chore: add test for RingBuffer

.github/workflows/main.yml

@@ -14,8 +14,7 @@ jobs:
   build:
     strategy:
       matrix:
-        # os: [ubuntu-latest, windows-latest, macos-latest]
-        os: [ubuntu-latest, windows-latest]
+        os: [ubuntu-latest, windows-latest, macos-latest]
     runs-on: ${{matrix.os}}
     steps:
       - uses: goto-bus-stop/setup-zig@v2
@@ -38,9 +37,9 @@ jobs:
             brew install sdl2
       - uses: actions/checkout@v3
         with:
-          submodules: recursive
+          submodules: true
       - name: build 
-        run: zig build -Doptimize=ReleaseSafe -Dcpu=baseline
+        run: zig build -Drelease-safe
       - name: upload
         uses: actions/upload-artifact@v3
         with:
@@ -51,7 +50,7 @@ jobs:
     steps: 
       - uses: actions/checkout@v3
         with:
-          submodules: recursive
+          submodules: true
       - uses: goto-bus-stop/setup-zig@v2
         with:
           version: master

.gitmodules

@@ -13,6 +13,3 @@
 [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

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

README.md

@@ -10,10 +10,10 @@ This is a simple (read: incomplete) for-fun long-term project. I hope to get "mo
 
 ### TODO
 
-- [x] Affine Sprites
+- [ ] Affine Sprites
 - [ ] Windowing (see [this branch](https://git.musuka.dev/paoda/zba/src/branch/window))
 - [ ] Audio Resampler (Having issues with SDL2's)
-- [x] Immediate Mode GUI (see [this branch](https://git.musuka.dev/paoda/zba/src/branch/imgui))
+- [ ] Immediate Mode GUI
 - [ ] Refactoring for easy-ish perf boosts
 
 ## Usage
@@ -77,7 +77,7 @@ arm7wrestler GBA Fixed | [destoer](https://github.com/destoer)
 
 ## Compiling
 
-Most recently built on Zig [v0.11.0-dev.1580+a5b34a61a](https://github.com/ziglang/zig/tree/a5b34a61a)
+Most recently built on Zig [0.11.0-dev.368+1829b6eab](https://github.com/ziglang/zig/tree/1829b6eab)
 
 ### Dependencies
 
@@ -91,17 +91,17 @@ zig-datetime | <https://github.com/frmdstryr/zig-datetime>
 `bitfields.zig` | [https://github.com/FlorenceOS/Florence](https://github.com/FlorenceOS/Florence/blob/aaa5a9e568/lib/util/bitfields.zig)
 `gl.zig` | <https://github.com/MasterQ32/zig-opengl>
 
-Use `git submodule update --init` from the project root to pull the git relevant git submodules
+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`
 
 Be sure to provide SDL2 using:
 
 - Linux: Your distro's package manager
-- macOS: ¯\\\_(ツ)_/¯ (try [this formula](https://formulae.brew.sh/formula/sdl2)?)
+- MacOS: ¯\\\_(ツ)_/¯
 - Windows: [`vcpkg`](https://github.com/Microsoft/vcpkg) (install `sdl2:x64-windows`)
 
 `SDL.zig` will provide a helpful compile error if the zig compiler is unable to find SDL2.
 
-Once you've got all the dependencies, execute `zig build -Doptimize=ReleaseSafe`. The executable is located at `zig-out/bin/`.
+Once you've got all the dependencies, execute `zig build -Drelease-fast`. The executable is located at `zig-out/bin/`.
 
 ## Controls
 

build.zig

@@ -1,55 +1,54 @@
 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 {
     // Minimum Zig Version
-    const min_ver = std.SemanticVersion.parse("0.11.0-dev.1580+a5b34a61a") catch return; // https://github.com/ziglang/zig/commit/a5b34a61a
+    const min_ver = std.SemanticVersion.parse("0.11.0-dev.323+30eb2a175") catch return; // https://github.com/ziglang/zig/commit/30eb2a175
     if (builtin.zig_version.order(min_ver).compare(.lt)) {
         std.log.err("{s}", .{b.fmt("Zig v{} does not meet the minimum version requirement. (Zig v{})", .{ builtin.zig_version, min_ver })});
         std.os.exit(1);
     }
 
+    // Standard target options allows the person running `zig build` to choose
+    // what target to build for. Here we do not override the defaults, which
+    // means any target is allowed, and the default is native. Other options
+    // for restricting supported target set are available.
     const target = b.standardTargetOptions(.{});
-    const optimize = b.standardOptimizeOption(.{});
 
-    const exe = b.addExecutable(.{
-        .name = "zba",
-        .root_source_file = .{ .path = "src/main.zig" },
-        .target = target,
-        .optimize = optimize,
-    });
+    // Standard release options allow the person running `zig build` to select
+    // between Debug, ReleaseSafe, ReleaseFast, and ReleaseSmall.
+    const mode = b.standardReleaseOptions();
 
+    const exe = b.addExecutable("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.addAnonymousModule("known_folders", .{ .source_file = .{ .path = "lib/known-folders/known-folders.zig" } });
+    exe.addPackagePath("known_folders", "lib/known-folders/known-folders.zig");
 
     // DateTime Library
-    exe.addAnonymousModule("datetime", .{ .source_file = .{ .path = "lib/zig-datetime/src/main.zig" } });
+    exe.addPackagePath("datetime", "lib/zig-datetime/src/main.zig");
 
     // Bitfield type from FlorenceOS: https://github.com/FlorenceOS/
-    exe.addAnonymousModule("bitfield", .{ .source_file = .{ .path = "lib/util/bitfield.zig" } });
+    // exe.addPackage(.{ .name = "bitfield", .path = .{ .path = "lib/util/bitfield.zig" } });
+    exe.addPackagePath("bitfield", "lib/util/bitfield.zig");
 
     // Argument Parsing Library
-    exe.addAnonymousModule("clap", .{ .source_file = .{ .path = "lib/zig-clap/clap.zig" } });
+    exe.addPackagePath("clap", "lib/zig-clap/clap.zig");
 
     // TOML Library
-    exe.addAnonymousModule("toml", .{ .source_file = .{ .path = "lib/zig-toml/src/toml.zig" } });
+    exe.addPackagePath("toml", "lib/zig-toml/src/toml.zig");
 
     // OpenGL 3.3 Bindings
-    exe.addAnonymousModule("gl", .{ .source_file = .{ .path = "lib/gl.zig" } });
-
-    // gdbstub
-    Gdbstub.link(exe);
+    exe.addPackagePath("gl", "lib/gl.zig");
 
     // Zig SDL Bindings: https://github.com/MasterQ32/SDL.zig
-    const sdk = Sdk.init(b, null);
+    const sdk = Sdk.init(b);
     sdk.link(exe, .dynamic);
-    exe.addModule("sdl2", sdk.getNativeModule());
+    exe.addPackage(sdk.getNativePackage("sdl2"));
 
+    exe.setBuildMode(mode);
     exe.install();
 
     const run_cmd = exe.run();
@@ -61,11 +60,9 @@ 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(.{
-        .root_source_file = .{ .path = "src/main.zig" },
-        .target = target,
-        .optimize = optimize,
-    });
+    const exe_tests = b.addTest("src/main.zig");
+    exe_tests.setTarget(target);
+    exe_tests.setBuildMode(mode);
 
     const test_step = b.step("test", "Run unit tests");
     test_step.dependOn(&exe_tests.step);

src/core/Bus.zig

@@ -60,8 +60,9 @@ 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] };
+    const read_table: *[table_len]?*const anyopaque = tables[0..table_len];
+    const left_write: *[table_len]?*anyopaque = tables[table_len .. 2 * table_len];
+    const right_write: *[table_len]?*anyopaque = tables[2 * table_len .. 3 * table_len];
 
     self.* = .{
         .pak = try GamePak.init(allocator, cpu, paths.rom, paths.save),
@@ -77,15 +78,18 @@ pub fn init(self: *Self, allocator: Allocator, sched: *Scheduler, cpu: *Arm7tdmi
         .sched = sched,
 
         .read_table = read_table,
-        .write_tables = write_tables,
+        .write_tables = .{ left_write, right_write },
         .allocator = allocator,
     };
 
-    self.fillReadTable(read_table);
+    // read_table, write_tables, and *Self are not restricted to the lifetime
+    // of this init function so we can initialize our tables here
+    fillReadTable(self, read_table);
 
-    // Internal Display Memory behaves differently on 8-bit reads
-    self.fillWriteTable(u32, write_tables[0]);
-    self.fillWriteTable(u8, write_tables[1]);
+    // Internal Display Memory behavious unusually on 8-bit reads
+    // so we have two different tables depending on whether there's an 8-bit read or not
+    fillWriteTable(u32, self, left_write);
+    fillWriteTable(u8, self, right_write);
 }
 
 pub fn deinit(self: *Self) void {
@@ -102,50 +106,50 @@ pub fn deinit(self: *Self) void {
     self.* = undefined;
 }
 
-fn fillReadTable(self: *Self, table: *[table_len]?*const anyopaque) void {
-    const vramMirror = @import("ppu/Vram.zig").mirror;
+fn fillReadTable(bus: *Self, table: *[table_len]?*const anyopaque) void {
+    const vramMirror = @import("ppu.zig").Vram.mirror;
 
-    for (table, 0..) |*ptr, i| {
-        const addr = @intCast(u32, page_size * i);
+    for (table) |*ptr, i| {
+        const addr = 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],
+            0x0200_0000...0x02FF_FFFF => &bus.ewram.buf[addr & 0x3FFFF],
+            0x0300_0000...0x03FF_FFFF => &bus.iwram.buf[addr & 0x7FFF],
             0x0400_0000...0x0400_03FF => null, // I/O
 
             // Internal Display Memory
-            0x0500_0000...0x05FF_FFFF => &self.ppu.palette.buf[addr & 0x3FF],
-            0x0600_0000...0x06FF_FFFF => &self.ppu.vram.buf[vramMirror(addr)],
-            0x0700_0000...0x07FF_FFFF => &self.ppu.oam.buf[addr & 0x3FF],
+            0x0500_0000...0x05FF_FFFF => &bus.ppu.palette.buf[addr & 0x3FF],
+            0x0600_0000...0x06FF_FFFF => &bus.ppu.vram.buf[vramMirror(addr)],
+            0x0700_0000...0x07FF_FFFF => &bus.ppu.oam.buf[addr & 0x3FF],
 
             // External Memory (Game Pak)
-            0x0800_0000...0x0DFF_FFFF => self.fillReadTableExternal(addr),
+            0x0800_0000...0x0DFF_FFFF => fillTableExternalMemory(bus, addr),
             0x0E00_0000...0x0FFF_FFFF => null, // SRAM
             else => null,
         };
     }
 }
 
-fn fillWriteTable(self: *Self, comptime T: type, table: *[table_len]?*const anyopaque) void {
+fn fillWriteTable(comptime T: type, bus: *Self, table: *[table_len]?*const anyopaque) void {
     comptime std.debug.assert(T == u32 or T == u16 or T == u8);
-    const vramMirror = @import("ppu/Vram.zig").mirror;
+    const vramMirror = @import("ppu.zig").Vram.mirror;
 
-    for (table, 0..) |*ptr, i| {
-        const addr = @intCast(u32, page_size * i);
+    for (table) |*ptr, i| {
+        const addr = 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],
+            0x0200_0000...0x02FF_FFFF => &bus.ewram.buf[addr & 0x3FFFF],
+            0x0300_0000...0x03FF_FFFF => &bus.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,
+            0x0500_0000...0x05FF_FFFF => if (T != u8) &bus.ppu.palette.buf[addr & 0x3FF] else null,
+            0x0600_0000...0x06FF_FFFF => if (T != u8) &bus.ppu.vram.buf[vramMirror(addr)] else null,
+            0x0700_0000...0x07FF_FFFF => if (T != u8) &bus.ppu.oam.buf[addr & 0x3FF] else null,
 
             // External Memory (Game Pak)
             0x0800_0000...0x0DFF_FFFF => null, // ROM
@@ -155,29 +159,24 @@ fn fillWriteTable(self: *Self, comptime T: type, table: *[table_len]?*const anyo
     }
 }
 
-fn fillReadTableExternal(self: *Self, addr: u32) ?*anyopaque {
+fn fillTableExternalMemory(bus: *Self, addr: usize) ?*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 end_addr = 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 gpio_data = start_addr <= 0x0800_00C4 and 0x0800_00C4 < end_addr;
+    const gpio_direction = start_addr <= 0x0800_00C6 and 0x0800_00C6 < end_addr;
+    const gpio_control = start_addr <= 0x0800_00C8 and 0x0800_00C8 < end_addr;
 
-        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 (bus.pak.gpio.device.kind != .None and (gpio_data or gpio_direction or gpio_control)) {
+        // We found a GPIO device, and this page a GPIO register. We want to handle this in slowmem
+        return null;
     }
 
-    if (self.pak.backup.kind == .Eeprom) {
-        if (self.pak.buf.len > 0x100_000) {
+    if (bus.pak.backup.kind == .Eeprom) {
+        if (bus.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
@@ -193,9 +192,49 @@ fn fillReadTableExternal(self: *Self, addr: u32) ?*anyopaque {
     // 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;
+    if (masked_addr >= bus.pak.buf.len) return null;
 
-    return &self.pak.buf[masked_addr];
+    return &bus.pak.buf[masked_addr];
+}
+
+// TODO: Take advantage of fastmem here too?
+pub fn dbgRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    return switch (page) {
+        // General Internal Memory
+        0x00 => blk: {
+            if (address < Bios.size)
+                break :blk self.bios.dbgRead(T, self.cpu.r[15], address);
+
+            break :blk self.openBus(T, address);
+        },
+        0x02 => self.ewram.read(T, address),
+        0x03 => self.iwram.read(T, address),
+        0x04 => self.readIo(T, address),
+
+        // Internal Display Memory
+        0x05 => self.ppu.palette.read(T, address),
+        0x06 => self.ppu.vram.read(T, address),
+        0x07 => self.ppu.oam.read(T, address),
+
+        // External Memory (Game Pak)
+        0x08...0x0D => self.pak.dbgRead(T, address),
+        0x0E...0x0F => blk: {
+            const value = self.pak.backup.read(unaligned_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),
+    };
 }
 
 fn readIo(self: *const Self, comptime T: type, address: u32) T {
@@ -277,7 +316,8 @@ pub fn read(self: *Self, comptime T: type, unaligned_address: u32) T {
     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));
+        const alignment = @alignOf(std.meta.Child(Ptr));
+        const ptr = @ptrCast(Ptr, @alignCast(alignment, some_ptr));
 
         // Note: We don't check array length, since we force align the
         // lower bits of the address as the GBA would
@@ -287,27 +327,6 @@ pub fn read(self: *Self, comptime T: type, unaligned_address: u32) 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);
 
@@ -318,7 +337,7 @@ fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T {
         // General Internal Memory
         0x00 => blk: {
             if (address < Bios.size)
-                break :blk self.bios.read(T, self.cpu.r[15], unaligned_address);
+                break :blk self.bios.read(T, self.cpu.r[15], address);
 
             break :blk self.openBus(T, address);
         },
@@ -333,40 +352,7 @@ fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T {
 
         // External Memory (Game Pak)
         0x08...0x0D => self.pak.read(T, address),
-        0x0E...0x0F => self.readBackup(T, unaligned_address),
-        else => self.openBus(T, address),
-    };
-}
-
-fn dbgSlowRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
-    const page = @truncate(u8, unaligned_address >> 24);
-    const address = forceAlign(T, unaligned_address);
-
-    return switch (page) {
-        // General Internal Memory
-        0x00 => blk: {
-            if (address < Bios.size)
-                break :blk self.bios.dbgRead(T, self.cpu.r[15], unaligned_address);
-
-            break :blk self.openBus(T, address);
-        },
-        0x02 => unreachable, // handled by fastmem
-        0x03 => unreachable, // handled by fastmem
-        0x04 => self.readIo(T, address),
-
-        // Internal Display Memory
-        0x05 => unreachable, // handled by fastmem
-        0x06 => unreachable, // handled by fastmem
-        0x07 => unreachable, // handled by fastmem
-
-        // External Memory (Game Pak)
-        0x08...0x0D => self.pak.dbgRead(T, address),
-        0x0E...0x0F => self.readBackup(T, unaligned_address),
-        else => self.openBus(T, address),
-    };
-}
-
-fn readBackup(self: *const Self, comptime T: type, unaligned_address: u32) T {
+        0x0E...0x0F => blk: {
             const value = self.pak.backup.read(unaligned_address);
 
             const multiplier = switch (T) {
@@ -376,7 +362,10 @@ fn readBackup(self: *const Self, comptime T: type, unaligned_address: u32) T {
                 else => @compileError("Backup: Unsupported read width"),
             };
 
-    return @as(T, value) * multiplier;
+            break :blk @as(T, value) * multiplier;
+        },
+        else => self.openBus(T, address),
+    };
 }
 
 pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
@@ -393,7 +382,8 @@ pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) vo
     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));
+        const alignment = @alignOf(std.meta.Child(Ptr));
+        const ptr = @ptrCast(Ptr, @alignCast(alignment, some_ptr));
 
         // Note: We don't check array length, since we force align the
         // lower bits of the address as the GBA would
@@ -406,34 +396,8 @@ pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) vo
     }
 }
 
-/// 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);
-
+pub 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);
 
@@ -456,41 +420,16 @@ fn slowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) vo
     }
 }
 
-fn dbgSlowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
-    @setCold(true);
-
-    const page = @truncate(u8, unaligned_address >> 24);
-    const address = forceAlign(T, unaligned_address);
-
-    switch (page) {
-        // General Internal Memory
-        0x00 => self.bios.write(T, address, value),
-        0x02 => unreachable, // completely handled by fastmem
-        0x03 => unreachable, // completely handled by fastmem
-        0x04 => return, // FIXME: Let debug writes mess with I/O
-
-        // Internal Display Memory
-        0x05 => self.ppu.palette.write(T, address, value),
-        0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, address, value),
-        0x07 => unreachable, // completely handled by fastmem
-
-        // External Memory (Game Pak)
-        0x08...0x0D => return, // FIXME: Debug Write to Backup/GPIO w/out messing with state
-        0x0E...0x0F => return, // FIXME: Debug Write to Backup w/out messing with state
-        else => {},
-    }
-}
-
 inline fn rotateBy(comptime T: type, address: u32) u32 {
     return switch (T) {
         u32 => address & 3,
         u16 => address & 1,
         u8 => 0,
-        else => @compileError("Unsupported write width"),
+        else => @compileError("Backup: Unsupported write width"),
     };
 }
 
-pub inline fn forceAlign(comptime T: type, address: u32) u32 {
+inline fn forceAlign(comptime T: type, address: u32) u32 {
     return switch (T) {
         u32 => address & ~@as(u32, 3),
         u16 => address & ~@as(u32, 1),

src/core/apu.zig

@@ -15,12 +15,10 @@ const SoundFifo = std.fifo.LinearFifo(u8, .{ .Static = 0x20 });
 const getHalf = util.getHalf;
 const setHalf = util.setHalf;
 const intToBytes = util.intToBytes;
+const RingBuffer = util.RingBuffer;
 
 const log = std.log.scoped(.APU);
 
-pub const host_rate = @import("../platform.zig").sample_rate;
-pub const host_format = @import("../platform.zig").sample_format;
-
 pub fn read(comptime T: type, apu: *const Apu, addr: u32) ?T {
     const byte_addr = @truncate(u8, addr);
 
@@ -246,17 +244,20 @@ pub const Apu = struct {
 
     sampling_cycle: u2,
 
-    stream: *SDL.SDL_AudioStream,
+    sample_queue: RingBuffer(u16),
     sched: *Scheduler,
 
     fs: FrameSequencer,
     capacitor: f32,
 
-    is_buffer_full: bool,
-
     pub const Tick = enum { Length, Envelope, Sweep };
 
     pub fn init(sched: *Scheduler) Self {
+        const NUM_CHANNELS: usize = 2;
+
+        const allocator = std.heap.c_allocator;
+        const sample_buf = allocator.alloc(u16, 0x800 * NUM_CHANNELS) catch @panic("failed to allocate sample buffer");
+
         const apu: Self = .{
             .ch1 = ToneSweep.init(sched),
             .ch2 = Tone.init(sched),
@@ -271,12 +272,11 @@ pub const Apu = struct {
             .bias = .{ .raw = 0x0200 },
 
             .sampling_cycle = 0b00,
-            .stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, 1 << 15, host_format, 2, host_rate).?,
+            .sample_queue = RingBuffer(u16).init(sample_buf),
             .sched = sched,
 
             .capacitor = 0,
             .fs = FrameSequencer.init(),
-            .is_buffer_full = false,
         };
 
         sched.push(.SampleAudio, apu.interval());
@@ -370,11 +370,6 @@ pub const Apu = struct {
     pub fn sampleAudio(self: *Self, late: u64) void {
         self.sched.push(.SampleAudio, self.interval() -| late);
 
-        // Whether the APU is busy or not is determined  by the main loop in emu.zig
-        // This should only ever be true (because this side of the emu is single threaded)
-        // When audio sync is disaabled
-        if (self.is_buffer_full) return;
-
         var left: i16 = 0;
         var right: i16 = 0;
 
@@ -430,23 +425,7 @@ pub const Apu = struct {
         const ext_left = (clamped_left << 5) | (clamped_left >> 6);
         const ext_right = (clamped_right << 5) | (clamped_right >> 6);
 
-        if (self.sampling_cycle != self.bias.sampling_cycle.read()) self.replaceSDLResampler();
-
-        _ = SDL.SDL_AudioStreamPut(self.stream, &[2]u16{ ext_left, ext_right }, 2 * @sizeOf(u16));
-    }
-
-    fn replaceSDLResampler(self: *Self) void {
-        @setCold(true);
-        const sample_rate = Self.sampleRate(self.bias.sampling_cycle.read());
-        log.info("Sample Rate changed from {}Hz to {}Hz", .{ Self.sampleRate(self.sampling_cycle), sample_rate });
-
-        // Sampling Cycle (Sample Rate) changed, Craete a new SDL Audio Resampler
-        // FIXME: Replace SDL's Audio Resampler with either a custom or more reliable one
-        const old_stream = self.stream;
-        defer SDL.SDL_FreeAudioStream(old_stream);
-
-        self.sampling_cycle = self.bias.sampling_cycle.read();
-        self.stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, @intCast(c_int, sample_rate), host_format, 2, host_rate).?;
+        self.sample_queue.push(ext_left, ext_right) catch {};
     }
 
     fn interval(self: *const Self) u64 {

src/core/bus/Bios.zig

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

src/core/bus/GamePak.zig

@@ -213,7 +213,6 @@ 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

@@ -137,7 +137,6 @@ 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;

src/core/bus/dma.zig

@@ -338,8 +338,11 @@ fn DmaController(comptime id: u2) type {
     };
 }
 
-pub fn onBlanking(bus: *Bus, comptime kind: DmaKind) void {
-    inline for (0..4) |i| bus.dma[i].poll(kind);
+pub fn pollDmaOnBlank(bus: *Bus, comptime kind: DmaKind) void {
+    comptime var i: usize = 0;
+    inline while (i < 4) : (i += 1) {
+        bus.dma[i].poll(kind);
+    }
 }
 
 const Adjustment = enum(u2) {

src/core/bus/gpio.zig

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

src/core/bus/io.zig

@@ -449,8 +449,6 @@ pub const BldY = extern union {
     raw: u16,
 };
 
-const u8WriteKind = enum { Hi, Lo };
-
 /// Write-only
 pub const WinH = extern union {
     x2: Bitfield(u16, 0, 8),
@@ -460,8 +458,6 @@ 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,
@@ -470,20 +466,20 @@ pub const WinV = extern union {
 pub const WinIn = extern union {
     w0_bg: Bitfield(u16, 0, 4),
     w0_obj: Bit(u16, 4),
-    w0_bld: Bit(u16, 5),
+    w0_colour: Bit(u16, 5),
     w1_bg: Bitfield(u16, 8, 4),
     w1_obj: Bit(u16, 12),
-    w1_bld: Bit(u16, 13),
+    w1_colour: Bit(u16, 13),
     raw: u16,
 };
 
 pub const WinOut = extern union {
     out_bg: Bitfield(u16, 0, 4),
     out_obj: Bit(u16, 4),
-    out_bld: Bit(u16, 5),
+    out_colour: Bit(u16, 5),
     obj_bg: Bitfield(u16, 8, 4),
     obj_obj: Bit(u16, 12),
-    obj_bld: Bit(u16, 13),
+    obj_colour: Bit(u16, 13),
     raw: u16,
 };
 

src/core/cpu.zig

@@ -39,12 +39,13 @@ pub const arm = struct {
     }
 
     fn populate() [0x1000]InstrFn {
-        comptime {
+        return comptime {
             @setEvalBranchQuota(0xE000);
-            var table = [_]InstrFn{und} ** 0x1000;
+            var ret = [_]InstrFn{und} ** 0x1000;
 
-            for (&table, 0..) |*handler, i| {
-                handler.* = switch (@as(u2, i >> 10)) {
+            var i: usize = 0;
+            while (i < ret.len) : (i += 1) {
+                ret[i] = switch (@as(u2, i >> 10)) {
                     0b00 => if (i == 0x121) blk: {
                         break :blk branchExchange;
                     } else if (i & 0xFCF == 0x009) blk: {
@@ -106,8 +107,8 @@ pub const arm = struct {
                 };
             }
 
-            return table;
-        }
+            return ret;
+        };
     }
 };
 
@@ -135,12 +136,13 @@ pub const thumb = struct {
     }
 
     fn populate() [0x400]InstrFn {
-        comptime {
+        return comptime {
             @setEvalBranchQuota(5025); // This is exact
-            var table = [_]InstrFn{und} ** 0x400;
+            var ret = [_]InstrFn{und} ** 0x400;
 
-            for (&table, 0..) |*handler, i| {
-                handler.* = switch (@as(u3, i >> 7 & 0x7)) {
+            var i: usize = 0;
+            while (i < ret.len) : (i += 1) {
+                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;
@@ -228,8 +230,8 @@ pub const thumb = struct {
                 };
             }
 
-            return table;
-        }
+            return ret;
+        };
     }
 };
 
@@ -383,7 +385,8 @@ pub const Arm7tdmi = struct {
         const now = getModeChecked(self, self.cpsr.mode.read());
 
         // Bank R8 -> r12
-        for (0..5) |i| {
+        var i: usize = 0;
+        while (i < 5) : (i += 1) {
             self.bank.fiq[Bank.fiqIdx(i, now)] = self.r[8 + i];
         }
 
@@ -401,7 +404,8 @@ pub const Arm7tdmi = struct {
         }
 
         // Grab R8 -> R12
-        for (0..5) |i| {
+        i = 0;
+        while (i < 5) : (i += 1) {
             self.r[8 + i] = self.bank.fiq[Bank.fiqIdx(i, next)];
         }
 
@@ -466,7 +470,8 @@ pub const Arm7tdmi = struct {
     }
 
     pub fn stepDmaTransfer(self: *Self) bool {
-        inline for (0..4) |i| {
+        comptime var i: usize = 0;
+        inline while (i < 4) : (i += 1) {
             if (self.bus.dma[i].in_progress) {
                 self.bus.dma[i].step(self);
                 return true;

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

src/core/cpu/arm/single_data_transfer.zig

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

src/core/cpu/thumb/alu.zig

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

src/core/cpu/thumb/block_data_transfer.zig

@@ -92,7 +92,8 @@ pub fn fmt15(comptime L: bool, comptime rb: u3) InstrFn {
 inline fn countRlist(opcode: u16) u32 {
     var count: u32 = 0;
 
-    inline for (0..8) |i| {
+    comptime var i: u4 = 0;
+    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: u1 = undefined;
+            var overflow: bool = 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: u1 = undefined;
+                var overflow: bool = undefined;
                 const result = add(&overflow, op1, op2);
                 cpu.r[rd] = result;
 
                 cpu.cpsr.n.write(result >> 31 & 1 == 1);
                 cpu.cpsr.z.write(result == 0);
-                cpu.cpsr.c.write(overflow == 0b1);
+                cpu.cpsr.c.write(overflow);
                 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: u1 = undefined;
+            var overflow: bool = 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 == 0b1);
+                    cpu.cpsr.c.write(overflow);
                     cpu.cpsr.v.write(((op1 ^ result) & (op2 ^ result)) >> 31 & 1 == 1);
                 },
             }

src/core/emu.zig

@@ -5,6 +5,7 @@ const config = @import("../config.zig");
 const Scheduler = @import("scheduler.zig").Scheduler;
 const Arm7tdmi = @import("cpu.zig").Arm7tdmi;
 const FpsTracker = @import("../util.zig").FpsTracker;
+const RingBuffer = @import("../util.zig").RingBuffer;
 
 const Timer = std.time.Timer;
 const Atomic = std.atomic.Atomic;
@@ -58,7 +59,7 @@ fn inner(comptime kind: RunKind, audio_sync: bool, quit: *Atomic(bool), schedule
 
             while (!quit.load(.Monotonic)) {
                 runFrame(scheduler, cpu);
-                audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+                audioSync(audio_sync, &cpu.bus.apu.sample_queue);
 
                 if (kind == .UnlimitedFPS) tracker.?.tick();
             }
@@ -77,7 +78,7 @@ fn inner(comptime kind: RunKind, audio_sync: bool, quit: *Atomic(bool), schedule
                 // the amount of time needed for audio to catch up rather than
                 // our expected wake-up time
 
-                audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+                audioSync(audio_sync, &cpu.bus.apu.sample_queue);
                 if (!audio_sync) spinLoop(&timer, wake_time);
                 wake_time = new_wake_time;
 
@@ -94,7 +95,7 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
         if (!cpu.stepDmaTransfer()) {
             if (cpu.isHalted()) {
                 // Fast-forward to next Event
-                sched.tick = sched.nextTimestamp();
+                sched.tick = sched.queue.peek().?.tick;
             } else {
                 cpu.step();
             }
@@ -104,22 +105,13 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
     }
 }
 
-fn audioSync(audio_sync: bool, stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
+fn audioSync(audio_sync: bool, sample_queue: *RingBuffer(u16)) 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;
+    // const sample_size = 2 * @sizeOf(u16);
+    // const max_buf_size: c_int = 0x400;
 
-    // Determine whether the APU is busy right at this moment
-    var still_full: bool = SDL.SDL_AudioStreamAvailable(stream) > sample_size * if (is_buffer_full.*) max_buf_size >> 1 else max_buf_size;
-    defer is_buffer_full.* = still_full; // Update APU Busy status right before exiting scope
-
-    // If Busy is false, there's no need to sync here
-    if (!still_full) return;
-
-    while (true) {
-        still_full = SDL.SDL_AudioStreamAvailable(stream) > sample_size * max_buf_size >> 1;
-        if (!audio_sync or !still_full) break;
-    }
+    _ = audio_sync;
+    _ = sample_queue;
 }
 
 fn videoSync(timer: *Timer, wake_time: u64) u64 {
@@ -146,8 +138,9 @@ 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;
 
-    for (0..times) |_| {
+    while (i < times) : (i += 1) {
         std.time.sleep(step);
 
         // Upon wakeup, check to see if this particular sleep was longer than expected
@@ -162,59 +155,3 @@ 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

@@ -1,40 +0,0 @@
-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

@@ -1,47 +0,0 @@
-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

@@ -1,60 +0,0 @@
-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

@@ -31,7 +31,7 @@ pub const Scheduler = struct {
     }
 
     pub fn handleEvent(self: *Self, cpu: *Arm7tdmi) void {
-        const event = self.queue.remove();
+        if (self.queue.removeOrNull()) |event| {
             const late = self.tick - event.tick;
 
             switch (event.kind) {
@@ -70,16 +70,22 @@ pub const Scheduler = struct {
                 .VBlank => cpu.bus.ppu.onHdrawEnd(cpu, late), // The end of a VBlank
             }
         }
+    }
 
     /// Removes the **first** scheduled event of type `needle`
     pub fn removeScheduledEvent(self: *Self, needle: EventKind) void {
-        for (self.queue.items, 0..) |event, i| {
+        var it = self.queue.iterator();
+
+        var i: usize = 0;
+        while (it.next()) |event| : (i += 1) {
             if (std.meta.eql(event.kind, needle)) {
 
-                // invalidates the slice we're iterating over
+                // This invalidates the iterator
                 _ = self.queue.removeIndex(i);
 
-                log.debug("Removed {?}@{}", .{ event.kind, event.tick });
+                // Since removing something from the PQ invalidates the iterator,
+                // this implementation can safely only remove the first instance of
+                // a Scheduled Event. Exit Early
                 break;
             }
         }

src/main.zig

@@ -4,17 +4,14 @@ 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 Allocator = std.mem.Allocator;
 const log = std.log.scoped(.Cli);
 const width = @import("core/ppu.zig").width;
 const height = @import("core/ppu.zig").height;
@@ -25,7 +22,6 @@ 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.
     \\
 );
@@ -91,49 +87,13 @@ pub fn main() void {
         cpu.fastBoot();
     }
 
-    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();
 
-    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});
-    }
+    gui.run(&cpu, &scheduler) catch |e| exitln("failed to run gui thread: {}", .{e});
 }
 
-fn handleArguments(allocator: Allocator, data_path: []const u8, result: *const clap.Result(clap.Help, &params, clap.parsers.default)) !FilePaths {
+pub fn handleArguments(allocator: Allocator, data_path: []const u8, result: *const clap.Result(clap.Help, &params, clap.parsers.default)) !FilePaths {
     const rom_path = romPath(result);
     log.info("ROM path: {s}", .{rom_path});
 

src/platform.zig

@@ -12,7 +12,7 @@ const FpsTracker = @import("util.zig").FpsTracker;
 const gba_width = @import("core/ppu.zig").width;
 const gba_height = @import("core/ppu.zig").height;
 
-pub const sample_rate = 1 << 15;
+pub const sample_rate = 1 << 16;
 pub const sample_format = SDL.AUDIO_U16;
 
 const default_title = "ZBA";
@@ -64,7 +64,7 @@ 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 {};
+        try gl.load(ctx, Self.glGetProcAddress);
         if (SDL.SDL_GL_SetSwapInterval(@boolToInt(config.config().host.vsync)) < 0) panic();
 
         const program_id = try compileShaders();
@@ -154,17 +154,9 @@ pub const Gui = struct {
         return tex_id;
     }
 
-    const RunOptions = struct {
-        quit: *std.atomic.Atomic(bool),
-        tracker: ?*FpsTracker = null,
-        cpu: *Arm7tdmi,
-        scheduler: *Scheduler,
-    };
-
-    pub fn run(self: *Self, opt: RunOptions) !void {
-        const cpu = opt.cpu;
-        const tracker = opt.tracker;
-        const quit = opt.quit;
+    pub fn run(self: *Self, cpu: *Arm7tdmi, scheduler: *Scheduler) !void {
+        var quit = std.atomic.Atomic(bool).init(false);
+        var tracker = FpsTracker.init();
 
         var buffer_ids = Self.generateBuffers();
         defer {
@@ -177,15 +169,13 @@ pub const Gui = struct {
         const tex_id = Self.generateTexture(cpu.bus.ppu.framebuf.get(.Renderer));
         defer gl.deleteTextures(1, &tex_id);
 
+        const thread = try std.Thread.spawn(.{}, emu.run, .{ &quit, scheduler, cpu, &tracker });
+        defer thread.join();
+
         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,
@@ -226,7 +216,7 @@ pub const Gui = struct {
                             SDL.SDLK_RSHIFT => keyinput.select.set(),
                             SDL.SDLK_i => {
                                 comptime std.debug.assert(sample_format == SDL.AUDIO_U16);
-                                log.err("Sample Count: {}", .{@intCast(u32, SDL.SDL_AudioStreamAvailable(cpu.bus.apu.stream)) / (2 * @sizeOf(u16))});
+                                log.err("Sample Count: {}", .{cpu.bus.apu.sample_queue.len() / 2});
                             },
                             // SDL.SDLK_j => log.err("Scheduler Capacity: {} | Scheduler Event Count: {}", .{ scheduler.queue.capacity(), scheduler.queue.count() }),
                             SDL.SDLK_k => {},
@@ -248,11 +238,9 @@ pub const Gui = struct {
             gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_INT, null);
             SDL.SDL_GL_SwapWindow(self.window);
 
-            if (tracker) |t| {
-                const dyn_title = std.fmt.bufPrintZ(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, t.value() }) catch unreachable;
+            const dyn_title = std.fmt.bufPrintZ(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, tracker.value() }) catch unreachable;
             SDL.SDL_SetWindowTitle(self.window, dyn_title.ptr);
         }
-        }
 
         quit.store(true, .Monotonic); // Terminate Emulator Thread
     }
@@ -311,7 +299,15 @@ const Audio = struct {
         const T = *Apu;
         const apu = @ptrCast(T, @alignCast(@alignOf(T), userdata));
 
-        _ = SDL.SDL_AudioStreamGet(apu.stream, stream, len);
+        comptime std.debug.assert(sample_format == SDL.AUDIO_U16);
+        const sample_buf = @ptrCast([*]u16, @alignCast(@alignOf(u16), stream))[0 .. @intCast(u32, len) / @sizeOf(u16)];
+
+        var previous: u16 = 0x8000;
+        for (sample_buf) |*sample| {
+            if (apu.sample_queue.pop()) |value| previous = value;
+
+            sample.* = previous;
+        }
     }
 };
 

src/util.zig

@@ -5,8 +5,6 @@ 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
@@ -125,7 +123,6 @@ 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),
 
@@ -184,6 +181,8 @@ pub const Logger = struct {
     }
 };
 
+const FmtArgTuple = struct { 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");
 
@@ -277,43 +276,109 @@ 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 Mutex = std.Thread.Mutex;
+
+pub fn RingBuffer(comptime T: type) type {
+    return struct {
         const Self = @This();
+        const Index = usize;
+        const max_capacity = (@as(Index, 1) << @typeInfo(Index).Int.bits - 1) - 1; // half the range of index type
 
-    layers: [2][]u8,
-    buf: []u8,
-    current: u1,
+        const log = std.log.scoped(.RingBuffer);
 
-    allocator: Allocator,
+        read: Index,
+        write: Index,
 
-    // TODO: Rename
-    const Device = enum { Emulator, Renderer };
+        buf: []T,
 
-    pub fn init(allocator: Allocator, comptime len: comptime_int) !Self {
-        const buf = try allocator.alloc(u8, len * 2);
-        std.mem.set(u8, buf, 0);
+        mutex: Mutex,
 
-        return .{
-            // Front and Back Framebuffers
-            .layers = [_][]u8{ buf[0..][0..len], buf[len..][0..len] },
-            .buf = buf,
-            .current = 0,
+        const Error = error{buffer_full};
 
-            .allocator = allocator,
+        pub fn init(buf: []T) Self {
+            std.mem.set(T, buf, 0);
+
+            std.debug.assert(std.math.isPowerOfTwo(buf.len)); // capacity must be a power of two
+            std.debug.assert(buf.len <= max_capacity);
+
+            return .{ .read = 0, .write = 0, .buf = buf, .mutex = .{} };
+        }
+
+        pub fn push(self: *Self, left: T, right: T) Error!void {
+            self.mutex.lock();
+            defer self.mutex.unlock();
+
+            try self._push(left);
+            self._push(right) catch |e| {
+                self.write -= 1; // undo the previous write;
+                return e;
             };
         }
 
-    pub fn deinit(self: *Self) void {
-        self.allocator.free(self.buf);
-        self.* = undefined;
+        pub fn pop(self: *Self) ?T {
+            self.mutex.lock();
+            defer self.mutex.unlock();
+
+            return self._pop();
         }
 
-    pub fn swap(self: *Self) void {
-        self.current = ~self.current;
+        pub fn len(self: *Self) Index {
+            self.mutex.lock();
+            defer self.mutex.unlock();
+
+            return self._len();
         }
 
-    pub fn get(self: *Self, comptime dev: Device) []u8 {
-        return self.layers[if (dev == .Emulator) self.current else ~self.current];
+        fn _push(self: *Self, value: T) Error!void {
+            if (self.isFull()) return error.buffer_full;
+            defer self.write += 1;
+
+            self.buf[self.mask(self.write)] = value;
+        }
+
+        fn _pop(self: *Self) ?T {
+            if (self.isEmpty()) return null;
+            defer self.read += 1;
+
+            return self.buf[self.mask(self.read)];
+        }
+
+        fn _len(self: *const Self) Index {
+            return self.write - self.read;
+        }
+
+        fn isFull(self: *const Self) bool {
+            return self._len() == self.buf.len;
+        }
+
+        fn isEmpty(self: *const Self) bool {
+            return self.read == self.write;
+        }
+
+        fn mask(self: *const Self, idx: Index) Index {
+            return idx & (self.buf.len - 1);
         }
     };
+}
+
+test "RingBuffer" {
+    const Queue = RingBuffer(u8);
+
+    var buf: [4]u8 = undefined;
+    var queue = Queue.init(&buf);
+
+    try queue.push(1, 2);
+    try std.testing.expectEqual(@as(?u8, 1), queue.pop());
+
+    try queue.push(3, 4);
+    try std.testing.expectError(Queue.Error.buffer_full, queue.push(5, 6));
+    try std.testing.expectEqual(@as(?u8, 2), queue.pop());
+
+    try queue.push(7, 8);
+
+    try std.testing.expectEqual(@as(?u8, 3), queue.pop());
+    try std.testing.expectEqual(@as(?u8, 4), queue.pop());
+    try std.testing.expectEqual(@as(?u8, 7), queue.pop());
+    try std.testing.expectEqual(@as(?u8, 8), queue.pop());
+    try std.testing.expectEqual(@as(?u8, null), queue.pop());
+}