To implement our parallel version of the Gauss algorithm we use the task framework provided by HPC-GAP. The structure of the source files reflects this by grouping our functions depending on how they make use of HPC-GAP's shared memory model.
main.gi: Contains the main function DoEchelonMatTransformationBlockwise, which is wrapped by EchelonMatBlockwise (2.1-2) and EchelonMatTransformationBlockwise (2.1-1). It is the function which schedules all tasks.
dependencies.g: The functions in this file compute the dependencies of the algorithm's subprograms between each other.
tasks.g: The functions in this file are scheduled as tasks by the main routine. They need to make sure that they only write read-only objects into the "shared" atomic lists.
thread-local.g: The functions in this file are called by functions from "tasks.g". In principle, these functions only work in a single thread-local region and don't need to know anything about other threads. These functions may only access read-only objects or objects from the executing thread's thread-local region. They may only emit or write into thread-local objects.
generated by GAPDoc2HTML