chore: reimplement alpha blending

window wrap now works (it's pretty slow though?)

src/core/Bus.zig

@@ -107,7 +107,7 @@ pub fn deinit(self: *Self) void {
 }
 
 fn fillReadTable(bus: *Self, table: *[table_len]?*const anyopaque) void {
-    const vramMirror = @import("ppu.zig").Vram.mirror;
+    const vramMirror = @import("ppu/Vram.zig").mirror;
 
     for (table) |*ptr, i| {
         const addr = page_size * i;
@@ -134,7 +134,7 @@ fn fillReadTable(bus: *Self, 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.zig").Vram.mirror;
+    const vramMirror = @import("ppu/Vram.zig").mirror;
 
     for (table) |*ptr, i| {
         const addr = page_size * i;

src/core/apu.zig

@@ -15,10 +15,12 @@ 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);
 
@@ -244,20 +246,17 @@ pub const Apu = struct {
 
     sampling_cycle: u2,
 
-    sample_queue: RingBuffer(u16),
+    stream: *SDL.SDL_AudioStream,
     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),
@@ -272,11 +271,12 @@ pub const Apu = struct {
             .bias = .{ .raw = 0x0200 },
 
             .sampling_cycle = 0b00,
-            .sample_queue = RingBuffer(u16).init(sample_buf),
+            .stream = SDL.SDL_NewAudioStream(SDL.AUDIO_U16, 2, 1 << 15, host_format, 2, host_rate).?,
             .sched = sched,
 
             .capacitor = 0,
             .fs = FrameSequencer.init(),
+            .is_buffer_full = false,
         };
 
         sched.push(.SampleAudio, apu.interval());
@@ -370,6 +370,11 @@ 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;
 
@@ -425,7 +430,23 @@ pub const Apu = struct {
         const ext_left = (clamped_left << 5) | (clamped_left >> 6);
         const ext_right = (clamped_right << 5) | (clamped_right >> 6);
 
-        self.sample_queue.push(ext_left, ext_right) catch {};
+        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).?;
     }
 
     fn interval(self: *const Self) u64 {

src/core/apu/ToneSweep.zig

@@ -94,9 +94,10 @@ pub fn sound1CntL(self: *const Self) u8 {
 pub fn setSound1CntL(self: *Self, value: u8) void {
     const new = io.Sweep{ .raw = value };
 
-    if (!new.direction.read()) {
-        // If at least one (1) sweep calculation has been made with
-        // the negate bit set (since last trigger), disable the channel
+    if (self.sweep.direction.read() and !new.direction.read()) {
+        // Sweep Negate bit has been cleared
+        // If At least 1 Sweep Calculation has been made since
+        // the last trigger, the channel is immediately disabled
 
         if (self.sweep_dev.calc_performed) self.enabled = false;
     }

src/core/apu/device/Sweep.zig

@@ -31,6 +31,7 @@ pub fn tick(self: *Self, ch1: *ToneSweep) void {
     if (self.timer == 0) {
         const period = ch1.sweep.period.read();
         self.timer = if (period == 0) 8 else period;
+        if (!self.calc_performed) self.calc_performed = true;
 
         if (self.enabled and period != 0) {
             const new_freq = self.calculate(ch1.sweep, &ch1.enabled);
@@ -51,10 +52,7 @@ pub fn calculate(self: *Self, sweep: io.Sweep, ch_enable: *bool) u12 {
     const shadow_shifted = shadow >> sweep.shift.read();
     const decrease = sweep.direction.read();
 
-    const freq = if (decrease) blk: {
-        self.calc_performed = true;
-        break :blk shadow - shadow_shifted;
-    } else shadow + shadow_shifted;
+    const freq = if (decrease) shadow - shadow_shifted else shadow + shadow_shifted;
     if (freq > 0x7FF) ch_enable.* = false;
 
     return freq;

src/core/bus/dma.zig

@@ -338,7 +338,7 @@ fn DmaController(comptime id: u2) type {
     };
 }
 
-pub fn pollDmaOnBlank(bus: *Bus, comptime kind: DmaKind) void {
+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);

src/core/bus/io.zig

@@ -449,6 +449,8 @@ pub const BldY = extern union {
     raw: u16,
 };
 
+const u8WriteKind = enum { Hi, Lo };
+
 /// Write-only
 pub const WinH = extern union {
     x2: Bitfield(u16, 0, 8),
@@ -458,6 +460,8 @@ pub const WinH = extern union {
 
 /// Write-only
 pub const WinV = extern union {
+    const Self = @This();
+
     y2: Bitfield(u16, 0, 8),
     y1: Bitfield(u16, 8, 8),
     raw: u16,
@@ -466,20 +470,20 @@ pub const WinV = extern union {
 pub const WinIn = extern union {
     w0_bg: Bitfield(u16, 0, 4),
     w0_obj: Bit(u16, 4),
-    w0_colour: Bit(u16, 5),
+    w0_bld: Bit(u16, 5),
     w1_bg: Bitfield(u16, 8, 4),
     w1_obj: Bit(u16, 12),
-    w1_colour: Bit(u16, 13),
+    w1_bld: Bit(u16, 13),
     raw: u16,
 };
 
 pub const WinOut = extern union {
     out_bg: Bitfield(u16, 0, 4),
     out_obj: Bit(u16, 4),
-    out_colour: Bit(u16, 5),
+    out_bld: Bit(u16, 5),
     obj_bg: Bitfield(u16, 8, 4),
     obj_obj: Bit(u16, 12),
-    obj_colour: Bit(u16, 13),
+    obj_bld: Bit(u16, 13),
     raw: u16,
 };
 

src/core/emu.zig

@@ -5,7 +5,6 @@ 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;
@@ -59,7 +58,7 @@ 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.sample_queue);
+                audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
 
                 if (kind == .UnlimitedFPS) tracker.?.tick();
             }
@@ -78,7 +77,7 @@ 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.sample_queue);
+                audioSync(audio_sync, cpu.bus.apu.stream, &cpu.bus.apu.is_buffer_full);
                 if (!audio_sync) spinLoop(&timer, wake_time);
                 wake_time = new_wake_time;
 
@@ -105,13 +104,22 @@ pub fn runFrame(sched: *Scheduler, cpu: *Arm7tdmi) void {
     }
 }
 
-fn audioSync(audio_sync: bool, sample_queue: *RingBuffer(u16)) void {
+fn audioSync(audio_sync: bool, stream: *SDL.SDL_AudioStream, is_buffer_full: *bool) void {
     comptime std.debug.assert(@import("../platform.zig").sample_format == SDL.AUDIO_U16);
-    // const sample_size = 2 * @sizeOf(u16);
-    // const max_buf_size: c_int = 0x400;
+    const sample_size = 2 * @sizeOf(u16);
+    const max_buf_size: c_int = 0x400;
 
-    _ = audio_sync;
-    _ = sample_queue;
+    // 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;
+    }
 }
 
 fn videoSync(timer: *Timer, wake_time: u64) u64 {

src/core/ppu/Oam.zig

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

src/core/ppu/Palette.zig

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

src/core/ppu/Vram.zig

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

src/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 << 16;
+pub const sample_rate = 1 << 15;
 pub const sample_format = SDL.AUDIO_U16;
 
 const default_title = "ZBA";
@@ -216,7 +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: {}", .{cpu.bus.apu.sample_queue.len() / 2});
+                                log.err("Sample Count: {}", .{@intCast(u32, SDL.SDL_AudioStreamAvailable(cpu.bus.apu.stream)) / (2 * @sizeOf(u16))});
                             },
                             // SDL.SDLK_j => log.err("Scheduler Capacity: {} | Scheduler Event Count: {}", .{ scheduler.queue.capacity(), scheduler.queue.count() }),
                             SDL.SDLK_k => {},
@@ -299,15 +299,7 @@ const Audio = struct {
         const T = *Apu;
         const apu = @ptrCast(T, @alignCast(@alignOf(T), userdata));
 
-        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;
-        }
+        _ = SDL.SDL_AudioStreamGet(apu.stream, stream, len);
     }
 };
 

src/util.zig

@@ -5,6 +5,8 @@ const config = @import("config.zig");
 const Log2Int = std.math.Log2Int;
 const Arm7tdmi = @import("core/cpu.zig").Arm7tdmi;
 
+const Allocator = std.mem.Allocator;
+
 // Sign-Extend value of type `T` to type `U`
 pub fn sext(comptime T: type, comptime U: type, value: T) T {
     // U must have less bits than T
@@ -123,6 +125,7 @@ pub const io = struct {
 
 pub const Logger = struct {
     const Self = @This();
+    const FmtArgTuple = std.meta.Tuple(&.{ u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32 });
 
     buf: std.io.BufferedWriter(4096 << 2, std.fs.File.Writer),
 
@@ -181,8 +184,6 @@ 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");
 
@@ -276,109 +277,43 @@ fn HalfInt(comptime T: type) type {
     return std.meta.Int(type_info.Int.signedness, type_info.Int.bits >> 1);
 }
 
-const Mutex = std.Thread.Mutex;
+/// Double Buffering Implementation
+pub const FrameBuffer = struct {
+    const Self = @This();
 
-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);
+    pub fn deinit(self: *Self) void {
+        self.allocator.free(self.buf);
+        self.* = undefined;
+    }
 
-            std.debug.assert(std.math.isPowerOfTwo(buf.len)); // capacity must be a power of two
-            std.debug.assert(buf.len <= max_capacity);
+    pub fn swap(self: *Self) void {
+        self.current = ~self.current;
+    }
 
-            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 pop(self: *Self) ?T {
-            self.mutex.lock();
-            defer self.mutex.unlock();
-
-            return self._pop();
-        }
-
-        pub fn len(self: *Self) Index {
-            self.mutex.lock();
-            defer self.mutex.unlock();
-
-            return self._len();
-        }
-
-        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());
-}
+    pub fn get(self: *Self, comptime dev: Device) []u8 {
+        return self.layers[if (dev == .Emulator) self.current else ~self.current];
+    }
+};