Optimized `_interpolate_colors` in `ShaderParamsReader` to use `torch.bucketize` and vectorized indexing instead of iterating over color stops. This improves performance from O(Stops) to O(1) in terms of kernel launches/Python overhead. Updated `CurlNoiseGenerator` to use the optimized shared implementation instead of its own slower local version. Also fixed a regression in `apply_shape_mask` hexgrid calculation.
60 KiB
60 KiB