The parallel intermediate language

The next challenge in the evolution of supercomputers will be the transition to exascale systems. However, while the move from terascale to petascale processing was considered evolutionary, it is widely believed that the leap to exascale supercomputers will require revolutionary advances. Simply scaling up current technology will not work. The projections for the exascale systems indicate that applications may have to support up to a billion separate threads to efficiently use the hardware, while the amount of memory per arithmetic functional unit will drop significantly. This implies the need for exploiting fine-grain parallelism with a programming model other than the currently used message passing or coarse-grain threads. As a response, the programming community is exploring data-driven runtimes. However, in order to utilize the new runtime systems, users will either need to rewrite all of their applications by hand in the new languages, or be provided with tools to help them move to the new languages. Requiring users to rewrite applications is very costly, time consuming, and error prone. We believe a better approach is to help ease users into new programming paradigms by providing them with both a way to utilize existing programming paradigms and applications, as well as providing them a way to write applications directly in the new programming notations. There is a disconnect between the high level languages such as HTAs that provide high levels of expressibility and programmability, and new data-driven runtimes, such as SCALE and OCR that provide high levels of control on supercomputers of the future. We want to bridge the gap between these notations with a Parallel Intermediate Language (PIL). As new runtimes are being developed to run on future supercomputers, we believe that a framework to help programmers target these new runtime systems is necessary. Thus, PIL should be retargetable, efficient, and should accept many high level languages as input. Such a framework can provide portability across many different machines and runtimes. Furthermore, we believe that when targeting a new runtime systems programmers can achieve increased productivity and performance through the utilization of multiresolution programming in their applications, while allowing a framework to ease the transition to new notations.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2016

Author dc:creator

• Smith, Adam Randall

Contributors dc:contributor

• Padua, David A
• Kale, Laxmikant
• Hwu, Wen-Mei
• Pinfold, Wilfread

• intermediate language
• compiler
• parallel

Statement dc:rights

• Copyright 2015 Adam R. Smith

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/88949

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/88949