         const key_code = event.key.keysym.sym;
+                        var keyinput = cpu.bus.io.keyinput.load(.Monotonic);
 
                         switch (key_code) {
-                            SDL.SDLK_UP => io.keyinput.up.set(),
+                            SDL.SDLK_UP => keyinput.up.set(),
-                            SDL.SDLK_DOWN => io.keyinput.down.set(),
+                            SDL.SDLK_DOWN => keyinput.down.set(),
-                            SDL.SDLK_LEFT => io.keyinput.left.set(),
+                            SDL.SDLK_LEFT => keyinput.left.set(),
-                            SDL.SDLK_RIGHT => io.keyinput.right.set(),
+                            SDL.SDLK_RIGHT => keyinput.right.set(),
-                            SDL.SDLK_x => io.keyinput.a.set(),
+                            SDL.SDLK_x => keyinput.a.set(),
-                            SDL.SDLK_z => io.keyinput.b.set(),
+                            SDL.SDLK_z => keyinput.b.set(),
-                            SDL.SDLK_a => io.keyinput.shoulder_l.set(),
+                            SDL.SDLK_a => keyinput.shoulder_l.set(),
-                            SDL.SDLK_s => io.keyinput.shoulder_r.set(),
+                            SDL.SDLK_s => keyinput.shoulder_r.set(),
-                            SDL.SDLK_RETURN => io.keyinput.start.set(),
+                            SDL.SDLK_RETURN => keyinput.start.set(),
-                            SDL.SDLK_RSHIFT => io.keyinput.select.set(),
+                            SDL.SDLK_RSHIFT => keyinput.select.set(),
-                            SDL.SDLK_i => log.err("Sample Count: {}", .{@intCast(u32, SDL.SDL_AudioStreamAvailable(cpu.bus.apu.stream)) / (2 * @sizeOf(u16))}),
+                            SDL.SDLK_i => {
-                            SDL.SDLK_j => log.err("Scheduler Capacity: {} | Scheduler Event Count: {}", .{ scheduler.queue.capacity(), scheduler.queue.count() }),
+                                comptime std.debug.assert(sample_format == SDL.AUDIO_U16);
-                            SDL.SDLK_k => {
+                                log.err("Sample Count: {}", .{@intCast(u32, SDL.SDL_AudioStreamAvailable(cpu.bus.apu.stream)) / (2 * @sizeOf(u16))});
-                                // Dump IWRAM to file
-                                log.info("PC: 0x{X:0>8}", .{cpu.r[15]});
-                                log.info("LR: 0x{X:0>8}", .{cpu.r[14]});
-                                // const iwram_file = try std.fs.cwd().createFile("iwram.bin", .{});
-                                // defer iwram_file.close();
-
-                                // try iwram_file.writeAll(cpu.bus.iwram.buf);
                             },
+                            // SDL.SDLK_j => log.err("Scheduler Capacity: {} | Scheduler Event Count: {}", .{ scheduler.queue.capacity(), scheduler.queue.count() }),
+                            SDL.SDLK_k => {},
                             else => {},
                         }
+
+                        cpu.bus.io.keyinput.store(keyinput.raw, .Monotonic);
                     },
                     else => {},
                 }
@@ -229,16 +248,17 @@ pub const Gui = struct {
             gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_INT, null);
             SDL.SDL_GL_SwapWindow(self.window);
 
-            const dyn_title = std.fmt.bufPrint(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, tracker.value() }) catch unreachable;
+            if (tracker) |t| {
-            SDL.SDL_SetWindowTitle(self.window, dyn_title.ptr);
+                const dyn_title = std.fmt.bufPrintZ(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, t.value() }) catch unreachable;
+                SDL.SDL_SetWindowTitle(self.window, dyn_title.ptr);
+            }
         }
 
-        quit.store(true, .SeqCst); // Terminate Emulator Thread
+        quit.store(true, .Monotonic); // Terminate Emulator Thread
     }
 
     pub fn deinit(self: *Self) void {
         self.audio.deinit();
-        // TODO: Buffer deletions
         gl.deleteProgram(self.program_id);
         SDL.SDL_GL_DeleteContext(self.ctx);
         SDL.SDL_DestroyWindow(self.window);
@@ -268,8 +288,8 @@ const Audio = struct {
         want.callback = Self.callback;
         want.userdata = apu;
 
-        std.debug.assert(sample_format == SDL.AUDIO_F32);
+        std.debug.assert(sample_format == SDL.AUDIO_U16);
-        log.info("Host Sample Rate: {}Hz, Host Format: SDL.AUDIO_F32", .{sample_rate});
+        log.info("Host Sample Rate: {}Hz, Host Format: SDL.AUDIO_U16", .{sample_rate});
 
         const device = SDL.SDL_OpenAudioDevice(null, 0, &want, &have, 0);
         if (device == 0) panic();
@@ -295,6 +315,28 @@ const Audio = struct {
     }
 };
 
+const shader = struct {
+    const Kind = enum { vertex, fragment };
+    const log = std.log.scoped(.Shader);
+
+    fn didCompile(id: gl.GLuint) bool {
+        var success: gl.GLint = undefined;
+        gl.getShaderiv(id, gl.COMPILE_STATUS, &success);
+
+        if (success == 0) err(id);
+
+        return success == 1;
+    }
+
+    fn err(id: gl.GLuint) void {
+        const buf_len = 512;
+        var error_msg: [buf_len]u8 = undefined;
+
+        gl.getShaderInfoLog(id, buf_len, 0, &error_msg);
+        log.err("{s}", .{std.mem.sliceTo(&error_msg, 0)});
+    }
+};
+
 fn panic() noreturn {
     const str = @as(?[*:0]const u8, SDL.SDL_GetError()) orelse "unknown error";
     @panic(std.mem.sliceTo(str, 0));

src/util.zig

@@ -5,6 +5,8 @@ const config = @import("config.zig");
 const Log2Int = std.math.Log2Int;
 const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
 
+const Allocator = std.mem.Allocator;
+
 // 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
@@ -47,7 +49,7 @@ pub const FpsTracker = struct {
 
     pub fn value(self: *Self) u32 {
         if (self.timer.read() >= std.time.ns_per_s) {
-            self.fps = self.count.swap(0, .SeqCst);
+            self.fps = self.count.swap(0, .Monotonic);
             self.timer.reset();
         }
 
@@ -66,57 +68,6 @@ pub fn intToBytes(comptime T: type, value: anytype) [@sizeOf(T)]u8 {
     return result;
 }
 
-/// The Title from the GBA Cartridge is an Uppercase ASCII string which is
-/// null-padded to 12 bytes
-///
-/// This function returns a slice of the ASCII string without the null terminator(s)
-/// (essentially, a proper Zig/Rust/Any modern language String)
-pub fn span(title: *const [12]u8) []const u8 {
-    const end = std.mem.indexOfScalar(u8, title, '\x00');
-    return title[0 .. end orelse title.len];
-}
-
-test "span" {
-    var example: *const [12]u8 = "POKEMON_EMER";
-    try std.testing.expectEqualSlices(u8, "POKEMON_EMER", span(example));
-
-    example = "POKEMON_EME\x00";
-    try std.testing.expectEqualSlices(u8, "POKEMON_EME", span(example));
-
-    example = "POKEMON_EM\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKEMON_EM", span(example));
-
-    example = "POKEMON_E\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKEMON_E", span(example));
-
-    example = "POKEMON_\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKEMON_", span(example));
-
-    example = "POKEMON\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKEMON", span(example));
-
-    example = "POKEMO\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKEMO", span(example));
-
-    example = "POKEM\x00\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKEM", span(example));
-
-    example = "POKE\x00\x00\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POKE", span(example));
-
-    example = "POK\x00\x00\x00\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "POK", span(example));
-
-    example = "PO\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "PO", span(example));
-
-    example = "P\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "P", span(example));
-
-    example = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00";
-    try std.testing.expectEqualSlices(u8, "", span(example));
-}
-
 /// Creates a copy of a title with all Filesystem-invalid characters replaced
 ///
 /// e.g. POKEPIN R/S to POKEPIN R_S
@@ -174,6 +125,7 @@ pub const io = struct {
 
 pub const Logger = struct {
     const Self = @This();
+    const FmtArgTuple = std.meta.Tuple(&.{ u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32 });
 
     buf: std.io.BufferedWriter(4096 << 2, std.fs.File.Writer),
 
@@ -196,7 +148,7 @@ pub const Logger = struct {
         if (cpu.cpsr.t.read()) {
             if (opcode >> 11 == 0x1E) {
                 // Instruction 1 of a BL Opcode, print in ARM mode
-                const low = cpu.bus.dbgRead(u16, cpu.r[15]);
+                const low = cpu.bus.dbgRead(u16, cpu.r[15] - 2);
                 const bl_opcode = @as(u32, opcode) << 16 | low;
 
                 self.print(arm_fmt, Self.fmtArgs(cpu, bl_opcode)) catch @panic("failed to write to log file");
@@ -232,8 +184,6 @@ 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");
 
@@ -302,8 +252,6 @@ pub inline fn getHalf(byte: u8) u4 {
     return @truncate(u4, byte & 1) << 3;
 }
 
-// TODO: Maybe combine SetLo and SetHi, use addr alignment to deduplicate code
-
 pub inline fn setHalf(comptime T: type, left: T, addr: u8, right: HalfInt(T)) T {
     const offset = @truncate(u1, addr >> if (T == u32) 1 else 0);
 
@@ -314,32 +262,12 @@ pub inline fn setHalf(comptime T: type, left: T, addr: u8, right: HalfInt(T)) T
         },
         u16 => switch (offset) {
             0b0 => (left & 0xFF00) | right,
-            0b1 => (right & 0x00FF) | @as(u16, right) << 8,
+            0b1 => (left & 0x00FF) | @as(u16, right) << 8,
         },
         else => @compileError("unsupported type"),
     };
 }
 
-/// Sets the high 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 & 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 setHi(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);
@@ -348,3 +276,44 @@ fn HalfInt(comptime T: type) type {
 
     return std.meta.Int(type_info.Int.signedness, type_info.Int.bits >> 1);
 }
+
+/// Double Buffering Implementation
+pub const FrameBuffer = struct {
+    const Self = @This();
+
+    layers: [2][]u8,
+    buf: []u8,
+    current: u1,
+
+    allocator: Allocator,
+
+    // TODO: Rename
+    const Device = enum { Emulator, Renderer };
+
+    pub fn init(allocator: Allocator, comptime len: comptime_int) !Self {
+        const buf = try allocator.alloc(u8, len * 2);
+        std.mem.set(u8, buf, 0);
+
+        return .{
+            // Front and Back Framebuffers
+            .layers = [_][]u8{ buf[0..][0..len], buf[len..][0..len] },
+            .buf = buf,
+            .current = 0,
+
+            .allocator = allocator,
+        };
+    }
+
+    pub fn deinit(self: *Self) void {
+        self.allocator.free(self.buf);
+        self.* = undefined;
+    }
+
+    pub fn swap(self: *Self) void {
+        self.current = ~self.current;
+    }
+
+    pub fn get(self: *Self, comptime dev: Device) []u8 {
+        return self.layers[if (dev == .Emulator) self.current else ~self.current];
+    }
+};