Date: Thu, 14 Nov 2002 07:11:48 -0800 (PST)
From: kris stewart <stewart@rohan.sdsu.edu>
To: cs575@rohan.sdsu.edu
Subject: Summary of Class Discussion - Group Report - Compiler Autopar

Hello CS 575 Students,

You have had the copy of the Stommel Model code since Lab 4 (October 2)
and we have been discussing in class (and via the web page notes) some 
of the background information that this Stommel Model is based on -

Jacobi Iteration

to approximate the solution of the steady-state Stommel Equation.

stc_00.c
stf_00.f

are the serial versions of this code in C and Fortran 90 and you have 
your choise of which language version to focus on depending on your own 
background.

The Goals is to explore the optimization and parallelization possible
using Rohan's Solaris compilers - without modification to the code.

Yesterday, we examined the sample code from Ch 11 in our text (p. 223)
that consisted of a simple code fragment with no loop dependencies.  
This is an ideal candidate for the compiler to be able to demonstrate 
its successful translation of the High Level Language into SPARC assembler 
that can use separate processors, specified by the user using

setenv PARALLEL #   where the example had # = 1, 2, 4

and compiling using f95 with the 

-xautopar

option.  I expect this same compiler option to work with cc

Your group assignment is to examine the information online from SUN
http://docs.sun.com/source/806-3566/plug_optimizing.html
to be able to apply the more general information our textbook has 
provided as background for "Programming Shared-Memory Multiprocessors"
to your specific computing platform Rohan.

As discussed yesterday in lecture, if one group member is more comfortable 
with Fortran and another more comfortable with C,  I would suggest each 
person individual 
1) explore the parallelization that can be delivered by appropriate compiler options
2) gather the timing evidence to support your conclusions on the run-time 
behavior of these compiler options

You will be expected to write a group report

assign4.html       (assignment 3 was you group presentation on security)

that will be available from the POC (point of contact, or spokesperson for 
your group) following our standard template, already used in your first 
two written reports.

Deadline: 27Nov02
for email notification that the webpage is ready to be viewed by the 
instructor as well as hardcopy of the report in class.

NOTE: Just as the "Computer/Internet Security" presentation that
groups made, you will individually be asked to email me how much you 
contributed to this group project.  You must also tell me how much the 
other groups members contribute.  THEREFORE, make sure that the members of
your group are aware and value your contribution to the overall effort.  

The grade on the final product (your report) will be distributed based on
the amount of reported effort by each individual group member.

I fell you will enjoy next week's Scientific Visualization Presentation in
lecture on Monday and the time to work as a group on Wednesday's class
time.

I will be attending and participating in Supercomputing 2002 in Baltimore, 
so that I can return with current information of the "state of the art" in 
High Performance Computing.

At the end of November, we began formal discussions on MPI and the San
Diego Supercomputer Center's Blue Horizon TERAFLOPS computer.  At that
time we will look into modifying the Stommel Model Code to include MPI 
directives to explicitly distribute the job across processors on the 
IBM machine.  

Recall, the final group project
assign5.html
will details the performance results you obtain using MPI on IBM Blue
Horizon at the San Diego Supercomputer and is due the day of your final
exam - Friday 20 Dec 02

Kris Stewart
stewart@rohan.sdsu.edu
bam223 Office Hours MW 1-2:30