Unit CastleGLImages

Return appropriate OpenGL format and type constants for given TCastleImage descendant. If you will pass here Img that is not a descendant of one of TextureImageClassesAll or PixelsImageClasses, they will raise EImageClassNotSupportedForOpenGL.

ImageGLInternalFormat works with TGPUCompressedImage classes also, returning appropriate enum, suitable for glCompressedTexImage2D.

Exceptions raised

EImageClassNotSupportedForOpenGL

When Img class is not supported by OpenGL.

If Tex <> 0 then it does glDeleteTextures on Tex and sets Tex to 0. In other words, this is a simple wrapper over glDeleteTextures that

1. checks if Tex really should be deleted

2. sets Tex to 0 to not free it once again

Call glTexParameterf to set GL_TEXTURE_MAX_ANISOTROPY_EXT on given texture target.

Takes care to check for appropriate OpenGL extension (if not present, does nothing), and to query OpenGL limit for Anisotropy (eventually clamping provided Anisotropy down).

Part of CastleGLImages unit: texture filtering (TTextureFilter and friends).

Set current texture minification and magnification filter.

This is just a thin wrapper for calling

glTexParameteri(Target, GL_TEXTURE_MIN_FILTER, ...);
glTexParameteri(Target, GL_TEXTURE_MAG_FILTER, ...);

Return wrap GL_CLAMP_TO_EDGE in both directions.

Resize the image to a size accepted as GL_TEXTURE_2D texture size for OpenGL. It tries to resize to a larger size, not smaller, to avoid losing image information.

It also makes texture have power-of-two size, if Sizing <> tsAny (or if GLFeatures.TextureNonPowerOfTwo = False). This is a must for normal textures, used for 3D rendering (with mipmapping and such). Using OpenGL non-power-of-2 textures is not good for such usage case, some OpenGLs crash (ATI), some are ultra slow (NVidia), some cause artifacts (Mesa). OpenGL ES explicitly limits what you can do with non-power-of-2. Sample model using non-power-of-2 is in inlined_textures.wrl.

Use Sizing = tsAny only for textures that you plan to use for drawing GUI images by TDrawableImage.

Does image have proper size for 2D OpenGL texture. See ResizeForTextureSize. Note that this checks glGet(GL_MAX_TEXTURE_SIZE), so requires initialized OpenGL context.

Load new texture to OpenGL. Generates new texture number by glGenTextures, then binds this texture, and loads it's data.

Takes care of UNPACK_ALIGNMENT inside (if needed, we'll change it and later revert back, so that the texture is correctly loaded).

Sets texture minification, magnification filters and wrap parameters.

Changes currently bound texture to this one (returned).

If mipmaps will be needed (this is decided looking at Filter.Minification) we will load them too.

1. As a first try, if CompositeForMipmaps is non-nil and has mipmaps (CompositeForMipmaps.Mipmaps), we will load these mipmaps. CompositeForMipmaps must be a normal 2D texture (CompositeType = ctTexture).

   Otherwise, we'll try to generate mipmaps, using various OpenGL mechanisms.

2. We will try using GenerateMipmap functionality to generate mipmaps on GPU. If not available, for uncompressed textures, we will generate mipmaps on CPU. For compressed textures, we will change minification filter to simple minLinear and make WritelnWarning.

Parameters

GUITexture

If True, the texture is not forced to have sizes as power-of-two. For TTextureSizing, it is like tsAny instead of tsScalablePowerOf2. Also, mipmaps are always disabled, regardless of requested Filter value, since mipmaps cannot work without power-of-two.

Exceptions raised

ETextureLoadError

If texture cannot be loaded for whatever reason. This includes ECannotLoadCompressedTexture if the compressed texture cannot be loaded for whatever reason. This includes EInvalidImageForOpenGLTexture if Image class is invalid for an OpenGL texture.

EImageClassNotSupportedForOpenGL

When Image class is not supported by OpenGL.

Load OpenGL texture into already reserved texture number. It uses existing OpenGL texture number (TextureId). Everything else works exactly the same as LoadGLTexture.

You can also use this to set "default unnamed OpenGL texture" parameters by passing TextureId = 0.

Exceptions raised

ETextureLoadError

Raised in the same situations as LoadGLTexture.

EImageClassNotSupportedForOpenGL

When Image class is not supported by OpenGL.

Decompress texture image by loading it to a temporary OpenGL(ES) texture and reading back. IOW, this does decompression by using current OpenGL(ES) context.

TODO: Note that the current implementation will always raise ECannotLoadCompressedTexture on OpenGLES, e.g. on mobile. Same as SaveTextureContents. This is fixable (we need to use FBO instead of glGetTexImage on OpenGLES), please submit a bugreport if you need it.

Exceptions raised

ECannotLoadCompressedTexture

If we cannot decompress, for example because we cannot load to OpenGL this compressed texture (for example because appropriate OpenGL extensions are not available, or such).

Save OpenGL texture contents to TCastleImage. Size of the texture must match the Image size, otherwise terrible things (writing over unallocated memory) will happen.

Exceptions raised

ECannotSaveTextureContents

If we cannot get texture contents (not supported on OpenGLES).

Comfortably load a 3D texture. Think about this as doing glTexImage3D(...) for you. It also sets texture minification, magnification filters and creates mipmaps if necessary.

It checks OpenGL 3D texture size requirements, and throws exceptions if not satisfied.

It takes care about OpenGL unpack parameters. Just don't worry about it.

If Filter uses mipmaps, then all mipmap levels will be loaded.

1. As a first try, if CompositeForMipmaps is non-nil and has mipmaps (CompositeForMipmaps.Mipmaps), we will load these mipmaps. CompositeForMipmaps must be a 3D texture (CompositeType = ctVolume).

2. Otherwise, we'll generate mipmaps.

   GenerateMipmap functionality will be required for this. When it is not available on this OpenGL implementation, we will change minification filter to simple linear and make WritelnWarning. So usually you just don't have to worry about this.

Pass TextureIdForProfiler only for profiling purposes (for TextureMemoryProfiler). This procedure assumes that the texture is already bound.

Exceptions raised

ETextureLoadError

If texture cannot be loaded for whatever reason, for example it's size is not correct for OpenGL 3D texture (we cannot automatically resize 3D textures, at least for now). Or it's compressed (although we support here TEncodedImage, OpenGL doesn't have any 3D texture compression available.)

EImageClassNotSupportedForOpenGL

When Image class is not supported by OpenGL.

Part of CastleGLImages unit: cubemaps sizing and loading.

Comfortably load all six cube map texture images. Think about this as doing glTexImage2D(Side, ...) for each cube side. It takes care of (almost?) everything you need to prepare OpenGL cube map texture.

It automatically takes care to adjust the texture size to appropriate size, honoring the "power of two" requirement and the GL_MAX_CUBE_MAP_TEXTURE_SIZE_ARB limit of OpenGL. So texture image may be resized (preferably up) internally before loading. Although, if texture is compressed, we cannot resize it — so ECannotLoadCompressedTexture will be raised if texture is not appropriate size.

It takes care about OpenGL unpack parameters. Just don't worry about it.

Pass TextureIdForProfiler only for profiling purposes (for TextureMemoryProfiler). This procedure assumes that the texture is already bound.

If mipmaps are requested:

1. First of all, if CompositeForMipmaps is non-nil and has mipmaps defined, we will load them from this CompositeForMipmaps image. CompositeForMipmaps must have CompositeType = ctCubeMap.

2. Otherwise, we'll try to generate images using OpenGL GenerateMipmap.

3. As a last resort, if GenerateMipmap is not available, we will fallback to generating mipmaps on CPU.

Exceptions raised

ETextureLoadError

If texture cannot be loaded for whatever reason. This includes ECannotLoadCompressedTexture if the compressed texture cannot be loaded for whatever reason (not availble appropriate extension, not correct texture size, mipmaps requested and CompositeForMipmaps/glGenerateMipmap not available). This includes EInvalidImageForOpenGLTexture if Image class is invalid for an OpenGL texture.

EImageClassNotSupportedForOpenGL

When Image class is not supported by OpenGL.

Save the current color buffer contents to image.

The suffix "NoFlush" is there to remind you that this function grabs the current buffer contents. Usually you want to redraw the screen to the back buffer, and call this function to capture back buffer before swapping, since this is the only reliable way to capture OpenGL screen. Just use TCastleWindow.SaveScreen to do it automatically.

Version with ImageClass can save to any image format from PixelsImageClasses.

Version with TCastleImage instance just uses this instance to save the image. You must pass here already created TCastleImage instance, it's class, Width and Height will be used when saving.

Exceptions raised

EImageClassNotSupportedForOpenGL

When Image class is not supported by OpenGL.

Captures current screen as a TDrawableImage instance, ready to be drawn on 2D screen.

Are we currently doing off-screen rendering. This is True if we're between TGLRenderToTexture.RenderBegin and TGLRenderToTexture.RenderEnd.

OpenGL texture memory profiler, to detect which textures use up the GPU texture memory. Especially useful on mobile devices, where texture memory is limited and your application must really optimize texture usage. Also useful to detect texture memory leaks.

Enable it at the very beginning of the application (for example in the main unit initialization clause), like this:

TextureMemoryProfiler.Enabled := true

Then at any point during the game you can query texture usage by TTextureMemoryProfiler.Summary method. For example dump it to the log like this:

WritelnLog('Textures', TextureMemoryProfiler.Summary);

The resulting output lists the currently allocated textures, in descending order of their size. This tells you which textures are worth optimizing. Maybe scale these textures down, maybe compress them, maybe remove alpha channel or convert them to grayscale...

The profiler counts the real texture size on GPU, knowing that some textures are resized to power of 2, that textures may be compressed on GPU, that some textures have mipmaps and so on. All kinds of textures (2D, 3D, cubemaps, float, depth etc.) are covered. Every type is reported — loaded from files, generated (like by GeneratedCubeMapTexture), embedded in code (by embedding images or fonts in code)... The textures loaded from disk are described by URL, for other textures we invent special URLs like:

• embedded-font:/CastleTextureFont_CanterBold_120

• embedded-image:/PanelSeparator

• generated:/GeneratedCubeMapTexture(GenTex2)

You can also use the profiler from view3dscene, run it with "–debug-texture-memory" command-line option, load your 3D model and then use menu option "Console -> Print Texture Memory Usage".

Save / restore OpenGL pixel store for unpacking / packing given TCastleImage. Before you pass this image to some OpenGL procedures (like glDrawPixels for unpacking, glReadPixels for packing), call BeforeXxx, and later call AfterXxx to restore original state. These will take care of setting / restoring pixel alignment.

type - don't add type Delphi can't parse that correctly

We recommend using CastleRenderOptions unit to get these types. But for backward compatibility, they are also available here.

Warning: this symbol is deprecated: use TDrawableImage

Warning: this symbol is deprecated: use TDrawableImage

Warning: this symbol is deprecated: use TDrawableImage

Warning: this symbol is deprecated: use TDrawableImage

Constraints on texture size, used by ResizeToTextureSize and IsTextureSized.

Values

• tsAny: Texture size does not have to be a power of two (unless GLFeatures.TextureNonPowerOfTwo = False, in which case all textures must have power-of-two, and then tsAny may be scaled to satisfy it (but it still will not be scaled for GLTextureScale)). It is not affected by GLTextureScale.
• tsRequiredPowerOf2: Texture size must be a power of two. It is not affected by GLTextureScale, because we cannot scale it.
• tsScalablePowerOf2: Texture size must be a power of two. It is affected by GLTextureScale, we can scale it.

Color buffer to grab, used by SaveScreen_NoFlush.

Values

• cbFront
• cbBack
• cbColorAttachment0

Values

• tbColor
• tbDepth
• tbColorAndDepth
• tbNone

List of sprites.

Part of CastleGLImages unit: miscellneous stuff.

We recommend using CastleRenderOptions unit to get these constants. But for backward compatibility, they are also available here.

Part of CastleGLImages unit: texture wrapping modes.

Scaling for all textures loaded to OpenGL. This allows you to conserve GPU memory. Each size (width, height, and (for 3D textures) depth) is scaled by 1 / 2ˆGLTextureScale. So value = 1 means no scaling, value = 2 means that each size is 1/2 (texture area is 1/4), value = 3 means that each size is 1/4 and so on.

Note that textures used for GUI, by TDrawableImage (more precisely: all non-power-of-2 textures) avoid this scaling entirely.

This allows to scale the textures at runtime. For an independent mechanism that allows to load already downscaled textures (and scale them as a preprocessing step when packaging your game), see TextureLoadingScale. The documentation of TextureLoadingScale lists all the differences between these two approaches to scaling.

Constraints the scaling done by GLTextureScale. Scaling caused by GLTextureScale cannot scale texture to something less than GLTextureMinSize. If texture size was already < GLTextureMinSize, it is not scaled at all by GLTextureScale. This must be a power of two.

Log (through CastleLog) all texture and image loading to GPU. If additionally TextureMemoryProfiler is enabled, then we will dump texture memory usage.