Module tiled 

Tiled raster processing for datasets exceeding VRAM budget.

This module provides automatic tile-based decomposition of large rasters that would otherwise overflow GPU VRAM. Key concepts:

• A tile is a rectangular sub-region of the full raster.
• An overlap halo (optional) extends each tile by overlap_pixels on every active edge, using edge-replication for boundary pixels.
• Splitting decomposes a flat f32 raster into Vec<RasterTile>.
• Stitching assembles processed tiles back into a full raster, discarding the halo and writing only the core interior.
• auto_tile_size finds a tile size that fits inside an available VRAM budget by halving the area until it fits.
• execute_tiled orchestrates split → per-tile tile_fn → stitch.

Structs

RasterTile

A single tile extracted from a raster, optionally padded with an overlap halo.

TiledConfig

Configuration for tiled raster processing.

Functions

auto_tile_size

Compute an auto-selected tile size that fits within the effective VRAM budget.

execute_tiled

High-level tiled executor: split → per-tile tile_fn → stitch.

split_into_tiles

Split a flat f32 raster (row-major) into tiles with an optional overlap halo.

stitch_tiles

Stitch processed tiles back into a full raster.

vram_per_tile

Estimate the VRAM required to process one tile.