Implementation of Parallel Algorithms
Abstract
Work is progressing in the design of a new, general intermediate representation for parallel code. We wish to provide a single compilation target that is executable on a wide variety of parallel machines and vector processors. The representation we are designing is an extension of VCODE, due to Blelloch at CMU, which is particularly suitable as a target only for data-parallel language. VCODE provides a rich set of vector operations, including powerful segmented prefix computations and permutation operations, which run efficiently on a wide range of architectures. However, VCODE provides only sequential control, which limits its application strictly to data-parallel programming. We have extended VCODE to form a new language called AVCODE which has operations for process creation, communication, and synchronization. AVCODE provides an avenue for efficiently targeting a parallel language to a number of different hardware systems through a single intermediate language.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 31, 1992
- Accession Number
- ADA248826
Entities
People
- John Reif
Organizations
- Duke University