chore: add non-working pipeline

src/bus/io.zig

@@ -163,9 +163,9 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
             0x0400_0034 => bus.ppu.aff_bg[1].writePcPd(value),
             0x0400_0038 => bus.ppu.aff_bg[1].setX(bus.ppu.dispstat.vblank.read(), value),
             0x0400_003C => bus.ppu.aff_bg[1].setY(bus.ppu.dispstat.vblank.read(), value),
-            0x0400_0040 => log.debug("Wrote 0x{X:0>8} to WIN0H and WIN1H", .{value}),
-            0x0400_0044 => log.debug("Wrote 0x{X:0>8} to WIN0V and WIN1V", .{value}),
-            0x0400_0048 => log.debug("Wrote 0x{X:0>8} to WININ and WINOUT", .{value}),
+            0x0400_0040 => bus.ppu.win.setH(value),
+            0x0400_0044 => bus.ppu.win.setV(value),
+            0x0400_0048 => bus.ppu.win.setIo(value),
             0x0400_004C => log.debug("Wrote 0x{X:0>8} to MOSAIC", .{value}),
             0x0400_0050 => {
                 bus.ppu.bldcnt.raw = @truncate(u16, value);
@@ -245,12 +245,12 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
             0x0400_003A => bus.ppu.aff_bg[1].x = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].x) & 0x0000_FFFF | (@as(u32, value) << 16)),
             0x0400_003C => bus.ppu.aff_bg[1].y = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].y) & 0xFFFF_0000 | value),
             0x0400_003E => bus.ppu.aff_bg[1].y = @bitCast(i32, @bitCast(u32, bus.ppu.aff_bg[1].y) & 0x0000_FFFF | (@as(u32, value) << 16)),
-            0x0400_0040 => log.debug("Wrote 0x{X:0>4} to WIN0H", .{value}),
-            0x0400_0042 => log.debug("Wrote 0x{X:0>4} to WIN1H", .{value}),
-            0x0400_0044 => log.debug("Wrote 0x{X:0>4} to WIN0V", .{value}),
-            0x0400_0046 => log.debug("Wrote 0x{X:0>4} to WIN1V", .{value}),
-            0x0400_0048 => log.debug("Wrote 0x{X:0>4} to WININ", .{value}),
-            0x0400_004A => log.debug("Wrote 0x{X:0>4} to WINOUT", .{value}),
+            0x0400_0040 => bus.ppu.win.h[0].raw = value,
+            0x0400_0042 => bus.ppu.win.h[1].raw = value,
+            0x0400_0044 => bus.ppu.win.v[0].raw = value,
+            0x0400_0046 => bus.ppu.win.v[1].raw = value,
+            0x0400_0048 => bus.ppu.win.in.raw = value,
+            0x0400_004A => bus.ppu.win.out.raw = value,
             0x0400_004C => log.debug("Wrote 0x{X:0>4} to MOSAIC", .{value}),
             0x0400_0050 => bus.ppu.bldcnt.raw = value,
             0x0400_0052 => bus.ppu.bldalpha.raw = value,
@@ -302,9 +302,9 @@ pub fn write(bus: *Bus, comptime T: type, address: u32, value: T) void {
             0x0400_0009 => bus.ppu.bg[0].cnt.raw = (@as(u16, value) << 8) | (bus.ppu.bg[0].cnt.raw & 0xFF),
             0x0400_000A => bus.ppu.bg[1].cnt.raw = (bus.ppu.bg[1].cnt.raw & 0xFF00) | value,
             0x0400_000B => bus.ppu.bg[1].cnt.raw = (@as(u16, value) << 8) | (bus.ppu.bg[1].cnt.raw & 0xFF),
-            0x0400_0048 => log.debug("Wrote 0x{X:0>2} to WININ_L", .{value}),
-            0x0400_0049 => log.debug("Wrote 0x{X:0>2} to WININ_H", .{value}),
-            0x0400_004A => log.debug("Wrote 0x{X:0>2} to WINOUT_L", .{value}),
+            0x0400_0048 => bus.ppu.win.setInL(value),
+            0x0400_0049 => bus.ppu.win.setInH(value),
+            0x0400_004A => bus.ppu.win.setOutL(value),
             0x0400_0054 => bus.ppu.bldy.raw = (bus.ppu.bldy.raw & 0xFF00) | value,
 
             // Sound
@@ -464,6 +464,40 @@ pub const BldY = extern union {
     raw: u16,
 };
 
+/// Write-only
+pub const WinH = extern union {
+    x2: Bitfield(u16, 0, 8),
+    x1: Bitfield(u16, 8, 8),
+    raw: u16,
+};
+
+/// Write-only
+pub const WinV = extern union {
+    y2: Bitfield(u16, 0, 8),
+    y1: Bitfield(u16, 8, 8),
+    raw: u16,
+};
+
+pub const WinIn = extern union {
+    w0_bg: Bitfield(u16, 0, 4),
+    w0_obj: Bit(u16, 4),
+    w0_colour: Bit(u16, 5),
+    w1_bg: Bitfield(u16, 8, 4),
+    w1_obj: Bit(u16, 12),
+    w1_colour: 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),
+    obj_bg: Bitfield(u16, 8, 4),
+    obj_obj: Bit(u16, 12),
+    obj_colour: Bit(u16, 13),
+    raw: u16,
+};
+
 /// Read / Write
 const InterruptRequest = extern union {
     vblank: Bit(u16, 0),

src/cpu.zig

@@ -60,6 +60,7 @@ pub const Arm7tdmi = struct {
     const Self = @This();
 
     r: [16]u32,
+    pipe: [2]u32,
     sched: *Scheduler,
     bus: Bus,
     cpsr: PSR,
@@ -82,6 +83,7 @@ pub const Arm7tdmi = struct {
     pub fn init(alloc: Allocator, sched: *Scheduler, paths: FilePaths) !Self {
         return Self{
             .r = [_]u32{0x00} ** 16,
+            .pipe = [_]u32{0xF000_0000} ** 2,
             .sched = sched,
             .bus = try Bus.init(alloc, sched, paths),
             .cpsr = .{ .raw = 0x0000_001F },
@@ -256,14 +258,14 @@ pub const Arm7tdmi = struct {
     }
 
     pub fn step(self: *Self) void {
-        if (self.cpsr.t.read()) {
-            const opcode = self.fetch(u16);
+        const opcode = self.pipe[0];
         if (enable_logging) if (self.log_file) |file| self.debug_log(file, opcode);
 
-            thumb_lut[thumbIdx(opcode)](self, &self.bus, opcode);
+        if (self.cpsr.t.read()) {
+            self.stepPipeline(u16);
+            thumb_lut[thumbIdx(@truncate(u16, opcode))](self, &self.bus, @truncate(u16, opcode));
         } else {
-            const opcode = self.fetch(u32);
-            if (enable_logging) if (self.log_file) |file| self.debug_log(file, opcode);
+            self.stepPipeline(u32);
 
             if (checkCond(self.cpsr, @truncate(u4, opcode >> 28))) {
                 arm_lut[armIdx(opcode)](self, &self.bus, opcode);
@@ -326,6 +328,27 @@ pub const Arm7tdmi = struct {
         }
     }
 
+    inline fn stepPipeline(self: *Self, comptime T: type) void {
+        self.pipe[0] = self.pipe[1]; // Decoded Instr will be executed next
+        self.pipe[1] = self.fetch(T); // Fetch new opcode to be decoded next
+    }
+
+    pub fn reloadPipeline(self: *Self, comptime T: type) void {
+        switch (T) {
+            u32 => {
+                self.pipe[0] = self.bus.read(u32, self.r[15]);
+                self.pipe[1] = self.bus.read(u32, self.r[15] + 4);
+                self.r[15] += 8;
+            },
+            u16 => {
+                self.pipe[0] = self.bus.read(u32, self.r[15]);
+                self.pipe[1] = self.bus.read(u32, self.r[15] + 2);
+                self.r[15] += 4;
+            },
+            else => @compileError("Pipeline Reload: unsupported read width"),
+        }
+    }
+
     inline fn fetch(self: *Self, comptime T: type) T {
         comptime std.debug.assert(T == u32 or T == u16); // Opcode may be 32-bit (ARM) or 16-bit (THUMB)
         defer self.r[15] += if (T == u32) 4 else 2;
@@ -461,7 +484,7 @@ pub const Arm7tdmi = struct {
                 log_str = try std.fmt.bufPrint(&buf, thumb_fmt, .{ r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15, c_psr, opcode });
             }
         } else {
-            log_str = try std.fmt.bufPrint(&buf, arm_fmt, .{ r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15, c_psr, opcode });
+            log_str = try std.fmt.bufPrint(&buf, arm_fmt, .{ r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15 - 4, c_psr, opcode });
         }
 
         _ = try file.writeAll(log_str);

src/cpu/arm/branch.zig

@@ -9,14 +9,24 @@ const sext = @import("../../util.zig").sext;
 pub fn branch(comptime L: bool) InstrFn {
     return struct {
         fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
-            if (L) cpu.r[14] = cpu.r[15];
-            cpu.r[15] = cpu.fakePC() +% (sext(u32, u24, opcode) << 2);
+            if (L) cpu.r[14] = cpu.r[15] - 4;
+
+            cpu.r[15] = cpu.r[15] +% (sext(u32, u24, opcode) << 2);
+            cpu.reloadPipeline(u32);
         }
     }.inner;
 }
 
 pub fn branchAndExchange(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
     const rn = opcode & 0xF;
-    cpu.cpsr.t.write(cpu.r[rn] & 1 == 1);
-    cpu.r[15] = cpu.r[rn] & 0xFFFF_FFFE;
+
+    if (cpu.r[rn] & 1 == 1) {
+        cpu.r[15] = cpu.r[rn] & ~@as(u32, 1);
+        cpu.cpsr.t.set();
+        cpu.reloadPipeline(u16);
+    } else {
+        cpu.r[15] = cpu.r[rn] & ~@as(u32, 3);
+        cpu.cpsr.t.unset();
+        cpu.reloadPipeline(u32);
+    }
 }

src/main.zig

@@ -24,7 +24,7 @@ const expected_rate = @import("emu.zig").frame_rate;
 
 const sample_rate = @import("apu.zig").host_sample_rate;
 
-pub const enable_logging: bool = false;
+pub const enable_logging: bool = true;
 const is_binary: bool = false;
 const log = std.log.scoped(.GUI);
 pub const log_level = if (builtin.mode != .Debug) .info else std.log.default_level;
@@ -72,7 +72,7 @@ pub fn main() anyerror!void {
     var cpu = try Arm7tdmi.init(alloc, &scheduler, paths);
     defer cpu.deinit();
     cpu.bus.attach(&cpu);
-    // cpu.fastBoot(); // Uncomment to skip BIOS
+    cpu.fastBoot(); // Uncomment to skip BIOS
 
     // Copy ROM title while Emulator still belongs to this thread
     const title = cpu.bus.pak.title;

src/ppu.zig

@@ -24,6 +24,7 @@ pub const Ppu = struct {
 
     // Registers
 
+    win: Window,
     bg: [4]Background,
     aff_bg: [2]AffineBackground,
 
@@ -64,6 +65,7 @@ pub const Ppu = struct {
             .alloc = alloc,
 
             // Registers
+            .win = Window.init(),
             .bg = [_]Background{Background.init()} ** 4,
             .aff_bg = [_]AffineBackground{AffineBackground.init()} ** 2,
             .dispcnt = .{ .raw = 0x0000 },
@@ -771,6 +773,53 @@ const Oam = struct {
     }
 };
 
+const Window = struct {
+    const Self = @This();
+
+    h: [2]io.WinH,
+    v: [2]io.WinV,
+
+    out: io.WinOut,
+    in: io.WinIn,
+
+    fn init() Self {
+        return .{
+            .h = [_]io.WinH{.{ .raw = 0 }} ** 2,
+            .v = [_]io.WinV{.{ .raw = 0 }} ** 2,
+
+            .out = .{ .raw = 0 },
+            .in = .{ .raw = 0 },
+        };
+    }
+
+    pub fn setH(self: *Self, value: u32) void {
+        self.h[0].raw = @truncate(u16, value);
+        self.h[1].raw = @truncate(u16, value >> 16);
+    }
+
+    pub fn setV(self: *Self, value: u32) void {
+        self.v[0].raw = @truncate(u16, value);
+        self.v[1].raw = @truncate(u16, value >> 16);
+    }
+
+    pub fn setIo(self: *Self, value: u32) void {
+        self.in.raw = @truncate(u16, value);
+        self.out.raw = @truncate(u16, value >> 16);
+    }
+
+    pub fn setInL(self: *Self, value: u8) void {
+        self.in.raw = (self.in.raw & 0xFF00) | value;
+    }
+
+    pub fn setInH(self: *Self, value: u8) void {
+        self.in.raw = (self.in.raw & 0x00FF) | (@as(u16, value) << 8);
+    }
+
+    pub fn setOutL(self: *Self, value: u8) void {
+        self.out.raw = (self.out.raw & 0xFF00) | value;
+    }
+};
+
 const Background = struct {
     const Self = @This();