Exploitation of Task Level Parallelism

Existing many systems were supporting task level parallelism usually involving the process of task creation and synchronization. The synchronization of task requires the clear definition about existing dependencies in a program or data-flow restraints among functions(tasks), or data usable information of the tasks. This thesis describes a method called Symbol-Table method which will used to exploits and detects the task level parallelism at inner level of sequential C-programs. This method is made up of two levels: a normal symbol table and an extended symbol table. A sequential program of C language is the input to the normal gcc compiler in which the procedures are defines as functions(tasks). Than we generate a normal symbol table with specific command by gcc compiler in Linux as an output. Then we use the information of that symbol table for generating the extended symbol table with additional information about variable's extended scopes and inner level function dependency. This extended symbol table is generated by the use of previously generated normal symbol table on the basis of variable's scope and L-value/R-value attributes. By that table we can identify the functions and variables those who are sharing the common variables and those who are accessing the different functions with extended scopes respectively. Then we can generate the program dependency graph by the using of that extended symbol table's information with a specific java program. A simple program for using this method has been implemented on a 64-bits linux based multiprocessor. Finally we can generate the Function graph for every variable in the program with the help of table's info and dependency graph's states. From that graph we can get the info about extended scoping of variables to identify and exploit the task level parallelism in the program. Then we can apply the parallelism with MPI or other parallel platforms to get optimized and error free parallelism