[ADOL-C] performance question

[Andrea Walther]

Dear all,

I forward the attached message since it was send to the
adolc mailing list but somehow became an uncaught bounce:

Hi all,

Now that my program works great with adol-c, I am looking deeper into  
its computational efficiency. To do so, I am using intel Vtune to  
analyse the hot spots. I tried 2 different implementation, with  
surprising results.

In the first implementation, I allocate big arrays to store some  
active variables. Imagine a cube and each of its faces. I allocate the  
memory for each faces.

In the second implementation, I allocate the memory for a single face,  
and overwrite the data for the subsequent faces. I was assuming that  
this second way would be faster due to smallest memory allocation.  
However, Vtune tells me that the 2nd way involves a costly  
merge_2_index_domains call, which makes the program to run slower than  
implementation 1.

Does this agree with your observations as well?
If you want, I could provide a simple test case for you to understand  
better...
Antoine

---- end of forwarded message -----

Best regards

Andrea

-- 
Prof. Dr. Andrea Walther
Lehrstuhl fuer Mathematik und ihre Anwendungen
Institut fuer Mathematik
Universitaet Paderborn
Warburger Str. 100
33098 Paderborn

Email: andrea.walther at uni-paderborn.de
Phone: ++49 5251 602721
        ++49 5251 602724 (sekr.)
Fax:   ++49 5251 603728

**********


-------------- next part --------------
An embedded message was scrubbed...
From: Antoine De Blois <antoine.deblois at aero.bombardier.com>
Subject: performance question
Date: Wed, 12 Jun 2013 17:16:13 +0000
Size: 6099
URL: <http://list.coin-or.org/pipermail/adol-c/attachments/20130612/30187580/attachment.mht>