fix: 8-bit writes to WIN PPU registers

Advance Wars depends on these registers similar to Mario Kart's 8-bit
writes to Affine Background registers:

src/Gui.zig

@@ -180,11 +180,11 @@ const Audio = struct {
 
     export fn callback(userdata: ?*anyopaque, stream: [*c]u8, len: c_int) void {
         const apu = @ptrCast(*Apu, @alignCast(@alignOf(*Apu), userdata));
-        const written = SDL.SDL_AudioStreamGet(apu.stream, stream, len);
+        _ = SDL.SDL_AudioStreamGet(apu.stream, stream, len);
 
         // If we don't write anything, play silence otherwise garbage will be played
         // FIXME: I don't think this hack to remove DC Offset is acceptable :thinking:
-        if (written == 0) std.mem.set(u8, stream[0..@intCast(usize, len)], 0x40);
+        // if (written == 0) std.mem.set(u8, stream[0..@intCast(usize, len)], 0x40);
     }
 };
 

src/core/apu.zig

@@ -163,6 +163,8 @@ pub const Apu = struct {
     fs: FrameSequencer,
     capacitor: f32,
 
+    is_buffer_full: bool,
+
     pub fn init(sched: *Scheduler) Self {
         const apu: Self = .{
             .ch1 = ToneSweep.init(sched),
@@ -178,11 +180,12 @@ pub const Apu = struct {
             .bias = .{ .raw = 0x0200 },
 
             .sampling_cycle = 0b00,
-            .stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, 1 << 15, SDL.AUDIO_U16, 2, host_sample_rate) orelse unreachable,
+            .stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, 1 << 15, SDL.AUDIO_U16, 2, host_sample_rate).?,
             .sched = sched,
 
             .capacitor = 0,
             .fs = FrameSequencer.init(),
+            .is_buffer_full = false,
         };
 
         sched.push(.SampleAudio, apu.sampleTicks());
@@ -277,6 +280,13 @@ pub const Apu = struct {
     }
 
     pub fn sampleAudio(self: *Self, late: u64) void {
+        self.sched.push(.SampleAudio, self.sampleTicks() -| 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;
 
@@ -325,28 +335,30 @@ pub const Apu = struct {
         left += bias;
         right += bias;
 
-        const tmp_left = std.math.clamp(@bitCast(u16, left), std.math.minInt(u11), std.math.maxInt(u11));
+        const clamped_left = std.math.clamp(@bitCast(u16, left), std.math.minInt(u11), std.math.maxInt(u11));
-        const tmp_right = std.math.clamp(@bitCast(u16, right), std.math.minInt(u11), std.math.maxInt(u11));
+        const clamped_right = std.math.clamp(@bitCast(u16, right), std.math.minInt(u11), std.math.maxInt(u11));
 
         // Extend to 16-bit signed audio samples
-        const final_left = (tmp_left << 5) | (tmp_left >> 6);
+        const ext_left = (clamped_left << 5) | (clamped_left >> 6);
-        const final_right = (tmp_right << 5) | (tmp_right >> 6);
+        const ext_right = (clamped_right << 5) | (clamped_right >> 6);
 
-        if (self.sampling_cycle != self.bias.sampling_cycle.read()) {
+        // FIXME: This rarely happens
-            const new_sample_rate = Self.sampleRate(self.bias.sampling_cycle.read());
+        if (self.sampling_cycle != self.bias.sampling_cycle.read()) self.replaceSDLResampler();
-            log.info("Sample Rate changed from {}Hz to {}Hz", .{ Self.sampleRate(self.sampling_cycle), new_sample_rate });
 
-            // Sample Rate Changed, Create a new Resampler since i can't figure out how to change
+        _ = SDL.SDL_AudioStreamPut(self.stream, &[2]u16{ ext_left, ext_right }, 2 * @sizeOf(u16));
-            // the parameters of the old one
+    }
-            const old = self.stream;
-            defer SDL.SDL_FreeAudioStream(old);
 
-            self.sampling_cycle = self.bias.sampling_cycle.read();
+    fn replaceSDLResampler(self: *Self) void {
-            self.stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, @intCast(c_int, new_sample_rate), SDL.AUDIO_U16, 2, host_sample_rate) orelse unreachable;
+        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 });
 
-        _ = SDL.SDL_AudioStreamPut(self.stream, &[2]u16{ final_left, final_right }, 2 * @sizeOf(u16));
+        // Sampling Cycle (Sample Rate) changed, Craete a new SDL Audio Resampler
-        self.sched.push(.SampleAudio, self.sampleTicks() -| late);
+        // 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), SDL.AUDIO_U16, 2, host_sample_rate).?;
     }
 
     fn sampleTicks(self: *const Self) u64 {

src/core/cpu.zig

@@ -28,9 +28,14 @@ pub const arm = struct {
     const multiply = @import("cpu/arm/multiply.zig").multiply;
     const multiplyLong = @import("cpu/arm/multiply.zig").multiplyLong;
 
+    /// Determine index into ARM InstrFn LUT
+    fn idx(opcode: u32) u12 {
+        return @truncate(u12, opcode >> 20 & 0xFF) << 4 | @truncate(u12, opcode >> 4 & 0xF);
+    }
+
     // Undefined ARM Instruction handler
     fn und(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
-        const id = armIdx(opcode);
+        const id = idx(opcode);
         cpu.panic("[CPU/Decode] ID: 0x{X:0>3} 0x{X:0>8} is an illegal opcode", .{ id, opcode });
     }
 
@@ -120,9 +125,14 @@ pub const thumb = struct {
     const swi = @import("cpu/thumb/software_interrupt.zig").fmt17;
     const branch = @import("cpu/thumb/branch.zig");
 
+    /// Determine index into THUMB InstrFn LUT 
+    fn idx(opcode: u16) u10 {
+        return @truncate(u10, opcode >> 6);
+    }
+
     /// Undefined THUMB Instruction Handler
     fn und(cpu: *Arm7tdmi, _: *Bus, opcode: u16) void {
-        const id = thumbIdx(opcode);
+        const id = idx(opcode);
         cpu.panic("[CPU/Decode] ID: 0b{b:0>10} 0x{X:0>2} is an illegal opcode", .{ id, opcode });
     }
 
@@ -420,13 +430,13 @@ pub const Arm7tdmi = struct {
             const opcode = self.fetch(u16);
             if (cpu_logging) self.logger.?.mgbaLog(self, opcode);
 
-            thumb.lut[thumbIdx(opcode)](self, self.bus, opcode);
+            thumb.lut[thumb.idx(opcode)](self, self.bus, opcode);
         } else {
             const opcode = self.fetch(u32);
             if (cpu_logging) self.logger.?.mgbaLog(self, opcode);
 
             if (checkCond(self.cpsr, @truncate(u4, opcode >> 28))) {
-                arm.lut[armIdx(opcode)](self, self.bus, opcode);
+                arm.lut[arm.idx(opcode)](self, self.bus, opcode);
             }
         }
     }
@@ -517,11 +527,11 @@ pub const Arm7tdmi = struct {
 
         if (self.cpsr.t.read()) {
             const opcode = self.bus.dbgRead(u16, self.r[15] - 4);
-            const id = thumbIdx(opcode);
+            const id = thumb.idx(opcode);
             std.debug.print("opcode: ID: 0x{b:0>10} 0x{X:0>4}\n", .{ id, opcode });
         } else {
             const opcode = self.bus.dbgRead(u32, self.r[15] - 4);
-            const id = armIdx(opcode);
+            const id = arm.idx(opcode);
             std.debug.print("opcode: ID: 0x{X:0>3} 0x{X:0>8}\n", .{ id, opcode });
         }
 
@@ -601,14 +611,6 @@ pub const Arm7tdmi = struct {
     }
 };
 
-inline fn armIdx(opcode: u32) u12 {
-    return @truncate(u12, opcode >> 20 & 0xFF) << 4 | @truncate(u12, opcode >> 4 & 0xF);
-}
-
-inline fn thumbIdx(opcode: u16) u10 {
-    return @truncate(u10, opcode >> 6);
-}
-
 pub fn checkCond(cpsr: PSR, cond: u4) bool {
     return switch (cond) {
         0x0 => cpsr.z.read(), // EQ - Equal

src/core/emu.zig

@@ -70,12 +70,21 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
     }
 }
 
-fn syncToAudio(cpu: *const Arm7tdmi) void {
+fn syncToAudio(stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
-    const stream = cpu.bus.apu.stream;
+    const sample_size = 2 * @sizeOf(u16);
-    const min_sample_count = 0x800;
+    const max_buf_size: c_int = 0x400;
 
-    // Busy Loop while we wait for the Audio system to catch up
+    // Determine whether the APU is busy right at this moment
-    while (SDL.SDL_AudioStreamAvailable(stream) > (@sizeOf(u16) * 2) * min_sample_count) {}
+    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 (!sync_audio or !still_full) break;
+    }
 }
 
 pub fn runUnsynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi, fps: ?*FpsTracker) void {
@@ -86,21 +95,21 @@ pub fn runUnsynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi,
 
         while (!quit.load(.SeqCst)) {
             runFrame(sched, cpu);
-            if (sync_audio) syncToAudio(cpu);
+            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
 
             tracker.tick();
         }
     } else {
         while (!quit.load(.SeqCst)) {
             runFrame(sched, cpu);
-            if (sync_audio) syncToAudio(cpu);
+            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
         }
     }
 }
 
 pub fn runSynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi, fps: ?*FpsTracker) void {
     log.info("Emulation thread w/ video sync", .{});
-    var timer = Timer.start() catch unreachable;
+    var timer = Timer.start() catch std.debug.panic("Failed to initialize std.timer.Timer", .{});
     var wake_time: u64 = frame_period;
 
     if (fps) |tracker| {
@@ -108,13 +117,14 @@ pub fn runSynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi, f
 
         while (!quit.load(.SeqCst)) {
             runFrame(sched, cpu);
-            const new_wake_time = syncToVideo(&timer, wake_time);
+            const new_wake_time = blockOnVideo(&timer, wake_time);
 
             // Spin to make up the difference of OS scheduler innacuracies
             // If we happen to also be syncing to audio, we choose to spin on
             // the amount of time needed for audio to catch up rather than
             // our expected wake-up time
-            if (sync_audio) syncToAudio(cpu) else spinLoop(&timer, wake_time);
+            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+            if (!sync_audio) spinLoop(&timer, wake_time);
             wake_time = new_wake_time;
 
             tracker.tick();
@@ -122,16 +132,17 @@ pub fn runSynchronized(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi, f
     } else {
         while (!quit.load(.SeqCst)) {
             runFrame(sched, cpu);
-            const new_wake_time = syncToVideo(&timer, wake_time);
+            const new_wake_time = blockOnVideo(&timer, wake_time);
-            // see above comment
-            if (sync_audio) syncToAudio(cpu) else spinLoop(&timer, wake_time);
 
+            // see above comment
+            syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+            if (!sync_audio) spinLoop(&timer, wake_time);
             wake_time = new_wake_time;
         }
     }
 }
 
-inline fn syncToVideo(timer: *Timer, wake_time: u64) u64 {
+inline fn blockOnVideo(timer: *Timer, wake_time: u64) u64 {
     // Use the OS scheduler to put the emulation thread to sleep
     const maybe_recalc_wake_time = sleep(timer, wake_time);
 
@@ -149,6 +160,8 @@ pub fn runBusyLoop(quit: *Atomic(bool), sched: *Scheduler, cpu: *Arm7tdmi) void
         runFrame(sched, cpu);
         spinLoop(&timer, wake_time);
 
+        syncToAudio(cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
+
         // Update to the new wake time
         wake_time += frame_period;
     }