A New-Generation Parallelizing Compiler System.
Abstract
Parallel computing is an enabling technology for many areas of science and engineering. Parallel computing is also our best investment to continue achieving performance gains as the limits of semiconductor technology are approached. Whereas the design of parallel computers is reasonably well-understood, the programming of these machines remains in its infancy. As a consequence, programming parallel computers is substantially more difficult than programming conventional uniprocessors. In fact, it is commonly said that the development of effective parallel programming techniques is the main challenge faced by high-performance computing today. Unless this challenge is met, the acceptance of parallel computers will remain slow, thereby hampering progress in computational science and engineering. An appealing strategy to meet this challenge is to use compilers to translate conventional programs into parallel form. Such compilers would enable the execution of existing programs on new parallel machines, thus allowing a seamless transition into parallel computing. With the support of such compilers, not only would legacy codes could be easily ported, but also new programs could be developed in the familiar sequential paradigm, thus liberating the programmer from the complexities of explicit, machine-oriented parallel programming.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1996
- Accession Number
- ADA313990
Entities
People
- David A. Padua
- Jay Hoeflinger
- Rudolf Eigenmann
Organizations
- University of Illinois Urbana–Champaign