Task Granularity Analysis Method Using Static Metrics of Source Code for C Programs

In the first phase of our automatic parallelizing translator for C program, a source code is decomposed into a set of tasks of the granularity of a statement level at the minimum. In the next phase, task scheduling which determines statically by which processor these tasks are processed is performed. Since this task scheduling is a combinatorial optimization problem, it is important for it to suppress the number of tasks which constitutes the program. Therefore, useless parallelism is removed using the information about the dependencies among tasks and task cost, and the task granularity analysis is required in order to make task granularity reasonable. However, since it is necessary to analyze the processing time of a task only using the information acquired from a source code, exact cost may be unable to be assigned in the time analysis using the conventional computational complexity analysis. So, in this paper, the method of aiming at the improvement in accuracy of execution time analysis is proposed by applying static metrics of source code