A Survey of Task-Based Parallel Programming Models
Xin Li · ASME Press eBooks · 2011
With the popular emergency of the multicore CPUs and manycore GPUs due to power usage and heat generation, developing task-level parallelism transparently of application software is the significant challenge for programmers to leverage the increasing number of processor cores or resources. It is difficult for programmers to manually match the execution requirements of different parts of the program as tasks with the computational capabilities of computing resources. People have proposed a variety of task-based programming models to well exploit inherent parallelism hidden in application software. This paper reviews those popular models and helps to understand gaps and opportunities provided in terms of performance and efficiency and benefits for the parallelism from different execution mechanisms. In future, we may great benefit from task-level parallel compilers with specific APIs to allow data dependences analysis and task scheduling automatically and efficiently at runtime.