Tang, PeiyiGao, Nianshu2003-07-102004-05-192011-01-052004-05-192011-01-051994http://hdl.handle.net/1885/40777http://digitalcollections.anu.edu.au/handle/1885/40777Data dependences between statements have long been used for detecting parallelism and converting sequential programs into parallel forms. However, some data dependences can be reversed and the transformed program still produces the same results. In this paper, we revisit vectorization and propose a new vectorization algorithm using reversible data dependences. The new algorithm can generate more or thicker vector statements than traditional algorithm. The techniques presented in this paper can be incorporated in all the existing vectorizing compilers for supercomputers.216645 bytes356 bytesapplication/pdfapplication/octet-streamen-AUparallelismvectorization algorithmsupercomputers1994