Optimisation à l'aide de traitement par lots

Introduction

Game engines have to send a set of instructions to the GPU to tell the GPU what and where to draw. These instructions are sent using common instructions called APIs. Examples of graphics APIs are OpenGL, OpenGL ES, and Vulkan.

Différentes API entraînent des coûts différents lors du dessin des objets. OpenGL gère beaucoup de travail pour l'utilisateur dans le pilote GPU, au prix de draw calls plus coûteux. Par conséquent, il est souvent possible d'accélérer les applications en réduisant le nombre dedraw calls.

Note

Le batching 2D n’est actuellement pris en charge que lors de l’utilisation du rendu GLES2.

Draw calls

En 2D, nous devons indiquer au GPU de rendre une série de primitives (rectangles, lignes, polygones, etc.). La technique la plus évidente consiste à dire au GPU de rendre une primitive à la fois, en lui donnant certaines informations telles que la texture utilisée, le matériau, la position, la taille, etc. puis en disant "Dessine !" (c'est ce qu'on appelle un draw call).

While this is conceptually simple from the engine side, GPUs operate very slowly when used in this manner. GPUs work much more efficiently if you tell them to draw a number of similar primitives all in one draw call, which we will call a "batch".

It turns out that they don't just work a bit faster when used in this manner; they work a lot faster.

As Godot is designed to be a general-purpose engine, the primitives coming into the Godot renderer can be in any order, sometimes similar, and sometimes dissimilar. To match Godot's general-purpose nature with the batching preferences of GPUs, Godot features an intermediate layer which can automatically group together primitives wherever possible and send these batches on to the GPU. This can give an increase in rendering performance while requiring few (if any) changes to your Godot project.

Comment ça marche

Instructions come into the renderer from your game in the form of a series of items, each of which can contain one or more commands. The items correspond to Nodes in the scene tree, and the commands correspond to primitives such as rectangles or polygons. Some items such as TileMaps and text can contain a large number of commands (tiles and glyphs respectively). Others, such as sprites, may only contain a single command (a rectangle).

Le traitement par lots utilise deux techniques principales pour regrouper les primitives :

  • Consecutive items can be joined together.
  • Consecutive commands within an item can be joined to form a batch.

Rupture de traitement par lots

Batching can only take place if the items or commands are similar enough to be rendered in one draw call. Certain changes (or techniques), by necessity, prevent the formation of a contiguous batch, this is referred to as "breaking batching".

Batching will be broken by (amongst other things):

  • Change of texture.
  • Change of material.
  • Change of primitive type (say, going from rectangles to lines).

Note

For example, if you draw a series of sprites each with a different texture, there is no way they can be batched.

Determining the rendering order

La question se pose : si seuls des objets similaires peuvent être rassemblés dans un lot, pourquoi ne pas examiner tous les objets d'une scène, regrouper tous les objets similaires et les dessiner ensemble ?

In 3D, this is often exactly how engines work. However, in Godot's 2D renderer, items are drawn in "painter's order", from back to front. This ensures that items at the front are drawn on top of earlier items when they overlap.

This also means that if we try and draw objects on a per-texture basis, then this painter's order may break and objects will be drawn in the wrong order.

In Godot, this back-to-front order is determined by:

  • The order of objects in the scene tree.
  • The Z index of objects.
  • The canvas layer.
  • YSort nodes.

Note

You can group similar objects together for easier batching. While doing so is not a requirement on your part, think of it as an optional approach that can improve performance in some cases. See the Diagnostiques section to help you make this decision.

Une astuce

And now, a sleight of hand. Even though the idea of painter's order is that objects are rendered from back to front, consider 3 objects A, B and C, that contain 2 different textures: grass and wood.

../../_images/overlap1.png

In painter's order they are ordered:

A - wood
B - grass
C - wood

Because of the texture changes, they can't be batched and will be rendered in 3 draw calls.

Cependant, l'ordre du peintre n'est nécessaire que dans l'hypothèse où ils seront dessinés sur le dessus les uns des autres. Si nous assouplissons cette hypothèse, c'est-à-dire si aucun de ces trois objets ne se chevauchent, il n'est pas nécessaire de préserver l'ordre du peintre. Le résultat rendu sera le même. Et si nous pouvions en tirer profit ?

Réorganisation des éléments

../../_images/overlap2.png

Il s'avère que nous pouvons réorganiser les éléments. Toutefois, nous ne pouvons le faire que si les éléments remplissent les conditions d'un test de chevauchement, afin de garantir que le résultat final sera le même que s'ils n'étaient pas réorganisés. Le test de chevauchement est très bon marché en termes de performance, mais il n'est pas absolument gratuit, de sorte qu'il y a un léger coût à tester pour décider si les éléments peuvent être réorganisés. Le nombre d'éléments à tester pour réorganisation peut être défini dans les paramètres du projet (voir ci-dessous), afin d'équilibrer les coûts et les bénéfices dans votre projet.

A - wood
C - wood
B - grass

Since the texture only changes once, we can render the above in only 2 draw calls.

Lumières

Although the batching system's job is normally quite straightforward, it becomes considerably more complex when 2D lights are used. This is because lights are drawn using additional passes, one for each light affecting the primitive. Consider 2 sprites A and B, with identical texture and material. Without lights, they would be batched together and drawn in one draw call. But with 3 lights, they would be drawn as follows, each line being a draw call:

../../_images/lights_overlap.png
A
A - light 1
A - light 2
A - light 3
B
B - light 1
B - light 2
B - light 3

That is a lot of draw calls: 8 for only 2 sprites. Now, consider we are drawing 1,000 sprites. The number of draw calls quickly becomes astronomical and performance suffers. This is partly why lights have the potential to drastically slow down 2D rendering.

Cependant, si vous vous souvenez de notre astuce de magicien lors de la réorganisation des objets, il s'avère que nous pouvons utiliser la même astuce pour contourner l'ordre du peintre pour les lumières !

If A and B are not overlapping, we can render them together in a batch, so the drawing process is as follows:

../../_images/lights_separate.png
AB
AB - light 1
AB - light 2
AB - light 3

That is only 4 draw calls. Not bad, as that is a 2× reduction. However, consider that in a real game, you might be drawing closer to 1,000 sprites.

  • Before: 1000 × 4 = 4,000 draw calls.
  • After: 1 × 4 = 4 draw calls.

C'est une diminution de 1000 fois des draw calls, et cela devrait donner une énorme augmentation des performances.

Test de chevauchement

However, as with the item reordering, things are not that simple. We must first perform the overlap test to determine whether we can join these primitives. This overlap test has a small cost. Again, you can choose the number of primitives to lookahead in the overlap test to balance the benefits against the cost. With lights, the benefits usually far outweigh the costs.

Also consider that depending on the arrangement of primitives in the viewport, the overlap test will sometimes fail (because the primitives overlap and therefore shouldn't be joined). In practice, the decrease in draw calls may be less dramatic than in a perfect situation with no overlapping at all. However, performance is usually far higher than without this lighting optimization.

Light scissoring

Batching can make it more difficult to cull out objects that are not affected or partially affected by a light. This can increase the fill rate requirements quite a bit and slow down rendering. Fill rate is the rate at which pixels are colored. It is another potential bottleneck unrelated to draw calls.

In order to counter this problem (and speed up lighting in general), batching introduces light scissoring. This enables the use of the OpenGL command glScissor(), which identifies an area outside of which the GPU won't render any pixels. We can greatly optimize fill rate by identifying the intersection area between a light and a primitive, and limit rendering the light to that area only.

Light scissoring is controlled with the scissor_area_threshold project setting. This value is between 1.0 and 0.0, with 1.0 being off (no scissoring), and 0.0 being scissoring in every circumstance. The reason for the setting is that there may be some small cost to scissoring on some hardware. That said, scissoring should usually result in performance gains when you're using 2D lighting.

The relationship between the threshold and whether a scissor operation takes place is not always straightforward. Generally, it represents the pixel area that is potentially "saved" by a scissor operation (i.e. the fill rate saved). At 1.0, the entire screen's pixels would need to be saved, which rarely (if ever) happens, so it is switched off. In practice, the useful values are close to 0.0, as only a small percentage of pixels need to be saved for the operation to be useful.

The exact relationship is probably not necessary for users to worry about, but is included in the appendix out of interest: Calcul du seuil de light scissoring

Light scissoring example diagram

Bottom right is a light, the red area is the pixels saved by the scissoring operation. Only the intersection needs to be rendered.

Pré-calcul de sommet

Le shader GPU reçoit des instructions sur ce qu'il faut dessiner de 2 manières principales :

  • Shader uniforms (e.g. modulate color, item transform).
  • Vertex attributes (vertex color, local transform).

However, within a single draw call (batch), we cannot change uniforms. This means that naively, we would not be able to batch together items or commands that change final_modulate or an item's transform. Unfortunately, that happens in an awful lot of cases. For instance, sprites are typically individual nodes with their own item transform, and they may have their own color modulate as well.

Pour contourner ce problème, le traitement par lots peut "pré-calculer" certaines des uniforms dans les attributs de vertex.

  • La transformation d'élément peut être combinée avec la transformation locale et envoyée dans un attribut de sommet.
  • La couleur modulée finale peut être combinée avec les couleurs des vertex, et envoyée dans un attribut de vertex.

In most cases, this works fine, but this shortcut breaks down if a shader expects these values to be available individually rather than combined. This can happen in custom shaders.

Custom shaders

As a result of the limitation described above, certain operations in custom shaders will prevent vertex baking and therefore decrease the potential for batching. While we are working to decrease these cases, the following caveats currently apply:

  • Reading or writing COLOR or MODULATE disables vertex color baking.
  • Reading VERTEX disables vertex position baking.

Paramètres du projet

To fine-tune batching, a number of project settings are available. You can usually leave these at default during development, but it's a good idea to experiment to ensure you are getting maximum performance. Spending a little time tweaking parameters can often give considerable performance gains for very little effort. See the on-hover tooltips in the Project Settings for more information.

rendu/traitement par lots/options

  • use_batching - Turns batching on or off.
  • use_batching_in_editor Turns batching on or off in the Godot editor. This setting doesn't affect the running project in any way.
  • single_rect_fallback - This is a faster way of drawing unbatchable rectangles. However, it may lead to flicker on some hardware so it's not recommended.

rendu/traitement par lots/paramètres

  • max_join_item_commands - L'une des façons les plus importantes de réaliser le traitement par lots est de joindre ensemble des éléments adjacents appropriés (nœuds), cependant ils ne peuvent être joints que si les commandes qu'ils contiennent sont compatibles. Le système doit donc faire une recherche parmi les commandes d'un élément pour déterminer s'il peut être joint. Le coût par commande est faible, et les éléments comportant un grand nombre de commandes ne valent pas la peine d'être joints, de sorte que la meilleure valeur peut dépendre du projet.
  • colored_vertex_format_threshold - Pré-calculer les couleurs dans les sommets donne un format de sommet plus grand. Cela ne vaut pas nécessairement la peine d'être fait, à moins qu'il y ait beaucoup de changements de couleur dans un élément joint. Ce paramètre représente la proportion de commandes contenant des changements de couleur / le total des commandes, au-dessus duquel il passe aux couleurs pré-calculées.
  • batch_buffer_size - Cela détermine la taille maximale d'un lot, cela n'a pas d'effet énorme sur les performances mais peut valoir la peine d'être diminuer pour les mobiles si la RAM une préoccupation.
  • item_reordering_lookahead - La réorganisation des éléments peut aider, en particulier pour les sprites entrelacés utilisant des textures différentes. Le lookahead pour le test de chevauchement a un faible coût, donc la meilleure valeur peut changer selon le projet.

rendu/traitement par lots/lumières

  • scissor_area_threshold - Voir light scissoring.
  • max_join_items - Joining items before lighting can significantly increase performance. This requires an overlap test, which has a small cost, so the costs and benefits may be project dependent, and hence the best value to use here.

rendu/traitement par lots/débogage

  • flash_batching - This is purely a debugging feature to identify regressions between the batching and legacy renderer. When it is switched on, the batching and legacy renderer are used alternately on each frame. This will decrease performance, and should not be used for your final export, only for testing.
  • diagnose_frame - This will periodically print a diagnostic batching log to the Godot IDE / console.

rendu/traitement par lots/précision

  • uv_contract - Sur certains matériels (notamment certains appareils Android), il a été signalé que des tuiles de tilemap dessinaient légèrement en dehors de leur plage d'UV, entraînant des artefacts de bord tels que des lignes autour des tuiles. Si vous constatez ce problème, essayez d'activer contrat uv. Cela provoque une petite contraction des coordonnées UV pour compenser les erreurs de précision des appareils.
  • uv_contract_amount - Hopefully, the default amount should cure artifacts on most devices, but this value remains adjustable just in case.

Diagnostiques

Although you can change parameters and examine the effect on frame rate, this can feel like working blindly, with no idea of what is going on under the hood. To help with this, batching offers a diagnostic mode, which will periodically print out (to the IDE or console) a list of the batches that are being processed. This can help pinpoint situations where batching isn't occurring as intended, and help you fix these situations to get the best possible performance.

Lecture d'un diagnostique

canvas_begin FRAME 2604
items
    joined_item 1 refs
            batch D 0-0
            batch D 0-2 n n
            batch R 0-1 [0 - 0] {255 255 255 255 }
    joined_item 1 refs
            batch D 0-0
            batch R 0-1 [0 - 146] {255 255 255 255 }
            batch D 0-0
            batch R 0-1 [0 - 146] {255 255 255 255 }
    joined_item 1 refs
            batch D 0-0
            batch R 0-2560 [0 - 144] {158 193 0 104 } MULTI
            batch D 0-0
            batch R 0-2560 [0 - 144] {158 193 0 104 } MULTI
            batch D 0-0
            batch R 0-2560 [0 - 144] {158 193 0 104 } MULTI
canvas_end

Ceci est un diagnostique typique.

  • joined_item: A joined item can contain 1 or more references to items (nodes). Generally, joined_items containing many references is preferable to many joined_items containing a single reference. Whether items can be joined will be determined by their contents and compatibility with the previous item.
  • batch R: A batch containing rectangles. The second number is the number of rects. The second number in square brackets is the Godot texture ID, and the numbers in curly braces is the color. If the batch contains more than one rect, MULTI is added to the line to make it easy to identify. Seeing MULTI is good as it indicates successful batching.
  • batch D: A default batch, containing everything else that is not currently batched.

Default batches

The second number following default batches is the number of commands in the batch, and it is followed by a brief summary of the contents:

l - line
PL - polyline
r - rect
n - ninepatch
PR - primitive
p - polygon
m - mesh
MM - multimesh
PA - particles
c - circle
t - transform
CI - clip_ignore

You may see "dummy" default batches containing no commands; you can ignore those.

Questions fréquentes

I don't get a large performance increase when enabling batching.

  • Essayez les diagnostiques, voyez dans quelle mesure le traitement par lots se fait, et si il peut être amélioré
  • Try changing batching parameters in the Project Settings.
  • Consider that batching may not be your bottleneck (see bottlenecks).

I get a decrease in performance with batching.

  • Try the steps described above to increase the number of batching opportunities.
  • Try enabling single_rect_fallback.
  • The single rect fallback method is the default used without batching, and it is approximately twice as fast. However, it can result in flickering on some hardware, so its use is discouraged.
  • Après avoir essayé ce qui précède, si votre scène se déroule toujours moins bien, envisagez d'arrêter le traitement par lots.

I use custom shaders and the items are not batching.

  • Les shaders personnalisés peuvent poser des problèmes pour la mise en lots, voir la section sur les shaders personnalisés

I am seeing line artifacts appear on certain hardware.

  • Voir le paramétrage du projet uv_contract qui peut être utilisé pour résoudre ce problème.

I use a large number of textures, so few items are being batched.

  • Consider using texture atlases. As well as allowing batching, these reduce the need for state changes associated with changing textures.

Annexe

Calcul du seuil de light scissoring

La proportion réelle de la zone de pixels de l'écran utilisée comme seuil est la valeur scissor_area_threshold à la puissance 4.

For example, on a screen size of 1920×1080, there are 2,073,600 pixels.

At a threshold of 1,000 pixels, the proportion would be:

1000 / 2073600 = 0.00048225
0.00048225 ^ (1/4) = 0.14819

So a scissor_area_threshold of 0.15 would be a reasonable value to try.

Going the other way, for instance with a scissor_area_threshold of 0.5:

0.5 ^ 4 = 0.0625
0.0625 * 2073600 = 129600 pixels

If the number of pixels saved is greater than this threshold, the scissor is activated.