v1.1.0

What's New

Performance Optimizations

• Cached ravel stride coefficients — eliminates repeated computation on every query
• Fused get_valid_ravel_indices — combines bounds checking, coordinate conversion, and raveling into a single pass
• Precomputed inverse resolution (_inv_resolution) — replaces division with multiplication in the hot path
• torch.nn.functional.grid_sample backend — 3-12x faster linear interpolation for 2D/3D float tensors (used automatically when applicable)

New Features

• BatchedViewLookup — vectorized lookup across multiple TorchMultidimView instances in a single call, useful for querying multiple cached SDFs simultaneously

Bug Fixes

• Fixed numpy backend: ensure_value_key, linear interpolation, and ravel_multi_index now work correctly with NumPy arrays
• Fixed infinite resolution edge case for size-1 grid dimensions

Other

• Added torch>=1.4 version constraint
• Added GitHub Actions CI (Python 3.9, 3.12)