[CSC 435] Class this Week

[Andrew J. Pounds]

I have heard from some of you that NONE of your code re-arrangement is
helping speed up the program and you are worried that you won't have
enough speedup runs for your paper.  Here is another way I will let you
create the data.  You can already do -O0, -O1, -O2, -O3, and -Ofast
(that's four speedup plots).  

For the remaining six find the two options at the -O1, -O2 or -O3 level
that give very good speedup. (look at "man gcc" and find -O1.   For
example, at first level you might compile with

gfortran -cpp -O0 -fmove-loop-invariants dusty.f

or maybe

gfortran -cpp -O0 -finline-functions-called-once dusty.f

both of the -f options are optimizations that are enabled at the -O1 level.

Then move to the next level and pick some options...

gfortran -cpp -O1 -fstrict_aliasing dusty.f

Try various options that give good speedup and then, in your paper,
explain why you think that particular optimization helped in the dusty code.


My plan is for you to have TWO workdays this week.   I will be with you
on Tuesday, but I will be out of the office Thursday morning.



-- 
Andrew J. Pounds, Ph.D.  (pounds_aj at mercer.edu)
Professor of Chemistry and Computer Science
Mercer University,  Macon, GA 31207   (478) 301-5627
http://faculty.mercer.edu/pounds_aj