Integrating speculative execution to task parallelism

Abstract

In addition to loops, there are other hot-code regions that parallel programming tries to parallelize. Task-based programming model simplifies it by providing annotations of dependences between tasks. Thus, a runtime system manages these dependences and schedules tasks to execute on cores. However, the programmer must use directives (e.g. taskwait) to ensure control dependences between tasks when these can even be data independent and therefore executed speculatively in parallel, or she/she must mark as in or out dependences to variables which are data dependent rarely and may be speculated.This proposal describes a novel speculation technique for the parallel execution of tasks. This technique, called Speculative Task Execution (STE), speculatively executes tasks of while iterations which have control dependences with previous tasks (Speculative While). However, STE can also be used to speculate data dependences between tasks of different iterations in a loop generated by the taskloop construct (Speculative Taskloop). We propose the implementation of a prototype to STE and the design of new speculative constructs and clauses to integrate STE into OpenMP task-based parallelization as: (a) the stl clause for the taskloop construct; (b) the while ste construct, and (c) the spec_in, spec_out, and spec_inout dependence-type modifiers for the depend clause. (AU)