Module dataflow 

Dataflow Graph Representation for Energy-Aware Execution

This module provides a dataflow graph (DFG) representation that enables:

• Implicit parallelism from data dependencies
• Energy-aware scheduling decisions
• RipTide-style ordered dataflow execution

Key Concepts

• Operators: Computation nodes that fire when inputs are ready
• Channels: Bounded FIFO queues connecting operators
• Steer: Conditional routing (RipTide’s control-flow primitive)
• Stream: Fused loop induction variable (27-52% operator reduction)
• Carry: Loop-carried dependencies

Energy-Aware Scheduling

Unlike traditional dataflow, Joule’s scheduler considers:

• Current thermal state
• Power/energy budgets
• Available parallelism

This allows the runtime to dynamically throttle parallelism when thermal or energy constraints are active.

Re-exports

pub use extract::DfgExtractor;

pub use optimize::DfgOptimizer;

pub use optimize::StreamFusion;

pub use schedule::EnergyAwareScheduler;

pub use schedule::SchedulingConfig;

pub use schedule::ThermalConstraint;

Modules

extract

MIR to Dataflow Graph Extraction

optimize

Dataflow Graph Optimizations

schedule

Energy-Aware Dataflow Scheduler

Structs

Channel

A channel connecting operators

ChannelId

Unique identifier for a channel between operators

DataflowGraph

A complete dataflow graph

DependencyAnalysis

Dependency analysis for a dataflow graph

EnergyEstimate

Energy estimate for an operator

OperatorId

Unique identifier for a dataflow operator

ScheduleStep

A step in the schedule (operators that can execute in parallel)

ScheduledDfg

A scheduled dataflow graph ready for execution

TokenConversionError

Error when converting a token value to an incompatible type

Enums

ComputeOp

Arithmetic/logic operations for Compute operators

DfOperator

A dataflow operator that fires when all inputs are ready

MemoryOp

Memory operations

TokenType

Type of token values

TokenValue

Value carried by a token through a channel