Module auto_parallel 

Automatic parallelization for computation graphs.

This module provides automatic detection and exploitation of parallelism opportunities:

• Dependency analysis: Build dependency graphs and detect parallelizable operations
• Cost modeling: Estimate execution costs and communication overhead
• Work partitioning: Dynamically partition work across threads/devices
• Load balancing: Balance work to minimize idle time
• Pipeline detection: Identify pipeline parallelism opportunities

Example

use tensorlogic_infer::{AutoParallelizer, ParallelizationStrategy, CostModel};

// Create auto-parallelizer with cost model
let parallelizer = AutoParallelizer::new()
    .with_strategy(ParallelizationStrategy::Aggressive)
    .with_cost_model(CostModel::ProfileBased);

// Analyze graph for parallelism
let analysis = parallelizer.analyze(&graph)?;
println!("Found {} parallelizable stages", analysis.num_stages);

// Generate parallel execution plan
let plan = parallelizer.generate_plan(&graph)?;

Structs

AutoParallelizer

Automatic parallelizer.

NodeInfo

Node information for parallelization analysis.

ParallelExecutionPlan

Parallel execution plan.

ParallelStage

Parallel stage containing nodes that can execute concurrently.

ParallelizationAnalysis

Parallelization analysis results.

WorkPartition

Work partition for a single worker.

Enums

AutoParallelError

Auto-parallelization errors.

CostModel

Cost model type.

DependencyType

Dependency type between nodes.

ParallelizationStrategy

Parallelization strategy.

Type Aliases

NodeId

Node ID in the computation graph.