Abstract:

In view of a large number of serial heritage programs unrunning efficiently on a highperformance parallel computer at present, a method on automatic parallelization of serial program is proposed, which is based on data dependence analysis. The serial programs’ automatic parallelization can be achieved through parallel processing on the first category basic blocks. In add
ition, the executing performance of the serial programs was improved when the serial programs run in parallel.

Key words: middle granularity parallel, basic block, parallel recognition, data dependence