This thread is locked.Only browsing is available.
Top Page > Browsing
Errors in installation of OpenMx3.7 and OpenMx3.8
Date: 2016/04/15 22:21
Name: Hawson   <shenghao_wu@163.com>

When I install OpenMx3.7 and OpenMx3.8 in 32 bit ubuntu machine, the same errors as follows come up. I set the CC and LIB in 'make le' as default. Could anybody help me overcome this problem. I would be thankful if You give me a guidance.
Thank you.
--------------------
Krylov.c: In function eKrylov_Col_trdf:
Krylov.c:5944:5: error: unknown type name e__m128df
__m128d mmSum00,mmSum01,mmSum10,mmSum11;
^
Krylov.c:5945:5: error: unknown type name e__m128df
__m128d mmSum20,mmSum21,mmSum30,mmSum31;
^
Krylov.c:5946:5: error: unknown type name e__m128df
__m128d mmTmp0, mmTmp1, mmTmp2, mmTmp3;
^
Krylov.c:5947:5: error: unknown type name e__m128df
__m128d mmTmp4, mmTmp5;
^
Krylov.c: In function eGenerate_pMatrix_trdf:
Krylov.c:6916:3: error: unknown type name e__m128df
__m128d mmSum00,mmSum01,mmSum10,mmSum11,mmSum20,mmSum21,mmSum30,mmSum31, mmTmp0, mmTmp1, mmTmp2, mmTmp3, mmTmp4, mmTmp5;
^
Krylov.c: In function eGenerate_pMatrix2_trdf:
Krylov.c:8394:3: error: unknown type name e__m128df
__m128d mmSum00,mmSum01,mmSum10,mmSum11,mmSum20,mmSum21,mmSum30,mmSum31, mmTmp0, mmTmp1, mmTmp2, mmTmp3, mmTmp4, mmTmp5;
^
Krylov.c: In function eEmbedding_Matrix_trdf:
Krylov.c:9014:3: error: unknown type name e__m128df
__m128d mmSum00,mmSum01,mmSum10,mmSum11,mmSum20,mmSum21,mmSum30,mmSum31, mmTmp0, mmTmp1;
^
Krylov.c: In function eS_orthonormalize_vec_trdf:
Krylov.c:10051:3: error: unknown type name e__m128df
__m128d mmSum00,mmSum01,mmSum10,mmSum11,mmSum20,mmSum21,mmSum30,mmSum31, mmTmp0, mmTmp1, mmTmp2, mmTmp3, mmTmp4, mmTmp5;
^
make: *** [Krylov.o] Error 1
e
Page: [1]

Re: Errors in installation of OpenMx3.7 and OpenMx3.8 ( No.1 )
Date: 2016/04/17 09:59
Name: T. Ozaki

Hi,

The error message implies that the compiler does not support SSE.
Please add '-Dnosse' as compiler option as discussed in
http://www.openmx-square.org/openmx_man3.8/node14.html

Regards,

TO
e
Re: Errors in installation of OpenMx3.7 and OpenMx3.8 ( No.2 )
Date: 2016/04/18 11:55
Name: Hawson  <shenghao_wu@163.com>

Dear Professor T. Ozaki

Many thanks for the reply.
When I set the compiler option as follows,
CC = mpicc -Dnoomp -Dnosse -O3 -I/usr/local/include
FC = mpif90 -Dnoomp -Dnosse -O3 -I/usr/local/include
LIB = -L/usr/local/lib -lfftw3 -llapack -lblas -lg2c -static

a new problem comes up,
mpicc -Dnoomp -Dnosse -O3 -I/usr/local/include -I./liberi-091216/source -c NBO_Krylov.c
NBO_Krylov.c:31:23: fatal error: mimic_omp.h: No such file or directory
#include "mimic_omp.h"

then, I move the 'mimic_omp.h' from openmx3.6 to openmx3.8,

root@ubuntu:/home/ubuntu# cp openmx3.6/source/mimic_omp.h openmx3.8/source/
root@ubuntu:/home/ubuntu/openmx3.8/source# make clean
root@ubuntu:/home/ubuntu/openmx3.8/source# make install

another bug comes up,

/usr/bin/ld: cannot find -lopen-rte
/usr/bin/ld: cannot find -lopen-pal
/usr/local/lib/libmpi.a(lib_libmpi_la-tcp_init.o): In function `MPID_nem_tcp_get_business_card':
tcp_init.c:(.text+0x553): warning: Using 'gethostbyname' in statically linked applications requires at runtime the shared libraries from the glibc version used for linking
collect2: error: ld returned 1 exit status
make: *** [openmx] Error 1

Could you give me another suggestion about this problem. Very thank for your help.

Sincerely
e
Re: Errors in installation of OpenMx3.7 and OpenMx3.8 ( No.3 )
Date: 2016/04/18 17:50
Name: T. Ozaki

Hi,

> then, I move the 'mimic_omp.h' from openmx3.6 to openmx3.8,

This is actually a problem. Thank you for your report.


> /usr/bin/ld: cannot find -lopen-rte
> /usr/bin/ld: cannot find -lopen-pal
> /usr/local/lib/libmpi.a(lib_libmpi_la-tcp_init.o): In function
> `MPID_nem_tcp_get_business_card':
> tcp_init.c:(.text+0x553): warning: Using 'gethostbyname' in statically linked applications
> requires at runtime the shared libraries from the glibc version used for linking
> collect2: error: ld returned 1 exit status
> make: *** [openmx] Error 1

This is not a bug, but a linking problem.
Please take a look at
http://www.openmx-square.org/tech_notes/OpenMX-Compile.pdf

Regards,

TO
e
Re: Errors in installation of OpenMx3.7 and OpenMx3.8 ( No.4 )
Date: 2016/05/18 13:19
Name: Hawson  <shenghao_wu@163.com>

Dear Professor T. Ozaki

Very sorry for this later reply.
I still cannot figure out my problem of installation in Ubuntu system, but I make success in OpenMx3.7 installation in CentOS system, and it very easy and simple. I should display my installation method for other users who also meet some problems in installation.

I install CentOS 7 both in my Virtual Machines and Server, and make the steps as follows,
1-Install the math libraries: lapack, blas, libgfortran and fftw3 (must be root):

# yum -y install blas lapack lapack-devel blas-devel fftw fftw-devel libgfortran

2 - Install openmpi to use openmp (must be root):

# yum -y install openmpi openmpi-devel

3
$ cd openmx3.7/source

4
CC = /usr/lib64/openmpi/bin/mpicc -O3 -I/usr/include -fopenmp -I/usr/include/openmpi-x86_64
FC = /usr/lib64/openmpi/bin/mpif90 -O3 -I/usr/include
LIB = -L/usr/lib64 -lfftw3 -llapack -lblas -lgfortran -L/usr/lib64/openmpi/lib -lmpi -lmpi_mpifh -lmpi_usempi


5 - save the file and compile:
$make

6- To run openmx with 3 cores:
$ /usr/lib/openmpi/bin/mpirun -np 1 openmx file.dat -nt 3
where file.dat is the input file of openmx.
succeed.

Hope this could be helpful for other users.

But now, I have some question of NEGF calculation,
I found it was very difficult to achieve a convergence even cannot get convergence in NEGF calculation no mater that I had set large values of History and StartPulay.
And I found the scf.maxIter in negf-example was about 1000 or more, so my question is, what value of scf.maxIter is proper in NEGF calculation above which the calculation cannot get convergence anymore?
Another question is, can I use the FRAC as the Atoms.SpeciesAndCoordinates.Unit in the lead and scatter NEGF calculation? Should I make a geometry optimization before NEGF calculation, and use the atoms Coordinates in *.xyz or *.out as the input in NEGF calculation?

Very thank for your suggestion.

Sincerely


e
Re: Errors in installation of OpenMx3.7 and OpenMx3.8 ( No.5 )
Date: 2016/05/19 14:52
Name: T. Ozaki

Hi,

> what value of scf.maxIter is proper in NEGF calculation above which the calculation cannot get
> convergence anymore?

Rather than taking care of scf.maxIter, one should care about scf.criterion.
For the transport calculations, using a rather large scf.criterion might be acceptable,
i.e., 1.0e-4 or 1.0e-5 is enough to get a convergent transmission in my experiences.

> can I use the FRAC as the Atoms.SpeciesAndCoordinates.Unit in the lead and scatter NEGF
> calculation?

The use of FRAC is not supported for NEGF calculations, as explained in
http://www.openmx-square.org/openmx_man3.8/node116.html

> Should I make a geometry optimization before NEGF calculation, and use the atoms Coordinates
> in *.xyz or *.out as the input in NEGF calculation?

It depends on what you want to study. It should be noted that geometry optimization
can be crucial to capture physics of system under investigation in some cases.

Regards,

TO
e

Page: [1]