#include <diff.h>

Inheritance diagram for fz::DifferenceStage< T, TOut >:

Public Member Functions
void	setInverse (bool inverse) override

void	setChunkSize (size_t bytes)

size_t	getRequiredInputAlignment () const override

void	execute (cudaStream_t stream, MemoryPool pool, const std::vector< void > &inputs, const std::vector< void * > &outputs, const std::vector< size_t > &sizes) override

std::string	getName () const override

std::vector< size_t >	estimateOutputSizes (const std::vector< size_t > &input_sizes) const override

std::unordered_map< std::string, size_t >	getActualOutputSizesByName () const override

size_t	getActualOutputSize (int index) const override

uint16_t	getStageTypeId () const override

uint8_t	getOutputDataType (size_t output_index) const override

uint8_t	getInputDataType (size_t) const override

size_t	serializeHeader (size_t output_index, uint8_t *buf, size_t max_size) const override

void	deserializeHeader (const uint8_t *buf, size_t size) override

size_t	getMaxHeaderSize (size_t output_index) const override

void	saveState () override

Public Member Functions inherited from fz::Stage
virtual std::vector< std::string >	getOutputNames () const

int	getOutputIndex (const std::string &name) const

virtual void	setDims (const std::array< size_t, 3 > &dims)

virtual void	onFinalize (size_t, MemoryPool *)

virtual size_t	estimateDeviceFootprintBytes (size_t) const

virtual size_t	estimatePinnedFootprintBytes (size_t) const

virtual void	postStreamSync (cudaStream_t stream)

virtual bool	isGraphCompatible () const

virtual size_t	estimateScratchBytes (const std::vector< size_t > &input_sizes) const

Detailed Description

template<typename T = float, typename TOut = T>
class fz::DifferenceStage< T, TOut >

Difference coding stage.

Note: Prior work: kernel follows the d_DIFFNB algorithm from the LC/PFPL framework (Burtscher et al., BSD-3-Clause). See THIRD_PARTY.md.

Forward (compression): first-order differences with optional negabinary output output[0] = input[0] output[i] = input[i] - input[i-1] (when TOut == T) output[i] = Negabinary<T>::encode(diff) (when TOut != T)

Inverse (decompression): cumulative sum with optional negabinary decode first

Optional chunking (setChunkSize > 0): The difference/cumsum resets at each chunk boundary. Each chunk is a fully independent context — the first element of each chunk is stored as-is (previous = 0 implied). This enables parallel decompression and is required for the PFPL pipeline where 16 KB chunks flow independently through Bitshuffle and RZE.

Template parameters: T — input element type (signed/unsigned integer or float). TOut — output element type (defaults to T). When TOut != T: T must be a signed integer and TOut its unsigned counterpart of the same width; negabinary encoding is fused at the final write of the forward kernel and the decode is the first step of the inverse kernel.

Serialized header layout (6 bytes): [0] DataType T (1 byte) [1] DataType TOut (1 byte) [2..5] chunk_size (uint32_t, little-endian, 0 = no chunking)

Member Function Documentation

◆ setInverse()

template<typename T = float, typename TOut = T>

void fz::DifferenceStage< T, TOut >::setInverse ( bool inverse )

inlineoverridevirtual

Switch between forward (compression) and inverse (decompression) mode. Affects getNumInputs()/getNumOutputs() for stages with asymmetric port counts.

Reimplemented from fz::Stage.

◆ setChunkSize()

template<typename T = float, typename TOut = T>

void fz::DifferenceStage< T, TOut >::setChunkSize ( size_t bytes )

inline

Set chunk size in bytes (default 0 = no chunking).

When > 0, differences and cumulative sums reset at each chunk boundary. Must be a positive multiple of sizeof(T). Pass 0 to disable chunking and process the whole array as a single context (the previous default).

◆ getRequiredInputAlignment()

template<typename T = float, typename TOut = T>

size_t fz::DifferenceStage< T, TOut >::getRequiredInputAlignment ( ) const

inlineoverridevirtual

Minimum input size alignment in bytes. Chunked stages return their chunk size; the pipeline uses the LCM of all stage alignments at finalize() to transparently zero-pad the input. Default: 1 (no alignment requirement).

Reimplemented from fz::Stage.

◆ execute()

template<typename T = float, typename TOut = T>

void fz::DifferenceStage< T, TOut >::execute	(	cudaStream_t	stream,
		MemoryPool *	pool,
		const std::vector< void * > &	inputs,
		const std::vector< void * > &	outputs,
		const std::vector< size_t > &	sizes
	)

overridevirtual

Execute the stage. Inputs, outputs, and sizes are device pointers/bytes.

Stages may call cudaStreamSynchronize(stream) or issue blocking D2H copies when the algorithm requires it (e.g. Huffman histogram readback for codebook construction, ANS renormalization tables). Such stages must return false from isGraphCompatible() and must document the sync points.

Note: the DAG dispatches sibling nodes (same topological level) via a sequential CPU loop, each enqueuing to its own stream. A sync inside execute() blocks the CPU from dispatching subsequent siblings until the synced stream is idle — this delays parallel branches in wide DAGs. In a linear pipeline there are no siblings and no extra cost.

Implements fz::Stage.

◆ getName()

template<typename T = float, typename TOut = T>

std::string fz::DifferenceStage< T, TOut >::getName ( ) const

inlineoverridevirtual

Human-readable name used in error messages and debug output.

Implements fz::Stage.

◆ estimateOutputSizes()

template<typename T = float, typename TOut = T>

std::vector< size_t > fz::DifferenceStage< T, TOut >::estimateOutputSizes ( const std::vector< size_t > & input_sizes ) const

inlineoverridevirtual

Estimate output buffer sizes given input sizes. Used for buffer allocation planning in PREALLOCATE mode — must be a safe upper bound; under-estimation causes buffer overruns.

Implements fz::Stage.

◆ getActualOutputSizesByName()

template<typename T = float, typename TOut = T>

std::unordered_map< std::string, size_t > fz::DifferenceStage< T, TOut >::getActualOutputSizesByName ( ) const

inlineoverridevirtual

Actual output sizes after execute(), keyed by output port name.

Implements fz::Stage.

◆ getActualOutputSize()

template<typename T = float, typename TOut = T>

size_t fz::DifferenceStage< T, TOut >::getActualOutputSize ( int index ) const

inlineoverridevirtual

Actual size of a single output by index after execute(). Avoids constructing the map for the common single-output case. Default delegates to getActualOutputSizesByName(); override to return directly from an internal field.

Reimplemented from fz::Stage.

◆ getStageTypeId()

template<typename T = float, typename TOut = T>

uint16_t fz::DifferenceStage< T, TOut >::getStageTypeId ( ) const

inlineoverridevirtual

Stage type identifier written into the FZM file header.

Implements fz::Stage.

◆ getOutputDataType()

template<typename T = float, typename TOut = T>

uint8_t fz::DifferenceStage< T, TOut >::getOutputDataType ( size_t output_index ) const

inlineoverridevirtual

DataType enum of the given output port.

Implements fz::Stage.

◆ getInputDataType()

template<typename T = float, typename TOut = T>

uint8_t fz::DifferenceStage< T, TOut >::getInputDataType ( size_t ) const

inlineoverridevirtual

Expected DataType of the given input port.

Used by Pipeline::finalize() to detect type mismatches between connected stages before any execution. Return DataType::UNKNOWN to opt out of checking — byte-transparent stages (Bitshuffle, RZE) and mock stages must return UNKNOWN; finalize() skips any connection where either side is UNKNOWN.

Reimplemented from fz::Stage.

◆ serializeHeader()

template<typename T = float, typename TOut = T>

size_t fz::DifferenceStage< T, TOut >::serializeHeader	(	size_t	output_index,
		uint8_t *	header_buffer,
		size_t	max_size
	)		const

inlineoverridevirtual

Serialize stage config into header_buffer (max 128 bytes) for the FZM file. Return the number of bytes written, or 0 if the stage has no config.

Reimplemented from fz::Stage.

◆ deserializeHeader()

template<typename T = float, typename TOut = T>

void fz::DifferenceStage< T, TOut >::deserializeHeader	(	const uint8_t *	header_buffer,
		size_t	size
	)

inlineoverridevirtual

Restore stage config from header_buffer during decompression.

Reimplemented from fz::Stage.

◆ getMaxHeaderSize()

template<typename T = float, typename TOut = T>

size_t fz::DifferenceStage< T, TOut >::getMaxHeaderSize ( size_t output_index ) const

inlineoverridevirtual

Maximum bytes this stage writes into its per-output FZM header slot.

Reimplemented from fz::Stage.

◆ saveState()

template<typename T = float, typename TOut = T>

void fz::DifferenceStage< T, TOut >::saveState ( )

inlineoverridevirtual

Save/restore config state around a decompression pass. deserializeHeader() overwrites the stage's forward-pass config; saveState() is called before and restoreState() after so the stage returns to its original configuration.

Reimplemented from fz::Stage.

Public Member Functions

Detailed Description

Member Function Documentation

◆ setInverse()

◆ setChunkSize()

◆ getRequiredInputAlignment()

◆ execute()

◆ getName()

◆ estimateOutputSizes()

◆ getActualOutputSizesByName()

◆ getActualOutputSize()

◆ getStageTypeId()

◆ getOutputDataType()

◆ getInputDataType()

◆ serializeHeader()

◆ deserializeHeader()

◆ getMaxHeaderSize()

◆ saveState()