Implementation of Parallelizing Compiler for Massively Parallel Computers

在这个研究项目中，我们首先从Parafrase-2中提取中间表示来开发我们的新并行编译器Narafrase的C++版本。同时，给出了数据划分图的定义，并设计了数据划分图作为通用的中间表示。通信开销的优化将通过我们提出的cc-COMA技术来解决。然后，我们结合用户界面库与通用中间表示Narafrase通过合作与PROMIS并行编译器开发的教授Polychronolous（调查）。我们还使用用户界面库在Narafrase上实现了几种并行化方法。现在，我们正在实现一个算法，同时分区的数据和程序的Narafrase。另一方面，我们已经研究和实现了一个并行化的支持工具，三维可视化。支持工具NaraView与Narafrase协作，以便用户可以有效且轻松地并行化其顺序程序。这样，我们已经开发了一个原型并行编译系统的大规模并行计算机。

In this research project, we first extract the intermediate representation from Parafrase-2 to develop a C++ version of our new parallelizing compiler called Narafrase. At the same time, we gave the definition of and designed the data partitioning graph as our universal intermediate representation. The optimization of communication overhead will be solved by our proposing cc-COMA technology. Then, we combined user interface library with the universal intermediate representation for Narafrase by collaborating with PROMIS parallelizing compiler developed by Prof. Polychronolous (investigator). We also implemented several parallelization methods on Narafrase with the use of the user interface library. Now we are implementing an algorithm for simultaneous partitioning of data and program for Narafrase. On the other hand, we have investigated and implemented a parallelization support tool by 3D visualization. The support tool, NaraView, collaborates with Narafrase so that users can parallelize their sequential programs effectively and easily. In this way, we have developed a prototype parallelizing compiler system for massively parallel computers.