[CPMD-list] Weird behavior of specific processor optimization

[Huiqun Zhou]

Dear CPMD users,

I recently built a serial version of CPMD 3.9.2 based on PC-IFC-P4 with tiny modifications for using Intel MKL 7.0.1,
and I used dynamic linking because the parallel version I wanted to create will run on ROCKS cluster, which enables
very easy distribution of Intel compiler to all compute nodes. I modified the compiler option -O to -O3, and everything
went OK. I tested the build with Si64 example, the elapsed time of run is 2'40.29'' on a P4 2.4GHz 1GB compute
node. I think the result is acceptable (please correct me if I'm wrong).

Then, I changed the compiler option for specific processor optimization, that is, added -xN option (-lsvml in LDFLAGS), 
and everything went OK again. But, when I tested the build with same Si64 example, the running time became 18'31.75'', 
over 8 times longer than the previous less optimized version! I noticed in the output that the further optimized version 
(with -O3 -xN) reset much more times in diagonization (DIIS) while the less optimized version reset only one time.

What's wrong here? I noticed that the distributed makefile templates are all lack of further optimization, so the weird 
behavior may have been a known fact. Please confirm me whether or not  we can use any complex optimization options 
in compiling CPMD. Please give me a pointer on how to compile an efficient parallel version because the parallel version
I created based on the less optimized version (with -O3, and add MPICH stuff) cost over 15 minutes on one compute 
node on Si64 example, too. 

Will static version be a cure?


Please HELP.


==========================
Huiqun Zhou, Doctor of Science
Earth Sciences
Nanjing University
22 Hankou Road, Gulou
Nanjing, 210093
China

e-mail: hqzhou at nju.edu.cn
Tel.: 86(25)8368-6750
mobil: 86-13182856800
==========================
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20050622/8bce5dc8/attachment.html