OpenMP Shared Memory Parallel Programming: International Workshop on OpenMP Applications and Tools, WOMPAT 2003, Toronto, Canada, June 26-27, 2003. ... (Lecture Notes in Computer Science, 2716)

✍ Scribed by Michael J. Voss (editor)

• Publisher: Springer
• Year: 2003
• Tongue: English
• Leaves: 280
• Category: Library

✦ Synopsis

This book contains the proceedings of the Workshop on OpenMP Applications and Tools, WOMPAT 2003. WOMPAT 2003 was held on June 26and 27, 2003 in Toronto, Canada. The workshop brought together the users and developers of the OpenMP API to meet, share ideas and experiences, and to discuss the latest developments in OpenMP and its applications. The OpenMP API is now a widely accepted standard for high-level shar- memory parallel programming. Since its introduction in 1997, OpenMP has gained support from the majority of high-performance compiler and hardware vendors. WOMPAT 2003 was the latest in a series of OpenMP-related workshops, which have included the annual o?erings of the Workshop on OpenMP Appli- tions and Tools (WOMPAT), the European Workshop on OpenMP (EWOMP) and the Workshop on OpenMP: Experiences and Implementations (WOMPEI). The WOMPAT 2003 program committee formally solicited papers. Extended abstracts were submitted by authors and all abstracts were reviewed by three members of the program committee. Of the 17 submitted abstracts, 15 were selected for presentation at the workshop. This book was published in time to be available at WOMPAT 2003, and therefore we hope that the papers contained herein are timely and useful for current developers and researchers. This book also contains selected papers from WOMPAT 2002. No formal proceedings had been created for this previous o?ering of the workshop, and so presenters were invited to prepare their papers for inclusion in this volume.

✦ Table of Contents

Lecture Notes in Computer Science
Preface
Table of Contents
OpenMP Support in the Intel® Thread Checker
1 Introduction
2 Correctness Analysis
2.1 Correctness: Kinds of Theaded Errors
2.2 Pros and Cons of Statis Analysis for Correctness
2.3 Pros and Cons of Dynamic Analysis for Correctness
3 Dynamic Correctness Analysis of OpenMP Applications
3.1 Relaxed Sequential Programming and Benefits
3.2 Projection Technology
4 Intel Tools for OpenMP
4.1 Creating Parallel Applications
4.2 Debugging Parallel Applications
5 Relaxed Sequential Programming: A Concrete Example
6 Sample Use of Intel Thread Checker
7 Conclusion
A C++ Infrastructure for Automatic Introduction and Translation of OpenMP Directives
1 Introduction
2 Infrastructure
2.1 Frontend
2.2 Midend
2.3 Backend
3 Semantics-Driven Inroduction of OpenMP Directives
3.1 User-Defined Abstractions
3.2 A++/P++ Serial and Parallel Array Class Library
3.3 Automated Insertion of OpenMP Directives
3.4 Example C++ Code
3.5 Discussion
4 Translation of OpenMP Directives
4.1 Translation Specification
4.2 Translation Algorithm
4.3 Discussion
5 Related Work
6 Conclusion and Future Work
Analyses for the Translation of OpenMP Codes into SPMD Style with Array Privatization
1 Introduction
2 Consistent Loop Scheduling for Data Reuse
2.1 Consistency Test
2.2 Examples of Consistent and Inconsistent Schedules
3 Interprocedural Analysis
3.1 Call Graph
3.2 Interprocedural Algorithm
4 Privatization Analysis
4.1 Privatization Algorithm
4.2 Special Case 1: LU
4.3 Special Case 2: ADI
5 Related Work
6 Conclusion and Future Work
A Runtime Optimization System for OpenMP
Introduction
Related Work in Runtime Optimization
OpenMP as a Target for Runtime Optimization
Runtime Characteristics of OpenMP Applications
Features of OpenMP Implementations
Optimization Opportunities
The stOMP Runtime Optimization System
The stOMP Compiler
The stOMP Runtime Library
The stOMP Dynamic Optimizer
A Preliminary Evaluation
The Performance of the EPCC Scheduling Microbenchmark
The Performance of Apsi, Art, and Equake
Conclusions
A Practical OpenMP Compiler for System on Chips
1 Introduction
2 Parallel Programs on 3SoC
2.1 Programming Different Parallel Processors
2.2 Programming Using PEs
2.3 Programming Using DSEs
3 Design of OpenMP Compiler/Translator
3.1 Synchronization
3.2 Scheduling and Computation Division
3.3 Data Attributes and Memory Allocation
4 Extensions to OpenMP
4.1 OpenMP Extensions for DSE Processors
4.2 OpenMP Extensions for Optimization on SOCs
5 Implementation
6 Performance Evaluation
7 Conclusion
Evaluation of OpenMP for the Cyclops Multithreaded Architecture
Introduction and Motivation
The Blue Gene/Cyclops Architecture
Towards OpenMP for BG/C
Benchmark Description
Micro-Benchmarks
NAS Benchmarks
Experimental Results
Micro-Benchmarks
NAS Benchmarks
Improving OpenMP Support for BG/C
Related Work
Conclusions
Busy-Wait Barrier Synchronization Using Distributed Counters with Local Sensor
Introduction
Overhead of a Barrier Synchronization
POWER4 SMP Architecture and Software
Testing Benchmark
Barrier Overhead on an SMP System
Design of Different Barriers
Barrier with Fetch-and-Add
Distributed Counter
Distributed Counter with Padding
Combined with Local Sensor
Performance Data and Analysis
Summary and Future Work
An OpenMP Implementation of Parallel FFT and Its Performance on IA-64 Processors
Introduction
A Recursive Three-Step FFT Algorithm
In-Cache FFT Algorithm and Parallelization
Performance Results
Performance Results on the DELL PowerEdge 7150
Performance Results on the hp Workstation zx6000
Conclusion
OpenMP and Compilation Issues in Embedded Applications
Introduction
EEMBC
Parallelization of EEMBC Benchmarks Using OpenMP
Experiment Methodology
Performance in OpenMP-Parallelized Codes
Compilation Issue
Conclusion
Parallelizing Parallel Rollout Algorithm for Solving Markov Decision Processes
Introduction
Background
Markov Decision Processes
Parallel Rollout
Parallelization of Parallel Rollout
Evaluation on Multi-class Scheduling
MDP Formulation
Simulation
Problem Setup
Measurement
Conclusion
DMPL: An OpenMP DLL Debugging Interface
1 Introduction
2 Background
2.1 Multiple Compilers, User Levels
2.2 Compiler Transformations
3 DMPL Objective
4DMPL Interface
4.1 DMPL Types
4.2 DMPL Functions
5 Discussion
Is the Schedule Clause Really Necessary in OpenMP?
Introduction
Motivation and Related Work
Dynamic Derivation of Loop Schedules
Current Implementation
Evaluation
Extended Overhead Analysis for OpenMP Performance Tuning
Introduction
Related Work on Overhead Analysis
Overhead Analysis: A Layered Model
Case Study: TLP and ILP Overlap
A New Overhead Classification Scheme
Parallelism Identification
Parallel Task Assignment/Scheduling
Synchronization Clause
Conclusion and Future Work
Supporting Realistic OpenMP Applications on a Commodity Cluster of Workstations
1 Introduction
2 Translating OpenMP Applications into Software DSM Programs
3 Benchmark Evaluation of Real Application Benchmarks
4 Advanced Optimizations
4.1 Computation Repartitioning
4.2 Page Aware Optimizations
4.3 Privatization Optimization
5 Results
6 Conclusions
OpenMP Runtime Support for Clusters of Multiprocessors
Introduction
OpenMP Runtime Library
Design
Architecture
Implementation Platforms
OpenMP Execution
Sharing the User-Level Thread Stacks
Sharing the Common Blocks
Relaxing the Protocol
Experimental Evaluation
Hardware Shared Memory
Distributed Memory - SMP Machine
Distributed Memory - Cluster of SMPs
Related Work
Ongoing Work
An Evaluation of MPI and OpenMP Paradigms for Multi-Dimensional Data Remapping
Introduction
Vacancy Tracking Algorithm
Parallel Paradigms on Cluster SMP Architectures
Multi-threaded Parallelism
Pure MPI Parallelism
Hybrid MPI/OpenMP Parallelism
Performance
Scheduling for OpenMP Parallelism
Pure MPI and Pure OpenMP Parallelisms within One Node
Pure MPI and Hybrid MPI/OpenMP Parallelisms Across Nodes
Conclusions
Experiences Using OpenMP Based on Compiler Directed Software DSM on a PC Cluster
Introduction
Programming Models
Message Passing
Shared Address Space
Hardware Platform and Software Description
Platform Description
SCore
SCASH
Omni OpenMP
Case Studies
Evaluation Strategy
The EP Benchmark Kernel
The CG Benchmark Kernel
The FT Kernel Benchmark
Problems Encountered
Related Work
Conclusions and Future Work
Managing C++ OpenMP Code and Its Exception Handling
1 Introduction
1.1 Current Standards Requirements
1.2 Examples
1.3 Expensive Runtime Diagnostic
2 Implementation Possibilities
2.1 Using an Exception Specification
2.2 Using a Try Block
3 Comparison of Results on EPCC OpenMP Microbenchmarks
4 Conclusions
5 Future Directions
Appendix: Debugging Uncaught Exception from a Parallel Region
Improving the Performance of OpenMP by Array Privatization
1 Introduction
2 Overview of SPMD Translation
2.1 SPMD Style with Array Privatization
2.2 General Strategy for SPMD Translation
3 Translating OpenMP Constructs
3.1 Other Issues in Loop Translation
3.2 Translation of Other OpenMP Constructs
4 Current Implementation
5 Related Work
6 Conclusion and Future Work
OpenMP Application Tuning Using Hardware Performance Counters
Introduction
The Experimental Setup
Test Cases Used
Computer Architectures and PAPI Implementations
Experimental Findings
Intel PentiumII
IBM Power3
SGI R10K/R14K
Discussion
Conclusion
Author Index