Colliding Rigidbody2Ds cause FPS drop even if their masks/layers don't match

[KoBeWi]

Godot version:

3.2.4 rc1

Minimal reproduction project:

rigidbody_mask_bug.zip
Compare Slow.tscn and NotSlow.tscn. In both the collision layers/masks are set to 0, but if the bodies are colliding, there is an extreme FPS drop.

[HenryWConklin]

From some quick profiling, looks like the culprit is BroadPhase2DHashGrid::_check_motion

  14.91%  godot.linuxbsd.  godot.linuxbsd.tools.64             [.] BroadPhase2DHashGrid::_check_motion
  10.23%  godot.linuxbsd.  godot.linuxbsd.tools.64             [.] Rect2::intersects
   8.04%  godot.linuxbsd.  godot.linuxbsd.tools.64             [.] CollisionObject2DSW::test_collision_mask

Looks like it's not short-circuiting on the collision masks, so it's always checking the AABB collisions:

godot/servers/physics_2d/broad_phase_2d_hash_grid.cpp

Lines 72 to 92 in 2abe996

	void BroadPhase2DHashGrid::_check_motion(Element *p_elem) {
	for (Map<Element *, PairData *>::Element *E = p_elem->paired.front(); E; E = E->next()) {
	bool physical_collision = p_elem->aabb.intersects(E->key()->aabb);
	bool logical_collision = p_elem->owner->test_collision_mask(E->key()->owner);
	if (physical_collision) {
	if (!E->get()->colliding || (logical_collision && !E->get()->ud && pair_callback)) {
	E->get()->ud = pair_callback(p_elem->owner, p_elem->subindex, E->key()->owner, E->key()->subindex, pair_userdata);
	} else if (E->get()->colliding && !logical_collision && E->get()->ud && unpair_callback) {
	unpair_callback(p_elem->owner, p_elem->subindex, E->key()->owner, E->key()->subindex, E->get()->ud, unpair_userdata);
	E->get()->ud = nullptr;
	}
	E->get()->colliding = true;
	} else { // No physcial_collision
	if (E->get()->colliding && unpair_callback) {
	unpair_callback(p_elem->owner, p_elem->subindex, E->key()->owner, E->key()->subindex, E->get()->ud, unpair_userdata);
	}
	E->get()->colliding = false;
	}
	}
	}

I might pick this up tomorrow if no one else takes it.

EDIT,next day:
Turns out the slow part is actually just dereferencing E->get(), so short circuiting there doesn't really help. Solution turns out to be a bit more complicated but I think I almost have it.