All squish API functions live in this namespace.

Typedefs
typedef unsigned char	u8
	Typedef a quantity that is a single unsigned byte.

Enumerations
enum	{ kDxt1 = ( 1 << 0 ) , kDxt3 = ( 1 << 1 ) , kDxt5 = ( 1 << 2 ) , kBc4 = ( 1 << 3 ) , kBc5 = ( 1 << 4 ) , kColourClusterFit = ( 1 << 5 ) , kColourRangeFit = ( 1 << 6 ) , kWeightColourByAlpha = ( 1 << 7 ) , kColourIterativeClusterFit = ( 1 << 8 ) , kSourceBGRA = ( 1 << 9 ) }

Functions
void	Compress (u8 const rgba, void block, int flags, float *metric=0)
	Compresses a 4x4 block of pixels. More...

void	CompressImage (u8 const rgba, int width, int height, int pitch, void blocks, int flags, float *metric=0)
	Compresses an image in memory. More...

void	CompressImage (u8 const rgba, int width, int height, void blocks, int flags, float *metric=0)

void	CompressMasked (u8 const rgba, int mask, void block, int flags, float *metric=0)
	Compresses a 4x4 block of pixels. More...

void	ComputeMSE (u8 const rgba, int width, int height, int pitch, u8 const dxt, int flags, double &colourMSE, double &alphaMSE)
	Computes MSE of an compressed image in memory. More...

void	ComputeMSE (u8 const rgba, int width, int height, u8 const dxt, int flags, double &colourMSE, double &alphaMSE)

void	Decompress (u8 rgba, void const block, int flags)
	Decompresses a 4x4 block of pixels. More...

void	DecompressImage (u8 rgba, int width, int height, int pitch, void const blocks, int flags)
	Decompresses an image in memory. More...

void	DecompressImage (u8 rgba, int width, int height, void const blocks, int flags)

int	GetStorageRequirements (int width, int height, int flags)
	Computes the amount of compressed storage required. More...

Enumeration Type Documentation

◆ anonymous enum

anonymous enum

Enumerator
kDxt1	Use DXT1 compression.
kDxt3	Use DXT3 compression.
kDxt5	Use DXT5 compression.
kBc4	Use BC4 compression.
kBc5	Use BC5 compression.
kColourClusterFit	Use a slow but high quality colour compressor (the default).
kColourRangeFit	Use a fast but low quality colour compressor.
kWeightColourByAlpha	Weight the colour by alpha during cluster fit (disabled by default).
kColourIterativeClusterFit	Use a very slow but very high quality colour compressor.
kSourceBGRA	Source is BGRA rather than RGBA.

Function Documentation

◆ Compress()

void squish::Compress	(	u8 const *	rgba,
		void *	block,
		int	flags,
		float *	metric = `0`
	)

inline

Parameters

rgba	The rgba values of the 16 source pixels.
block	Storage for the compressed DXT block.
flags	Compression flags.
metric	An optional perceptual metric.

The source pixels should be presented as a contiguous array of 16 rgba values, with each component as 1 byte each. In memory this should be:

{ r1, g1, b1, a1, .... , r16, g16, b16, a16 }

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. When using DXT1 compression, 8 bytes of storage are required for the compressed DXT block. DXT3 and DXT5 compression require 16 bytes of storage per block.

The flags parameter can also specify a preferred colour compressor to use when fitting the RGB components of the data. Possible colour compressors are: kColourClusterFit (the default), kColourRangeFit (very fast, low quality) or kColourIterativeClusterFit (slowest, best quality).

When using kColourClusterFit or kColourIterativeClusterFit, an additional flag can be specified to weight the importance of each pixel by its alpha value. For images that are rendered using alpha blending, this can significantly increase the perceived quality.

The metric parameter can be used to weight the relative importance of each colour channel, or pass NULL to use the default uniform weight of { 1.0f, 1.0f, 1.0f }. This replaces the previous flag-based control that allowed either uniform or "perceptual" weights with the fixed values { 0.2126f, 0.7152f, 0.0722f }. If non-NULL, the metric should point to a contiguous array of 3 floats.

This method is an inline that calls CompressMasked with a mask of 0xffff, provided for compatibility with older versions of squish.

◆ CompressImage() [1/2]

void squish::CompressImage	(	u8 const *	rgba,
		int	width,
		int	height,
		int	pitch,
		void *	blocks,
		int	flags,
		float *	metric = `0`
	)

Parameters

rgba	The pixels of the source.
width	The width of the source image.
height	The height of the source image.
pitch	The pitch of the source image.
blocks	Storage for the compressed output.
flags	Compression flags.
metric	An optional perceptual metric.

The source pixels should be presented as a contiguous array of width*height rgba values, with each component as 1 byte each. In memory this should be:

{ r1, g1, b1, a1, .... , rn, gn, bn, an } for n = width*height

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. When using DXT1 compression, 8 bytes of storage are required for each compressed DXT block. DXT3 and DXT5 compression require 16 bytes of storage per block.

The flags parameter can also specify a preferred colour compressor to use when fitting the RGB components of the data. Possible colour compressors are: kColourClusterFit (the default), kColourRangeFit (very fast, low quality) or kColourIterativeClusterFit (slowest, best quality).

When using kColourClusterFit or kColourIterativeClusterFit, an additional flag can be specified to weight the importance of each pixel by its alpha value. For images that are rendered using alpha blending, this can significantly increase the perceived quality.

The metric parameter can be used to weight the relative importance of each colour channel, or pass NULL to use the default uniform weight of { 1.0f, 1.0f, 1.0f }. This replaces the previous flag-based control that allowed either uniform or "perceptual" weights with the fixed values { 0.2126f, 0.7152f, 0.0722f }. If non-NULL, the metric should point to a contiguous array of 3 floats.

Internally this function calls squish::CompressMasked for each block, which allows for pixels outside the image to take arbitrary values. The function squish::GetStorageRequirements can be called to compute the amount of memory to allocate for the compressed output.

Note on compression quality: When compressing textures with libsquish it is recommended to apply a gamma-correction beforehand. This will reduce the blockiness in dark areas. The level of necessary gamma-correction is platform dependent. For example, a gamma correction with gamma = 0.5 before compression and gamma = 2.0 after decompression yields good results on the Windows platform but for other platforms like MacOS X a different gamma value may be more suitable.

◆ CompressImage() [2/2]

void squish::CompressImage	(	u8 const *	rgba,
		int	width,
		int	height,
		void *	blocks,
		int	flags,
		float *	metric = `0`
	)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

◆ CompressMasked()

void squish::CompressMasked	(	u8 const *	rgba,
		int	mask,
		void *	block,
		int	flags,
		float *	metric = `0`
	)

Parameters

rgba	The rgba values of the 16 source pixels.
mask	The valid pixel mask.
block	Storage for the compressed DXT block.
flags	Compression flags.
metric	An optional perceptual metric.

The source pixels should be presented as a contiguous array of 16 rgba values, with each component as 1 byte each. In memory this should be:

{ r1, g1, b1, a1, .... , r16, g16, b16, a16 }

The mask parameter enables only certain pixels within the block. The lowest bit enables the first pixel and so on up to the 16th bit. Bits beyond the 16th bit are ignored. Pixels that are not enabled are allowed to take arbitrary colours in the output block. An example of how this can be used is in the CompressImage function to disable pixels outside the bounds of the image when the width or height is not divisible by 4.

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. When using DXT1 compression, 8 bytes of storage are required for the compressed DXT block. DXT3 and DXT5 compression require 16 bytes of storage per block.

The flags parameter can also specify a preferred colour compressor to use when fitting the RGB components of the data. Possible colour compressors are: kColourClusterFit (the default), kColourRangeFit (very fast, low quality) or kColourIterativeClusterFit (slowest, best quality).

When using kColourClusterFit or kColourIterativeClusterFit, an additional flag can be specified to weight the importance of each pixel by its alpha value. For images that are rendered using alpha blending, this can significantly increase the perceived quality.

The metric parameter can be used to weight the relative importance of each colour channel, or pass NULL to use the default uniform weight of { 1.0f, 1.0f, 1.0f }. This replaces the previous flag-based control that allowed either uniform or "perceptual" weights with the fixed values { 0.2126f, 0.7152f, 0.0722f }. If non-NULL, the metric should point to a contiguous array of 3 floats.

◆ ComputeMSE() [1/2]

void squish::ComputeMSE	(	u8 const *	rgba,
		int	width,
		int	height,
		int	pitch,
		u8 const *	dxt,
		int	flags,
		double &	colourMSE,
		double &	alphaMSE
	)

Parameters

rgba	The original image pixels.
width	The width of the source image.
height	The height of the source image.
pitch	The pitch of the source image.
dxt	The compressed dxt blocks
flags	Compression flags.
colourMSE	The MSE of the colour values.
alphaMSE	The MSE of the alpha values.

The colour MSE and alpha MSE are computed across all pixels. The colour MSE is averaged across all rgb values (i.e. colourMSE = sum sum_k ||dxt.k - rgba.k||/3)

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. All other flags are ignored.

Internally this function calls squish::Decompress for each block.

◆ ComputeMSE() [2/2]

void squish::ComputeMSE	(	u8 const *	rgba,
		int	width,
		int	height,
		u8 const *	dxt,
		int	flags,
		double &	colourMSE,
		double &	alphaMSE
	)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

◆ Decompress()

void squish::Decompress	(	u8 *	rgba,
		void const *	block,
		int	flags
	)

Parameters

rgba	Storage for the 16 decompressed pixels.
block	The compressed DXT block.
flags	Compression flags.

The decompressed pixels will be written as a contiguous array of 16 rgba values, with each component as 1 byte each. In memory this is:

{ r1, g1, b1, a1, .... , r16, g16, b16, a16 }

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. All other flags are ignored.

◆ DecompressImage() [1/2]

void squish::DecompressImage	(	u8 *	rgba,
		int	width,
		int	height,
		int	pitch,
		void const *	blocks,
		int	flags
	)

Parameters

rgba	Storage for the decompressed pixels.
width	The width of the source image.
height	The height of the source image.
pitch	The pitch of the decompressed pixels.
blocks	The compressed DXT blocks.
flags	Compression flags.

The decompressed pixels will be written as a contiguous array of width*height 16 rgba values, with each component as 1 byte each. In memory this is:

{ r1, g1, b1, a1, .... , rn, gn, bn, an } for n = width*height

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. All other flags are ignored.

Internally this function calls squish::Decompress for each block.

◆ DecompressImage() [2/2]

void squish::DecompressImage	(	u8 *	rgba,
		int	width,
		int	height,
		void const *	blocks,
		int	flags
	)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

◆ GetStorageRequirements()

int squish::GetStorageRequirements	(	int	width,
		int	height,
		int	flags
	)

Parameters

width	The width of the image.
height	The height of the image.
flags	Compression flags.

The flags parameter should specify kDxt1, kDxt3, kDxt5, kBc4, or kBc5 compression, however, DXT1 will be used by default if none is specified. All other flags are ignored.

Most DXT images will be a multiple of 4 in each dimension, but this function supports arbitrary size images by allowing the outer blocks to be only partially used.

Typedefs

Enumerations

Functions

Enumeration Type Documentation

◆ anonymous enum

Function Documentation

◆ Compress()

◆ CompressImage() [1/2]

◆ CompressImage() [2/2]

◆ CompressMasked()

◆ ComputeMSE() [1/2]

◆ ComputeMSE() [2/2]

◆ Decompress()

◆ DecompressImage() [1/2]

◆ DecompressImage() [2/2]

◆ GetStorageRequirements()