fz::LorenzoQuantConfig Struct Reference

#include <lorenzo_quant.h>

Public Attributes
float	error_bound
	Absolute bound after mode conversion (used by decompressor).
uint32_t	quant_radius
	Quantization radius.
uint32_t	num_elements
	Total element count.
uint32_t	outlier_count
	Actual number of outliers.
DataType	input_type
	Original input type (1B).
DataType	code_type
	Quantization code type (1B).
uint8_t	ndim
	Spatial dimensionality 1/2/3 (0 treated as 1).
uint8_t	eb_mode
	ErrorBoundMode cast to uint8_t.
uint32_t	dim_x
	X (fast) dimension; 0 = infer from num_elements.
uint32_t	dim_y
	Y dimension (1 for 1-D).
uint32_t	dim_z
	Z dimension (1 for 1-D/2-D).
float	user_eb
	Original user-specified error bound value.
float	value_base
	value_range (NOA) or max(|data|) (REL) used in conversion.
uint8_t	zigzag_codes
	1 if codes are zigzag-encoded, else 0.
uint8_t	centering
	1 if per-tile mean centering is enabled, else 0.
uint8_t	reserved [2]
double	error_bound_f64
double	value_base_f64
	Full-precision value_base; 0 in pre-2026-08-08 headers.

Detailed Description

Serialized Lorenzo predictor configuration stored in FZMBufferEntry.stage_config.

Written by serializeHeader() at compression time; read back by deserializeHeader() to reconstruct decompressor state. Fits within the 128-byte FZM_STAGE_CONFIG_SIZE limit.

Member Data Documentation

reserved

uint8_t fz::LorenzoQuantConfig::reserved[2]

Must be zero.

error_bound_f64

double fz::LorenzoQuantConfig::error_bound_f64

Full-precision absolute bound. error_bound above is a narrow copy kept for older readers and CANNOT carry an f64 bound: the inverse reconstructs by prefix-summing quantized residuals and scaling by 2*abs_eb, so for a field like S3D/N2 (sum ~ 3.3e8) a 6e-08 relative error in the stored bound lands as a 4.4e-08 absolute error – 40x the bound itself. 0 in headers written before 2026-08-08; readers fall back to error_bound.