Meeting Agenda

Wednesday, April 30^th, 1997

Location:

DC 1304

Time:

13:30

Chair:

Dan Milgram

1. Adoption of the Agenda - additions or deletions

2. Coffee Hour

Coffee hour this week:

Celine Latulipe

Coffee hour next week:

Matthew Davidchuk

3. Next meeting

Date:

May 7^th, 1997

Location:

DC 1304

Time:

13:30

Chair:

Liddy Olds

Technical presentation:

Marryat Ma

4. Forthcoming

Chairs:

1. Mark Riddell (5/14)
2. Navid Sadikali (5/21)
3. Clara Tsang (5/28)

Tech Presenters:

1. Marryat Ma (5/7)
2. Dan Milgram (5/14)
3. Liddy Olds (5/21)

5. Technical Presentation

Presenter:

Ashraf Michail

Title:

Rendering Scenes with High Dynamic Ranges

Abstract:

Rendering systems typically produce images that have large intensity
ranges although the target display device often has a significantly
smaller range. Traditionally, the computed intensity values have been
mapped onto the target device range by using ad hoc scaling functions.
I will discuss some other approaches including the constraint-based
rendering system which I am currently working on.

6.General Discussion Items

7. Action List

• Alias Brown Bag Seminar Series
• Ivan Sutherland Visit
• CGL Photo

8. Director's Meeting

9. Seminars

• Master's Thesis Presentation - Profiling User Level Threads


Robert Denda, graduate student, Wirtschaftsinformatik" at the University of Mannheim, Germany, speak on ``Profiling User-Level Threads''.

DATE: Thursday, May 1, 1997

TIME: 10:00-11:00 a.m.

ROOM: DC 1304

ABSTRACT

The larger and more complex a program becomes, the greater the need to understand its dynamic behaviour, both to locate problems and to optimize performance.

One of the most important tools for locating dynamic problems and performance bottlenecks is a profiler. A profiler monitors a program's dynamic behaviour and reveals information about the program's execution, possibly at multiple levels of abstraction. For example, one essential step in performance optimization is to detect a program's "hot spots" for potential optimization. Detecting these spots is usually non-trivial and time consuming without appropriate
performance tools.

Concurrency adds substantially to the complexity of a program. Profiling concurrent programs entails many problems not present in sequential programs, such as thread communication and access to shared resources. A profiler has to deal with these multiple executing threads of control all potentially introducing errors and performance problems.

In this talk, I will discuss the important aspects and problems of profiling, and present an implementation of a prototype profiler for the uC++ thread-library, which provides multiple performance metrics on a per thread basis in a shared-memory concurrent environment.

10. Lab Cleanup (until 14:30 or 5 minutes)