[CSC 435] Threads in ATLAS

[Andrew J. Pounds]

Well your assignment just got a little easier...

The newest version of the ATLAS libraries that we are using sets the 
number of threads to be used (in the parallel versions) based on the 
processors available in the system.  It sets this at compile time and it 
cannot be changed (even using calls to set_omp_num_threadsor by using 
the OMP_NUM_THREADSenvironment variables).

So, on your serial performance graphs (megaflops vs problem size) 
include results for the single threaded ATLAS.

On your parallel speedup curves the ATLAS will always be running using 8 
cores and should therefore be a constant value.  Plot that (but do not 
fit it to Amdahls law) as a line on your plot.   Your plots will then 
have a speedup curve for your code and then a horizontal line 
(presumably above your speedup curve) for the ATLAS function.

So, just to be clear, you are not building speedup curves for ATLAS but 
you are still comparing your performance to it.

-- 
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

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://theochem.mercer.edu/pipermail/csc435/attachments/20160410/f4a88c55/attachment.html>