Summary: Using the Run­Time Sizes of Data Structures
to Guide Parallel­Thread Creation
Lorenz Huelsbergen
AT&T Bell Laboratories
lorenz@research.att.com
James R. Larus
University of Wisconsin--Madison
larus@cs.wisc.edu
Alexander Aiken
University of California--Berkeley
aiken@cs.berkeley.edu
Abstract
Dynamic granularity estimation is a new technique for
automatically identifying expressions in functional lan­
guages for parallel evaluation. Expressions with lit­
tle computation relative to thread­creation costs should
evaluate sequentially for maximum performance. Static
identification of such threads is however difficult. There­
fore, dynamic granularity estimation has compile­time
and run­time components: Abstract interpretation stat­

Source: Aiken, Alex - Department of Computer Science, Stanford University

Collections: Computer Technologies and Information Sciences