allocator: std.mem.Allocator = undefined, timeLeft: f64 = 5.0, pub var NeonObjectTable: core.EngineObjectVTable = core.EngineObjectVTable.from(@This(), "Headless"); pub fn create(allocator: std.mem.Allocator) !*@This() { const self = try allocator.create(@This()); self.* = .{ .allocator = allocator, }; return self; } pub fn prepare(self: *@This()) !void { _ = self; for(0..120_000) |i| { _ = i; const e = try core.createEntity(); const x = e.addComponent(core.Scene).?; _ = x; } // try self.testThing(); } pub fn tick(self: *@This(), dt: f64) void { // std.Thread.sleep(1000_000); self.timeLeft -= dt; const start = core.getEngineTime(); core.parallelJob(UpdateWorldTransformsJob{}, false, 1) catch unreachable; // wait 10 ms while (core.getEngineTime() - start < 0.010) { // std.Thread.sleep(10_000_000); } if (self.timeLeft < 0) core.exitNow(); } pub fn deinit(self: *@This()) void { self.allocator.destroy(self); } const Src = std.builtin.SourceLocation; pub fn Barrier(src: Src) type { return struct { pub const Src = src; pub var count: u32 = 0; pub var generation: u32 = 0; pub var mutex: std.Thread.Mutex = .{}; pub var cv: std.Thread.Condition = .{}; pub fn sync(threadId: u32, barrierCount: u32) !void { _ = threadId; mutex.lock(); defer mutex.unlock(); count += 1; if (count == barrierCount) { // 2 second timeout count = 0; generation += 1; cv.broadcast(); } else { try cv.timedWait(&mutex, 1000 * 1000 * 1000 * 2); } } }; } const MultiJob = struct { threadId: u32 = 0, threadCount: u32 = 1, threadName: []const u8, pub fn func(ctx: @This(), job: *core.JobContext) void { _ = job; core.tracy.SetThreadName(@ptrCast(ctx.threadName.ptr)); ctx.loop() catch {}; if (core.getEngine().isShuttingDown()) {} Barrier(@src()).sync(ctx.threadId, ctx.threadCount) catch unreachable; } pub fn loop(ctx: @This()) !void { while (!core.getEngine().isShuttingDown()) { try Barrier(@src()).sync(ctx.threadId, ctx.threadCount); try ctx.tick(); if (ctx.threadId == 0) { const z = core.tracy.ZoneN(@src(), "Thread 0 sleep"); defer z.End(); core.waitSeconds(0.01); } } } pub fn tick(ctx: @This()) !void { const z = core.tracy.ZoneN(@src(), "MultiJob Tick"); defer z.End(); // std.Thread.sleep(100_000 * ctx.threadId); try Barrier(@src()).sync(ctx.threadId, ctx.threadCount); } }; const UpdateWorldTransformsJob = struct { world: ?*anyopaque = null, pub fn func(self: @This(), thread: *core.ThreadContext) void { const z = core.tracy.ZoneN(@src(), "Transform Hierarchy - wide"); defer z.End(); thread.barrier(@src()) catch unreachable; var outputs: std.ArrayList(core.math.Transform) = .{}; var outputList: std.ArrayList(usize) = .{}; self.updateTransformsHierarchy(thread, &outputs, &outputList) catch |err| switch (err) { error.OutOfMemory => { unreachable; }, // else => { // thread.abort(@src(), "unknown error", err); // return; // }, }; // sync results of output List thread.barrier(@src()) catch unreachable; // write out all results, there likely is no issue with false sharing as transforms are exactly 64 bytes // const dense = self.world.denseScenes(); // for (outputList.items) |i| { // // dense[i].transform = outputs.items[i]; // } } pub fn updateTransformsHierarchy(self: @This(), thread: *core.ThreadContext, outputs: *std.ArrayList(core.math.Transform), outputList: *std.ArrayList(usize)) !void { _ = outputList; _ = self; //const dense = self.world.denseScenes(); const densePosRot = core.Scene.SceneObjectContainer.denseItems(.posRot); // everything allocated with thread.scratch is blown away when the thread is complete try outputs.resize(thread.scratch(), densePosRot.len); // scan and mark all root nodes for update const split = thread.splitSlice(core.scene.ScenePosRot, densePosRot); const locals:[]core.math.Transform = try thread.scratch().alloc(core.math.Transform, split.slice.len); for (split.slice, 0..) |posRot, i| { const index = split.startIndex + i; _ = index; locals[i] = posRot.toTransform(); } //for (split.slice, 0..) |repr, i| { //} } }; // new idea that im thinking i want to do... // // game is now responsible for scheduling order of work // // eg. prepare_game now has to return a struct that defines a list of phases. eg. the default list looks like, // // engine.setEngineTickPhases(&.{ // core.tick, // // ecs.tick // }); // // core. comes with a bunch of prebuilt phases. pub fn testThing(self: *@This()) !void { _ = self; const workerCount = 16; core.setBarrierCount(workerCount); for (0..workerCount) |i| { const name = try std.fmt.allocPrintSentinel(std.heap.c_allocator, "MultiJob{d}", .{i}, 0); try core.dispatchJob(MultiJob{ .threadName = name, .threadId = @intCast(i), .threadCount = workerCount }); } // workerCountHint=0 // async==false // will block until the job is complete, the active thread will also construct a threadContext that picks up one of the parallel tasks (todo implement workstealing) // try core.dispatchMulti(UpdateWorldTransformsJob{ .world = self.mainWorld }, null, false); // null == use max workers, } pub fn main() anyerror!void { var spec = try backlog.getSpec("headless"); // try spec.put("useGPA", .{ .boolean = false }); _ = backlog.startEngine(&NeonObjectTable, &spec); } const std = @import("std"); const backlog = @import("backlog"); const core = backlog.core;