Add allocator benchmarks

benchmarks/gdscript/alloc.gd

@@ -0,0 +1,57 @@
+extends Benchmark
+
+const ITERATIONS = 100_000
+
+
+func benchmark_deep_tree():
+	var rt := Node.new()
+	for i in ITERATIONS:
+		var n := Node.new()
+		n.add_child(rt)
+		rt = n
+
+	# Avoid triggering a stack overflow with rt.free()
+	while rt.get_child_count():
+		var n := rt.get_child(0)
+		rt.remove_child(n)
+		rt.free()
+		rt = n
+	rt.free()
+
+func benchmark_wide_tree():
+	var rt := Node.new()
+	for i in ITERATIONS:
+		rt.add_child(Node.new())
+	rt.free()
+
+func benchmark_fragmentation():
+	var top := Node.new()
+	for i in 5:
+		top.add_child(Node.new())
+
+	for k in 10:
+		for i in ITERATIONS:
+			# Attempt to scatter children in memory by assigning newly created nodes to a random parent
+			var idx := randi() % top.get_child_count()
+			top.get_child(idx).add_child(Node.new())
+
+		var tmp := top.get_child(0)
+		top.remove_child(tmp)
+		# Since nodes in the tree are scattered in memory,
+		# freeing subtrees this way should maximize fragmentation.
+		tmp.free()
+		top.add_child(Node.new())
+		#print("Iteration %d: %.3f MB" % [k, Performance.get_monitor(Performance.MEMORY_STATIC) / 1e6])
+
+	top.free()
+
+func benchmark_duplicate():
+	var rt := Node.new()
+	for i in 16:
+		var n := Node.new()
+		n.add_child(rt.duplicate())
+		n.add_child(rt.duplicate())
+		rt.free()
+		rt = n
+		#print("Iteration %d: %.3f MB" % [i, Performance.get_monitor(Performance.MEMORY_STATIC) / 1e6])
+	rt.free()