Implementing Coarse Grained Task Parallelism Using OpenMP

Abstract — Parallelization is an important technique to increase the performance of software programs. Parallel programs are written to make efficient use of multiple cores. Most of the existing legacy applications are sequential without any multithreading or parallel programming. Manual efforts required to parallelize these applications are huge and complicated. So there is a need for automatic parallelization tools. The proposed system implements coarse grained task parallelism with the insertion of OpenMP directives in an input C program. The output is a multithreaded C program which can utilize multiple cores on multi-core shared memory systems. The system is placed between application and compiler and generated output needs to be compiled like a normal C program. This is a source to source conversion tool as the input and output are both C source code. To extend beyond the reach of loop level parallelism, task level parallelism is proposed and provides better scalability. Depending on the nature of the program, parallelization may have a performance improvement