Workshop : Compiler Optimizations on Distributed Memory Systems

Santosh Pande <>
Tue, 29 Aug 1995 20:41:47 GMT

          From comp.compilers

Related articles
Workshop : Compiler Optimizations on Distributed Memory Systems (Santosh Pande) (1995-08-29)
| List of all articles for this month |

Newsgroups: comp.compilers
From: Santosh Pande <>
Keywords: conference, storage, parallel
Organization: Compilers Central
Date: Tue, 29 Aug 1995 20:41:47 GMT

              Workshop : Compiler Optimizations on Distributed Memory Systems
              -------- ----------------------------------------------------

                                                  Santosh Pande, Co-ordinator
                      School of Electrical Engineering and Computer Science
                                            Division of Computer Science
                                                        Ohio University
                                                    Athens, Ohio 45701
                                      E-mail :
                                                Telephone : (614)-593-1251

A half day workshop on `Compiler Optimizations on Distributed Memory
Systems' is planned at the 7th IEEE Symposium
on Parallel and Distributed Processing (Dallas Symposium) at
San Antonio, TX.
The workshop will be conducted
on October 25th (Wednesday) from 1:30 pm to 5:00 pm.

The general format of the workshop is as follows:

1. Introduction of the theme, the issues and the speakers
2. Invited talks each for a duration of 25 minutes including
3. Panel discussion of the speakers
4. Conclusion.

It is intended that the workshop presents inter-relationships
and issues in different compilation phases from dependence detection
to mapping and code generation.
The following topics are of special interest in the workshop:

    -- Exact dependence analysis
    -- Interprocedural dataflow analysis
    -- Communication/data locality optimizations
    -- Optimizing loop transformations in HPF framework
    -- Code generation issues such as address and communication generation
    -- Mapping task parallelism

The goal of the workshop is to present a comprehensive, integrated
picture of the area by discussing the present research status, open
directions and the projected future work on the above aspects of
compiling on distributed memory systems. Another important objective
is to bring out similarities and differences between compiling on
shared memory systems and on distributed memory systems to identify
the applicability of the techniques already developed for shared
memory to the distributed memory.

First, there will be six invited talks by distinguished researchers
in the field on topics ranging from dependence detection to mapping
and code generation. At the end of the talks, there will be a panel
discussion on `How difficult is automatic parallelization on
distributed memory systems?' (or in other words, `Is data parallel
programming only an interim solution or a long term one?')

Following is the program of the workshop :

Session I : 1:30 pm to 2:45 pm (Ballroom C)

  (1) "Efficient and Exact Dependence Analysis Techniques in Practice"
            Kleanthis Psarris, University of Texas

  (2) "The Effectiveness of Interprocedural Automatic Parallelization"
            Mary Hall, California Institute of Technology

  (3) "Communication Analysis for Shared and Distributed Memory Machines"
            Chau -Wen Tseng, University of Maryland, College Park

Break : 2:45 pm to 3:15 pm

Session II : 3:15 to 4:30 pm (Ballroom C)

  (4) "Loop Transformations for Optimizing HPF Programs Prior to SPMD Code
          Vivek Sarkar, Application Development Technology Institute (ADTI)
                                        IBM Software Solutions Division.

  (5) "Address and Communication Generation for Cyclic(k) Distributions"
            J. Ramanujam, Louisiana State University, Baton Rouge

  (6) "Exploiting Loop and Task Parallelism in Scheduling Iterative Task
            Tao Yang, University of California, Santa Barbara

          4:30 pm to 5:00 pm (Ballroom C)

          Panel discussion of above speakers
          Co-ordinator : Prof. Dharma P. Agrawal, North Carolina State University

          Topic : `How difficult is automatic parallelization
                                      on distributed memory systems?'
                                      (or, `Is data parallel programming only
                                      an interim solution or a long term one?')

Those interested in attending, please contact the co-ordinator
Santosh Pande for registration and other information.


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.