Octave 3.1.51 (latest on hg) under Linux (SUSE 11.1) problem

[Khaled]

 Dear Maintainers,

  As you will realize later, I an not the kind of professional to get through the developing cycle of octave. However, I was in the need for the generalized eigenvalue problem and the SVDS (provided by arpack). Therefore, I tried to compile the development source (3.1.51) yesterday!!

  Compiling went perfect (except two issues I reported before with inv.cc and CmplxQRP.cc). I tried a work around and some how it works:). I already compiled octave 3.0.3 (and I am still using it on my laptop) using exactly the same command with the same the non-commercial intel math kernel (mkl 10.0.015) as follows:

"./configure CFLAGS=-fPIC FFLAGS=-ff2c --with-blas="-lmkl -lguide -lpthread -lm" --with-lapack="-lmkl_lapack" --enable-64 --enable-shared --enable-static --enable-bounds-check"

I am using gfortran and I prefer not compile with ifort... however my final build is >>10 times slower than 3.0.3. I don't know exactly if BLAS is configured as mkl or not. Even the configuration declared that it is not configured as BLAS since it is not compiled with gfortran. However, the configuration suggests for adding -ff2c in the compilation option, which I did as you see.

1- my first question now is: why is it much slower than ver 3.0.3 (final release)!!

2- Second question: the svds and arpack does not work properly!! on testing SVDS with this command:

svds(eye(2));

"Octave terminate and gives this error:
panic: Segmentation fault -- stopping myself...
attempting to save variables to `octave-core'...
save to `octave-core' complete
Segmentation fault "

I used this package for installing my octave:

http://mathema.tician.de/dl/software/arpack-autotools

I selected version (0.96)!

However, I do suggest to include a working arpack into octave source or compile it with octave !!

Thanks for your help in advance.

Here is my final configuration status:

"
  Installation prefix:  /usr/local
  C compiler:           gcc   -Wall -W -Wshadow -fPIC
  C++ compiler:         g++  -I/usr/include/freetype2  -Wall -W -Wshadow -Wold-style-cast -g -O2
  Fortran compiler:     gfortran -ff2c
  Fortran libraries:     -L/usr/lib64/gcc/x86_64-suse-linux/4.3 -L/usr/lib64/gcc/x86_64-suse-linux/4.3/../../../../lib64 -L/lib/../lib64 -L/usr/lib/../lib64 -L/usr/lib64/gcc/x86_64-suse-linux/4.3/../../../../x86_64-suse-linux/lib -L/usr/lib64/gcc/x86_64-suse-linux/4.3/../../.. -lhdf5 -lz -lgfortranbegin -lgfortran -lm
  BLAS libraries:      
  FFTW libraries:       -lfftw3 -lfftw3f
  GLPK libraries:       -lglpk
  UMFPACK libraries:    -lumfpack
  AMD libraries:        -lamd
  CAMD libraries:       -lcamd
  COLAMD libraries:     -lcolamd
  CCOLAMD libraries:    -lccolamd
  CHOLMOD libraries:    -lcholmod
  CXSPARSE libraries:   -lcxsparse
  ARPACK libraries:     -larpack
  HDF5 libraries:       -lhdf5
  CURL libraries:       -lcurl
  REGEX libraries:      -L/usr/lib64 -lpcre
  QHULL libraries: -lqhull
  LIBS:                 -lreadline  -lncurses -ldl -lmkl -lguide -lpthread -lm -lhdf5 -lz -lm
  Default pager:        less
  gnuplot:              gnuplot
  OPENGL libraries:     -L/usr/X11R6/lib -lGL -lGLU -lfreetype -lz -lftgl
  FLTK backend libs:    -L/usr/lib64 -Wl,-rpath,/usr/lib64 -lfltk_gl -lGLU -lGL -lfltk -lpthread -ldl -lm -lXext -lX11
  Magick config:        GraphicsMagick++-config

  Do internal array bounds checking:  true
  Build static libraries:             true
  Build shared libraries:             true
  Dynamic Linking:                    true (dlopen)
  Include support for GNU readline:   true
  64-bit array dims and indexing:     true

configure: WARNING: A BLAS library was detected but found incompatible with your Fortran 77 compiler.  The reference BLAS implementation will be used. To improve performance, consider using a different Fortran compiler or a switch like -ff2c to make your Fortran compiler use a calling convention compatible with the way your BLAS library was compiled, or use a different BLAS library.
"

[Jaroslav Hajek-2]

On Mon, Jan 5, 2009 at 12:39 PM, Khaled <kshawky@...> wrote:

>
>  Dear Maintainers,
>
>  As you will realize later, I an not the kind of professional to get
> through the developing cycle of octave. However, I was in the need for the
> generalized eigenvalue problem and the SVDS (provided by arpack). Therefore,
> I tried to compile the development source (3.1.51) yesterday!!
>
>  Compiling went perfect (except two issues I reported before with inv.cc
> and CmplxQRP.cc). I tried a work around and some how it works:). I already
> compiled octave 3.0.3 (and I am still using it on my laptop) using exactly
> the same command with the same the non-commercial intel math kernel (mkl
> 10.0.015) as follows:
>
> "./configure CFLAGS=-fPIC FFLAGS=-ff2c --with-blas="-lmkl -lguide -lpthread
> -lm" --with-lapack="-lmkl_lapack" --enable-64 --enable-shared
> --enable-static --enable-bounds-check"
>
> I am using gfortran and I prefer not compile with ifort... however my final
> build is >>10 times slower than 3.0.3. I don't know exactly if BLAS is
> configured as mkl or not. Even the configuration declared that it is not
> configured as BLAS since it is not compiled with gfortran. However, the
> configuration suggests for adding -ff2c in the compilation option, which I
> did as you see.
>
> 1- my first question now is: why is it much slower than ver 3.0.3 (final
> release)!!
>

Because MKL was not used. You have the (slow) reference BLAS, as
indicated by configure's result.

Intel MKL is best suited to be used with ifort. It is compatible with
gfortran, but you need to link in the Intel's runtime libraries (see
Intel manuals).

regards


--
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz

[Khaled]

Thanks Jaroslav for your reply!

  But I don´t Know how to compile Octave using ifort!! I did not found the suitable linking command on the internet. For example, the ifort is installed under

/opt/intel/Compiler/bin/intel64/ifort

and the libs are listed under

/opt/intel/Compiler/lib/intel64/

Best

Khaled

Jaroslav Hajek-2 wrote:

On Mon, Jan 5, 2009 at 12:39 PM, Khaled <[e-mail]> wrote:
>
>  Dear Maintainers,
>
>  As you will realize later, I an not the kind of professional to get
> through the developing cycle of octave. However, I was in the need for the
> generalized eigenvalue problem and the SVDS (provided by arpack). Therefore,
> I tried to compile the development source (3.1.51) yesterday!!
>
>  Compiling went perfect (except two issues I reported before with inv.cc
> and CmplxQRP.cc). I tried a work around and some how it works:). I already
> compiled octave 3.0.3 (and I am still using it on my laptop) using exactly
> the same command with the same the non-commercial intel math kernel (mkl
> 10.0.015) as follows:
>
> "./configure CFLAGS=-fPIC FFLAGS=-ff2c --with-blas="-lmkl -lguide -lpthread
> -lm" --with-lapack="-lmkl_lapack" --enable-64 --enable-shared
> --enable-static --enable-bounds-check"
>
> I am using gfortran and I prefer not compile with ifort... however my final
> build is >>10 times slower than 3.0.3. I don't know exactly if BLAS is
> configured as mkl or not. Even the configuration declared that it is not
> configured as BLAS since it is not compiled with gfortran. However, the
> configuration suggests for adding -ff2c in the compilation option, which I
> did as you see.
>
> 1- my first question now is: why is it much slower than ver 3.0.3 (final
> release)!!
>

Because MKL was not used. You have the (slow) reference BLAS, as
indicated by configure's result.

Intel MKL is best suited to be used with ifort. It is compatible with
gfortran, but you need to link in the Intel's runtime libraries (see
Intel manuals).

regards


--
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz

[John W. Eaton-3]

On  5-Jan-2009, Khaled wrote:

|
| Thanks Jaroslav for your reply!
|
|   But I don´t Know how to compile Octave using ifort!! I did not found the
| suitable linking command on the internet. For example, the ifort is
| installed under
|
| /opt/intel/Compiler/bin/intel64/ifort
|
| and the libs are listed under
|
| /opt/intel/Compiler/lib/intel64/

Why not simplify things and use ATLAS instead of mkl?

jwe

[dbateman2]

John W. Eaton wrote:

> On  5-Jan-2009, Khaled wrote:
>
> |
> | Thanks Jaroslav for your reply!
> |
> |   But I don´t Know how to compile Octave using ifort!! I did not found the
> | suitable linking command on the internet. For example, the ifort is
> | installed under
> |
> | /opt/intel/Compiler/bin/intel64/ifort
> |
> | and the libs are listed under
> |
> | /opt/intel/Compiler/lib/intel64/
>
> Why not simplify things and use ATLAS instead of mkl?
>
> jwe
>
>
>  

Or as you just want SVDS use a pre-built 3.0.3 from your distribution
and compile only ARPACK and the arpack octave-forge package..

D.

--
David Bateman                                dbateman@...
35 rue Gambetta                              [phone] (Home)
92100 Boulogne-Billancourt FRANCE            [phone] (Mob)

[Khaled]

I compared ATLAS vs MKL, found that MKL is factor faster than ATLAS!!



Of course this not an absolute statement!!! I only compared the things I do need in my work ... like matrix inversion, mul, QR, SVD, EIG ...

Best

Khaled

 

John W. Eaton-3 wrote:

On  5-Jan-2009, Khaled wrote:

|
| Thanks Jaroslav for your reply!
|
|   But I don´t Know how to compile Octave using ifort!! I did not found the
| suitable linking command on the internet. For example, the ifort is
| installed under
|
| /opt/intel/Compiler/bin/intel64/ifort
|
| and the libs are listed under
|
| /opt/intel/Compiler/lib/intel64/

Why not simplify things and use ATLAS instead of mkl?

jwe

[John W. Eaton-3]

On  5-Jan-2009, Khaled wrote:

| I compared ATLAS vs MKL, found that MKL is factor faster than ATLAS!!
|
| :confused:

What does "factor faster than" mean to you?

In any case, I would recommend using ATLAS because it is free (as in
freedom) software.  MKL is not.  Likewise for ifort.

Are you interested in helping the community of free software users?
If so, then please consider using and promoting free software tools
rather than proprietary tools like ifort and MKL.

jwe