v0.1.7

• Enhance integer dtype handling: promote narrow integers to float32 and wide integers to float64 for GPU and CPU operations. Add tests for integer binary operations to verify compute-dtype dispatch.
• Fix compute-dtype selection for integer rasters and use native dtypes in benchmarks
• Add streamed windowed IO for tile-by-tile file processing (vibeSpatial-fx3.6)
• Add multi-pass tiling for CCL and distance transform (vibeSpatial-fx3.5)
• Add accumulator map-reduce tiling for histogram/zonal ops (vibeSpatial-fx3.4)
• Add halo-aware tiling for stencil/focal operations (vibeSpatial-fx3.3)
• Add tiling execution engine for pointwise operations (vibeSpatial-fx3.2)
• Add RasterPlan analyzer for VRAM-aware spatial tiling (vibeSpatial-fx3.1)
• Make Tier B OOM-safe pool the default and fix D→H memory init path
• Add tiered RMM memory pool manager (ADR-0040) to fix shared-GPU OOM
• Fused NVRTC binary kernel replacing 7-kernel CuPy chain + multiband dispatch
• Fix raster_apply() auto-dispatch: set threshold=0 for user-provided functions
• Fix stale tests for multiband morphology dispatch and expression shape mismatch
• Phase 4: Comprehensive multiband test suite
• Add spectral index convenience functions (raster_ndvi, raster_band_ratio, raster_band_math)
• Expression engine band indexing parser
• Add band-fused expression kernel template
• Multiband zonal stats dispatch and CPU fallbacks for algebra operations
• Multiband dispatch for distance transform, hydrology, and histogram
• Multiband dispatch for CCL and morphology operations
• Multiband dispatch for resampling operations
• Multiband dispatch for focal and terrain operations (vibeSpatial-2to.3.1)
• Add band dispatch executors for per-band GPU and CPU processing
• Phase 1 band dispatch infrastructure: VRAM budget and OwnedRasterArray band methods
• Emit warnings on silent multiband squeeze in focal/algebra/label/hydrology
• Add band selection parameter to polygonize
• Add multiband validation guard to zonal stats
• Add multiband validation guard to histogram functions

v0.1.6

• Add Python engineer agent
• Complete BUG_SWEEP: device residency, zonal tests, test fixes, tolerance tightening
• Fix four bugs from sweep: pad_value collision, implicit syncs, host-side XOR, non-atomic writes
• Fix four dispatch/overflow/ping-pong bugs from parallel worktree sweep
• Fix Gaussian filter sigma<=0 causing division by zero and NaN kernel weights
• Fix focal std ddof inconsistency: align to population std (ddof=0)
• Fix GPU terrain kernel boundary handling: use edge replication instead of 0-fill
• Fix raster_divide double-applied nodata masking (Bug #10)
• Fix hydrology convergence returning unconverged results silently (Bug #9)
• Fix histogram equalize nodata information loss for non-uint8 (Bug #8)
• Fix CPU histogram equalize NaN nodata filtering (Bug #7)
• Fix focal stats crash on multi-band rasters (Bug #6: 3D nodata_mask vs 2D data)
• Fix terrain derivative nodata metadata mismatch (Bug #5)
• Fix histogram equalize kernel parameter type mismatch (Bug Sweep #4)
• Fix _ensure_host_state silently ignoring CuPy import failure with device data
• Fix nodata_mask reading stale host data on device-resident rasters
• Fix resample kernel parameter type mismatch for float32/int16/uint8 with nodata
• Add sphinx-autoapi for auto-generated API reference docs
• Fix review findings: eliminate D→H→D ping-pongs, remove implicit syncs, update docs
• Fix distance transform D→H→D ping-pong: use device residency APIs
• Fix terrain derivative GPU function: add missing return and diagnostics
• Merge branch 'worktree-agent-af380a3a'
• Merge branch 'worktree-agent-a43e064f'
• Merge branch 'worktree-agent-acbb2b3f'
• Merge branch 'worktree-agent-ae031f8c'
• Merge branch 'worktree-agent-a3c42e5f'
• Merge branch 'worktree-agent-a30f0bde'
• Merge branch 'worktree-agent-a003b8df'
• Merge branch 'worktree-agent-a450f50c'
• Merge branch 'worktree-agent-a7362140'
• Merge branch 'worktree-agent-a3046107'
• Add GPU sink/depression fill for DEM conditioning
• Add GPU histogram, CDF, equalization, and percentile operations
• Add GPU resampling/warp with nearest, bilinear, and bicubic kernels
• Add GPU Euclidean distance transform via Jump Flooding Algorithm
• Add fused element-wise algebra via runtime NVRTC expression compilation
• Add GPU focal statistics: min/max/mean/std/range/variety kernels
• Add GPU terrain derivatives: TRI, TPI, and curvature kernels
• Add GPU hillshade kernel fusing Horn-method slope/aspect with illumination
• Add arbitrary NxN structuring elements for GPU morphology
• Add GPU sieve filter kernel for zero-copy CCL pipeline
• Add benchmarking script for vibespatial.raster operations
• Deep CUDA optimization pass across all GPU kernels and dispatch paths
• Add Claude Code skill, agent, and hook infrastructure
• Fix all pre-existing test failures and cap Python version

v0.1.5

• Fix 12 bugs found during deep codebase audit

v0.1.4

• Merge PyPI publish into release workflow to fix GITHUB_TOKEN trigger issue

v0.1.3

• Generate release notes from commit subjects instead of GitHub auto-notes
• Add AGENTS.md for AI agent onboarding
• Add NVRTC disk cache re-exports and tests
• Add pre-commit hooks with ruff lint/format and zero-copy compliance checks
• Add cuda-python and cuda-cccl to cu12/cu13 extras
• Add cu12/cu13 optional dependency extras
• Add automated release workflow

v0.1.2

Full Changelog: v0.1.1...v0.1.2

Release v0.1.1

Metadata updates for pypi and docs

Release v0.1.0

vibespatial-raster

GPU-first raster processing for vibeSpatial. Custom NVRTC kernels and CCCL primitives for every operation — algebra, focal, zonal stats, labeling, rasterize, polygonize — with CPU fallbacks via scipy and rasterio.

Operations

Category	Operations	GPU Backend
IO	GeoTIFF, COG, JPEG2000 read/write	nvImageCodec (zero-copy decode to device)
Algebra	add, subtract, multiply, divide, apply, where, classify	CuPy element-wise
Focal	convolution, Gaussian filter, slope, aspect	NVRTC shared-memory stencil kernels
Zonal	count, sum, mean, min, max, std, median	CCCL segmented reduce
Rasterize	vector-to-raster (point-in-polygon)	NVRTC per-pixel PIP kernel
Label	connected components (4/8 connectivity)	NVRTC union-find
Morphology	erode, dilate, open, close, sieve	NVRTC tiled kernels
Polygonize	raster-to-vector (marching squares)	NVRTC classify + emit edges