This page gives a brief overview of how images are drawn using OpenGL and how libGDX simplifies and optimizes the task through the SpriteBatch class.

Drawing images

An image that has been decoded from its original format (e.g., PNG) and uploaded to the GPU is called a texture. To draw a texture, geometry is described and the texture is applied by specifying where each vertex in the geometry corresponds on the texture. For example, the geometry could be a rectangle and the texture could be applied so that each corner of the rectangle corresponds to a corner of the texture. A rectangle that is a subset of a texture is called a texture region.

To do the actual drawing, first the texture is bound (i.e., made the current texture), then the geometry is given to OpenGL to draw. The size and position on the screen that the texture is drawn is determined by both the geometry and how the OpenGL viewport is configured. Many 2D games configure the viewport to match the screen resolution. This means that the geometry is specified in pixels, which makes it easy to draw textures in the appropriate size and position on the screen.

It is very common to draw a texture mapped to rectangular geometry. It is also very common to draw the same texture or various regions of that texture many times. It would be inefficient to send each rectangle one at a time to the GPU to be drawn. Instead, many rectangles for the same texture can be described and sent to the GPU all at once. This is what the SpriteBatch class does.

SpriteBatch is given a texture and coordinates for each rectangle to be drawn. It collects the geometry without submitting it to the GPU. If it is given a texture different than the last texture, then it binds the last texture, submits the collected geometry to be drawn, and begins collecting geometry for the new texture.

Changing textures every few rectangles that are drawn prevents SpriteBatch from batching much geometry. Also, binding a texture is a somewhat expensive operation. For these reasons, it is common to store many smaller images in a larger image and then draw regions of the larger image to both maximize geometry batching and avoid texture changes. See TexturePacker for more information.

SpriteBatch

Using SpriteBatch (source) in an application looks like this:

public class Game implements ApplicationAdapter {
    private SpriteBatch batch;
    public void create () {
        batch = new SpriteBatch();
    }
    public void render () {
        Gdx.gl.glClear(GL20.GL_COLOR_BUFFER_BIT); // This cryptic line clears the screen.
        batch.begin();
        // Drawing goes here!
        batch.end();
    }
}

All SpriteBatch drawing calls must be made between the begin and end methods. Non-SpriteBatch drawing cannot occur between begin and end.

SpriteBatch assumes the active texture unit is 0. When using custom shaders and binding textures yourself, you can reset this with the following code:

Gdx.gl.glActiveTexture(GL20.GL_TEXTURE0);

Texture

The Texture class decodes an image file and loads it into GPU memory. The image file should be placed in the “assets” folder. The image’s dimensions should be powers of two (16x16, 64x256, etc) for compatibility and performance reasons.

private Texture texture;
...
texture = new Texture(Gdx.files.internal("image.png"));
...
batch.begin();
batch.draw(texture, 10, 10);
batch.end();

Here a texture is created and passed to a SpriteBatch to be drawn. The texture will be drawn in a rectangle positioned at 10,10 with a width and height equal to the size of the texture. SpriteBatch has many methods for drawing a texture:

TextureRegion

The TextureRegion class (source) describes a rectangle inside a texture and is useful for drawing only a portion of the texture.

private TextureRegion region;
...
texture = new Texture(Gdx.files.internal("image.png"));
region = new TextureRegion(texture, 20, 20, 50, 50);
...
batch.begin();
batch.draw(region, 10, 10);
batch.end();

Here the 20, 20, 50, 50 describes the portion of the texture, which is then drawn at 10,10. The same can be achieved by passing the Texture and other parameters to SpriteBatch, but TextureRegion makes it convenient to have a single object that describes both.

SpriteBatch has many methods for drawing a texture region:

Sprite

The Sprite class (source) describes both a texture region, the geometry where it will be drawn, and the color it will be drawn.

private Sprite sprite;
...
texture = new Texture(Gdx.files.internal("image.png"));
sprite = new Sprite(texture, 20, 20, 50, 50);
sprite.setPosition(10, 10);
sprite.setRotation(45);
...
batch.begin();
sprite.draw(batch);
batch.end();

Here the 20, 20, 50, 50 describes the portion of the texture, which is rotated 45 degrees and then drawn at 10,10. The same can be achieved by passing the Texture or a TextureRegion and other parameters to SpriteBatch, but Sprite makes it convenient to have a single object that describes everything. Also, because Sprite stores the geometry and only recomputes it when necessary, it is slightly more efficient if the scale, rotation, or other properties are unchanged between frames.

Note that Sprite mixes model information (position, rotation, etc) with view information (the texture being drawn). This makes Sprite inappropriate when applying a design pattern that wishes to strictly separate the model from the view. In that case, using Texture or TextureRegion may make more sense.

Also note that there is no Sprite constructor that is related to the position of the Sprite. calling Sprite(Texture, int, int, int, int) does not edit the position. It is necessary to call Sprite#setPosition(float,float) or else the sprite will be drawn at the default position of 0,0.

Tinting

When a texture is drawn, it can be tinted a color:

private Texture texture;
private TextureRegion region;
private Sprite sprite;
...
texture = new Texture(Gdx.files.internal("image.png"));
region = new TextureRegion(texture, 20, 20, 50, 50);
sprite = new Sprite(texture, 20, 20, 50, 50);
sprite.setPosition(100, 10);
sprite.setColor(0, 0, 1, 1);
...
batch.begin();
batch.setColor(1, 0, 0, 1);
batch.draw(texture, 10, 10);
batch.setColor(0, 1, 0, 1);
batch.draw(region, 50, 10);
sprite.draw(batch);
batch.end();

This shows how to draw a texture, region, and sprite with a tint color. The color values here are described using RGBA values between 1 and 0. Alpha is ignored if blending is disabled.

Blending

Blending is enabled by default. This means that when a texture is drawn, translucent portions of the texture are merged with pixels already on the screen at that location.

When blending is disabled, anything already on the screen at that location is replaced by the texture. This is more efficient, so blending should always be disabled unless it is needed. E.g., when drawing a large background image over the whole screen, a performance boost can be gained by first disabling blending:

Gdx.gl.glClear(GL20.GL_COLOR_BUFFER_BIT); // This cryptic line clears the screen.
batch.begin();
batch.disableBlending();
backgroundSprite.draw(batch);
batch.enableBlending();
// Other drawing here.
batch.end();

Note: Be sure to clear the screen each frame. If this is not done, a texture with alpha can be drawn on top of itself hundreds of times, making it appear opaque. Also, some GPU architectures perform better when the screen is cleared each frame, even if opaque images are being drawn over the entire screen.

Viewport

SpriteBatch manages its own projection and transformation matrixes. When a SpriteBatch is created, it uses the current application size to setup an orthographic projection using a y-up coordinate system. When begin is called, it sets up the viewport.

Performance tuning

SpriteBatch has a constructor that sets the maximum number of sprites that can be buffered before sending to the GPU. If this is too low, it will cause extra calls to the GPU. If this is too high, the SpriteBatch will be using more memory than is necessary.

SpriteBatch has a public int field named maxSpritesInBatch. This indicates the highest number of sprites that were sent to the GPU at once over the lifetime of the SpriteBatch. Setting a very large SpriteBatch size and then checking this field can help determine the optimum SpriteBatch size. It should be sized equal to or slightly more than maxSpritesInBatch. This field may be set to zero to reset it at any time.

SpriteBatch has a public int field named renderCalls. After end is called, this field indicates how many times a batch of geometry was sent to the GPU between the last calls to begin and end. This occurs when a different texture must be bound, or when the SpriteBatch has cached enough sprites to be full. If the SpriteBatch is sized appropriately and renderCalls is large (more than maybe 15-20), it indicates that many texture binds are occurring.

SpriteBatch has an additional constructor that takes a size and a number of buffers. This is an advanced feature that causes vertex buffer objects (VBOs) to be used rather than the usual vertex arrays (VAs). A list of buffers is kept, and each render call uses the next buffer in the list (wrapping around). When maxSpritesInBatch is low and renderCalls is large, this may provide a small performance boost.