Doctor of Philosophy (PhD)
Electrical and Computer Engineering
Writing high performance code has steadily become more challenging since the design of computing systems has moved toward parallel processors in forms of multi and many-core architectures. This trend has resulted in exceedingly more heterogeneous architectures and programming models. Moreover, the prevalence of distributed systems, especially in fields relying on supercomputers, has caused the programming of such diverse environment more difficulties. To mitigate such challenges, an assortment of tools and programming models have been introduced in the past decade or so. Some efforts focused on the characteristics of the code, such as polyhedral compilers (e.g. Pluto, PPCG, etc.) while others took in consideration the aspects of the application domain and proposed domain specific languages (DSLs). DSLs are developed either in the form of a stand-alone language, like Halide for image processing, or as a part of a general purpose language (e.g., Firedrake- a DSL embedded in Python for solving PDEs using FEM.) called embedded. All these approaches attempt to provide the best input to the underlying common programming models like MPI and OpenMP for distributed and shared memory systems respectively. This dissertation introduces Kaashi, a high-level run-time system, embedded in C++ language, designed to manage memory and execution order of programs with large input data and complex dependencies. Kaashi provides a uniform front-end to multiple back-ends focusing on distributed systems. Kaashi abstractions allows the programmer to define the problem’s data domain as a collection of sets and relations between pairs of such sets. The aforesaid level of abstraction could enable series of optimizations which, otherwise, are very expensive to detect or not feasible at all. Furthermore, Kaashi’s API helps novice programmers to write their code more structurally without getting involved in details of data management and communication.
Document Availability at the Time of Submission
Secure the entire work for patent and/or proprietary purposes for a period of one year. Student has submitted appropriate documentation which states: During this period the copyright owner also agrees not to exercise her/his ownership rights, including public use in works, without prior authorization from LSU. At the end of the one year period, either we or LSU may request an automatic extension for one additional year. At the end of the one year secure period (or its extension, if such is requested), the work will be released for access worldwide.
Tohid (Rastegar Tohid, Mohammed), M., "Analysis and Optimization of Scientific Applications through Set and Relation Abstractions" (2017). LSU Doctoral Dissertations. 4404.
Available for download on Saturday, February 23, 2019