Nikhil Jindal, Akhil Jindal, Divyashikha Sethia
Abstract
In this research thesis, an auto-parallelisation tool for C codes is presented. The proposed tool takes a serial C code as input and generates an equivalent parallel code as output. Due to its widespread use, NVIDIA’s Compute Unified Device Architecture (CUDA) is used as the target platform for running the output parallel code. The tool works in two phases. It first automatically identifies parallelisable portions of input code and then generates the equivalent CUDA code. hiCUDA is used as the intermediate language between the two phases. The tool has been written in Perl. The tool is aimed at parallelising the vast repository of legacy C codes which at present is unable to take any advantage of the significant improvements in the field of high performance computing. Preliminary results show that the tool is able to parallelise a large percentage of the usual C codes and provides a significant speedup of ~10 times in the process.
Publication
"Automated Tool to Generate Parallel CUDA Code from a Serial C Code” has been published Applications is pleased to inform that your manuscript is published in IJCA (International Journal of Computer Applications) July 2012 Edition with effect from July 28, 2012.