tmp: progress towards unifying scheduler

This commit is contained in:
Rekai Nyangadzayi Musuka 2023-09-24 14:47:08 -05:00
parent 2f06f178e2
commit 75a1fde938
10 changed files with 123 additions and 117 deletions

View File

@ -1,6 +1,7 @@
const std = @import("std");
const Bus = @import("Bus.zig");
const nds9 = @import("nds9.zig");
const nds7 = @import("nds7.zig");
const PriorityQueue = std.PriorityQueue(Event, void, Event.lessThan);
const Allocator = std.mem.Allocator;
@ -19,6 +20,12 @@ pub fn push(self: *@This(), kind: Event.Kind, offset: u64) void {
self.queue.add(.{ .kind = kind, .tick = self.tick + offset }) catch unreachable;
}
pub inline fn peek(self: *const @This()) Event {
@setRuntimeSafety(false);
return self.queue.items[0];
}
pub fn deinit(self: @This()) void {
self.queue.deinit();
}
@ -37,18 +44,33 @@ pub fn reset(self: *@This()) void {
self.tick = 0;
}
pub fn handle(self: *@This(), bus: *Bus) void {
pub fn handle7(self: *@This(), bus: *nds7.Bus) void {
_ = bus;
const event = self.queue.remove();
// const late = self.tick - event.tick;
switch (event.kind) {
.nds7 => {},
.heat_death => unreachable,
.nds9 => unreachable, // handled by handle7
}
}
pub fn handle9(self: *@This(), bus: *nds9.Bus) void {
const event = self.queue.remove();
const late = self.tick - event.tick;
switch (event.kind) {
.heat_death => unreachable,
.nds9 => |ev| switch (ev) {
.draw => {
bus.ppu.drawScanline(bus);
bus.ppu.onHdrawEnd(self, late);
},
.hblank => bus.ppu.onHblankEnd(self, late),
.vblank => bus.ppu.onHblankEnd(self, late),
},
.heat_death => unreachable,
.nds7 => unreachable, // handled by handle7
}
}
@ -56,13 +78,20 @@ pub const Event = struct {
tick: u64,
kind: Kind,
pub const Kind = enum {
heat_death,
pub const Event7 = enum {};
pub const Event9 = enum {
draw,
hblank,
vblank,
};
pub const Kind = union(enum) {
heat_death: void,
nds9: Event9,
nds7: Event7,
};
fn lessThan(_: void, left: @This(), right: @This()) std.math.Order {
return std.math.order(left.tick, right.tick);
}

View File

@ -4,6 +4,7 @@ const nds7 = @import("nds7.zig");
const Header = @import("cartridge.zig").Header;
const SharedIo = @import("io.zig").Io;
const Scheduler = @import("Scheduler.zig");
const Arm946es = nds9.Arm946es;
const Allocator = std.mem.Allocator;
@ -68,47 +69,83 @@ const dot_clock = 5585664; // 5.585664 Hz
const arm7_clock = bus_clock;
const arm9_clock = bus_clock * 2;
pub fn runFrame(nds7_group: nds7.Group, nds9_group: nds9.Group) void {
// TODO: might be more efficient to run them both in the same loop?
{
const scheduler = nds9_group.scheduler;
pub fn runFrame(scheduler: *Scheduler, arm7: nds7.Group, arm9: nds9.Group) void {
const nds9_frame_end = blk: {
const cycles_per_dot = arm9_clock / dot_clock + 1;
comptime std.debug.assert(cycles_per_dot == 12);
const cycles_per_frame = 355 * 263 * cycles_per_dot;
const frame_end = scheduler.tick + cycles_per_frame;
break :blk scheduler.tick + cycles_per_frame;
};
const cpu = nds9_group.cpu;
const bus = nds9_group.bus;
// const nds7_frame_end = blk: {
// const cycles_per_dot = arm7_clock / dot_clock + 1;
// comptime std.debug.assert(cycles_per_dot == 6);
while (scheduler.tick < frame_end) {
cpu.step();
// const cycles_per_frame = 355 * 263 * cycles_per_dot;
// const frame_end = scheduler.tick + cycles_per_frame;
// };
if (scheduler.tick >= scheduler.next()) scheduler.handle(bus);
// TODO: Figure out Syncing
while (scheduler.tick < nds9_frame_end) {
arm7.cpu.step();
arm9.cpu.step();
arm9.cpu.step();
const next_ev = scheduler.peek();
if (scheduler.tick >= next_ev.tick) {
switch (next_ev.kind) {
.heat_death => unreachable,
.nds9 => scheduler.handle9(arm9.bus),
.nds7 => scheduler.handle7(arm7.bus),
}
}
{
const scheduler = nds7_group.scheduler;
const cycles_per_dot = arm7_clock / dot_clock + 1;
comptime std.debug.assert(cycles_per_dot == 6);
const cycles_per_frame = 355 * 263 * cycles_per_dot;
const frame_end = scheduler.tick + cycles_per_frame;
const cpu = nds7_group.cpu;
const bus = nds7_group.bus;
while (scheduler.tick < frame_end) {
cpu.step();
if (scheduler.tick >= scheduler.next()) scheduler.handle(bus);
}
}
}
// pub fn runFrame(nds7_group: nds7.Group, nds9_group: nds9.Group) void {
// // TODO: might be more efficient to run them both in the same loop?
// {
// const scheduler = nds9_group.scheduler;
// const cycles_per_dot = arm9_clock / dot_clock + 1;
// comptime std.debug.assert(cycles_per_dot == 12);
// const cycles_per_frame = 355 * 263 * cycles_per_dot;
// const frame_end = scheduler.tick + cycles_per_frame;
// const cpu = nds9_group.cpu;
// const bus = nds9_group.bus;
// while (scheduler.tick < frame_end) {
// cpu.step();
// if (scheduler.tick >= scheduler.next()) scheduler.handle(bus);
// }
// }
// {
// const scheduler = nds7_group.scheduler;
// const cycles_per_dot = arm7_clock / dot_clock + 1;
// comptime std.debug.assert(cycles_per_dot == 6);
// const cycles_per_frame = 355 * 263 * cycles_per_dot;
// const frame_end = scheduler.tick + cycles_per_frame;
// const cpu = nds7_group.cpu;
// const bus = nds7_group.bus;
// while (scheduler.tick < frame_end) {
// cpu.step();
// if (scheduler.tick >= scheduler.next()) scheduler.handle(bus);
// }
// }
// }
// FIXME: Perf win to allocating on the stack instead?
pub const SharedContext = struct {
const MiB = 0x100000;

View File

@ -2,7 +2,6 @@ const std = @import("std");
pub const Bus = @import("nds7/Bus.zig");
pub const io = @import("nds7/io.zig");
pub const Scheduler = @import("nds7/Scheduler.zig");
pub const Arm7tdmi = @import("arm32").Arm7tdmi;
const Allocator = std.mem.Allocator;
@ -11,11 +10,9 @@ const Allocator = std.mem.Allocator;
pub const Group = struct {
cpu: *Arm7tdmi,
bus: *Bus,
scheduler: *Scheduler,
/// Responsible for deallocated the ARM7 CPU, Bus and Scheduler
pub fn deinit(self: @This(), allocator: Allocator) void {
self.bus.deinit(allocator);
self.scheduler.deinit();
}
};

View File

@ -1,7 +1,7 @@
const std = @import("std");
const io = @import("io.zig");
const Scheduler = @import("Scheduler.zig");
const Scheduler = @import("../Scheduler.zig");
const SharedIo = @import("../io.zig").Io;
const SharedContext = @import("../emu.zig").SharedContext;
const forceAlign = @import("../emu.zig").forceAlign;

View File

@ -1,59 +0,0 @@
const std = @import("std");
const Bus = @import("Bus.zig");
const PriorityQueue = std.PriorityQueue(Event, void, Event.lessThan);
const Allocator = std.mem.Allocator;
tick: u64 = 0,
queue: PriorityQueue,
pub fn init(allocator: Allocator) !@This() {
var queue = PriorityQueue.init(allocator, {});
try queue.add(.{ .tick = std.math.maxInt(u64), .kind = .heat_death });
return .{ .queue = queue };
}
pub fn push(self: *@This(), kind: Event.Kind, offset: u64) void {
self.queue.add(.{ .kind = kind, .tick = self.tick + offset }) catch unreachable;
}
pub fn deinit(self: @This()) void {
self.queue.deinit();
}
pub fn now(self: @This()) u64 {
return self.tick;
}
pub fn next(self: @This()) u64 {
@setRuntimeSafety(false);
return self.queue.items[0].tick;
}
pub fn reset(self: *@This()) void {
self.tick = 0;
}
pub fn handle(self: *@This(), bus: *Bus) void {
_ = bus;
const event = self.queue.remove();
const late = self.tick - event.tick;
_ = late;
switch (event.kind) {
.heat_death => unreachable,
}
}
pub const Event = struct {
tick: u64,
kind: Kind,
pub const Kind = enum { heat_death };
fn lessThan(_: void, left: @This(), right: @This()) std.math.Order {
return std.math.order(left.tick, right.tick);
}
};

View File

@ -3,7 +3,6 @@ const std = @import("std");
pub const Bus = @import("nds9/Bus.zig");
pub const Cp15 = @import("nds9/Cp15.zig");
pub const io = @import("nds9/io.zig");
pub const Scheduler = @import("nds9/Scheduler.zig");
pub const Arm946es = @import("arm32").Arm946es;
const Allocator = std.mem.Allocator;
@ -12,11 +11,9 @@ const Allocator = std.mem.Allocator;
pub const Group = struct {
cpu: *Arm946es,
bus: *Bus,
scheduler: *Scheduler,
/// Responsible for deallocating the ARM9 CPU, Bus and Scheduler
pub fn deinit(self: @This(), allocator: Allocator) void {
self.bus.deinit(allocator);
self.scheduler.deinit();
}
};

View File

@ -2,7 +2,7 @@ const std = @import("std");
const io = @import("io.zig");
const Ppu = @import("../ppu.zig").Ppu;
const Scheduler = @import("Scheduler.zig");
const Scheduler = @import("../Scheduler.zig");
const SharedContext = @import("../emu.zig").SharedContext;
const forceAlign = @import("../emu.zig").forceAlign;
@ -27,7 +27,7 @@ pub fn init(allocator: Allocator, scheduler: *Scheduler, shared_ctx: SharedConte
@memset(vram1_mem, 0);
const dots_per_cycle = 3; // ARM946E-S runs twice as fast as the ARM7TDMI
scheduler.push(.draw, 256 * dots_per_cycle);
scheduler.push(.{ .nds9 = .draw }, 256 * dots_per_cycle);
return .{
.main = shared_ctx.main,

View File

@ -2,6 +2,7 @@ const std = @import("std");
const Allocator = std.mem.Allocator;
const nds9 = @import("nds9.zig");
const Scheduler = @import("Scheduler.zig");
pub const screen_width = 256;
pub const screen_height = 192;
@ -54,7 +55,7 @@ pub const Ppu = struct {
}
/// HDraw -> HBlank
pub fn onHdrawEnd(self: *@This(), nds9_scheduler: *nds9.Scheduler, late: u64) void {
pub fn onHdrawEnd(self: *@This(), scheduler: *Scheduler, late: u64) void {
const dots_in_hblank = 99;
std.debug.assert(self.io.dispstat.hblank.read() == false);
std.debug.assert(self.io.dispstat.vblank.read() == false);
@ -62,10 +63,10 @@ pub const Ppu = struct {
// TODO: Signal HBlank IRQ
self.io.dispstat.hblank.set();
nds9_scheduler.push(.hblank, dots_in_hblank * cycles_per_dot -| late);
scheduler.push(.{ .nds9 = .hblank }, dots_in_hblank * cycles_per_dot -| late);
}
pub fn onHblankEnd(self: *@This(), nds9_scheduler: *nds9.Scheduler, late: u64) void {
pub fn onHblankEnd(self: *@This(), scheduler: *Scheduler, late: u64) void {
const scanline_count = 192 + 71;
const prev_scanline = self.io.vcount.scanline.read();
@ -85,7 +86,7 @@ pub const Ppu = struct {
// Draw Another Scanline
const dots_in_hdraw = 256;
return nds9_scheduler.push(.draw, dots_in_hdraw * cycles_per_dot -| late);
return scheduler.push(.{ .nds9 = .draw }, dots_in_hdraw * cycles_per_dot -| late);
}
if (scanline == 192) {
@ -100,7 +101,7 @@ pub const Ppu = struct {
std.debug.assert(self.io.dispstat.vblank.read() == (scanline != 262));
const dots_in_scanline = 256 + 99;
nds9_scheduler.push(.hblank, dots_in_scanline * cycles_per_dot -| late);
scheduler.push(.{ .nds9 = .hblank }, dots_in_scanline * cycles_per_dot -| late);
}
};

View File

@ -5,6 +5,8 @@ const nds9 = @import("core/nds9.zig");
const nds7 = @import("core/nds7.zig");
const emu = @import("core/emu.zig");
const Scheduler = @import("core/Scheduler.zig");
const IBus = @import("arm32").Bus;
const IScheduler = @import("arm32").Scheduler;
const ICoprocessor = @import("arm32").Coprocessor;
@ -41,23 +43,24 @@ pub fn main() !void {
const shared_ctx = try SharedContext.init(allocator);
defer shared_ctx.deinit(allocator);
var scheduler = try Scheduler.init(allocator);
defer scheduler.deinit();
const nds9_group: nds9.Group = blk: {
var scheduler = try nds9.Scheduler.init(allocator);
var bus = try nds9.Bus.init(allocator, &scheduler, shared_ctx);
var cp15 = nds9.Cp15{};
var arm946es = nds9.Arm946es.init(IScheduler.init(&scheduler), IBus.init(&bus), ICoprocessor.init(&cp15));
break :blk .{ .cpu = &arm946es, .bus = &bus, .scheduler = &scheduler };
break :blk .{ .cpu = &arm946es, .bus = &bus };
};
defer nds9_group.deinit(allocator);
const nds7_group: nds7.Group = blk: {
var scheduler = try nds7.Scheduler.init(allocator);
var bus = try nds7.Bus.init(allocator, &scheduler, shared_ctx);
var arm7tdmi = nds7.Arm7tdmi.init(IScheduler.init(&scheduler), IBus.init(&bus));
break :blk .{ .cpu = &arm7tdmi, .bus = &bus, .scheduler = &scheduler };
break :blk .{ .cpu = &arm7tdmi, .bus = &bus };
};
defer nds7_group.deinit(allocator);
@ -67,7 +70,7 @@ pub fn main() !void {
defer ui.deinit(allocator);
ui.setTitle(rom_title);
try ui.run(nds7_group, nds9_group);
try ui.run(&scheduler, nds7_group, nds9_group);
}
fn handlePositional(result: ClapResult) ![]const u8 {

View File

@ -11,6 +11,7 @@ const imgui = @import("ui/imgui.zig");
const emu = @import("core/emu.zig");
const KeyInput = @import("core/nds9/io.zig").KeyInput;
const Scheduler = @import("core/Scheduler.zig");
const Allocator = std.mem.Allocator;
@ -86,7 +87,7 @@ pub const Ui = struct {
return SDL.SDL_GL_GetProcAddress(proc.ptr);
}
pub fn run(self: *Self, nds7_group: nds7.Group, nds9_group: nds9.Group) !void {
pub fn run(self: *Self, scheduler: *Scheduler, nds7_group: nds7.Group, nds9_group: nds9.Group) !void {
// TODO: Sort this out please
const vao_id = opengl_impl.vao();
@ -108,7 +109,7 @@ pub const Ui = struct {
var event: SDL.SDL_Event = undefined;
emu_loop: while (true) {
emu.runFrame(nds7_group, nds9_group);
emu.runFrame(scheduler, nds7_group, nds9_group);
while (SDL.SDL_PollEvent(&event) != 0) {
_ = zgui.backend.processEvent(&event);