[Pw_forum] Difficulties compiling comp_resp_mat.j to optimize the U

Gabriele Sclauzero sclauzer at sissa.it
Mon May 25 18:00:25 CEST 2009



Axel Kohlmeyer wrote:
> On Mon, 2009-05-25 at 15:48 +0100, pc229 at kent.ac.uk wrote:
>> I do apologize, 
>> I made a mistake  when I paste the string that I use to compile it.
>> Anyway I tried both the static and the dynamics compilation using
>> either this string 
>>
>> and the output is : 
>>
>>
>> and 
>> ifort -o resp_mat.x resp_mat.f90 \
>> -L/opt/intel/mkl/10.1.2.024/lib/32 -lmkl_lapack64 -lmkl -lguide
>> -lpthread
> 
> why suddenly use now -lmkl_lapack64 ???
> there is not libmkl_lapack64.so in the 32bit version of mkl.

In my machine at SISSA there is, but maybe it is a rather exotic installation... (actually 
I was surprised myself...)
[sclauzer at brenta:/usr/local/intel_mkl_90/lib/32]$ ls -l
total 110268
-r--r--r-- 1 root root   267450 Oct  6  2006 libguide.a
-r-xr-xr-x 1 root root   226328 Oct  6  2006 libguide.so
-r-xr-xr-x 1 root root  4744646 Oct  6  2006 libmkl_def.so
-r--r--r-- 1 root root 47116850 Oct  6  2006 libmkl_ia32.a
-r-xr-xr-x 1 root root   663086 Oct  6  2006 libmkl_ias.so
-r-xr-xr-x 1 root root  2967647 Oct  6  2006 libmkl_lapack32.so
-r-xr-xr-x 1 root root  3014308 Oct  6  2006 libmkl_lapack64.so
-r--r--r-- 1 root root 10418470 Oct  6  2006 libmkl_lapack.a
-r-xr-xr-x 1 root root  5645669 Oct  6  2006 libmkl_p3.so
-r-xr-xr-x 1 root root  6511253 Oct  6  2006 libmkl_p4m.so
-r-xr-xr-x 1 root root  6462455 Oct  6  2006 libmkl_p4p.so
-r-xr-xr-x 1 root root  6586091 Oct  6  2006 libmkl_p4.so
-r-xr-xr-x 1 root root   768161 Oct  6  2006 libmkl.so
-r--r--r-- 1 root root  1333138 Oct  6  2006 libmkl_solver.a
-r-xr-xr-x 1 root root  2189200 Oct  6  2006 libmkl_vml_def.so
-r-xr-xr-x 1 root root  2595590 Oct  6  2006 libmkl_vml_p3.so
-r-xr-xr-x 1 root root  3623813 Oct  6  2006 libmkl_vml_p4m.so
-r-xr-xr-x 1 root root  3624853 Oct  6  2006 libmkl_vml_p4p.so
-r-xr-xr-x 1 root root  3640224 Oct  6  2006 libmkl_vml_p4.so
-r-xr-xr-x 1 root root   476207 Oct  6  2006 libvml.so

In fact libmkl_lapack32.so contains single precision LAPACK routines, while 
libmkl_lapack64.so the double precision ones.

Anyway I agree that the better way to understand how to link MKL is to read the local 
guide or check Intel's website.

GS


> it should work with plain -lmkl_lapack
> 
> cheers,
>    axel.
> 
> 
>> I cannot manage this problem. If it doesn't matter, can you please
>> send me, privately, your compiled version? I am struggling a lot.
>> Thank you, Piero
> 
> _please_ rather then doing random stuff that does not make sense,
> spend some time reading the MKL documentation. it is quite detailed,
> useful and explains _everything_ about linking MKL dynamically and
> statically. you have to keep in mind that the linux linker always
> prefers shared libraries (.so) over static libraries. you should
> also see from the documentation that the static version or lapack
> is already included in libmkl_ia32.a. in conclusion: either link with
> 
> -openmp -lmkl_lapack -lmkl
> 
> or 
> 
> -i-static -openmp -lmkl_ia32
> 
> or 
> 
> -lmkl_lapack -lmkl_intel -lmkl_sequential -lmkl_core 
> 
> cheers,
>    axel.
> 
>> ----- Original Message -----
>> From: Gabriele Sclauzero <sclauzer at sissa.it>
>> Date: Monday, May 25, 2009 3:18 pm
>> Subject: Re: [Pw_forum] Difficulties compiling comp_resp_mat.j to
>> optimize the U
>> To: PWSCF Forum <pw_forum at pwscf.org>
>>
>>> I thought Axel already replied to this post...
>>>
>>> Pieremanuele Canepa wrote:
>>>> Dear all,
>>>> I'm trying to compile the comp_resp_mat.j concerning the 
>>> DFT+U. Then, I 
>>>> tried to use the default settings (ifort-o r.x resp_mat.f90 
>>>> -L/soft/local/intel/mkl72/lib/32 -lmkl_lapack \ -lmkl_ia32 -
>>> lguide 
>>>> -lpthread) and what I got, is reported below. Do you have any 
>>> suggestions?> Probably  I am making some stupid mistake. 
>>> have somebody already 
>>>> compiled this program on 32 bit machine?
>>>> I don't have any experience compiling with lapack libraries as 
>>> you can 
>>>> probably realize.
>>>> Cheers, Piero 
>>> I don't understand if you're trying to link against mkl 7.2 or 
>>> 10.1. Do you know at least 
>>> which version is installed on your machine and where? Try to 
>>> find out this first.
>>>
>>> On my 32 bit Intel machine I've got MKL 9.0 installed in 
>>> /usr/local/intel_mkl_90/ and this 
>>> compilation command works (for static linking):
>>>
>>> ifort -o resp_mat.x resp_mat.f90 \
>>>     -L"/usr/local/intel_mkl_90/lib/32" -
>>> lmkl_lapack -lmkl_ia32 -lguide -lpthread
>>>
>>> ldd resp_mat.x
>>>          linux-gate.so.1 
>>> =>  (0x00af3000)
>>>          libguide.so => 
>>> /usr/local/intel_mkl_90/lib/32/libguide.so (0x00a55000)
>>>          libm.so.6 => 
>>> /lib/libm.so.6 (0x00aa4000)
>>>          libc.so.6 => 
>>> /lib/libc.so.6 (0x00111000)
>>>          libgcc_s.so.1 
>>> => /lib/libgcc_s.so.1 (0x00d4c000)
>>>          libdl.so.2 => 
>>> /lib/libdl.so.2 (0x00acd000)
>>>          libpthread.so.0 
>>> => /lib/libpthread.so.0 (0x00ad3000)
>>>          /lib/ld-
>>> linux.so.2 (0x00941000)
>>>
>>>
>>> If you want to link dynamically use:
>>>
>>> ifort -o resp_mat.x resp_mat.f90 \
>>>     -L"/usr/local/intel_mkl_90/lib/32" -
>>> lmkl_lapack64 -lmkl -lguide -lpthread
>>>
>>> ldd resp_mat.x
>>>          linux-gate.so.1 
>>> =>  (0x00741000)
>>>          
>>> libmkl_lapack64.so => 
>>> /usr/local/intel_mkl_90/lib/32/libmkl_lapack64.so (0x00c32000)
>>>          libmkl.so => 
>>> /usr/local/intel_mkl_90/lib/32/libmkl.so (0x00eba000)
>>>          libguide.so => 
>>> /usr/local/intel_mkl_90/lib/32/libguide.so (0x00111000)
>>>          libpthread.so.0 
>>> => /lib/libpthread.so.0 (0x00ad3000)
>>>          libm.so.6 => 
>>> /lib/libm.so.6 (0x00aa4000)
>>>          libc.so.6 => 
>>> /lib/libc.so.6 (0x0095f000)
>>>          libgcc_s.so.1 
>>> => /lib/libgcc_s.so.1 (0x0015e000)
>>>          libdl.so.2 => 
>>> /lib/libdl.so.2 (0x00acd000)
>>>          /lib/ld-
>>> linux.so.2 (0x00941000)
>>>
>>> Obviously you must change the path to the exact location of your 
>>> MKL installation
>>>
>>>
>>> HTH
>>>
>>> GS
>>>
>>>
>>>> resp_mat.f90(92): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(93): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(119): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(188): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(235): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(235): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(235): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(235): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(252): (col. 10) remark: PERMUTED LOOP WAS VECTORIZED.
>>>> resp_mat.f90(265): (col. 16) remark: PERMUTED LOOP WAS VECTORIZED.
>>>> resp_mat.f90(322): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(339): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(365): (col. 16) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(377): (col. 7) remark: LOOP WAS VECTORIZED.
>>>> resp_mat.f90(483): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(483): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(483): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(483): (col. 13) remark: PARTIAL LOOP WAS VECTORIZED.
>>>> resp_mat.f90(500): (col. 10) remark: PERMUTED LOOP WAS VECTORIZED.
>>>> resp_mat.f90(513): (col. 16) remark: PERMUTED LOOP WAS VECTORIZED.
>>>> /tmp/ifortMUANg9.o: In function `invmat_.':
>>>> resp_mat.f90:(.text+0x9c33): undefined reference to `dcopy_'
>>>> resp_mat.f90:(.text+0x9c4a): undefined reference to `dgetrf_'
>>>> resp_mat.f90:(.text+0x9c6a): undefined reference to `dgetri_'
>>>> /opt/intel/mkl/10.1.2.024/lib/32/libmkl_lapack.so 
>>>> <http://10.1.2.20/lib/32/libmkl_lapack.so>: undefined 
>>> reference to 
>>>> `mkl_pds_c_blklu_unsym_omp_pardiso'
>>>> /opt/intel/mkl/10.1.2.024/lib/32/libmkl_lapack.so 
>>>> <http://10.1.2.20/lib/32/libmkl_lapack.so>: undefined 
>>> reference to 
>>>> `mkl_lapack_cpptrs'
>>>> /opt/intel/mkl/10.1.2.024/lib/32/libmkl_lapack.so 
>>>> <http://10.1.2.20/lib/32/libmkl_lapack.so>: undefined 
>>> reference to 
>>>> `mkl_serv_strcat2'
>>>> /opt/intel/mkl/10.1.2.024/lib/32/libmkl_lapack.so 
>>>> <http://10.1.2.20/lib/32/libmkl_lapack.so>: undefined 
>>> reference to 
>>>> `mkl_pds_blklu_omp_pardiso'
>>>> and so on 
>>>> -- 
>>>> Pieremanuele Canepa
>>>> Room 230
>>>> School of Physical Sciences, Ingram Building,
>>>> University of Kent, Canterbury, Kent,
>>>> CT2 7NH
>>>> United Kingdom
>>>> -----------------------------------------------------------
>>>>
>>>>
>>>> ---------------------------------------------------------------
>>> ---------
>>>> _______________________________________________
>>>> Pw_forum mailing list
>>>> Pw_forum at pwscf.org
>>>> http://www.democritos.it/mailman/listinfo/pw_forum
>>> -- 
>>>
>>>
>>> o ------------------------------------------------ o
>>> | Gabriele Sclauzero, PhD 
>>> Student                  |
>>> | c/o:   SISSA & CNR-INFM 
>>> Democritos,              |
>>> |        via Beirut 2-4, 
>>> 34014 Trieste (Italy)     |
>>> | email: 
>>> sclauzer at sissa.it                         |
>>> | phone: +39 040 3787 
>>> 511                          |
>>> | skype: 
>>> gurlonotturno                             |
>>> o ------------------------------------------------ o
>>> _______________________________________________
>>> Pw_forum mailing list
>>> Pw_forum at pwscf.org
>>> http://www.democritos.it/mailman/listinfo/pw_forum
>>>
>> ---
>> Pieremanuele Canepa
>> Room 230
>> School of Physical Sciences, Ingram Building, 
>> University of Kent, Canterbury, Kent, 
>> CT2 7NH
>> United Kingdom
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
> 

-- 


o ------------------------------------------------ o
| Gabriele Sclauzero, PhD Student                  |
| c/o:   SISSA & CNR-INFM Democritos,              |
|        via Beirut 2-4, 34014 Trieste (Italy)     |
| email: sclauzer at sissa.it                         |
| phone: +39 040 3787 511                          |
| skype: gurlonotturno                             |
o ------------------------------------------------ o


More information about the Pw_forum mailing list