Reusable GPU query batches and resident bounded k-NN/radius neighborhoods

CHANGELOG.md

@@ -21,6 +21,13 @@ removed no sooner than the next major (see `docs/API_STABILITY.md`).
 
 ### Changed
 
+- Add explicitly uploaded reusable GPU query batches and resident KD-tree
+  nearest-k/radius neighborhoods under `gpu-wgpu`. k=0..=64 retains CPU
+  selection and stable distance order; radius rows retain traversal order and
+  count all matches even when capacity is exceeded. Complete readback access
+  rejects overflow; named truncated access and count-only search are explicit.
+  Owned result buffers can feed subsequent GPU kernels without readback.
+
 - Extend explicit GPU correspondence reuse to GICP under `register-gicp-gpu`,
   retaining unbounded matching followed by the original host distance gate,
   ordered Hessian/gradient reduction and CPU pose solving. Optional radius

README.md

@@ -649,6 +649,13 @@ GICP exposes the same explicit correspondence entry point under
 and share the transfer receipt. Its Hessian/gradient accumulation and pose
 solver stay on CPU. See [GICP measurements](notes/2026-10-12_gpu_gicp_correspondences.md).
 
+GPU k-NN/radius search can retain its results for later device passes with
+`GpuQueryBatch::upload` and
+`GpuKdTree::{nearest_k_resident,radius_neighbors_resident}` (`gpu-wgpu`). Radius
+capacity overflow is counted explicitly and complete readback rejects it.
+This provides the neighborhood foundation for feature compute; FPFH/ISS/MLS
+remain CPU algorithms. See [contracts and measurements](notes/2026-10-12_gpu_bounded_neighborhoods.md).
+
 ### Library
 
 Load or save by file extension:

crates/spatialrust-gpu/src/kernels/mod.rs

@@ -51,7 +51,10 @@ mod voxel_gather;
 
 pub(crate) mod cluster_union;
 pub(crate) mod nearest;
-pub use nearest::GpuKdTree;
+pub use nearest::{
+    GpuKdTree, GpuNeighborhoodKind, GpuNeighborhoodReadback, GpuNeighborhoods, GpuQueryBatch,
+    MAX_GPU_K_NEIGHBORS,
+};
 #[cfg(feature = "gpu-wgpu")]
 mod euclidean_cluster;
 

crates/spatialrust-gpu/src/kernels/nearest.rs

@@ -7,6 +7,14 @@ use wgpu::util::DeviceExt;
 
 use crate::{scratch::ScratchBuffers, WgpuRuntime};
 
+#[path = "neighborhoods.rs"]
+mod neighborhoods;
+pub(crate) use neighborhoods::NeighborhoodPipelines;
+pub use neighborhoods::{
+    GpuNeighborhoodKind, GpuNeighborhoodReadback, GpuNeighborhoods, GpuQueryBatch,
+    MAX_GPU_K_NEIGHBORS,
+};
+
 #[repr(C)]
 #[derive(Clone, Copy, Pod, Zeroable)]
 struct Node {