Module parallel_orchestrator 

Parallel solver orchestration for multi-stage physics pipelines.

This module provides a dependency-aware scheduler that organizes solver stages into parallel waves via topological sorting. Stages within the same wave have no mutual dependencies and can conceptually execute in parallel, while stages in later waves depend on earlier ones.

Architecture

The orchestrator works in three phases:

1. Registration - stages and their dependencies are added via ParallelOrchestrator::add_stage.
2. Scheduling - ParallelOrchestrator::compute_schedule performs a topological sort to group independent stages into waves (Kahn’s algorithm with cycle detection).
3. Execution - ParallelOrchestrator::execute runs each wave sequentially, executing stages within each wave. Per-stage wall-clock timings are accumulated.

Future: rayon-based parallelism

Currently stages within a wave run sequentially. When rayon is added as a workspace dependency, intra-wave parallelism can be enabled behind a feature flag (e.g. parallel-rayon) by replacing the sequential loop with rayon::scope or rayon::join.

Structs

ParallelOrchestrator

Orchestrates the execution of multiple solver stages respecting dependencies.

PipelineSchedule

A computed execution schedule produced by topological sorting.

StageDependency

Describes ordering constraints for a single stage.

Enums

OrchestratorError

Errors that can occur during orchestration scheduling or execution.

Traits

SolverStage

A single stage in a physics solver pipeline.

Functions

topological_sort

Performs a topological sort of n stages using Kahn’s algorithm.