The Method of Parallel Optimization and Parallel Recognition Based on Data Dependence

摘要

For application programs in scientific and technological fields have grown increasingly large and complex, it is becoming more difficult to parallelize these programs by hand using message-passing libraries. To reduce this difficulty, we are researching the compilation technology for serial program automatic parallelization. In this paper, the author puts forward a kind of parallel recognition algorithm in parallelization compiler. In the algorithm the author adopts the idea of the medium grain parallel. Through this algorithm, the parallelization compiler can identify all of the parallelizable blocks. So that the application programs can be speeded up and the execution ability can be improved when the blocks execute on multi-processors. Parallel processing often can make the runtime of application programs shorter than serial processing, but if the radio of parallel workload to overhead about creating parallel thread or the radio of parallel workload to parallel thread number is small, parallel execution can degrades program performance. To solve this problem, the author proposes several parallel optimization approaches in the end of the paper.