Functions
void	av1_alloc_txb_buf (AV1_COMP *cpi)
	Allocate the memory resources for all the macro blocks in the current coding frame.

void	av1_free_txb_buf (AV1_COMP *cpi)
	Free the memory resources for all the macro blocks in the current coding frame.

void	av1_write_coeffs_txb (const AV1_COMMON const cm, MACROBLOCK const x, aom_writer *w, int blk_row, int blk_col, int plane, int block, TX_SIZE tx_size)
	Write quantized coefficients in a transform block into bitstream using entropy coding.

void	av1_write_intra_coeffs_mb (const AV1_COMMON const cm, MACROBLOCK x, aom_writer *w, BLOCK_SIZE bsize)
	Write quantized coefficients of all transform blocks in an intra macroblock into the bitstream using entropy coding.

uint8_t	av1_get_txb_entropy_context (const tran_low_t qcoeff, const SCAN_ORDER scan_order, int eob)
	Pack the context info of the current transform block into an uint8_t.

void	av1_update_intra_mb_txb_context (const AV1_COMP cpi, ThreadData td, RUN_TYPE dry_run, BLOCK_SIZE bsize, uint8_t allow_update_cdf)
	Update the probability model (cdf) and the entropy context related to coefficient coding for all transform blocks in the intra macroblock.

void	av1_update_and_record_txb_context (int plane, int block, int blk_row, int blk_col, BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg)
	Update the probability model (cdf) and the entropy context related to coefficient coding for a transform block.

void	av1_record_txb_context (int plane, int block, int blk_row, int blk_col, BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg)
	Update the entropy context related to coefficient coding for a transform block.

CB_COEFF_BUFFER *	av1_get_cb_coeff_buffer (const struct AV1_COMP *cpi, int mi_row, int mi_col)
	Get the corresponding CB_COEFF_BUFFER of the current macro block.

static int	av1_cost_skip_txb (const CoeffCosts coeff_costs, const TXB_CTX const txb_ctx, int plane, TX_SIZE tx_size)
	Returns the entropy cost associated with skipping the current transform block.

int	av1_optimize_txb (const struct AV1_COMP cpi, MACROBLOCK x, int plane, int block, TX_SIZE tx_size, TX_TYPE tx_type, const TXB_CTX const txb_ctx, int rate_cost, int sharpness)
	Adjust the magnitude of quantized coefficients to achieve better rate-distortion (RD) trade-off.

int	av1_cost_coeffs_txb (const MACROBLOCK x, const int plane, const int block, const TX_SIZE tx_size, const TX_TYPE tx_type, const TXB_CTX const txb_ctx, int reduced_tx_set_used)
	Compute the entropy cost of coding coefficients in a transform block.

int	av1_cost_coeffs_txb_laplacian (const MACROBLOCK x, const int plane, const int block, const TX_SIZE tx_size, const TX_TYPE tx_type, const TXB_CTX const txb_ctx, const int reduced_tx_set_used, const int adjust_eob)
	Estimate the entropy cost of coding a transform block using Laplacian distribution.

int	av1_cost_coeffs_txb_estimate (const MACROBLOCK *x, const int plane, const int block, const TX_SIZE tx_size, const TX_TYPE tx_type)
	Estimate the entropy cost of transform coefficients using Laplacian distribution.

Detailed Description

This module describes the algorithms of transform coefficient coding and optimization in AV1. More details will be added.

Function Documentation

◆ av1_alloc_txb_buf()

void av1_alloc_txb_buf ( AV1_COMP * cpi )

Allocate the memory resources for all the macro blocks in the current coding frame.

Each macro block will need a CB_COEFF_BUFFER to store information for rate-distortion optimization and entropy coding of transform coefficients.

Parameters

[in] cpi Top-level encoder structure

◆ av1_free_txb_buf()

void av1_free_txb_buf ( AV1_COMP * cpi )

Free the memory resources for all the macro blocks in the current coding frame.

See av1_alloc_txb_buf and CB_COEFF_BUFFER for more details.

Parameters

[in] cpi Top-level encoder structure

◆ av1_write_coeffs_txb()

void av1_write_coeffs_txb	(	const AV1_COMMON *const	cm,
		MACROBLOCK *const	x,
		aom_writer *	w,
		int	blk_row,
		int	blk_col,
		int	plane,
		int	block,
		TX_SIZE	tx_size )

Write quantized coefficients in a transform block into bitstream using entropy coding.

This function will write the quantized coefficients in a transform block into the bitstream using entropy coding.

The coding steps are as follows.

1) Code the end of block position "eob", which is the scan index of the last non-zero coefficient plus one.

2) Code the lower magnitude level (<= COEFF_BASE_RANGE + NUM_BASE_LEVELS) for each coefficient in reversed scan order.

3) Code the sign and higher magnitude level (> COEFF_BASE_RANGE + NUM_BASE_LEVELS) in forward scan order.

Parameters

[in]	cm	Top-level structure shared by encoder and decoder
[in]	x	Pointer to structure holding the data for the current encoding macroblock
[in]	w	Entropy coding write pointer
[in]	blk_row	The row index of the current transform block in the macroblock. Each unit has 4 pixels in y plane
[in]	blk_col	The col index of the current transform block in the macroblock. Each unit has 4 pixels in y plane
[in]	plane	The index of the current plane
[in]	block	The index of the current transform block in the macroblock. It's defined by number of 4x4 units that have been coded before the currernt transform block
[in]	tx_size	The given transform size

◆ av1_write_intra_coeffs_mb()

void av1_write_intra_coeffs_mb	(	const AV1_COMMON *const	cm,
		MACROBLOCK *	x,
		aom_writer *	w,
		BLOCK_SIZE	bsize )

Write quantized coefficients of all transform blocks in an intra macroblock into the bitstream using entropy coding.

All transform blocks in the intra macroblock share the same transform size.

This function use av1_write_coeffs_txb() to code each transform block in raster order.

Parameters

[in]	cm	Top-level structure shared by encoder and decoder
[in]	x	Pointer to structure holding the data for the current encoding macroblock
[in]	w	Entropy coding write pointer
[in]	bsize	Block size of the current macroblock

◆ av1_get_txb_entropy_context()

uint8_t av1_get_txb_entropy_context	(	const tran_low_t *	qcoeff,
		const SCAN_ORDER *	scan_order,
		int	eob )

Pack the context info of the current transform block into an uint8_t.

This context info will be collected and consolidated by its neighbor transform blocks for coding transform block skip flag (tx_skip) and the sign of DC coefficient (dc_sign).

Parameters

[in]	qcoeff	Buffer of quantized coefficients
[in]	scan_order	Coding order of coefficients in the transform block
[in]	eob	The scan index of last non-zero coefficient plus one

◆ av1_update_intra_mb_txb_context()

void av1_update_intra_mb_txb_context	(	const AV1_COMP *	cpi,
		ThreadData *	td,
		RUN_TYPE	dry_run,
		BLOCK_SIZE	bsize,
		uint8_t	allow_update_cdf )

Update the probability model (cdf) and the entropy context related to coefficient coding for all transform blocks in the intra macroblock.

This function will go through each transform block in the intra macorblock and call av1_update_and_record_txb_context to update the probability model and entropy context properly.

Parameters

[in]	cpi	Top-level encoder structure
[in]	td	Top-level multithreading structure
[in]	dry_run	Whether this is a dry run.
[in]	bsize	Block size of the current macroblock
[in]	allow_update_cdf	Allowed to update probability model (cdf) or not.

◆ av1_update_and_record_txb_context()

void av1_update_and_record_txb_context	(	int	plane,
		int	block,
		int	blk_row,
		int	blk_col,
		BLOCK_SIZE	plane_bsize,
		TX_SIZE	tx_size,
		void *	arg )

Update the probability model (cdf) and the entropy context related to coefficient coding for a transform block.

There are regular mode and dry run for this funtion.

Regular mode:

The probability model (cdf) for each coding symbol in the transform block will be updated.

The entropy context of this transform block will be updated.

Dry run:

The probability model update will be skipped.

The entropy context of this transform block will be updated.

Parameters

[in]	plane	The index of the current plane.
[in]	block	The index of the current transform block in the macroblock. It's defined by number of 4x4 units that have been coded before the currernt transform block.
[in]	blk_row	The row index of the current transform block in the macroblock. Each unit has 4 pixels in y plane.
[in]	blk_col	The col index of the current transform block in the macroblock. Each unit has 4 pixels in y plane.
[in]	plane_bsize	Block size for this plane. When the video source uses chroma subsampling, the block size of UV planes will be smaller than the block size of Y plane.
[in]	tx_size	The given transform size.
[in]	arg	This parameter will be translated into tokenize_b_args, in which RUN_TYPE indicates using regular mode or dry run.

◆ av1_record_txb_context()

void av1_record_txb_context	(	int	plane,
		int	block,
		int	blk_row,
		int	blk_col,
		BLOCK_SIZE	plane_bsize,
		TX_SIZE	tx_size,
		void *	arg )

Update the entropy context related to coefficient coding for a transform block.

There are regular mode and dry run for this function.

Regular mode:

The entropy context of this transform block will be updated.

Dry run:

The probability model update will be skipped.

The entropy context of this transform block will be updated.

Parameters

[in]	plane	The index of the current plane.
[in]	block	The index of the current transform block in the macroblock. It's defined by number of 4x4 units that have been coded before the currernt transform block.
[in]	blk_row	The row index of the current transform block in the macroblock. Each unit has 4 pixels in y plane.
[in]	blk_col	The col index of the current transform block in the macroblock. Each unit has 4 pixels in y plane.
[in]	plane_bsize	Block size for this plane. When the video source uses chroma subsampling, the block size of UV planes will be smaller than the block size of Y plane.
[in]	tx_size	The given transform size.
[in]	arg	This parameter will be translated into tokenize_b_args, in which RUN_TYPE indicates using regular mode or dry run.

◆ av1_get_cb_coeff_buffer()

CB_COEFF_BUFFER * av1_get_cb_coeff_buffer	(	const struct AV1_COMP *	cpi,
		int	mi_row,
		int	mi_col )

Get the corresponding CB_COEFF_BUFFER of the current macro block.

The macroblock's location is described by mi_row and mi_col, row and column mi indexes in the coding frame.

Each mi unit is a 4x4 pixel block.

Parameters

[in]	cpi	Top-level encoder structure.
[in]	mi_row	Row mi index of the current transform block in the frame.
[in]	mi_col	Column mi index of the current transform block in the frame.

Returns: CB_COEFF_BUFFER* Pointer of CB_COEFF_BUFFER associated to this macroblock.

Referenced by encode_sb_row().

◆ av1_cost_skip_txb()

static int av1_cost_skip_txb	(	const CoeffCosts *	coeff_costs,
		const TXB_CTX *const	txb_ctx,
		int	plane,
		TX_SIZE	tx_size )

inlinestatic

Returns the entropy cost associated with skipping the current transform block.

Parameters

[in]	coeff_costs	Table of entropy cost for coefficient coding.
[in]	txb_ctx	Context info for entropy coding transform block skip flag (tx_skip) and the sign of DC coefficient (dc_sign).
[in]	plane	The index of the current plane
[in]	tx_size	The transform size

References CoeffCosts::coeff_costs, and LV_MAP_COEFF_COST::txb_skip_cost.

◆ av1_optimize_txb()

int av1_optimize_txb	(	const struct AV1_COMP *	cpi,
		MACROBLOCK *	x,
		int	plane,
		int	block,
		TX_SIZE	tx_size,
		TX_TYPE	tx_type,
		const TXB_CTX *const	txb_ctx,
		int *	rate_cost,
		int	sharpness )

Adjust the magnitude of quantized coefficients to achieve better rate-distortion (RD) trade-off.

This function goes through each coefficient and greedily choose to lower the coefficient magnitude by 1 or not based on the RD score.

The coefficients are processing in reversed scan order.

Note that, the end of block position (eob) may change if the original last coefficient is lowered to zero.

Parameters

[in]	cpi	Top-level encoder structure
[in]	x	Pointer to structure holding the data for the current encoding macroblock
[in]	plane	The index of the current plane
[in]	block	The index of the current transform block in the
[in]	tx_size	The transform size
[in]	tx_type	The transform type
[in]	txb_ctx	Context info for entropy coding transform block skip flag (tx_skip) and the sign of DC coefficient (dc_sign).
[out]	rate_cost	The entropy cost of coding the transform block after adjustment of coefficients.
[in]	sharpness	When sharpness > 0, the function will be less aggressive towards lowering the magnitude of coefficients. In this way, the transform block will contain more high-frequency coefficients and therefore will preserve the sharpness of the reconstructed block.

◆ av1_cost_coeffs_txb()

int av1_cost_coeffs_txb	(	const MACROBLOCK *	x,
		const int	plane,
		const int	block,
		const TX_SIZE	tx_size,
		const TX_TYPE	tx_type,
		const TXB_CTX *const	txb_ctx,
		int	reduced_tx_set_used )

Compute the entropy cost of coding coefficients in a transform block.

Parameters

[in]	x	Pointer to structure holding the data for the current encoding macroblock.
[in]	plane	The index of the current plane.
[in]	block	The index of the current transform block in the macroblock. It's defined by number of 4x4 units that have been coded before the currernt transform block.
[in]	tx_size	The transform size.
[in]	tx_type	The transform type.
[in]	txb_ctx	Context info for entropy coding transform block skip flag (tx_skip) and the sign of DC coefficient (dc_sign).
[in]	reduced_tx_set_used	Whether the transform type is chosen from a reduced set.

◆ av1_cost_coeffs_txb_laplacian()

int av1_cost_coeffs_txb_laplacian	(	const MACROBLOCK *	x,
		const int	plane,
		const int	block,
		const TX_SIZE	tx_size,
		const TX_TYPE	tx_type,
		const TXB_CTX *const	txb_ctx,
		const int	reduced_tx_set_used,
		const int	adjust_eob )

Estimate the entropy cost of coding a transform block using Laplacian distribution.

This function compute the entropy costs of the end of block position (eob) and the transform type (tx_type) precisely.

Then using av1_cost_coeffs_txb_estimate to estimate the entropy costs of coefficients in the transform block.

In the end, the function returns the sum of entropy costs of end of block position (eob), transform type (tx_type) and coefficients.

Compared to av1_cost_coeffs_txb, this function is much faster but less accurate.

Parameters

[in]	x	Pointer to structure holding the data for the current encoding macroblock
[in]	plane	The index of the current plane
[in]	block	The index of the current transform block in the macroblock. It's defined by number of 4x4 units that have been coded before the currernt transform block
[in]	tx_size	The transform size
[in]	tx_type	The transform type
[in]	txb_ctx	Context info for entropy coding transform block skip flag (tx_skip) and the sign of DC coefficient (dc_sign).
[in]	reduced_tx_set_used	Whether the transform type is chosen from a reduced set.
[in]	adjust_eob	Whether to adjust the end of block position (eob) or not.

Returns: int Estimated entropy cost of coding the transform block.

◆ av1_cost_coeffs_txb_estimate()

int av1_cost_coeffs_txb_estimate	(	const MACROBLOCK *	x,
		const int	plane,
		const int	block,
		const TX_SIZE	tx_size,
		const TX_TYPE	tx_type )

Estimate the entropy cost of transform coefficients using Laplacian distribution.

This function assumes each transform coefficient is of its own Laplacian distribution and the coefficient is the only observation of the Laplacian distribution.

Based on that, each coefficient's coding cost can be estimated by computing the entropy of the corresponding Laplacian distribution.

This function then return the sum of the estimated entropy cost for all coefficients in the transform block.

Note that the entropy cost of end of block (eob) and transform type (tx_type) are not included.

Parameters

[in]	x	Pointer to structure holding the data for the current encoding macroblock
[in]	plane	The index of the current plane
[in]	block	The index of the current transform block in the macroblock. It's defined by number of 4x4 units that have been coded before the currernt transform block
[in]	tx_size	The transform size
[in]	tx_type	The transform type

Returns: int Estimated entropy cost of coefficients in the transform block.

Functions

Detailed Description

Function Documentation

◆ av1_alloc_txb_buf()

◆ av1_free_txb_buf()

◆ av1_write_coeffs_txb()

◆ av1_write_intra_coeffs_mb()

◆ av1_get_txb_entropy_context()

◆ av1_update_intra_mb_txb_context()

◆ av1_update_and_record_txb_context()

◆ av1_record_txb_context()

◆ av1_get_cb_coeff_buffer()

◆ av1_cost_skip_txb()

◆ av1_optimize_txb()

◆ av1_cost_coeffs_txb()

◆ av1_cost_coeffs_txb_laplacian()

◆ av1_cost_coeffs_txb_estimate()