TC: Optimizing compiler

In scientific codes, particularly those used in simulations, the majority of execution time is consumed by the processing of program loops. However, the ordering of statement instances constituting a loop nest, which arises from its naive, typically hand-written implementation, usually leads to inefficient and suboptimal utilization of hardware resources. Over the past decades, significant research efforts have been devoted to developing loop nest transformations aimed at enhancing thread- and instruction-level parallelism, as well as improving memory hierarchy usage, with the affine transformations framework becoming the canonical approach.

This paper presents TC – a source-to-source optimizing compiler for program loop nests. Unlike state-of-the-art solutions, TC extends the polyhedral compilation pipeline with a data dependence analysis based on the transitive closure relation, significantly broadening the range of optimization opportunities.