Source-Level Transformation of Legacy Sequential Programs into Scalable Thread-Parallel Code

The number of cores on a CPU chip is currently doubling every two years, in a manner consistent with Moore’s Law. This is leading to an era of many-core computing [1, 2, 3], in which thread parallelism is desirable for the HPC community for the following reasons. First, those sequential algorithms, which were previously hard to parallelize on distributed-memory computer clusters