Module gpu 

GPU-accelerated and parallel graph algorithms.

This module provides parallel graph traversal and shortest-path algorithms designed with a GPU-ready interface. Current implementations use CPU-parallel execution via Rayon-compatible patterns and are ready to be backed by actual GPU kernels in future releases.

Algorithms

• algorithms::gpu_bfs — Parallel BFS (level-synchronous, frontier-based)
• algorithms::gpu_sssp_bellman_ford — Bellman-Ford SSSP (GPU-friendly, detects negative cycles)
• algorithms::gpu_sssp_delta_stepping — Delta-stepping SSSP (highly parallel, better cache behavior than Dijkstra)

Graph Format

Algorithms in this module accept graphs in Compressed Sparse Row (CSR) format for BFS/Bellman-Ford, or adjacency-list format for delta-stepping. CSR is the preferred format for GPU workloads due to coalesced memory access.

use scirs2_graph::gpu::algorithms::{gpu_bfs, GpuBfsConfig};

// 3-node path: 0 -> 1 -> 2
let row_ptr = vec![0, 1, 2, 2];
let col_idx = vec![1, 2];
let config = GpuBfsConfig::default();
let dist = gpu_bfs(&row_ptr, &col_idx, 0, &config).expect("bfs failed");
assert_eq!(dist[0], 0);
assert_eq!(dist[1], 1);
assert_eq!(dist[2], 2);

Re-exports

pub use algorithms::gpu_bfs;

pub use algorithms::gpu_sssp_bellman_ford;

pub use algorithms::gpu_sssp_delta_stepping;

pub use algorithms::GpuBfsConfig;

pub use algorithms::GpuGraphBackend;

Modules

algorithms

Parallel BFS and SSSP algorithms with a GPU-ready interface.