[Omp] Threadprivate variables

[Patricia Bittencourt Sampaio]

>When making comments like this, it would be helpful
>if you specified the type of machine, operating
>system, and compiler you are using.  I don't
>think that this is accurate as a general statement
>and may well vary considerably from one compiler
>version (or machine, or OS, etc) to the next.

Sorry, my mistake. This experiment was made at at
Altix Itanium 2 architecture, IA-64, using linux
version 2.4.21, with Intel Compiler Version 8.1 for
Itanium. 

However I have made the same experiment on Pentium4.
2.80GHz, with linux version 2.6.7, with Intel Compiler
Version 8.1, and it increases just 20% of time from
executing without openmp directives to executing with
it, using just one thread. What shows from the results
I have posted latter that this increase it is not so
critical as the ones I reported before. Considering
this, I guess it is probably a factor of the
architecture in use. 

The variables used as threadprivate were all scalar
ones and declared as static double. It is kind of
difficulty saying the amount of work, cause it depends
on what you consider as a unit of work. However the
threadprivate variables are accessed on a loop that
ranges from 0 to 134217728.


thanks for the interest,


Patricia Bittencourt Sampaio
====================================
Federal University of Rio de Janeiro




		
_______________________________________________________
Yahoo! Acesso Grátis - Internet rápida e grátis. Instale o discador agora!
http://br.acesso.yahoo.com