Return to CLIP Main Page

phv

Automatic Parallelization of Irregular
and Pointer-Based Computations:
Perspectives from Logic and Constraint Programming



Manuel Hermenegildo
herme at fi.upm.es
http://www.clip.dia.fi.upm.es/

School of Computer Science
Technical University of Madrid

Work supported in part by ESPRIT projects PEPMA and PARFORCE and by CICYT/MEC grants. The UPM work presented is a joint effort of the members of the CLIP group at the UPM School of Computer Science: Francisco Bueno, Daniel Cabeza, Manuel Carro, María José García de la Banda, Manuel Hermenegildo, Pedro López, and Germán Puebla.

 
• Introduction / Motivation
• Constraint Logic Programming: Variables and Statements
• Constraint Logic Programming: Procedures
• Constraint Logic Programming: Conditionals/Choices
• Constraint Logic Programming: Recursion
• Constraint Logic Programming: Data Structures
• LP and CLP - Parallelization
• Parallelism in Constraint Logic Programs
• Parallelization: Correctness and Efficiency
• Independence - Strict Independence
• Independence - Non-Strict Independence
• Independence - Constraint Independence
• Parallelization Process
• Parallelization Process
• &-Prolog/CIAO Parallelizer Overview
• ``Sharing'' Abstraction (Groundness + Set Sharing)
• Visualization of And-parallelism - (small) qsort, 1 processor
• Visualization of And-parallelism - (small) qsort, 4 processors
• Sharing + Freeness Domain
• Some Speedups
• Some CLP Results (Run-time System)
• Dealing with Speculation
• Dealing with Overheads, Irregularity
• Dealing with Overheads, Irregularity: Granularity Control
• Granularity Control Example
• Granularity Control: Some Optimizations
• Granularity Control System Output Example
• Granularity Control: Experimental Results
• Wrap-up: strong points
• Wrap-up: areas for improvement