Transformation of non-strict structure-data access into scheduled strict access in non-strict dataflow languages

In order to show the feasibility of a fine-grain dataflow computation scheme, we are implementing a fine-grain non-strict dataflow language, called V, on stock parallel machines, using a fine-grain multithreading approach. Fine-grain non-strict data structures such as I-structures provide high level abstraction to easily write programs with potentially high parallelism due to the eager evaluation of non-strict functions and non-strict structure-data. Non-strict data. structures, however, require frequent dynamic scheduling at a fine-grain level, which offsets the gain of latency hiding. In addition to the dynamic scheduling at a fine-grain level, asynchronous access to structure-data using non-strict data structures cause heavy overhead. In order to solve this problem, we employ a method to transform non-strict access to scheduled strict access by dependency analysis. The performance evaluation results indicate that this technique is effective to improve the performance.