Simplified High Level Parallelism Expression on Heterogeneous Systems through Data Partition Pattern Description.

This article focuses on DACL (Data Associated Computing Language), a high-level programming language designed to simplify heterogeneous parallel programming and improve application portability across diverse hardware architectures. DACL introduces data partition patterns as a core abstraction to express parallelism independently of runtime details, separating parallelism expression from computing process logic through an association structure. The language features C-compatible syntax with extensions for data declaration, association structures, and calculations, enabling modular programming and reuse of existing code. A prototype compilation system translates DACL programs into OpenMP and OpenCL code, supporting cross-platform execution on CPUs, GPUs, and MIC coprocessors. Experimental evaluation using benchmarks from Parboil and Rodinia suites demonstrates that DACL reduces programming effort compared to OpenCL, achieves comparable or better performance than manually optimized code, and facilitates portability across heterogeneous systems.