Languages and compilers for high performance computing

Mehr zum Buch

This collection of works explores various advanced topics in compiler design and optimization techniques. It includes experiences with source-to-source transformations using Cetus and provides a tutorial on the LLVM Compiler Framework. The content covers the Open Research Compiler and Trimaran, focusing on instruction-level parallelism. Key studies discuss phase-based miss rate prediction, speculative subword register allocation, and empirical performance-model driven data layout optimization. Innovative approaches for parallel numerical algorithms using hierarchically tiled arrays and geometric partitioning of N-dimensional iteration spaces are presented. The analysis of divide-and-conquer algorithms is exemplified through JuliusC, while memory operation parallelism for code optimization is also examined. The text delves into locality optimization via an ILP-based approach and introduces a code isolator for managing code fragments. Additionally, it discusses trace utilization for inlining in Java programs and provides insights into MHP information analysis for concurrent Java applications. Further topics include high-level language compilation for vector architectures, high-level optimization of FFTs, and loop optimizations for object-oriented abstractions. The performance of the OSCAR multigrain parallelizing compiler and experiences with Co-array Fortran on shared memory platforms are also highlighted, alongside autom