De-interleaved filtering steps implementation class. More...

#include <avir_dil.h>

Inheritance diagram for avir::CImageResizerFilterStepDIL< fptype, fptypesimd >:

Public Member Functions
void	calcScanlineBias (const fptype const p0, const int SrcLen, fptype const ElBiases) const

void	convertVtoH (const fptype ip, fptype op, const int SrcLen, const int SrcIncr) const

void	doFilter (const fptype const Src, fptype Dst, const int DstIncr) const

void	doResize (const fptype SrcLine, fptype DstLine, int DstLineIncr, const fptype const ElBiases, fptype const xx) const

void	doResize2 (const fptype SrcLine, fptype DstLine, int DstLineIncr, const fptype const ElBiases, fptype const xx) const

void	doUpsample (const fptype Src, fptype Dst) const

template<class Tin>
void	packScanline (const Tin const ip0, fptype const op0, const int l) const

void	prepareInBuf (fptype *Src) const

void	unbiasScanline (fptype const p0, const int l, const fptype const ElBiases) const

Static Public Member Functions
static void	applySRGBGamma (fptype *const p0, const int l, const CImageResizerVars &Vars0)

template<class Tout>
static void	unpackScanline (const fptype const ip0, Tout const op0, const int l, const CImageResizerVars &Vars0)

Additional Inherited Members
Public Attributes inherited from avir::CImageResizerFilterStep< fptype, fptype >
double	DCGain
	DC gain which was applied to the filter. Not defined if ResampleFactor = 0.

int	EdgePixelCount
	The number of edge pixels added. Affects the initial position within the input scanline, used to produce edge pixels. This variable is used and should be defined when IsUpsample=false and ResampleFactor>0. When assigning this variable it is also necessary to update InPrefix, OutLen and Vars.o variables.

CBuffer< fptype >	Flt
	Filter to use at this step.

CDSPFracFilterBankLin< fptype > *	FltBank
	Filter bank in use by *this resizing step.

int	FltLatency
	Filter's latency (group delay, shift) in pixels.

CFltBuffer	FltOrig
	Originally-designed filter. This buffer may not be assigned. Assigned by filters that precede the resizing step if such filter is planned to be embedded into the interpolation filter as "external" filter. If IsUpsample=true and this filter buffer is not empty, the upsampling step will not itself apply any filtering over upsampled input scanline.

int	InBuf
	Input buffer index, 0 or 1.

int	InElIncr
	Pixel element increment within the input buffer, used during de-interleaved processing: in this case each image's channel is stored independently, InElIncr elements apart.

int	InLen
	Input scanline's length in pixels.

int	InPrefix
	Required input prefix pixels. These prefix pixels will be filled with source scanline's first pixel value. If IsUpsample is "true", this is the additional number of times the first pixel will be filtered before processing scanline, this number is also reflected in the OutPrefix.

int	InSuffix
	Required input suffix pixels. These suffix pixels will be filled with source scanline's last pixel value. If IsUpsample is "true", this is the additional number of times the last pixel will be filtered before processing scanline, this number is also reflected in the OutSuffix.

bool	IsUpsample
	"True" if this step is an upsampling step, "false" if downsampling step. Should be set to "false" if ResampleFactor equals 0.

int	OutBuf
	Output buffer index. 0 or 1; 2 for the last step.

int	OutElIncr
	Pixel element increment within the output buffer, used during de-interleaved processing. Equals to the InBufElIncr of the next step.

int	OutLen
	Length of the resulting scanline.

int	OutPrefix
	Required output prefix pixels. These prefix pixels will not be pre-filled with any values. Value is valid only if IsUpsample equals "true".

int	OutSuffix
	Required input suffix pixels. These suffix pixels will not be pre-filled with any values. Value is valid only if IsUpsample equals "true".

CBuffer< fptype >	PrefixDC
	DC component fluctuations added at the start of the resulting scanline, used when IsUpsample equals "true".

int	ResampleFactor
	Resample factor (>=1). If 0, this is a resizing step. This value should be >1 if IsUpsample equals "true".

CRPosBuf *	RPosBuf
	Resizing positions buffer. Used when ResampleFactor equals 0 (resizing step).

CBuffer< fptype >	SuffixDC
	DC component fluctuations added at the end of the resulting scanline, used when IsUpsample equals "true".

const CImageResizerVars *	Vars
	Image resizing-related variables.

Static Public Attributes inherited from avir::CImageResizerFilterStep< fptype, fptype >
static const int	EdgePixelCountDef
	The default number of pixels additionally produced at scanline edges during filtering. This is required to reduce edge artifacts.

Detailed Description

template<class fptype, class fptypesimd>
class avir::CImageResizerFilterStepDIL< fptype, fptypesimd >

De-interleaved filtering steps implementation class.

This class implements scanline filtering functions in de-interleaved mode. This means that pixels are processed in groups.

Template Parameters

fptype	Floating point type to use for storing pixel elements. SIMD types cannot be used.
fptypesimd	The SIMD type used to store a pack of "fptype" values.

Member Function Documentation

◆ applySRGBGamma()

template<class fptype, class fptypesimd>

static void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::applySRGBGamma	(	fptype *const	p0,
		const int	l,
		const CImageResizerVars &	Vars0 )

static

Function applies Linear to sRGB gamma correction to the specified scanline.

Parameters

p	Scanline.
l	The number of pixels to de-linearize.
Vars0	Image resizing-related variables.

◆ calcScanlineBias()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::calcScanlineBias	(	const fptype *const	p0,
		const int	SrcLen,
		fptype *const	ElBiases ) const

Function calculates scanline's DC gain for each channel, further used to "unbias" the scanline.

Parameters

	p0	Source scanline.
	SrcLen	Source scanline's length.
[out]	ElBiases	Resuling biases.

◆ convertVtoH()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::convertVtoH	(	const fptype *	ip,
		fptype *	op,
		const int	SrcLen,
		const int	SrcIncr ) const

Function converts vertical scanline to horizontal scanline. This function is called by the image resizer when image is resized vertically. This means that the vertical scanline is stored in the same format produced by the packScanline() and maintained by other filtering functions.

Parameters

ip	Input vertical scanline, pixel elements are grouped, SrcLen elements apart.
op	Output buffer (temporary buffer used during resizing), pixel elements are grouped, "l" elements apart.
SrcLen	The number of pixels in the input scanline, also used to calculate input buffer increment.
SrcIncr	Input buffer increment to the next vertical pixel.

◆ doFilter()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::doFilter	(	const fptype *const	Src,
		fptype *	Dst,
		const int	DstIncr ) const

Function peforms scanline filtering with optional downsampling. Function makes use of the symmetry of the filter.

Parameters

Src	Source scanline buffer (length = this -> InLen). Source scanline increment will be equal to 1.
Dst	Destination scanline buffer.
DstIncr	Destination scanline buffer increment, used for horizontal or vertical scanline stepping.

◆ doResize()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::doResize	(	const fptype *	SrcLine,
		fptype *	DstLine,
		int	DstLineIncr,
		const fptype *const	ElBiases,
		fptype *const	xx ) const

Function performs resizing of a single scanline. This function does not "know" about the length of the source scanline buffer. This buffer should be padded with enough pixels so that ( SrcPos - FilterLenD2 ) is always >= 0 and ( SrcPos + ( DstLineLen - 1 ) * k + FilterLenD2 + 1 ) does not exceed source scanline's buffer length. SrcLine's increment is assumed to be equal to 1.

Parameters

SrcLine	Source scanline buffer.
DstLine	Destination (resized) scanline buffer.
DstLineIncr	Destination scanline position increment, used for horizontal or vertical scanline stepping.
ElBiases	Bias values to add to the resulting scanline.
xx	Temporary buffer, of size FltBank -> getFilterLen(), must be aligned by fpclass :: fpalign.

◆ doResize2()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::doResize2	(	const fptype *	SrcLine,
		fptype *	DstLine,
		int	DstLineIncr,
		const fptype *const	ElBiases,
		fptype *const	xx ) const

Same as doResize(). No specific 2x filter-less upsampling optimization.

◆ doUpsample()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::doUpsample	(	const fptype *	Src,
		fptype *	Dst ) const

Function peforms scanline upsampling with filtering.

Parameters

Src	Source scanline buffer (length = this -> InLen). Source scanline increment will be equal to ElCount.
Dst	Destination scanline buffer.

◆ packScanline()

template<class fptype, class fptypesimd>

template<class Tin>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::packScanline	(	const Tin *const	ip0,
		fptype *const	op0,
		const int	l ) const

Function performs "packing" (de-interleaving) of a scanline and type conversion. If required, the sRGB gamma correction is applied.

Parameters

ip0	Input scanline, pixel elements interleaved.
op0	Output scanline, pixel elements are grouped, "l" elements apart.
l	The number of pixels to "pack".

◆ prepareInBuf()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::prepareInBuf ( fptype * Src ) const

Function prepares input scanline buffer for *this filtering step. Left- and right-most pixels are replicated to make sure no buffer overrun happens. Such approach also allows to bypass any pointer range checks.

Parameters

Src	Source buffer.

◆ unbiasScanline()

template<class fptype, class fptypesimd>

void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::unbiasScanline	(	fptype *const	p0,
		const int	l,
		const fptype *const	ElBiases ) const

Function applies "unbiasing" to the scanline, by subtracting the previously calculated bias (DC gain) values.

Parameters

p0	Scanline.
l	Scanline's length.
ElBiases	Biases to subtract, for each channel.

◆ unpackScanline()

template<class fptype, class fptypesimd>

template<class Tout>

static void avir::CImageResizerFilterStepDIL< fptype, fptypesimd >::unpackScanline	(	const fptype *const	ip0,
		Tout *const	op0,
		const int	l,
		const CImageResizerVars &	Vars0 )

static

Function performs "unpacking" of a scanline and type conversion (truncation is used when floating point is converted to integer). The unpacking function assumes that scanline is stored in the style produced by the packScanline() function.

Parameters

ip0	Input scanline, pixel elements are grouped, "l" elements apart.
op0	Output scanline, pixel elements are interleaved.
l	The number of pixels to "unpack".
Vars0	Image resizing-related variables. ElCount is assumed to be equal to ElCountIO.

Public Member Functions

Static Public Member Functions

Additional Inherited Members

Detailed Description

Member Function Documentation

◆ applySRGBGamma()

◆ calcScanlineBias()

◆ convertVtoH()

◆ doFilter()

◆ doResize()

◆ doResize2()

◆ doUpsample()

◆ packScanline()

◆ prepareInBuf()

◆ unbiasScanline()

◆ unpackScanline()