chore: reimplement alpha blending

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

lib/SDL.zig

@@ -1 +1 @@
-Subproject commit 00b43568854f14e3bab340a4e062776ecb44a727
+Subproject commit 2fbd4b228516bf08348a3173f1446c7e8d75540a

lib/zig-clap

@@ -1 +1 @@
-Subproject commit a1b01ffeab452790790034b8a0e97aa30bbeb800
+Subproject commit 8a38c14266d1d00e0c3aee16b2e2b69675209147

src/core/Bus.zig

@@ -197,8 +197,29 @@ fn fillTableExternalMemory(bus: *Self, addr: usize) ?*anyopaque {
     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 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 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
+        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);
 
@@ -210,29 +231,18 @@ pub fn dbgRead(self: *const Self, comptime T: type, unaligned_address: u32) T {
 
             break :blk self.openBus(T, address);
         },
-        0x02 => self.ewram.read(T, address),
+        0x02 => unreachable, // handled by fastmem
-        0x03 => self.iwram.read(T, address),
+        0x03 => unreachable, // handled by fastmem
         0x04 => self.readIo(T, address),
 
         // Internal Display Memory
-        0x05 => self.ppu.palette.read(T, address),
+        0x05 => unreachable, // handled by fastmem
-        0x06 => self.ppu.vram.read(T, address),
+        0x06 => unreachable, // handled by fastmem
-        0x07 => self.ppu.oam.read(T, address),
+        0x07 => unreachable, // handled by fastmem
 
         // External Memory (Game Pak)
         0x08...0x0D => self.pak.dbgRead(T, address),
-        0x0E...0x0F => blk: {
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
-            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),
     };
 }
@@ -352,22 +362,24 @@ fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T {
 
         // External Memory (Game Pak)
         0x08...0x0D => self.pak.read(T, address),
-        0x0E...0x0F => blk: {
+        0x0E...0x0F => self.readBackup(T, unaligned_address),
-            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 readBackup(self: *const Self, comptime T: type, unaligned_address: u32) T {
+    const value = self.pak.backup.read(unaligned_address);
+
+    const multiplier = switch (T) {
+        u32 => 0x01010101,
+        u16 => 0x0101,
+        u8 => 1,
+        else => @compileError("Backup: Unsupported read width"),
+    };
+
+    return @as(T, value) * multiplier;
+}
+
 pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) void {
     const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits;
     const page = unaligned_address >> bits;

src/core/apu/ToneSweep.zig

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

src/core/apu/device/Sweep.zig

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