FZGPUModules 2.0
GPU-accelerated modular compression pipelines
Loading...
Searching...
No Matches
tiled_lorenzo_stage.h File Reference

Dimension-aware (tiled separable) Lorenzo predictor — cuSZp3 delta. Lossless. More...

#include "stage/stage.h"
#include "fzm_format.h"
#include <cuda_runtime.h>
#include <algorithm>
#include <array>
#include <cstdint>
#include <cstring>
#include <stdexcept>
#include <string>
#include <type_traits>
#include <unordered_map>
#include <vector>

Go to the source code of this file.

Classes

struct  fz::TiledLorenzoConfig
 
class  fz::TiledLorenzoStage< T >
 

Namespaces

namespace  fz
 

Detailed Description

Dimension-aware (tiled separable) Lorenzo predictor — cuSZp3 delta. Lossless.

Within each fixed-size tile (8x8 in 2-D, 4x4x4 in 3-D by default) the predictor applies a separable integer Lorenzo: the leading x-column predicts down y, the leading y/z edges predict down z, and every other element predicts from its x neighbor. The forward pass emits its output in tile-major order (each tile's elements contiguous, edge tiles zero-padded) so a downstream AdaptiveBitpackStage(block_size = tile_elems) packs each tile as exactly one block — reproducing cuSZp3's compression ratio. The inverse un-tiles back to the original natural (row-major) order.

Supported element types: signed int16_t, int32_t (linear quantizer codes).