Affine calculator in GCC

• Main Authors: Ying-Jie Wang, 王穎杰
• Other Authors: 廖世偉
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/29960853231347657366

碩士 === 國立臺灣大學 === 資訊工程學研究所 === 97 === Multicore processors have become pervasive in these days. But, it is still difficult to program these architectures to effectively utilize the computation power of multiple processing units. There are several research topics addressing this issue, one of the strong and at the same time very hard approaches is automatic parallelization. Affine partitioning provides a systematic framework to find asymptotically optimal computation and data decomposition for multicore processors. Affine partitioning is a powerful unifying theory, the framework uniformly models a large class of high level optimizations such as loop interchange, reversal, skewing, fusion, fission, re-indexing, scaling. Based on this unified framework, that maximize parallelism while minimizing communication in programs with arbitrary loop nestings and affine data accesses. In this papaer, we proposed a compiler pass in GCC called “affine calculator pass” which integrates affine partitioning framework and GCC to achieve the goal of automatic parallelization. We have successfully compiled some programs in C language which composed of different types of array access dependencies in arbitrary nested loop. And the outputs of the compilation are executables. Each of these executables can execute in parallel using GOMP library to utilize the processing units of multicore processor.