chore: update dependencies

This will not improve perf in any way because this code only gets run
one time a second orz

.github/workflows/main.yml

@@ -14,7 +14,8 @@ jobs:
   build:
     strategy:
       matrix:
-        os: [ubuntu-latest, windows-latest, macos-latest]
+        # os: [ubuntu-latest, windows-latest, macos-latest]
+        os: [ubuntu-latest, windows-latest]
     runs-on: ${{matrix.os}}
     steps:
       - uses: goto-bus-stop/setup-zig@v2
@@ -39,7 +40,7 @@ jobs:
         with:
           submodules: true
       - name: build 
-        run: zig build -Doptimize=ReleaseSafe
+        run: zig build -Doptimize=ReleaseSafe -Dcpu=baseline
       - name: upload
         uses: actions/upload-artifact@v3
         with:

.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

README.md

@@ -96,7 +96,7 @@ Use `git submodule update --init` from the project root to pull the git relevant
 Be sure to provide SDL2 using:
 
 - Linux: Your distro's package manager
-- MacOS: `brew` (install [this formula](https://formulae.brew.sh/formula/sdl2))
+- 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.

build.zig

@@ -1,6 +1,8 @@
 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
@@ -40,6 +42,9 @@ pub fn build(b: *std.build.Builder) void {
     // OpenGL 3.3 Bindings
     exe.addAnonymousModule("gl", .{ .source_file = .{ .path = "lib/gl.zig" } });
 
+    // gdbstub
+    Gdbstub.link(exe);
+
     // Zig SDL Bindings: https://github.com/MasterQ32/SDL.zig
     const sdk = Sdk.init(b, null);
     sdk.link(exe, .dynamic);

lib/zba-gdbstub

@@ -0,0 +1 @@
+Subproject commit acb59994fcfb83d36efef47810a9dc791c6e542e

lib/zig-clap

@@ -1 +1 @@
-Subproject commit 272d8e2088b2cae037349fb260dc05ec46bba422
+Subproject commit 861de651f3e1314973b1273ac7856e96b2625ff3

lib/zig-datetime

@@ -1 +1 @@
-Subproject commit 932d2845210644ca736faf35f5bea31eb1a15465
+Subproject commit bf0ae0c27cfe92fdd9a92c8f1ac6d1939ae60c77

src/core/Bus.zig

@@ -105,7 +105,7 @@ pub fn deinit(self: *Self) void {
 fn fillReadTable(self: *Self, table: *[table_len]?*const anyopaque) void {
     const vramMirror = @import("ppu/Vram.zig").mirror;
 
-    for (table) |*ptr, i| {
+    for (table, 0..) |*ptr, i| {
         const addr = @intCast(u32, page_size * i);
 
         ptr.* = switch (addr) {
@@ -132,7 +132,7 @@ fn fillWriteTable(self: *Self, comptime T: type, table: *[table_len]?*const anyo
     comptime std.debug.assert(T == u32 or T == u16 or T == u8);
     const vramMirror = @import("ppu/Vram.zig").mirror;
 
-    for (table) |*ptr, i| {
+    for (table, 0..) |*ptr, i| {
         const addr = @intCast(u32, page_size * i);
 
         ptr.* = switch (addr) {
@@ -198,55 +198,6 @@ fn fillReadTableExternal(self: *Self, addr: u32) ?*anyopaque {
     return &self.pak.buf[masked_addr];
 }
 
-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 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 readIo(self: *const Self, comptime T: type, address: u32) T {
     return io.read(self, T, address) orelse self.openBus(T, address);
 }
@@ -336,6 +287,27 @@ 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);
 
@@ -366,6 +338,34 @@ fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T {
     };
 }
 
+fn dbgSlowRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const page = @truncate(u8, unaligned_address >> 24);
+    const address = forceAlign(T, unaligned_address);
+
+    return switch (page) {
+        // General Internal Memory
+        0x00 => blk: {
+            if (address < Bios.size)
+                break :blk self.bios.dbgRead(T, self.cpu.r[15], unaligned_address);
+
+            break :blk self.openBus(T, address);
+        },
+        0x02 => unreachable, // handled by fastmem
+        0x03 => unreachable, // handled by fastmem
+        0x04 => self.readIo(T, address),
+
+        // Internal Display Memory
+        0x05 => unreachable, // handled by fastmem
+        0x06 => unreachable, // handled by fastmem
+        0x07 => unreachable, // handled by fastmem
+
+        // External Memory (Game Pak)
+        0x08...0x0D => self.pak.dbgRead(T, address),
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
+        else => self.openBus(T, address),
+    };
+}
+
 fn readBackup(self: *const Self, comptime T: type, unaligned_address: u32) T {
     const value = self.pak.backup.read(unaligned_address);
 
@@ -406,6 +406,31 @@ 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);
 
@@ -431,6 +456,31 @@ 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,

src/core/bus/GamePak.zig

@@ -213,6 +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

@@ -137,6 +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;

src/core/bus/dma.zig

@@ -339,10 +339,7 @@ fn DmaController(comptime id: u2) type {
 }
 
 pub fn onBlanking(bus: *Bus, comptime kind: DmaKind) void {
-    comptime var i: usize = 0;
-    inline while (i < 4) : (i += 1) {
-        bus.dma[i].poll(kind);
-    }
+    inline for (0..4) |i| 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(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);
+        self.year = bcd(@intCast(u8, now.date.year - 2000));
+        self.month = @truncate(u5, bcd(now.date.month));
+        self.day = @truncate(u6, bcd(now.date.day));
+        self.weekday = @truncate(u3, bcd((now.date.weekday() + 1) % 7)); // API is Monday = 0, Sunday = 6. We want Sunday = 0, Saturday = 6
+        self.hour = @truncate(u6, bcd(now.time.hour));
+        self.minute = @truncate(u7, bcd(now.time.minute));
+        self.second = @truncate(u7, bcd(now.time.second));
     }
 
     fn step(self: *Self, value: Data) u4 {
@@ -449,16 +449,8 @@ pub const Clock = struct {
     }
 };
 
-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);
+/// Converts an 8-bit unsigned integer to its BCD representation.
+/// Note: Algorithm only works for values between 0 and 99 inclusive.
+fn bcd(value: u8) u8 {
+    return ((value / 10) << 4) + (value % 10);
 }

src/core/cpu.zig

@@ -39,13 +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;
-            while (i < ret.len) : (i += 1) {
-                ret[i] = switch (@as(u2, i >> 10)) {
+            for (&table, 0..) |*handler, i| {
+                handler.* = switch (@as(u2, i >> 10)) {
                     0b00 => if (i == 0x121) blk: {
                         break :blk branchExchange;
                     } else if (i & 0xFCF == 0x009) blk: {
@@ -107,8 +106,8 @@ pub const arm = struct {
                 };
             }
 
-            return ret;
-        };
+            return table;
+        }
     }
 };
 
@@ -136,13 +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;
-            while (i < ret.len) : (i += 1) {
-                ret[i] = switch (@as(u3, i >> 7 & 0x7)) {
+            for (&table, 0..) |*handler, i| {
+                handler.* = switch (@as(u3, i >> 7 & 0x7)) {
                     0b000 => if (i >> 5 & 0x3 == 0b11) blk: {
                         const I = i >> 4 & 1 == 1;
                         const is_sub = i >> 3 & 1 == 1;
@@ -230,8 +228,8 @@ pub const thumb = struct {
                 };
             }
 
-            return ret;
-        };
+            return table;
+        }
     }
 };
 
@@ -385,8 +383,7 @@ pub const Arm7tdmi = struct {
         const now = getModeChecked(self, self.cpsr.mode.read());
 
         // Bank R8 -> r12
-        var i: usize = 0;
-        while (i < 5) : (i += 1) {
+        for (0..5) |i| {
             self.bank.fiq[Bank.fiqIdx(i, now)] = self.r[8 + i];
         }
 
@@ -404,8 +401,7 @@ pub const Arm7tdmi = struct {
         }
 
         // Grab R8 -> R12
-        i = 0;
-        while (i < 5) : (i += 1) {
+        for (0..5) |i| {
             self.r[8 + i] = self.bank.fiq[Bank.fiqIdx(i, next)];
         }
 
@@ -470,8 +466,7 @@ pub const Arm7tdmi = struct {
     }
 
     pub fn stepDmaTransfer(self: *Self) bool {
-        comptime var i: usize = 0;
-        inline while (i < 4) : (i += 1) {
+        inline for (0..4) |i| {
             if (self.bus.dma[i].in_progress) {
                 self.bus.dma[i].step(self);
                 return true;

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 while (i < 8) : (i += 1) {
+    inline for (0..8) |i| {
         if (opcode >> (7 - i) & 1 == 1) count += 1;
     }
 

src/core/emu.zig

@@ -146,9 +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
@@ -163,3 +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.zig

@@ -625,8 +625,7 @@ pub const Ppu = struct {
                 const framebuf_base = width * @as(usize, scanline);
                 if (obj_enable) self.fetchSprites();
 
-                var layer: usize = 0;
-                while (layer < 4) : (layer += 1) {
+                for (0..4) |layer| {
                     self.drawSprites(@truncate(u2, layer));
                     if (layer == self.bg[0].cnt.priority.read() and bg_enable & 1 == 1) self.drawBackground(0);
                     if (layer == self.bg[1].cnt.priority.read() and bg_enable >> 1 & 1 == 1) self.drawBackground(1);
@@ -640,8 +639,7 @@ pub const Ppu = struct {
                 const framebuf_base = width * @as(usize, scanline);
                 if (obj_enable) self.fetchSprites();
 
-                var layer: usize = 0;
-                while (layer < 4) : (layer += 1) {
+                for (0..4) |layer| {
                     self.drawSprites(@truncate(u2, layer));
                     if (layer == self.bg[0].cnt.priority.read() and bg_enable & 1 == 1) self.drawBackground(0);
                     if (layer == self.bg[1].cnt.priority.read() and bg_enable >> 1 & 1 == 1) self.drawBackground(1);
@@ -654,8 +652,7 @@ pub const Ppu = struct {
                 const framebuf_base = width * @as(usize, scanline);
                 if (obj_enable) self.fetchSprites();
 
-                var layer: usize = 0;
-                while (layer < 4) : (layer += 1) {
+                for (0..4) |layer| {
                     self.drawSprites(@truncate(u2, layer));
                     if (layer == self.bg[2].cnt.priority.read() and bg_enable >> 2 & 1 == 1) self.drawAffineBackground(2);
                     if (layer == self.bg[3].cnt.priority.read() and bg_enable >> 3 & 1 == 1) self.drawAffineBackground(3);
@@ -671,7 +668,7 @@ pub const Ppu = struct {
                 const vram_buf = @ptrCast([*]const u16, @alignCast(@alignOf(u16), self.vram.buf));
                 const framebuf = @ptrCast([*]u32, @alignCast(@alignOf(u32), self.framebuf.get(.Emulator)));
 
-                for (vram_buf[vram_base .. vram_base + width]) |bgr555, i| {
+                for (vram_buf[vram_base .. vram_base + width], 0..) |bgr555, i| {
                     framebuf[framebuf_base + i] = rgba888(bgr555);
                 }
             },
@@ -685,7 +682,7 @@ pub const Ppu = struct {
                 const pal_buf = @ptrCast([*]const u16, @alignCast(@alignOf(u16), self.palette.buf));
                 const framebuf = @ptrCast([*]u32, @alignCast(@alignOf(u32), self.framebuf.get(.Emulator)));
 
-                for (self.vram.buf[vram_base .. vram_base + width]) |pal_id, i| {
+                for (self.vram.buf[vram_base .. vram_base + width], 0..) |pal_id, i| {
                     framebuf[framebuf_base + i] = rgba888(pal_buf[pal_id]);
                 }
             },
@@ -701,8 +698,7 @@ pub const Ppu = struct {
                 const vram_buf = @ptrCast([*]const u16, @alignCast(@alignOf(u16), self.vram.buf));
                 const framebuf = @ptrCast([*]u32, @alignCast(@alignOf(u32), self.framebuf.get(.Emulator)));
 
-                var i: usize = 0;
-                while (i < width) : (i += 1) {
+                for (0..width) |i| {
                     const bgr555 = if (scanline < m5_height and i < m5_width) vram_buf[vram_base + i] else self.palette.backdrop();
                     framebuf[framebuf_base + i] = rgba888(bgr555);
                 }
@@ -718,7 +714,7 @@ pub const Ppu = struct {
         // FIXME: @ptrCast between slices changing the length isn't implemented yet
         const framebuf = @ptrCast([*]u32, @alignCast(@alignOf(u32), self.framebuf.get(.Emulator)));
 
-        for (self.scanline.top()) |maybe_top, i| {
+        for (self.scanline.top(), 0..) |maybe_top, i| {
             const maybe_btm = self.scanline.btm()[i];
 
             const bgr555 = self.getBgr555(maybe_top, maybe_btm);
@@ -990,8 +986,7 @@ pub const Ppu = struct {
             cpu.handleInterrupt();
         }
 
-        // See if HBlank DMA is present and not enabled
-
+        // If we're not also in VBlank, attempt to run any pending DMA Reqs
         if (!self.dispstat.vblank.read())
             dma.onBlanking(cpu.bus, .HBlank);
 

src/core/scheduler.zig

@@ -73,7 +73,7 @@ pub const Scheduler = struct {
 
     /// Removes the **first** scheduled event of type `needle`
     pub fn removeScheduledEvent(self: *Self, needle: EventKind) void {
-        for (self.queue.items) |event, i| {
+        for (self.queue.items, 0..) |event, i| {
             if (std.meta.eql(event.kind, needle)) {
 
                 // invalidates the slice we're iterating over

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;
@@ -22,6 +25,7 @@ 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.
     \\
 );
@@ -87,13 +91,49 @@ 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();
 
-    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});
 

src/platform.zig

@@ -154,9 +154,17 @@ pub const Gui = struct {
         return tex_id;
     }
 
-    pub fn run(self: *Self, cpu: *Arm7tdmi, scheduler: *Scheduler) !void {
-        var quit = std.atomic.Atomic(bool).init(false);
-        var tracker = FpsTracker.init();
+    const RunOptions = struct {
+        quit: *std.atomic.Atomic(bool),
+        tracker: ?*FpsTracker = null,
+        cpu: *Arm7tdmi,
+        scheduler: *Scheduler,
+    };
+
+    pub fn run(self: *Self, opt: RunOptions) !void {
+        const cpu = opt.cpu;
+        const tracker = opt.tracker;
+        const quit = opt.quit;
 
         var buffer_ids = Self.generateBuffers();
         defer {
@@ -169,13 +177,15 @@ 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,
@@ -238,8 +248,10 @@ pub const Gui = struct {
             gl.drawElements(gl.TRIANGLES, 6, gl.UNSIGNED_INT, null);
             SDL.SDL_GL_SwapWindow(self.window);
 
-            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);
+            if (tracker) |t| {
+                const dyn_title = std.fmt.bufPrintZ(&title_buf, "ZBA | {s} [Emu: {}fps] ", .{ self.title, t.value() }) catch unreachable;
+                SDL.SDL_SetWindowTitle(self.window, dyn_title.ptr);
+            }
         }
 
         quit.store(true, .Monotonic); // Terminate Emulator Thread