Executive Summary : | High performance is one of the most important requirements of large modern day systems. This requirement reaches newer dimensions in the context of modern hardware landscape (multicore systems), and workloads (large data sets). This project aims to build program optimizations and analysis techniques to address this problem, which is of paramount importance to all the clients of HPC (defence, weather modeling, molecular biology, bio-technology, and so on). Two complimentary approaches can be addressed to address these issues: i) traditional sequential programs are parallelized using auto-parallelizing compil- ers, and ii) parallel programs are designed using explicitly parallel languages that depict the underlying true (parallel) program logic, and the compiler is expected to compile the code into efficient parallel code that ensures efficient utilization of all the available resources (many core systems, shared / distributed memory, communication channels and so on). Considering the limitations of the auto parallelization techniques in the past, we pursue research in the latter area. This project plans to devise new analysis and optimization schemes on top of our recent work, the IMOP compiler framework, which was partly funded by a prior CRG grant (ending in Jun 2022). The IMOP compiler framework is a source-to-source language translator, which can be used to conveniently write program analysis and optimizations techniques for OpenMP C programs. The goal of this project is to design program analysis and optimizations that take ad- vantage of the specific hardware resources, and program nature to optimize for speed. In addition, we want to devise new language features to help improve the programmability, for HPC applications, while ensuring improved performance. |