[📜일지] 오늘 한일) 성능 최적화

[익명(hotkey02)]

1206번에 23.4ms? 끔찍하네

----
수정 후

2932번 호출에 8.6 ms

---

func _physics_process(delta):

var s_delta = TimeManager.scale_delta(delta)

var prev_pos := global_position

if velocity.length() > 0:

var collision: KinematicCollision2D = move_and_collide(velocity * s_delta)

if collision:

var collider = collision.get_collider()

if collider is Wall:

velocity = velocity.bounce(collision.get_normal()) * bounce

if is_being_pulled:

if _check_collect_path(prev_pos, global_position):

_collect()

return

else:

velocity = velocity.move_toward(Vector2.ZERO, friction * s_delta)

if velocity == Vector2.ZERO and spawn_scale >= 1.0:

set_physics_process(false)

이게 기존코드인데 is_being_pulled 상태에는 물리 충돌이 필요하지 않으니까 로직을 _process로 이전하고 move_and_collide 대신
global_position += velocity * delta 사용으로 수정하기로 하고 _process 로 넘어감

func _process(delta):

# 아이템 줍는거에는 일반 시간 사용

# var s_delta = TimeManager.scale_delta(delta)

if spawn_scale < 1.0:

spawn_scale = move_toward(spawn_scale, 1.0, SPAWN_SCALE_SPEED * delta)

scale = Vector2.ONE * spawn_scale

else:

if not is_being_pulled:

set_process(false)

if is_being_pulled:

if not _has_valid_target_magnet():

is_being_pulled = false

return

var target_pos = target_magnet.global_position

var distance = global_position.distance_to(target_pos)

if distance < COLLECT_RANGE or velocity.length() > FORCED_COLLECT_SPEED:

_collect()

return

if velocity.length() > 10.0:

var target_rotation = velocity.angle() + PI/2

rotation = lerp_angle(rotation, target_rotation, 2 * delta)

var direction = (target_pos - global_position).normalized()

pull_speed += PULL_STRENTGTH * delta

# 기본 속도는 계속 증가

var base_speed = pow(pull_speed, 2) * 0.01

# 가까울 때만 추가 부스트 (150px 이하일 때)

var distance_boost = 1.0

if distance < 50.0:

distance_boost += (50.0 - distance) / 50.0

var target_velocity = direction * base_speed * distance_boost

var speed_factor = min(velocity.length() / 1000.0, 1.0)

var distance_factor = 1.0 - min(distance / 200.0, 1.0)

var dynamic_turn_speed = TURN_SPEED + (speed_factor * 0.3) + (distance_factor * 0.2)

velocity = velocity.lerp(target_velocity, dynamic_turn_speed)

일단 모든 거리 관련 연산을 squared 기반으로 변경 이러면 제곱근 계산이 빠져서 성능적으로 좋아보임

var distance_sq := global_position.distance_squared_to(target_pos)

if distance_sq < COLLECT_RANGE_SQ or velocity.length_squared() > FORCED_COLLECT_SPEED_SQ:

_collect()

return

그다음 _process 수정하기 전에 _check_collect_path를 확인해보자
해당 함수는 drop의 속도가 빨라 획득하지 못하고 오버슛 했을경우를 막아주는 함수이다

func _check_collect_path(start_pos: Vector2, end_pos: Vector2) -> bool:

if is_collected or not _has_valid_target_magnet():

return false

var target_pos := target_magnet.global_position

var collect_range_sq := COLLECT_RANGE * COLLECT_RANGE

if start_pos.distance_squared_to(target_pos) <= collect_range_sq:

return true

if end_pos.distance_squared_to(target_pos) <= collect_range_sq:

return true

var segment := end_pos - start_pos

var segment_len_sq := segment.length_squared()

if segment_len_sq <= 0.0001:

return false

var t := clampf((target_pos - start_pos).dot(segment) / segment_len_sq, 0.0, 1.0)

var closest_point := start_pos + segment * t

return closest_point.distance_squared_to(target_pos) <= collect_range_sq

매번 target_magnet 을 들고오는데 _process로 이동했으니 이미 들고있는 값을 쓸 수 있을것 같다
collect_range_sq도 상수로 빼고 하면

func _is_collect_path_crossed(

start_pos: Vector2,

end_pos: Vector2,

target_pos: Vector2,

start_distance_sq: float,

end_distance_sq: float

) -> bool:

if start_distance_sq <= COLLECT_RANGE_SQ:

return true

if end_distance_sq <= COLLECT_RANGE_SQ:

return true

var segment := end_pos - start_pos

var segment_len_sq := segment.length_squared()

if segment_len_sq <= 0.0001:

return false

var t := clampf((target_pos - start_pos).dot(segment) / segment_len_sq, 0.0, 1.0)

var closest_point := start_pos + segment * t

return closest_point.distance_squared_to(target_pos) <= COLLECT_RANGE_SQ

이정도로 가능할것 같다
사용하는 값 계산은 전부 외부에서 하고 내부는 확인만 하는정도

그래서 최종 코드

func _physics_process(delta):

if is_being_pulled:

set_physics_process(false)

return # 끌려가는 중에는 물리 이동 처리 안함, _process에서 처리

var s_delta = TimeManager.scale_delta(delta)

if velocity != Vector2.ZERO:

var collision: KinematicCollision2D = move_and_collide(velocity * s_delta)

if collision:

var collider = collision.get_collider()

if collider is Wall:

velocity = velocity.bounce(collision.get_normal()) * bounce

velocity = velocity.move_toward(Vector2.ZERO, friction * s_delta)

if velocity == Vector2.ZERO and spawn_scale >= 1.0:

set_physics_process(false)

func _process(delta):

# 아이템 줍는거에는 일반 시간 사용

if spawn_scale < 1.0:

spawn_scale = move_toward(spawn_scale, 1.0, SPAWN_SCALE_SPEED * delta)

scale = Vector2.ONE * spawn_scale

else:

if not is_being_pulled:

set_process(false)

if is_being_pulled:

if not _has_valid_target_magnet():

is_being_pulled = false

return

var start_pos := global_position

var target_pos := target_magnet.global_position

var to_target := target_pos - start_pos

var distance_sq := to_target.length_squared()

var velocity_len_sq := velocity.length_squared()

if distance_sq < COLLECT_RANGE_SQ or velocity_len_sq > FORCED_COLLECT_SPEED_SQ:

_collect()

return

if velocity_len_sq > 100.0:

var target_rotation = velocity.angle() + PI / 2

rotation = lerp_angle(rotation, target_rotation, 2 * delta)

var distance := sqrt(distance_sq)

var direction := to_target / distance

pull_speed += PULL_STRENTGTH * delta

# 기본 속도는 계속 증가

var base_speed = pull_speed * pull_speed * 0.01

# 가까울 때만 추가 부스트 (150px 이하일 때)

var distance_boost = 1.0

if distance < 50.0:

distance_boost += (50.0 - distance) / 50.0

var target_velocity = direction * base_speed * distance_boost

var speed_factor = min(sqrt(velocity_len_sq) / 1000.0, 1.0)

var distance_factor = 1.0 - min(distance / 200.0, 1.0)

var dynamic_turn_speed = TURN_SPEED + (speed_factor * 0.3) + (distance_factor * 0.2)

velocity = velocity.lerp(target_velocity, dynamic_turn_speed)

global_position += velocity * delta

var end_distance_sq := global_position.distance_squared_to(target_pos)

if _is_collect_path_crossed(start_pos, global_position, target_pos, distance_sq, end_distance_sq):

_collect()

return

16gb m1 mac 에서 이정도로 돌아가면 일반 pc 에서는 지연이 체감되지 않을거라 생각함
오늘의 최적화는 여기서 끝

[211214(tomatoss)]

[엔록(whale0407)]

GOSU