[Wien] problems with wien2k 11 installation

[Gavin Abo]

If the error is coming from composerxe, there are two possible causes 
that I can think of:

1) Software bug(s)
2) Instruction set compiler option

More details are given below.  Though, the problem is less likely to be 
caused by composerxe since the TiC runs without error.

-------------------------------------------------------------------------------------------------------------------
In case you don't know, there is a fairly extensive list of composerxe 
bug fixes that have been made since the update 5 you are using. You can 
check out the full list of bugs since update 5 at:

http://software.intel.com/en-us/articles/intel-composer-xe-2011-compilers-fixes-list/

The bug fixes in Update 9 that I see that 'might' have a significant 
effect on Wien2k:

DPD200167060     Fortran                     Intel Fortran Compiler 12.0 
fails to vectorize a loop which works in version 11
DPD200174066     Fortran                     Incorrect results with the 
optimization turned on
DPD200261402     C, C++, Fortran      Memory leak for function 
parameters with size >64k

According to the mkl library list at:

http://software.intel.com/sites/products/documentation/hpc/composerxe/en-us/mklxe/mkl_userguide_lnx/MKL_UG_Appdx__C/Directory_Structure_intel64_dyn.htm

libmkl_avx.so is the "Kernel optimized for the Intel(R) Advanced Vector 
Extensions (Intel(R) AVX)"

According to the composerxe user and reference guide:

http://software.intel.com/sites/products/documentation/hpc/composerxe/en-us/2011Update/fortran/lin/index.htm

The default option when -x option is not specified is SSE2.  If you used 
the recommended compiler options by siteconfig in Wien2k, then I believe 
-xcode is not specified and uses the SSE2 instruction set.

As described at:

http://software.intel.com/en-us/articles/performance-tools-for-software-developers-intel-compiler-options-for-sse-generation-and-processor-specific-optimizations/

You can specify the compiler option AVX instruction set depending on the 
generation of the i7 processor:

-xCORE-AVX2 (for future generation)
-xCORE-AVX-I (for 3rd generation)
-xAVX (for 2nd generation)
-xSSE4.2 (for previous generation)

or

-xHost (should select this for you, the highest instruction set for the 
processor)

So I wonder if the libmkl_avx.so error is coming from not specifying the 
AVX instruction set in the Wienk2 compiler options.  This is something 
you would have to test, since I don't have an i7 system.

An alternative to changing the instruction set might be to not link 
libmkl_avx.so.  I don't know whether you manually specified -lmkl_avx in 
your mkl options or if it was automatically selected by one of your 
given libraries (my guess would be by -lmkl_core).

On 4/25/2012 11:58 AM, Laurence Marks wrote:
> I assume that you are using the release which works (as against the
> ones which have bugs). Please check the mailing list.
>
> Assuming that you are, almost certainly nobody can help you without
> more information than you have provided.
>
> 2012/4/25 Ramkumar Thapa<r.k.thapa at gmail.com>:
>> Dear Profs.  Gerhard Fetcher, Peter and Lawrence,
>> I had installed wien2k 11.1 ver in my corei7 machine with ifort
>> composerxe-2011.5.220  (mkl 10.3). I could run TiC very well and results of
>> density,DOS and bandstructues are ok. However, when I run it for rare earth
>> compounds like EuFeP, it gives errors like :SIGSEGV, segmentation fault
>> occurred, Image:libmkl_avx.so. SCF stops after LAPW0 END. Can you please
>> help me to correctly run the SCF without errors. My operating system is
>> CENTOS 6.0 and 64 bits.
>> Thanking you,
>> Prof.R.K.Thapa
>> Mizoram University
>> India
>>
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>
>
>