Struct SimdPageRank 

pub struct SimdPageRank;

SIMD-accelerated operations for PageRank power iteration.

These functions accelerate the inner loops of PageRank computation: distributing rank contributions, computing convergence (L1 norm of the difference vector), and normalizing the rank vector.

Implementations

impl SimdPageRank

pub fn power_iteration_step( transition_matrix: &Array2<f64>, old_rank: &Array1<f64>, damping: f64, ) -> Result<Array1<f64>>

Perform one PageRank power iteration step using SIMD.

Computes new_rank = (1 - damping) / n + damping * M * old_rank where M is the column-stochastic transition matrix.

Arguments

• transition_matrix - Row-stochastic transition matrix (n x n)
• old_rank - Current rank vector (length n)
• damping - Damping factor (typically 0.85)

Returns

The new rank vector after one iteration.

pub fn convergence_l1( new_rank: &ArrayView1<'_, f64>, old_rank: &ArrayView1<'_, f64>, ) -> f64

Compute L1 norm of the difference between two rank vectors (convergence check).

Returns sum(|new_rank[i] - old_rank[i]|) using SIMD acceleration.

Arguments

• new_rank - New rank vector
• old_rank - Previous rank vector

Returns

The L1 norm of the difference.

pub fn has_converged( new_rank: &ArrayView1<'_, f64>, old_rank: &ArrayView1<'_, f64>, tolerance: f64, ) -> bool

Check if PageRank has converged by comparing L1 norm against tolerance.

Arguments

• new_rank - New rank vector
• old_rank - Previous rank vector
• tolerance - Convergence threshold

Returns

true if the L1 norm of the difference is below tolerance.

pub fn scale_rank(rank: &ArrayView1<'_, f64>, scalar: f64) -> Array1<f64>

SIMD-accelerated vector-scalar multiply for rank distribution.

Computes result[i] = rank[i] * scalar for distributing a node’s rank equally among its neighbors.

Arguments

• rank - The rank vector
• scalar - The scalar multiplier (e.g., damping / degree)

Returns

The scaled vector.

pub fn normalize_rank(rank: &Array1<f64>) -> Result<Array1<f64>>

Normalize a rank vector so it sums to 1.0.

Arguments

• rank - The rank vector to normalize

Returns

The normalized rank vector, or an error if the sum is zero.

pub fn dangling_node_contribution( rank: &ArrayView1<'_, f64>, is_dangling: &[bool], damping: f64, ) -> f64

Compute damping-adjusted teleportation vector.

For dangling nodes (nodes with no outgoing edges), their rank is distributed uniformly. This function computes the dangling node contribution: (damping / n) * sum(rank[i] for dangling node i).

Arguments

• rank - Current rank vector
• is_dangling - Boolean mask: true for dangling nodes
• damping - Damping factor

Returns

The uniform teleportation contribution per node.

pub fn compute_pagerank( transition_matrix: &Array2<f64>, damping: f64, tolerance: f64, max_iterations: usize, ) -> Result<(Array1<f64>, usize)>

Full SIMD-accelerated PageRank computation.

Runs the PageRank power iteration until convergence or max iterations.

Arguments

• transition_matrix - Row-stochastic transition matrix (n x n)
• damping - Damping factor (typically 0.85)
• tolerance - Convergence threshold
• max_iterations - Maximum number of iterations

Returns

A tuple of (rank vector, iterations used).