This work introduces MDH-DSL, a reduction-aware DSL for data-parallel computations grounded in the algebraic formalism of Multi-Dimensional Homomorphisms (MDH). MDH provides a principled model for expressing data-parallel computations in terms of structured input views, an operator-rich multi-dimensional homomorphism, and explicit output views. In contrast to existing DSLs, MDH-DSL treats reduction operators as first-class constructs, supports user-defined reductions, and naturally expresses nested and multi-dimensional reduction patterns. The DSL directly mirrors the minimal MDH high-level representation, consisting of three composable higher-order constructs - inp_view, md_hom, and out_view - making DSL programs concise, expressive, and algebraically meaningful.
Ercüment Kaya Technical University of München and Leibniz Supercomputing Centre, Hossam Ahmed Technical University of München and Leibniz Supercomputing Centre, Martin Schulz Technical University of Munich