[Pw_forum] MKL 10.2 not recognized by configure script / Xeon=amd64?

[Matthias Timmer]

Dear readers,

thank you for your replies and great help. Just to answer and comment on 
a few things:
>> 2) When the configure-script runs, it always tells me, that the
>> "x86_64"-architecture of the Xeons we have is "amd64" (see line 1443 in the
>> configure-script). I think this is highly irritating.
>>     
>
> On Debian systems, x86_64 is often called amd64:
> http://alioth.debian.org/docman/view.php/30192/21/debian-amd64-howto.html#id250846
>
> Not sure if you have a Debian (or Ubuntu) system or not, but that
> might help explain what you are seeing.
>   
According to my last information, we have scientific linux.
>
> BLAS_LIBS      = -lmkl_blas95_lp64 -lmkl_intel_lp64 
> -lmkl_intel_sequential -lmkl_core
>
> LAPACK_LIBS    =  -lfftw3xf_intel -lmkl_lapack95_lp64 -lmkl_intel_lp64 
> -lmkl_intel_sequential -lmkl_core
Since the linker does not recognize

-lmkl_intel_sequential


I needed to change this towards -lmkl_sequential.

I also added -lpthread for BLAS_LIBS, following

http://software.intel.com/en-us/articles/intel-mkl-link-line-advisor/

pointed out by Baris.

For the moment it works perfectly, and I get a very considerable 
reduction in calculation time (seems to be >50% after first tests on our 
Xeons, just because of the MKL library instead of PWSCF internal - can 
that be true?).
Compared to our Opteron cluster (ifort 9.1 / mkl 9.0), the Xeons (ifort 
11.1 / mkl 10.2.1.017) save > 50% calculation time.

Many thanks to all of you who tried to help!

Kind regards,

Matthias Timmer,
University of Duisburg-Essen
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Matthias_Timmer.vcf
Type: text/x-vcard
Size: 318 bytes
Desc: not available
Url : http://www.democritos.it/pipermail/pw_forum/attachments/20091020/87cdae12/attachment-0001.vcf