[molpro-user] Molpro 2008 - clearspeed

[Lorenzo Gontrani]

Hello all..

We have recently installed a couple of Clearspeed accelerator boards
in our Opteron nodes..
I compiled the 2008.1 version with the instructions found in the manual 
(-clearspeed and -clearspeedbase option at configure time).
During the building phase, the Cn sources were compiled, and the main 
executable was produced in the binaries directory, as well as a file named 
csdfti.csx (no errors reported during "make").
I tuned the code as usual (make tuning), ran the tests (all successful) and 
then tried to run the benchmarks.

Apparently, there is no advantage at all (elapsed times are worse, actually) 
in the clearspeed version, if compared to the normal (host) version.

Am I missing something? How should the code be invoked?

According to an example I found on the site  
http://www2.warwick.ac.uk/fac/sci/eng/staff/ymc/research/japan/tebiki-eng.pdf, 
and supposing that the clearspeed options were implemented in DFT methods 
(benchmarks on big molecular systems are reported on the internet),
I tried to "play" with the environment variable  
CS_BLAS_HOST_ASSIST_PERCENTAGE, but it definetly had no effect. I tried other 
tests (non DFT) also, with the same result.

Where can I find help and information on this issue? Who could I ask to?

Thanks a lot

Lorenzo

-- 
=======================================

Lorenzo Gontrani

CASPUR - http://www.caspur.it
Consorzio per le Applicazioni di Supercalcolo Per Università e Ricerca
via dei Tizii, 6 - 00185 Roma
Tel +39 06 44486812 
Fax +39 06 4957083 
GSM +39 338 7615798
Email l DOT gontrani AT caspur DOT it

=======================================