Return fh1h2h3 csep 524 parallel computaion winter

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Winter 2013: Chamberlain 64 Par33oned Global Address Space (PGAS) Programming Models Partitioned Global Address Space Languages (Or perhaps: partitioned global namespace languages) •  abstract concept: –  support a shared namespace on distributed memory •  permit any parallel task to access any lexically visible variable •  doesn’t matter if it’s local or remote –  establish a strong sense of ownership •  every variable has a well-defined location •  local variables are cheaper to access than remote ones The commercial: “Your shared memory convenience is in my distributed memory locality model!” “Mmmmmmmm” SIAM PP10 - Chamberlain (66) Traditional PGAS Languages •  Founding fathers: UPC, Co-Array Fortran, Titanium –  extensions to C, Fortran, and Java, respectively –  details vary, but potential for: •  arrays that are decomposed across nodes •  pointers that refer to remote objects –  note that earlier languages could also be considered PGAS, but that the term didn’t exist yet •  If I had a week to spare, we would spend some time on these first before getting to Chapel –  instead, we’ll do Chapel this week and come back to CAF and UPC next week SIAM PP10 - Chamberlain (67) Distributed Memory Chapel (ahem… “mul3- locale Chapel”) (switch to other slide decks) Using Chapel for distributed memory •  Primary change: –  CHPL_COMM=none...
View Full Document

This document was uploaded on 04/04/2014.

Ask a homework question - tutors are online