Chapter 15. Parallel Schedulers

• 15.1 ParNode
  • 15.1.1 ParNode protected members
  • 15.1.2 Other ParNode public members
  • 15.1.3 Iterators for ParNode

• 15.10 Hu Level Scheduler
  • 15.10.1 Class HuNode
  • 15.10.2 Class HuGraph
  • 15.10.3 Class HuScheduler
  • 15.10.4 Class HuParProcs

• 15.2 Class ParGraph
  • 15.2.1 Other ParGraph protected members
  • 15.2.2 Other ParGraph public members
  • 15.2.3 Class NodePair

• 15.3 Class ParScheduler
  • 15.3.1 compileRun method
  • 15.3.2 Other ParScheduler protected members
  • 15.3.3 Other ParScheduler public members

• 15.4 class ParProcessors
  • 15.4.1 Other ParProcessors protected members
  • 15.4.2 Other ParProcessors public members

• 15.5 UniProcessor
  • 15.5.1 Class NodeSchedule
  • 15.5.2 Members for scheduling
  • 15.5.3 Sub-Universe creation
    • Spread and Collect stars
  • 15.5.4 Members for code generation
  • 15.5.5 Other UniProcessor protected members
  • 15.5.6 Other UniProcessor public members
  • 15.5.7 Iterator for UniProcessor

• 15.6 Dynamic Level Scheduler

• 15.7 Class DLGraph

• 15.8 class DLScheduler

• 15.9 Class DLParProcs

• 15.11 Declustering Scheduler
  • 15.11.1 Class DCNode
  • 15.11.2 Classes DCArc and DCArcList
  • 15.11.3 Class DCGraph
  • 15.11.4 Class DCCluster
  • 15.11.5 Class DCClusterList
  • 15.11.6 Class DCClustArc and class DCClustArcList
  • 15.11.7 Class DCParProcs

Authors: Soonhoi Ha

The base scheduler object is ParScheduler. Since it is derived from class SDFScheduler, it inherits many methods and members from the SDFScheduler class. The ParScheduler class has a ParProcessors class that has member methods to implement the main scheduling algorithm. The ParProcessors class has an array of UniProcessor classes. The UniProcessor class, privately derived from class DoubleLinkList, is mapped to a processing element in the target architecture.

Note that all parallel scheduling algorithms are retargettable: they do not assume any specific topology while they take the effect of topology into account to estimate the interprocessor communication overhead.

Refer to class ParScheduler, to see the overall procedure of parallel scheduling. Refer to class UniProcessor, to see the procedure of sub-universe generations.