high performance computing on graphics processing units: hgpu.org

In this paper, we discuss a novel approach of improving the performance and accuracy of parallelization compilers by utilizing massively parallel processing power of GPGPU (General Purpose Graphical Processing Units). With the advent of multi core processors, it has become necessary to use parallel programming methodologies. However, parallel programmers need tools that can provide support at the time of program development. Such tools require program analysis after every logical completion of the program. This process is time consuming. Existing solutions, provided by researchers, uses the incremental program analysis approach to avoid reanalysis of the code every time. This approach is practical but compromises accuracy and does not give real time performance. GPGPU is the revolutionary advancement in parallel computing area and its architecture designed to suit data parallel applications. The program analysis algorithms, required for developing parallel programming tools, are invariably not data parallel. However, it is possible to redesign those algorithms to exploit the parallel processing power of GPGPU. In this paper, we discuss the design of a novel data parallel side effect analysis (DP-SEA) and reported a 44 times speed up over the CPU implementation.