Module gpu_accel 

CPU-backed scaffold for future GPU acceleration of spatial algorithms

This module is a placeholder/scaffold for GPU-accelerated spatial algorithms. GPU compute is not yet implemented: every compute method in this module currently runs on the crate’s optimized CPU SIMD fallback and returns correct results — none of them dispatch any work to a graphics card.

The cuda, rocm, and vulkan Cargo features, when enabled, currently gate only GPU capability detection (probing nvidia-smi, rocm-smi, and vulkaninfo). They do not unlock any GPU compute path; the methods below always delegate to the CPU SIMD implementations regardless.

A future release may add a real GPU backend (for example via the COOLJAPAN oxicuda-* ecosystem). Until then, for production GPU-free work the CPU equivalents in this crate are what actually run — see the simd_distance module and the AdvancedSimdKMeans / AdvancedSimdNearestNeighbors types.

Planned backends (scaffold — not yet functional)

These are the intended targets once a real GPU path lands; none of them perform GPU compute today:

• NVIDIA GPUs (CUDA backend)
• AMD GPUs (ROCm/HIP backend)
• Intel GPUs (Level Zero backend)
• Vulkan compute for cross-platform support

Examples

The example uses the GPU-named API, but the computation transparently runs on the CPU SIMD fallback today — no GPU is required or used.

use scirs2_spatial::gpu_accel::{GpuDistanceMatrix, GpuKMeans};
use scirs2_core::ndarray::array;

// Distance matrix computation (runs on the CPU SIMD fallback)
let points = array![[0.0, 0.0], [1.0, 0.0], [0.0, 1.0], [1.0, 1.0]];

let gpu_matrix = GpuDistanceMatrix::new()?;
let distances = gpu_matrix.compute_parallel(&points.view()).await?;
println!("GPU distance matrix: {:?}", distances);

// K-means clustering (runs on the CPU SIMD fallback)
let gpu_kmeans = GpuKMeans::new(2)?;
let (centroids, assignments) = gpu_kmeans.fit(&points.view()).await?;
println!("GPU centroids: {:?}", centroids);

Structs

GpuCapabilities

GPU device capabilities and information

GpuDevice

GPU device management and capability detection

GpuDistanceMatrix

GPU-accelerated distance matrix computation

GpuKMeans

GPU-accelerated K-means clustering

GpuNearestNeighbors

GPU-accelerated nearest neighbor search

HybridProcessor

Hybrid CPU-GPU workload distribution

Enums

GpuBackend

Supported GPU backends

ProcessingStrategy

Processing strategy for workload distribution

Functions

get_gpu_capabilities

Get GPU capabilities

global_gpu_device

Get the global GPU device instance

is_gpu_acceleration_available

Check if GPU acceleration is available globally

report_gpu_status

Report GPU acceleration status