Drawing performance between 3 and 4 branches.

[boruok]

Noticed visible difference in performance drawing gui-nodes between 3 and 4 branches.

extends Node

func _process(delta: float) -> void:
	var drawcalls := Performance.get_monitor(Performance.RENDER_2D_DRAW_CALLS_IN_FRAME)
	var fps := Performance.get_monitor(Performance.TIME_FPS)
	$CanvasLayer/Label.text = "%s %s %s" % [get_child_count()-1, drawcalls, fps]

func _input(event: InputEvent) -> void:
	if event is InputEventKey:
		if event.pressed:
			var rect := get_viewport().get_visible_rect()
			for i in 500:
				var node := Button.new()
				node.rect_size = Vector2(64, 64)
				node.rect_position = Vector2(randf() * rect.size.x, randf() * rect.size.y)
				add_child(node)

Here stats after instantiating default Button into tree (Label, Panel, CheckButton, ColorRect, etc. stats are relatively same.

Renderer	Drawcalls	FPS	Nodes
GLES2	67	57	30000
GLES3	67	55	30000
Forward+	30000	35	30000
Mobile	30000	30	30000
Compatibility	30000	8	30000

stats taken on:
Pop!_OS 22.04 LTS, NVIDIA GTX1060 6Gb, 570.133.07

test_draw3.zip
test_draw4.zip

1. Why each node consumes 1 drawcall? Is batching exists?
2. Does this mean that for gui-heavy game is better stick to 3x-branch?

[Calinou]

Why each node consumes 1 drawcall? Is batching exists?

Yes, there is 2D batching in 4.4 and later in all rendering methods. In 4.3 and prior, it's only in OpenGL.

It sounds like you've reached a case where batching does not work for some reason. Remember that nodes must share some characteristics for batching to work (e.g. they must use the same shader/material, not unique copies of it).

Also, I'm not sure if batching for nine-patch rects (which StyleBoxFlat uses) is implemented in 4.x. Most GUI-heavy games use a lot of text, but not a lot of individual nine-patch rects.