The Algorithm of Parallel Recognition Based on Data Dependence

Abstract

In this paper, the author puts forward a kind of parallel recognition algorithm in implementation of parallelization compiler[3]. Application programs in scientific and technological fields have grown increasingly large and complex, thus, 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. The algorithm uses the conceptions of medium grain parallel[2] and data dependence to realize the target of the program automatic parallelization. At the first, the author defines the data dependence of blocks by variables that the blocks have accesses at the same time. For another, by this definition, the parallelization compiler can identify all of the parallelizable blocks in the serial program and can provide a certain possibility to implement the application programs automatic parallelization. The goal to speed up the program and to improve the execution ability of the program can be realized when the blocks parallel execute on multi-processors. At the last, in the end of the paper, the author analyzes the performance of the proposed algorithm, so the author puts forward two strategies of parallel optimization.

A Machine Learning and Compiler-based Approach to Automatically Parallelize Serial Programs Using OpenMP

“…Application Development Model [22] 2.3. "The Algorithm of Parallel Recognition Based on Data Dependence" [27] In this paper, the authors investigated and presented an algorithm that would be used to recognize parallelizable regions or blocks of code in serial programs based on data dependence. Motivated from the fact that it is difficult and takes time to manually parallelize large and complex serial programs, the goal of the algorithm is to provide a mechanism to be integrated into modern compilers to achieve automatic program parallelization or to design an automatic parallel compiler.…”

“…The character of the blocks is that the whole codes of blocks either must be executed or must not be executed." [27] Data dependence between blocks is recognized based on input and output variables that the blocks have access at the same time. If output variables of a block depend on input variables of another block and vice versa, these blocks have data dependence and cannot be executed or computed in parallel.…”