摘要 |
PROBLEM TO BE SOLVED: To obtain a loop parallelization method which parallelizes a loop in which data dependence over loop repetition exists which is conventionally difficult. SOLUTION: A pipeline parallelization conversion part 15 generates loops for issuing barrier synchronization before and after multiplex loops to the multiplex loops in which the data dependence over the loop repetition exists, and also generates a sentence for issuing the barrier synchronization immediately after the divided loops. Thus, the multiplex loop is executed in parallel like a pipeline in a multi-processor. COPYRIGHT: (C)2006,JPO&NCIPI
|